Projects - User contributions [en]

Projects:2019s2-20001 Using Machine Learning to Determine Deposit Height and Defects for Wire + Arc Additive Manufacture (3D printing)

2020-03-17T02:00:08Z

A1713568: Replaced content with "Hi."

Hi.

Projects:2019s2-20001 Using Machine Learning to Determine Deposit Height and Defects for Wire + Arc Additive Manufacture (3D printing)

2019-10-16T02:29:03Z

A1713568: /* Result discussion */

[[Category:Projects]]
[[Category:Final Year Projects]]
[[Category:2019s2|24501]]
== Introduction ==
AML3D is one of the leading company in the Wire Arc Additive Manufacturing (WAAM) industry. In short, WAAM is the Metal 3D printing. The technology has the ability to reduce up to 75% of manufacture time, cost, and material. Currently, WAAM is benefiting other industry such as: mining, defence, marine, and aviation.

<gallery mode="slideshow">
File:Welding baby.gif|center|Welding using robotic arm
File:Laser baby.gif|center|User laser sensor to find the CTWD
</gallery>

[[File:aml3d_logo.jpg|center|100px]]
=== Project team ===
==== Project students ====
* Anh Tran
* Nhat Nguyen
==== Supervisors ====
* Dr. Brian Ng
* Dr. Paul Colegrove (AML3D, sponsored company)
=== Objectives ===
The objective of this project is to increase the efficiency of the manufacture process at AML3D. In order to do so, the team will investigate into the possibility of automating and optimising the quality control processes. The two quality control processes that are currently being implemented at AML3D are measuring layer height using laser sensors, and human supervision for detecting defects. These processes add overhead into production time and usage of human resource, which is not desired. To achieve the goal, machine learning methods will be used extensively to analyse the electrical signatures of the weld process.

== Background ==
=== Wire Arc Additive Manufacturing ===
Wire and Arc Additive Manufacturing (WAAM) is a type of additive manufacturing that uses electric arc as the heat source and material wire to feed the manufacture process <ref name="WAAM"> S. W. Williams, F. Martina, A. C. Addison, J. Ding, G. Pardal & P. Colegrove (2016) Wire + Arc Additive Manufacturing, Materials Science and Technology, 32:7, 641-647, DOI: 10.1179/1743284715Y.0000000073 </ref>. WAAM has been investigated since the 1990s <ref name="WAAM" />, but only recently that it received more attention from the manufacture world.

Its significant comes from the ability to manufacture complex model with less time and less material. Figure 1a and 1b show real parts that was manufactured by AML3D. Such custom made parts might takes months before be ready to be shipped, but with WAAM, the production time can reduce down to weeks. Currently, the industries that benefit the most from WAAM are maritime and aerospace.

Similar to other additive manufacture methods, WAAM achieves such results by sliding the models into multiple layers, and then build the model layer by layer. The movement control is normally handled by a robotic arm (AML3D uses ABB's Arc Welder robot), and the welding path is generated by a Computer Aid Manufacture (CAM) software.

[[File:welding_part1.JPG|thumb|Figure 1a: A propeller]]
[[File:welding_part2.jpg|thumb|Figure 1b: Another part of the ship]]
[[File:abb_robot.jpg|thumb|Figure 2: ABB's IRB 1520ID Arc Welder]]

=== Gas metal arc welding ===
<div><ul>
<li style="display: inline-block;"> [[File:gmaw_welding.png|thumb|500px|none|Figure 2: GMAW welding process <ref name="welding_book"> Norrish, J. (2006). Advanced welding processes. Elsevier. </ref>]] </li>
</ul></div>

Gas metal arc welding (GMAW), or also known as metal inert gas (MIG), is the welding process used at AML3D for WAAM. Due to its high deposition rate and economic benefits, GMAW has became more popular. We will explore two variants of GMAW, which are Pulse Multi Control and Cold Metal Transfer. Both processes are developed by Fronius, an Austrian welding company. Note that it is not required to understand the welding physics to follow this wiki. The next two following sections' purpose is to highlight the high variability, high dynamic nature in welding.

[[File:fronius.png|thumb|Figure 3: Fronius logo]]

==== Pulse Multi Control ====
Pulse Multi Control (PMC) welding is Fronius' modified Pulsed GMAW. The advantage of Pulsed GMAW is the ability to control the metal droplet transfer in welding. Note that from figure 4, the metal droplet is characterised by a downward slope of the current pulse. Figure 5 shows the current signal captured from experiments conducted at AML3D (PMC is used).

<div><ul>
<li style="display: inline-block;"> [[File:Pulsed_gmaw_transfer_mode.png|thumb|500px|none|Figure 4: Relationship between pulse current and droplet transfer<ref name="welding_book" />]] </li>
<li style="display: inline-block;"> [[File:Pmc_current_in_time_domain.png|thumb|650px|none|Figure 5: Current signal captured from experiments conducted at AML3D (PMC process)]] </li>
</ul></div>
==== Cold Metal Transfer ====
Cold Metal Transfer (CMT) is a complex welding process, also developed by Fronius. By detecting a short circuit (mode C and D in figure 8), CMT can make adjustment such as retracting the welding material to cool down, and therefore create a smoother, more stable weld. The complexity of CMT can be seen in the current signal of the process (figure 9). Compare to PMC, CMT's signal has more variation during one signal period.

<div><ul>
<li style="display: inline-block;"> [[File:Transfer_mode.png|thumb|800px|none|Figure 8: Mechanism of metal droplet transfer mode]] </li>
<li style="display: inline-block;"> [[File:Cmt1_current_in_time_domain.png|thumb|735px|none|Figure 9: Current signal captured from experiments conducted at AML3D (CMT process)]] </li>
</ul></div>
=== Machine Learning ===
Machine learning (or statistical learning) refer to a set of tools to give data scientists an insight into the data and make better decisions based on those information <ref name="introduction_ml_book"> James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An introduction to statistical learning (Vol. 112, p. 18). New York: Springer. </ref>. In this project, we will explore the capabilities of classical machine learning methods (i.e anything but Deep Learning) due to the limitation of our available data. By the time of writing this wiki, we have only explored Support Vector Machine.

==== Support Vector Machine ====
The main idea of Support Vector Machine (SVM) is to draw hyperplanes that best separate multiple classes of data. The original SVM judges best decision hyperplane based on the separation margin between data classes (figure 10). However, there are multiple variants of SVM, such as hard/soft margin SVM, Nu-SVM, One class SVM (to solve anomaly detection problem). In our project, we choose to use a soft margin implementation of SVM. The benefit of soft margin SVM (as oppose to hard margin SVM) is that by allowing some misclassifications, we get a wider margin and therefore the solution becomes more "generalise" (i.e work better with a new set of dataset that is independent from the training dataset).

<div><ul>
<li style="display: inline-block;"> [[File:svm_margin.png|thumb|400px|none|Figure 10: Separation margin <ref name="bishop_book"> Bishop, C. M. (2006). Pattern recognition and machine learning. Springer. </ref>]] </li>
</ul></div>

==== Kernel tricks ====
Figure 10 is of linearly separable cases for SVM. For dataset that is not easy to separate with a hyperplane in their original features space (such as figure 11), the data often get projected into a higher dimension space where it is easier to separate (figure 14). Such technique is called the Kernel Tricks. It is important to note that Kernel Tricks does not actually "map" the data from '''X''' dimension to '''Z''' dimension, but rather pairwise compare the similarity of the input space in '''Z''' dimension.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_nonseparable.png |thumb|350px|none|Figure 11: Nonseparable dataset <ref name="bishop_book"></ref>]] </li>
<li style="display: inline-block;"> [[File:Svm_kernel_trick.png |thumb|650px|none|Figure 12: Simplified example of the Kernel tricks]] </li>
</ul></div>

== Previous studies ==
There are many researches into defect detection in GMAW process. The three most dominant analysis methods are spectroscopic analysis, acoustic analysis, and electrical signal analysis.

=== Spectroscopic analysis ===
In <ref name="spectroscopic_analysis">D. Bebiano and S. Alfaro, “A weld defects detection system based on a spectrometer,” Sensors, vol. 9, no. 4, pp. 2851–2861, 2009.</ref>, the authors captured the intensity of radiation emission of electric arc. It is obvious from figure 13 that good weld's intensity stays stable while defect one's is characterised by a sudden peak. Therefore, a threshold is set to detect the defect welds.

<div><ul>
<li style="display: inline-block;"> [[File:spectroscopic_analysis.png|thumb|400px|none|Figure 13: Spectroscopic analysis <ref name="spectroscopic_analysis"></ref>]] </li>
</ul></div>

Other research into using spectroscopic data to monitor that can be of interest is <ref>D. Naso, B. Turchiano, and P. Pantaleo, “A fuzzy-logic based optical sensor for online weld defect-detection,” IEEE transactions on Industrial Informatics, vol. 1, no. 4, pp. 259–273, 2005.</ref>.

=== Acoustic analysis ===
With <ref name="acoustic_analysis">E. Cayo and S. C. Alfaro, “A non-intrusive GMA welding process quality monitoring system using acoustic sensing,” Sensors, vol. 9, no. 9, pp. 7150–7166, 2009.</ref>, the author discovered that arc ignition is characterised by a distinct sound and therefore it is possible to capture the frequency of ignition through acoustic sensors. A tolerance band was set and both ignition frequency and sound pressure level were used to monitor the quality of the weld.

<div><ul>
<li style="display: inline-block;"> [[File:acoustic_analysis.png|thumb|400px|none|Figure 14: Acoustic analysis <ref name="acoustic_analysis"></ref>]] </li>
</ul></div>

Other research into using acoustic signal to monitor that can be of interest are <ref>M. Fidali, “Detection of welding process instabilities using acoustic signals,” in International Congress on Technical Diagnostic, pp. 191–201, Springer, 2016.</ref>, <ref>L. Zhang, A. C. Basantes-Defaz, D. Ozevin, and E. Indacochea, “Real-time monitoring of welding process using air-coupled ultrasonics and acoustic emission,” The International Journal of Advanced Manufacturing Technology, vol. 101, no. 5-8, pp. 1623–1634, 2019.</ref>.

=== Electrical signal analysis ===
With <ref name="electrical_analysis_1">A. Sumesh, K. Rameshkumar, A. Raja, K. Mohandas, A. Santhakumari, and R. Shyambabu, “Establishing correlation between current and voltage signatures of the arc and weld defects in GMAW process,” Arabian Journal for Science and Engineering, vol. 42, no. 11, pp. 4649–4665, 2017.</ref>, the author plot the probability density distribution of voltage signal and discovered that the stable weld's signal concentrate in one region while unstable signal spread out. While in the paper, the authors did not developed a monitoring method using their findings, it shows that there is a significant differences between stable and unstable weld's electrical signature.

<div><ul>
<li style="display: inline-block;"> [[File:Electrical analysis pdd.png|thumb|400px|none|Figure 15: Voltage PDD <ref name="electrical_analysis_1"></ref>]] </li>
</ul></div>

In <ref name="electrical_analysis_2">S. Simpson, “Signature images for arc welding fault detection,” Science and Technology of Welding and Joining, vol. 12, no. 6, pp. 481–486, 2007.</ref>, the authors shows that by plotting the signature image of voltage and current scatter plot, both arcing and short-circuiting region's signature image can notify a defects welds. A more visual representation of short circuiting and arcing region can be found in [[#Gas metal arc welding|Gas metal arc welding section]].

<div><ul>
<li style="display: inline-block;"> [[File:Electrical voltage signature.png|thumb|400px|none|Figure 16a: Electrical voltage signature of stable weld <ref name="electrical_analysis_2"></ref>]] </li>
<li style="display: inline-block;"> [[File:Electrical signature 2.png|thumb|400px|none|Figure 16b: Electrical voltage signature of unstable weld <ref name="electrical_analysis_2"></ref>]] </li>
</ul></div>

Other research into using electrical signal to monitor that can be of interest are <ref>R. Madigan, “Arc sensing for defects in constant-voltage gas metal arc welding,” Welding Journal, vol. 78, pp. 322S–328S, 1999.</ref>, <ref>Y. Huang, K. Wang, Z. Zhou, X. Zhou, and J. Fang, “Stability evaluation of short-circuiting gas metal arc welding based on ensemble empirical mode decomposition,” Measurement Science and Technology, vol. 28, no. 3, p. 035006, 2017.</ref>,<ref>Z. Zhang, X. Chen, H. Chen, J. Zhong, and S. Chen, “Online welding quality monitoring based on feature extraction of arc voltage signal,” The International Journal of Advanced Manufacturing Technology, vol. 70, no. 9-12, pp. 1661–1671, 2014</ref>,<ref>X. Li and S. Simpson, “Parametric approach to positional fault detection in short arc welding,” Science and Technology of welding and joining, vol. 14, no. 2, pp. 146–151, 2009.</ref>, and <ref>E. Wei, D. Farson, R. Richardson, and H. Ludewig, “Detection of weld surface porosity by statistical analysis of arc current in gas metal arc welding,” Journal of Manufacturing Processes, vol. 3, no. 1, pp. 50–59, 2001.</ref>.

== Experiment setup ==
The welding system uses TPS 320i C Pulse (figure 15a), produced by Fronius, to supply power, feed wire, and system cooling. For movement control of the welding torch, the system use ABB’s IRB 1520ID (figure 15b), which is a dedicated arc welding robot. The Data Acquisition device is NI's DAQ. Each channel (voltage and current) is sampled at 25kHz while the dominant frequency for voltage and current signal are around 130Hz (for CMT process) and 190Hz (for PMC process).

[[File:Fronius_power_supply.jpg |thumb|Figure 17a: Fronius' Power Supply]]
[[File:Welding_robot2.jpg |thumb|Figure 17b: ABB's welding robot]]

[[File:Ni_daq.jpg|thumb|Figure 16: NI's DAQ device]]

== Statistical Analysis ==

=== Preprocessing ===
As the captured signal coming from Fronius' power supply, the signals also pick up switching power noise. The noise will interfere with the accuracy of the analysis if not filtered. In this project, we applied a 1000Hz digital Low Pass Butterworth filter. The effect of the filter can be seen in figure 18, while the comparation of using different filters can be seen in the gallery.

<div><ul>
<li style="display: inline-block;"> [[File:Filter_effect.png |thumb|750px|none|Figure 18: Effect of 1000Hz low pass filter]] </li>
<li style="display: inline-block;"> [[File:Freq_response.png |thumb|750px|none|Figure 19: Frequency response of the in-use filter]] </li>
</ul></div>

<gallery mode="packed-overlay" caption="Gallery: Comparison of different filtering method">
File:Effect_of_filtering2.png|widths:75px|heights:75px|In use filter vs 300Hz low pass filter
File:Effect_of_filtering3.png|widths:75px|heights:75px|In use filter vs 500Hz low pass filter
File:Effect_of_filtering4.png|widths:75px|heights:75px|In use filter vs 750Hz low pass filter
File:Effect_of_filtering5.png|widths:75px|heights:75px|In use filter vs 2000Hz low pass filter
</gallery>

=== Power Spectral Density ===
Our goal for this project is to detect the defect welds and predict the layer heights (the contact tip to work piece distance, CTWD, to be more accurate) in real time (refer to [[#Relating Contact Tip to Work-piece Distance to layer height|this section]] to understand the relationship between Contact Tip to Work-piece Distance and Layer height). An analysis of the current power spectral density (PSD) shows that there is an inverse proportional relationship between the CTWD of a weld and the frequency where its '''current signal''' PSD reach the peak, and also the magnitude of the peak. Since the peak of PSD can be fairly sensitive, we only consider the frequency of the peak PSD.

Figure 20 shows the PSD plot of voltage, current, and power signal of the weld using PMC process. In this experiment, we created multiple one-line, 8 seconds welds. These eight-second welds then get segmented into smaller two-second chunk to increase the number of data and decrease the processing time.

<div><ul>
<li style="display: inline-block;"> [[File:Psd pmc good bad all lowfreq.png |thumb|750px|none|Figure 20: Power spectral density of PMC process]] </li>
<li style="display: inline-block;"> [[File:One line weld.jpg |thumb|578px|none|Figure 21: Simple one-line weld experiment]] </li>
<li style="display: inline-block;"> [[File:Cmt_psd_stable_allCTWD_zoomed.png |thumb|750px|none|Figure 22: Simple one-line weld experiment]] </li>

</ul></div>

It seems like the inverse proportional relationship holds true for most cases. Other experiments showed that for a short period at the beginning and ending of the weld, the signal can behave differently relative to the rest of the weld. Currently, to migrate with this problem, data will not be collected and analysed (both offline and online) at these periods. However, the above results seem to only hold true for PMC processes. With CMT process, the current PSD plot indicates no relationship between the CTWD and the frequency at peak (figure 24).

For the next stage of the project, our plan is to develop a linear regression model to predict the CTWD for PMC process. With CMT process, we will have to look into different analysis techniques.

=== Support Vector Machine ===
By looking at the average voltage and average current of each segment of weld while making a simple straight path (PMC process), we realised there is a visible separation margin between stable and unstable welds. Using SVM with RBF kernel, we were able to correctly classify the quality of the weld with accuracy of 98%.

However, with CMT process, the separation cannot be observed.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_pmcNoOxide_rbf.png |thumb|750px|none|Figure 23: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:CmtNoOxide_average_voltage_current.png |thumb|750px|none|Figure 24: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

=== CTWD Regression ===
With CTWD prediction, the result will only be useful if it is done on-the-line while making the part. And therefore, as oppose to defect detection, we cannot have a short calibration process every constant time to check for worn contact tip. By analysing the data while building a real part, we found that the average voltage (for PMC process) and average current (for CMT process) seems to have the strongest linear relationship with the CTWD. The acceptable error for CTWD prediction is +/- 1 mm (according to AML3D requirement). For both PMC and CMT process, our root means square error is less than 1 (PMC: 0.68 and CMT: 0.83), which satisfied the requirement.

<div><ul>
<li style="display: inline-block;"> [[File:Linear_regression_pmc_wall.png |thumb|750px|none|Figure 25: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:Linear_regression_cmt_wall.png |thumb|750px|none|Figure 26: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

= Result discussion =
These results show that there is a strong possibility of defect detection and CTWD prediction in welding. However, these initial findings are still a proof of concept. In the next stage of our research, we will refine the algorithm and integrate into AML's pipeline.

== References ==
<references />

Projects:2019s2-20001 Using Machine Learning to Determine Deposit Height and Defects for Wire + Arc Additive Manufacture (3D printing)

2019-10-15T05:28:40Z

A1713568: /* Preprocessing */

[[Category:Projects]]
[[Category:Final Year Projects]]
[[Category:2019s2|24501]]
== Introduction ==
AML3D is one of the leading company in the Wire Arc Additive Manufacturing (WAAM) industry. In short, WAAM is the Metal 3D printing. The technology has the ability to reduce up to 75% of manufacture time, cost, and material. Currently, WAAM is benefiting other industry such as: mining, defence, marine, and aviation.

<gallery mode="slideshow">
File:Welding baby.gif|center|Welding using robotic arm
File:Laser baby.gif|center|User laser sensor to find the CTWD
</gallery>

[[File:aml3d_logo.jpg|center|100px]]
=== Project team ===
==== Project students ====
* Anh Tran
* Nhat Nguyen
==== Supervisors ====
* Dr. Brian Ng
* Dr. Paul Colegrove (AML3D, sponsored company)
=== Objectives ===
The objective of this project is to increase the efficiency of the manufacture process at AML3D. In order to do so, the team will investigate into the possibility of automating and optimising the quality control processes. The two quality control processes that are currently being implemented at AML3D are measuring layer height using laser sensors, and human supervision for detecting defects. These processes add overhead into production time and usage of human resource, which is not desired. To achieve the goal, machine learning methods will be used extensively to analyse the electrical signatures of the weld process.

== Background ==
=== Wire Arc Additive Manufacturing ===
Wire and Arc Additive Manufacturing (WAAM) is a type of additive manufacturing that uses electric arc as the heat source and material wire to feed the manufacture process <ref name="WAAM"> S. W. Williams, F. Martina, A. C. Addison, J. Ding, G. Pardal & P. Colegrove (2016) Wire + Arc Additive Manufacturing, Materials Science and Technology, 32:7, 641-647, DOI: 10.1179/1743284715Y.0000000073 </ref>. WAAM has been investigated since the 1990s <ref name="WAAM" />, but only recently that it received more attention from the manufacture world.

Its significant comes from the ability to manufacture complex model with less time and less material. Figure 1a and 1b show real parts that was manufactured by AML3D. Such custom made parts might takes months before be ready to be shipped, but with WAAM, the production time can reduce down to weeks. Currently, the industries that benefit the most from WAAM are maritime and aerospace.

Similar to other additive manufacture methods, WAAM achieves such results by sliding the models into multiple layers, and then build the model layer by layer. The movement control is normally handled by a robotic arm (AML3D uses ABB's Arc Welder robot), and the welding path is generated by a Computer Aid Manufacture (CAM) software.

[[File:welding_part1.JPG|thumb|Figure 1a: A propeller]]
[[File:welding_part2.jpg|thumb|Figure 1b: Another part of the ship]]
[[File:abb_robot.jpg|thumb|Figure 2: ABB's IRB 1520ID Arc Welder]]

=== Gas metal arc welding ===
<div><ul>
<li style="display: inline-block;"> [[File:gmaw_welding.png|thumb|500px|none|Figure 2: GMAW welding process <ref name="welding_book"> Norrish, J. (2006). Advanced welding processes. Elsevier. </ref>]] </li>
</ul></div>

Gas metal arc welding (GMAW), or also known as metal inert gas (MIG), is the welding process used at AML3D for WAAM. Due to its high deposition rate and economic benefits, GMAW has became more popular. We will explore two variants of GMAW, which are Pulse Multi Control and Cold Metal Transfer. Both processes are developed by Fronius, an Austrian welding company. Note that it is not required to understand the welding physics to follow this wiki. The next two following sections' purpose is to highlight the high variability, high dynamic nature in welding.

[[File:fronius.png|thumb|Figure 3: Fronius logo]]

==== Pulse Multi Control ====
Pulse Multi Control (PMC) welding is Fronius' modified Pulsed GMAW. The advantage of Pulsed GMAW is the ability to control the metal droplet transfer in welding. Note that from figure 4, the metal droplet is characterised by a downward slope of the current pulse. Figure 5 shows the current signal captured from experiments conducted at AML3D (PMC is used).

<div><ul>
<li style="display: inline-block;"> [[File:Pulsed_gmaw_transfer_mode.png|thumb|500px|none|Figure 4: Relationship between pulse current and droplet transfer<ref name="welding_book" />]] </li>
<li style="display: inline-block;"> [[File:Pmc_current_in_time_domain.png|thumb|650px|none|Figure 5: Current signal captured from experiments conducted at AML3D (PMC process)]] </li>
</ul></div>
==== Cold Metal Transfer ====
Cold Metal Transfer (CMT) is a complex welding process, also developed by Fronius. By detecting a short circuit (mode C and D in figure 8), CMT can make adjustment such as retracting the welding material to cool down, and therefore create a smoother, more stable weld. The complexity of CMT can be seen in the current signal of the process (figure 9). Compare to PMC, CMT's signal has more variation during one signal period.

<div><ul>
<li style="display: inline-block;"> [[File:Transfer_mode.png|thumb|800px|none|Figure 8: Mechanism of metal droplet transfer mode]] </li>
<li style="display: inline-block;"> [[File:Cmt1_current_in_time_domain.png|thumb|735px|none|Figure 9: Current signal captured from experiments conducted at AML3D (CMT process)]] </li>
</ul></div>
=== Machine Learning ===
Machine learning (or statistical learning) refer to a set of tools to give data scientists an insight into the data and make better decisions based on those information <ref name="introduction_ml_book"> James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An introduction to statistical learning (Vol. 112, p. 18). New York: Springer. </ref>. In this project, we will explore the capabilities of classical machine learning methods (i.e anything but Deep Learning) due to the limitation of our available data. By the time of writing this wiki, we have only explored Support Vector Machine.

==== Support Vector Machine ====
The main idea of Support Vector Machine (SVM) is to draw hyperplanes that best separate multiple classes of data. The original SVM judges best decision hyperplane based on the separation margin between data classes (figure 10). However, there are multiple variants of SVM, such as hard/soft margin SVM, Nu-SVM, One class SVM (to solve anomaly detection problem). In our project, we choose to use a soft margin implementation of SVM. The benefit of soft margin SVM (as oppose to hard margin SVM) is that by allowing some misclassifications, we get a wider margin and therefore the solution becomes more "generalise" (i.e work better with a new set of dataset that is independent from the training dataset).

<div><ul>
<li style="display: inline-block;"> [[File:svm_margin.png|thumb|400px|none|Figure 10: Separation margin <ref name="bishop_book"> Bishop, C. M. (2006). Pattern recognition and machine learning. Springer. </ref>]] </li>
</ul></div>

==== Kernel tricks ====
Figure 10 is of linearly separable cases for SVM. For dataset that is not easy to separate with a hyperplane in their original features space (such as figure 11), the data often get projected into a higher dimension space where it is easier to separate (figure 14). Such technique is called the Kernel Tricks. It is important to note that Kernel Tricks does not actually "map" the data from '''X''' dimension to '''Z''' dimension, but rather pairwise compare the similarity of the input space in '''Z''' dimension.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_nonseparable.png |thumb|350px|none|Figure 11: Nonseparable dataset <ref name="bishop_book"></ref>]] </li>
<li style="display: inline-block;"> [[File:Svm_kernel_trick.png |thumb|650px|none|Figure 12: Simplified example of the Kernel tricks]] </li>
</ul></div>

== Previous studies ==
There are many researches into defect detection in GMAW process. The three most dominant analysis methods are spectroscopic analysis, acoustic analysis, and electrical signal analysis.

=== Spectroscopic analysis ===
In <ref name="spectroscopic_analysis">D. Bebiano and S. Alfaro, “A weld defects detection system based on a spectrometer,” Sensors, vol. 9, no. 4, pp. 2851–2861, 2009.</ref>, the authors captured the intensity of radiation emission of electric arc. It is obvious from figure 13 that good weld's intensity stays stable while defect one's is characterised by a sudden peak. Therefore, a threshold is set to detect the defect welds.

<div><ul>
<li style="display: inline-block;"> [[File:spectroscopic_analysis.png|thumb|400px|none|Figure 13: Spectroscopic analysis <ref name="spectroscopic_analysis"></ref>]] </li>
</ul></div>

Other research into using spectroscopic data to monitor that can be of interest is <ref>D. Naso, B. Turchiano, and P. Pantaleo, “A fuzzy-logic based optical sensor for online weld defect-detection,” IEEE transactions on Industrial Informatics, vol. 1, no. 4, pp. 259–273, 2005.</ref>.

=== Acoustic analysis ===
With <ref name="acoustic_analysis">E. Cayo and S. C. Alfaro, “A non-intrusive GMA welding process quality monitoring system using acoustic sensing,” Sensors, vol. 9, no. 9, pp. 7150–7166, 2009.</ref>, the author discovered that arc ignition is characterised by a distinct sound and therefore it is possible to capture the frequency of ignition through acoustic sensors. A tolerance band was set and both ignition frequency and sound pressure level were used to monitor the quality of the weld.

<div><ul>
<li style="display: inline-block;"> [[File:acoustic_analysis.png|thumb|400px|none|Figure 14: Acoustic analysis <ref name="acoustic_analysis"></ref>]] </li>
</ul></div>

Other research into using acoustic signal to monitor that can be of interest are <ref>M. Fidali, “Detection of welding process instabilities using acoustic signals,” in International Congress on Technical Diagnostic, pp. 191–201, Springer, 2016.</ref>, <ref>L. Zhang, A. C. Basantes-Defaz, D. Ozevin, and E. Indacochea, “Real-time monitoring of welding process using air-coupled ultrasonics and acoustic emission,” The International Journal of Advanced Manufacturing Technology, vol. 101, no. 5-8, pp. 1623–1634, 2019.</ref>.

=== Electrical signal analysis ===
With <ref name="electrical_analysis_1">A. Sumesh, K. Rameshkumar, A. Raja, K. Mohandas, A. Santhakumari, and R. Shyambabu, “Establishing correlation between current and voltage signatures of the arc and weld defects in GMAW process,” Arabian Journal for Science and Engineering, vol. 42, no. 11, pp. 4649–4665, 2017.</ref>, the author plot the probability density distribution of voltage signal and discovered that the stable weld's signal concentrate in one region while unstable signal spread out. While in the paper, the authors did not developed a monitoring method using their findings, it shows that there is a significant differences between stable and unstable weld's electrical signature.

<div><ul>
<li style="display: inline-block;"> [[File:Electrical analysis pdd.png|thumb|400px|none|Figure 15: Voltage PDD <ref name="electrical_analysis_1"></ref>]] </li>
</ul></div>

In <ref name="electrical_analysis_2">S. Simpson, “Signature images for arc welding fault detection,” Science and Technology of Welding and Joining, vol. 12, no. 6, pp. 481–486, 2007.</ref>, the authors shows that by plotting the signature image of voltage and current scatter plot, both arcing and short-circuiting region's signature image can notify a defects welds. A more visual representation of short circuiting and arcing region can be found in [[#Gas metal arc welding|Gas metal arc welding section]].

<div><ul>
<li style="display: inline-block;"> [[File:Electrical voltage signature.png|thumb|400px|none|Figure 16a: Electrical voltage signature of stable weld <ref name="electrical_analysis_2"></ref>]] </li>
<li style="display: inline-block;"> [[File:Electrical signature 2.png|thumb|400px|none|Figure 16b: Electrical voltage signature of unstable weld <ref name="electrical_analysis_2"></ref>]] </li>
</ul></div>

Other research into using electrical signal to monitor that can be of interest are <ref>R. Madigan, “Arc sensing for defects in constant-voltage gas metal arc welding,” Welding Journal, vol. 78, pp. 322S–328S, 1999.</ref>, <ref>Y. Huang, K. Wang, Z. Zhou, X. Zhou, and J. Fang, “Stability evaluation of short-circuiting gas metal arc welding based on ensemble empirical mode decomposition,” Measurement Science and Technology, vol. 28, no. 3, p. 035006, 2017.</ref>,<ref>Z. Zhang, X. Chen, H. Chen, J. Zhong, and S. Chen, “Online welding quality monitoring based on feature extraction of arc voltage signal,” The International Journal of Advanced Manufacturing Technology, vol. 70, no. 9-12, pp. 1661–1671, 2014</ref>,<ref>X. Li and S. Simpson, “Parametric approach to positional fault detection in short arc welding,” Science and Technology of welding and joining, vol. 14, no. 2, pp. 146–151, 2009.</ref>, and <ref>E. Wei, D. Farson, R. Richardson, and H. Ludewig, “Detection of weld surface porosity by statistical analysis of arc current in gas metal arc welding,” Journal of Manufacturing Processes, vol. 3, no. 1, pp. 50–59, 2001.</ref>.

== Experiment setup ==
The welding system uses TPS 320i C Pulse (figure 15a), produced by Fronius, to supply power, feed wire, and system cooling. For movement control of the welding torch, the system use ABB’s IRB 1520ID (figure 15b), which is a dedicated arc welding robot. The Data Acquisition device is NI's DAQ. Each channel (voltage and current) is sampled at 25kHz while the dominant frequency for voltage and current signal are around 130Hz (for CMT process) and 190Hz (for PMC process).

[[File:Fronius_power_supply.jpg |thumb|Figure 17a: Fronius' Power Supply]]
[[File:Welding_robot2.jpg |thumb|Figure 17b: ABB's welding robot]]

[[File:Ni_daq.jpg|thumb|Figure 16: NI's DAQ device]]

== Statistical Analysis ==

=== Preprocessing ===
As the captured signal coming from Fronius' power supply, the signals also pick up switching power noise. The noise will interfere with the accuracy of the analysis if not filtered. In this project, we applied a 1000Hz digital Low Pass Butterworth filter. The effect of the filter can be seen in figure 18, while the comparation of using different filters can be seen in the gallery.

<div><ul>
<li style="display: inline-block;"> [[File:Filter_effect.png |thumb|750px|none|Figure 18: Effect of 1000Hz low pass filter]] </li>
<li style="display: inline-block;"> [[File:Freq_response.png |thumb|750px|none|Figure 19: Frequency response of the in-use filter]] </li>
</ul></div>

<gallery mode="packed-overlay" caption="Gallery: Comparison of different filtering method">
File:Effect_of_filtering2.png|widths:75px|heights:75px|In use filter vs 300Hz low pass filter
File:Effect_of_filtering3.png|widths:75px|heights:75px|In use filter vs 500Hz low pass filter
File:Effect_of_filtering4.png|widths:75px|heights:75px|In use filter vs 750Hz low pass filter
File:Effect_of_filtering5.png|widths:75px|heights:75px|In use filter vs 2000Hz low pass filter
</gallery>

=== Power Spectral Density ===
Our goal for this project is to detect the defect welds and predict the layer heights (the contact tip to work piece distance, CTWD, to be more accurate) in real time (refer to [[#Relating Contact Tip to Work-piece Distance to layer height|this section]] to understand the relationship between Contact Tip to Work-piece Distance and Layer height). An analysis of the current power spectral density (PSD) shows that there is an inverse proportional relationship between the CTWD of a weld and the frequency where its '''current signal''' PSD reach the peak, and also the magnitude of the peak. Since the peak of PSD can be fairly sensitive, we only consider the frequency of the peak PSD.

Figure 20 shows the PSD plot of voltage, current, and power signal of the weld using PMC process. In this experiment, we created multiple one-line, 8 seconds welds. These eight-second welds then get segmented into smaller two-second chunk to increase the number of data and decrease the processing time.

<div><ul>
<li style="display: inline-block;"> [[File:Psd pmc good bad all lowfreq.png |thumb|750px|none|Figure 20: Power spectral density of PMC process]] </li>
<li style="display: inline-block;"> [[File:One line weld.jpg |thumb|578px|none|Figure 21: Simple one-line weld experiment]] </li>
<li style="display: inline-block;"> [[File:Cmt_psd_stable_allCTWD_zoomed.png |thumb|750px|none|Figure 22: Simple one-line weld experiment]] </li>

</ul></div>

It seems like the inverse proportional relationship holds true for most cases. Other experiments showed that for a short period at the beginning and ending of the weld, the signal can behave differently relative to the rest of the weld. Currently, to migrate with this problem, data will not be collected and analysed (both offline and online) at these periods. However, the above results seem to only hold true for PMC processes. With CMT process, the current PSD plot indicates no relationship between the CTWD and the frequency at peak (figure 24).

For the next stage of the project, our plan is to develop a linear regression model to predict the CTWD for PMC process. With CMT process, we will have to look into different analysis techniques.

=== Support Vector Machine ===
By looking at the average voltage and average current of each segment of weld while making a simple straight path (PMC process), we realised there is a visible separation margin between stable and unstable welds. Using SVM with RBF kernel, we were able to correctly classify the quality of the weld with accuracy of 98%.

However, with CMT process, the separation cannot be observed.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_pmcNoOxide_rbf.png |thumb|750px|none|Figure 23: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:CmtNoOxide_average_voltage_current.png |thumb|750px|none|Figure 24: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

=== CTWD Regression ===
With CTWD prediction, the result will only be useful if it is done on-the-line while making the part. And therefore, as oppose to defect detection, we cannot have a short calibration process every constant time to check for worn contact tip. By analysing the data while building a real part, we found that the average voltage (for PMC process) and average current (for CMT process) seems to have the strongest linear relationship with the CTWD. The acceptable error for CTWD prediction is +/- 1 mm (according to AML3D requirement). For both PMC and CMT process, our root means square error is less than 1 (PMC: 0.68 and CMT: 0.83), which satisfied the requirement.

<div><ul>
<li style="display: inline-block;"> [[File:Linear_regression_pmc_wall.png |thumb|750px|none|Figure 25: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:Linear_regression_cmt_wall.png |thumb|750px|none|Figure 26: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

== Result discussion ==
These results show that there is a strong possibility of defect detection and CTWD prediction in welding. However, these initial findings are still a proof of concept. In the next stage of our research, we will refine the algorithm and integrate into AML's pipeline.

== References ==
<references />

Projects:2019s2-20001 Using Machine Learning to Determine Deposit Height and Defects for Wire + Arc Additive Manufacture (3D printing)

2019-10-15T05:28:21Z

A1713568: /* Preprocessing */

[[Category:Projects]]
[[Category:Final Year Projects]]
[[Category:2019s2|24501]]
== Introduction ==
AML3D is one of the leading company in the Wire Arc Additive Manufacturing (WAAM) industry. In short, WAAM is the Metal 3D printing. The technology has the ability to reduce up to 75% of manufacture time, cost, and material. Currently, WAAM is benefiting other industry such as: mining, defence, marine, and aviation.

<gallery mode="slideshow">
File:Welding baby.gif|center|Welding using robotic arm
File:Laser baby.gif|center|User laser sensor to find the CTWD
</gallery>

[[File:aml3d_logo.jpg|center|100px]]
=== Project team ===
==== Project students ====
* Anh Tran
* Nhat Nguyen
==== Supervisors ====
* Dr. Brian Ng
* Dr. Paul Colegrove (AML3D, sponsored company)
=== Objectives ===
The objective of this project is to increase the efficiency of the manufacture process at AML3D. In order to do so, the team will investigate into the possibility of automating and optimising the quality control processes. The two quality control processes that are currently being implemented at AML3D are measuring layer height using laser sensors, and human supervision for detecting defects. These processes add overhead into production time and usage of human resource, which is not desired. To achieve the goal, machine learning methods will be used extensively to analyse the electrical signatures of the weld process.

== Background ==
=== Wire Arc Additive Manufacturing ===
Wire and Arc Additive Manufacturing (WAAM) is a type of additive manufacturing that uses electric arc as the heat source and material wire to feed the manufacture process <ref name="WAAM"> S. W. Williams, F. Martina, A. C. Addison, J. Ding, G. Pardal & P. Colegrove (2016) Wire + Arc Additive Manufacturing, Materials Science and Technology, 32:7, 641-647, DOI: 10.1179/1743284715Y.0000000073 </ref>. WAAM has been investigated since the 1990s <ref name="WAAM" />, but only recently that it received more attention from the manufacture world.

Its significant comes from the ability to manufacture complex model with less time and less material. Figure 1a and 1b show real parts that was manufactured by AML3D. Such custom made parts might takes months before be ready to be shipped, but with WAAM, the production time can reduce down to weeks. Currently, the industries that benefit the most from WAAM are maritime and aerospace.

Similar to other additive manufacture methods, WAAM achieves such results by sliding the models into multiple layers, and then build the model layer by layer. The movement control is normally handled by a robotic arm (AML3D uses ABB's Arc Welder robot), and the welding path is generated by a Computer Aid Manufacture (CAM) software.

[[File:welding_part1.JPG|thumb|Figure 1a: A propeller]]
[[File:welding_part2.jpg|thumb|Figure 1b: Another part of the ship]]
[[File:abb_robot.jpg|thumb|Figure 2: ABB's IRB 1520ID Arc Welder]]

=== Gas metal arc welding ===
<div><ul>
<li style="display: inline-block;"> [[File:gmaw_welding.png|thumb|500px|none|Figure 2: GMAW welding process <ref name="welding_book"> Norrish, J. (2006). Advanced welding processes. Elsevier. </ref>]] </li>
</ul></div>

Gas metal arc welding (GMAW), or also known as metal inert gas (MIG), is the welding process used at AML3D for WAAM. Due to its high deposition rate and economic benefits, GMAW has became more popular. We will explore two variants of GMAW, which are Pulse Multi Control and Cold Metal Transfer. Both processes are developed by Fronius, an Austrian welding company. Note that it is not required to understand the welding physics to follow this wiki. The next two following sections' purpose is to highlight the high variability, high dynamic nature in welding.

[[File:fronius.png|thumb|Figure 3: Fronius logo]]

==== Pulse Multi Control ====
Pulse Multi Control (PMC) welding is Fronius' modified Pulsed GMAW. The advantage of Pulsed GMAW is the ability to control the metal droplet transfer in welding. Note that from figure 4, the metal droplet is characterised by a downward slope of the current pulse. Figure 5 shows the current signal captured from experiments conducted at AML3D (PMC is used).

<div><ul>
<li style="display: inline-block;"> [[File:Pulsed_gmaw_transfer_mode.png|thumb|500px|none|Figure 4: Relationship between pulse current and droplet transfer<ref name="welding_book" />]] </li>
<li style="display: inline-block;"> [[File:Pmc_current_in_time_domain.png|thumb|650px|none|Figure 5: Current signal captured from experiments conducted at AML3D (PMC process)]] </li>
</ul></div>
==== Cold Metal Transfer ====
Cold Metal Transfer (CMT) is a complex welding process, also developed by Fronius. By detecting a short circuit (mode C and D in figure 8), CMT can make adjustment such as retracting the welding material to cool down, and therefore create a smoother, more stable weld. The complexity of CMT can be seen in the current signal of the process (figure 9). Compare to PMC, CMT's signal has more variation during one signal period.

<div><ul>
<li style="display: inline-block;"> [[File:Transfer_mode.png|thumb|800px|none|Figure 8: Mechanism of metal droplet transfer mode]] </li>
<li style="display: inline-block;"> [[File:Cmt1_current_in_time_domain.png|thumb|735px|none|Figure 9: Current signal captured from experiments conducted at AML3D (CMT process)]] </li>
</ul></div>
=== Machine Learning ===
Machine learning (or statistical learning) refer to a set of tools to give data scientists an insight into the data and make better decisions based on those information <ref name="introduction_ml_book"> James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An introduction to statistical learning (Vol. 112, p. 18). New York: Springer. </ref>. In this project, we will explore the capabilities of classical machine learning methods (i.e anything but Deep Learning) due to the limitation of our available data. By the time of writing this wiki, we have only explored Support Vector Machine.

==== Support Vector Machine ====
The main idea of Support Vector Machine (SVM) is to draw hyperplanes that best separate multiple classes of data. The original SVM judges best decision hyperplane based on the separation margin between data classes (figure 10). However, there are multiple variants of SVM, such as hard/soft margin SVM, Nu-SVM, One class SVM (to solve anomaly detection problem). In our project, we choose to use a soft margin implementation of SVM. The benefit of soft margin SVM (as oppose to hard margin SVM) is that by allowing some misclassifications, we get a wider margin and therefore the solution becomes more "generalise" (i.e work better with a new set of dataset that is independent from the training dataset).

<div><ul>
<li style="display: inline-block;"> [[File:svm_margin.png|thumb|400px|none|Figure 10: Separation margin <ref name="bishop_book"> Bishop, C. M. (2006). Pattern recognition and machine learning. Springer. </ref>]] </li>
</ul></div>

==== Kernel tricks ====
Figure 10 is of linearly separable cases for SVM. For dataset that is not easy to separate with a hyperplane in their original features space (such as figure 11), the data often get projected into a higher dimension space where it is easier to separate (figure 14). Such technique is called the Kernel Tricks. It is important to note that Kernel Tricks does not actually "map" the data from '''X''' dimension to '''Z''' dimension, but rather pairwise compare the similarity of the input space in '''Z''' dimension.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_nonseparable.png |thumb|350px|none|Figure 11: Nonseparable dataset <ref name="bishop_book"></ref>]] </li>
<li style="display: inline-block;"> [[File:Svm_kernel_trick.png |thumb|650px|none|Figure 12: Simplified example of the Kernel tricks]] </li>
</ul></div>

== Previous studies ==
There are many researches into defect detection in GMAW process. The three most dominant analysis methods are spectroscopic analysis, acoustic analysis, and electrical signal analysis.

=== Spectroscopic analysis ===
In <ref name="spectroscopic_analysis">D. Bebiano and S. Alfaro, “A weld defects detection system based on a spectrometer,” Sensors, vol. 9, no. 4, pp. 2851–2861, 2009.</ref>, the authors captured the intensity of radiation emission of electric arc. It is obvious from figure 13 that good weld's intensity stays stable while defect one's is characterised by a sudden peak. Therefore, a threshold is set to detect the defect welds.

<div><ul>
<li style="display: inline-block;"> [[File:spectroscopic_analysis.png|thumb|400px|none|Figure 13: Spectroscopic analysis <ref name="spectroscopic_analysis"></ref>]] </li>
</ul></div>

Other research into using spectroscopic data to monitor that can be of interest is <ref>D. Naso, B. Turchiano, and P. Pantaleo, “A fuzzy-logic based optical sensor for online weld defect-detection,” IEEE transactions on Industrial Informatics, vol. 1, no. 4, pp. 259–273, 2005.</ref>.

=== Acoustic analysis ===
With <ref name="acoustic_analysis">E. Cayo and S. C. Alfaro, “A non-intrusive GMA welding process quality monitoring system using acoustic sensing,” Sensors, vol. 9, no. 9, pp. 7150–7166, 2009.</ref>, the author discovered that arc ignition is characterised by a distinct sound and therefore it is possible to capture the frequency of ignition through acoustic sensors. A tolerance band was set and both ignition frequency and sound pressure level were used to monitor the quality of the weld.

<div><ul>
<li style="display: inline-block;"> [[File:acoustic_analysis.png|thumb|400px|none|Figure 14: Acoustic analysis <ref name="acoustic_analysis"></ref>]] </li>
</ul></div>

Other research into using acoustic signal to monitor that can be of interest are <ref>M. Fidali, “Detection of welding process instabilities using acoustic signals,” in International Congress on Technical Diagnostic, pp. 191–201, Springer, 2016.</ref>, <ref>L. Zhang, A. C. Basantes-Defaz, D. Ozevin, and E. Indacochea, “Real-time monitoring of welding process using air-coupled ultrasonics and acoustic emission,” The International Journal of Advanced Manufacturing Technology, vol. 101, no. 5-8, pp. 1623–1634, 2019.</ref>.

=== Electrical signal analysis ===
With <ref name="electrical_analysis_1">A. Sumesh, K. Rameshkumar, A. Raja, K. Mohandas, A. Santhakumari, and R. Shyambabu, “Establishing correlation between current and voltage signatures of the arc and weld defects in GMAW process,” Arabian Journal for Science and Engineering, vol. 42, no. 11, pp. 4649–4665, 2017.</ref>, the author plot the probability density distribution of voltage signal and discovered that the stable weld's signal concentrate in one region while unstable signal spread out. While in the paper, the authors did not developed a monitoring method using their findings, it shows that there is a significant differences between stable and unstable weld's electrical signature.

<div><ul>
<li style="display: inline-block;"> [[File:Electrical analysis pdd.png|thumb|400px|none|Figure 15: Voltage PDD <ref name="electrical_analysis_1"></ref>]] </li>
</ul></div>

In <ref name="electrical_analysis_2">S. Simpson, “Signature images for arc welding fault detection,” Science and Technology of Welding and Joining, vol. 12, no. 6, pp. 481–486, 2007.</ref>, the authors shows that by plotting the signature image of voltage and current scatter plot, both arcing and short-circuiting region's signature image can notify a defects welds. A more visual representation of short circuiting and arcing region can be found in [[#Gas metal arc welding|Gas metal arc welding section]].

<div><ul>
<li style="display: inline-block;"> [[File:Electrical voltage signature.png|thumb|400px|none|Figure 16a: Electrical voltage signature of stable weld <ref name="electrical_analysis_2"></ref>]] </li>
<li style="display: inline-block;"> [[File:Electrical signature 2.png|thumb|400px|none|Figure 16b: Electrical voltage signature of unstable weld <ref name="electrical_analysis_2"></ref>]] </li>
</ul></div>

Other research into using electrical signal to monitor that can be of interest are <ref>R. Madigan, “Arc sensing for defects in constant-voltage gas metal arc welding,” Welding Journal, vol. 78, pp. 322S–328S, 1999.</ref>, <ref>Y. Huang, K. Wang, Z. Zhou, X. Zhou, and J. Fang, “Stability evaluation of short-circuiting gas metal arc welding based on ensemble empirical mode decomposition,” Measurement Science and Technology, vol. 28, no. 3, p. 035006, 2017.</ref>,<ref>Z. Zhang, X. Chen, H. Chen, J. Zhong, and S. Chen, “Online welding quality monitoring based on feature extraction of arc voltage signal,” The International Journal of Advanced Manufacturing Technology, vol. 70, no. 9-12, pp. 1661–1671, 2014</ref>,<ref>X. Li and S. Simpson, “Parametric approach to positional fault detection in short arc welding,” Science and Technology of welding and joining, vol. 14, no. 2, pp. 146–151, 2009.</ref>, and <ref>E. Wei, D. Farson, R. Richardson, and H. Ludewig, “Detection of weld surface porosity by statistical analysis of arc current in gas metal arc welding,” Journal of Manufacturing Processes, vol. 3, no. 1, pp. 50–59, 2001.</ref>.

== Experiment setup ==
The welding system uses TPS 320i C Pulse (figure 15a), produced by Fronius, to supply power, feed wire, and system cooling. For movement control of the welding torch, the system use ABB’s IRB 1520ID (figure 15b), which is a dedicated arc welding robot. The Data Acquisition device is NI's DAQ. Each channel (voltage and current) is sampled at 25kHz while the dominant frequency for voltage and current signal are around 130Hz (for CMT process) and 190Hz (for PMC process).

[[File:Fronius_power_supply.jpg |thumb|Figure 17a: Fronius' Power Supply]]
[[File:Welding_robot2.jpg |thumb|Figure 17b: ABB's welding robot]]

[[File:Ni_daq.jpg|thumb|Figure 16: NI's DAQ device]]

== Statistical Analysis ==

=== Preprocessing ===
As the captured signal coming from Fronius' power supply, the signals also pick up switching power noise. The noise will interfere with the accuracy of the analysis if not filtered. In this project, we applied a 1000Hz digital Low Pass Butterworth filter. The effect of the filter can be seen in figure 18, while the comparation of using different filters can be seen in the gallery.

<div><ul>
<li style="display: inline-block;"> [[File:Filter_effect.png |thumb|750px|none|Figure 18: Effect of 1000Hz low pass filter]] </li>
<li style="display: inline-block;"> [[File:Freq_response.png |thumb|750px|none|Figure 19: Frequency response of the in-use filter]] </li>
</ul></div>

<gallery mode="packed-overlay" caption="Gallery: Comparison of different filtering method" widths=75px heights=75px>
File:Effect_of_filtering2.png|widths:75px|heights:75px|In use filter vs 300Hz low pass filter
File:Effect_of_filtering3.png|widths:75px|heights:75px|In use filter vs 500Hz low pass filter
File:Effect_of_filtering4.png|widths:75px|heights:75px|In use filter vs 750Hz low pass filter
File:Effect_of_filtering5.png|widths:75px|heights:75px|In use filter vs 2000Hz low pass filter
</gallery>

=== Power Spectral Density ===
Our goal for this project is to detect the defect welds and predict the layer heights (the contact tip to work piece distance, CTWD, to be more accurate) in real time (refer to [[#Relating Contact Tip to Work-piece Distance to layer height|this section]] to understand the relationship between Contact Tip to Work-piece Distance and Layer height). An analysis of the current power spectral density (PSD) shows that there is an inverse proportional relationship between the CTWD of a weld and the frequency where its '''current signal''' PSD reach the peak, and also the magnitude of the peak. Since the peak of PSD can be fairly sensitive, we only consider the frequency of the peak PSD.

Figure 20 shows the PSD plot of voltage, current, and power signal of the weld using PMC process. In this experiment, we created multiple one-line, 8 seconds welds. These eight-second welds then get segmented into smaller two-second chunk to increase the number of data and decrease the processing time.

<div><ul>
<li style="display: inline-block;"> [[File:Psd pmc good bad all lowfreq.png |thumb|750px|none|Figure 20: Power spectral density of PMC process]] </li>
<li style="display: inline-block;"> [[File:One line weld.jpg |thumb|578px|none|Figure 21: Simple one-line weld experiment]] </li>
<li style="display: inline-block;"> [[File:Cmt_psd_stable_allCTWD_zoomed.png |thumb|750px|none|Figure 22: Simple one-line weld experiment]] </li>

</ul></div>

It seems like the inverse proportional relationship holds true for most cases. Other experiments showed that for a short period at the beginning and ending of the weld, the signal can behave differently relative to the rest of the weld. Currently, to migrate with this problem, data will not be collected and analysed (both offline and online) at these periods. However, the above results seem to only hold true for PMC processes. With CMT process, the current PSD plot indicates no relationship between the CTWD and the frequency at peak (figure 24).

For the next stage of the project, our plan is to develop a linear regression model to predict the CTWD for PMC process. With CMT process, we will have to look into different analysis techniques.

=== Support Vector Machine ===
By looking at the average voltage and average current of each segment of weld while making a simple straight path (PMC process), we realised there is a visible separation margin between stable and unstable welds. Using SVM with RBF kernel, we were able to correctly classify the quality of the weld with accuracy of 98%.

However, with CMT process, the separation cannot be observed.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_pmcNoOxide_rbf.png |thumb|750px|none|Figure 23: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:CmtNoOxide_average_voltage_current.png |thumb|750px|none|Figure 24: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

=== CTWD Regression ===
With CTWD prediction, the result will only be useful if it is done on-the-line while making the part. And therefore, as oppose to defect detection, we cannot have a short calibration process every constant time to check for worn contact tip. By analysing the data while building a real part, we found that the average voltage (for PMC process) and average current (for CMT process) seems to have the strongest linear relationship with the CTWD. The acceptable error for CTWD prediction is +/- 1 mm (according to AML3D requirement). For both PMC and CMT process, our root means square error is less than 1 (PMC: 0.68 and CMT: 0.83), which satisfied the requirement.

<div><ul>
<li style="display: inline-block;"> [[File:Linear_regression_pmc_wall.png |thumb|750px|none|Figure 25: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:Linear_regression_cmt_wall.png |thumb|750px|none|Figure 26: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

== Result discussion ==
These results show that there is a strong possibility of defect detection and CTWD prediction in welding. However, these initial findings are still a proof of concept. In the next stage of our research, we will refine the algorithm and integrate into AML's pipeline.

== References ==
<references />

Projects:2019s2-20001 Using Machine Learning to Determine Deposit Height and Defects for Wire + Arc Additive Manufacture (3D printing)

2019-10-15T05:27:33Z

A1713568: /* Preprocessing */

[[Category:Projects]]
[[Category:Final Year Projects]]
[[Category:2019s2|24501]]
== Introduction ==
AML3D is one of the leading company in the Wire Arc Additive Manufacturing (WAAM) industry. In short, WAAM is the Metal 3D printing. The technology has the ability to reduce up to 75% of manufacture time, cost, and material. Currently, WAAM is benefiting other industry such as: mining, defence, marine, and aviation.

<gallery mode="slideshow">
File:Welding baby.gif|center|Welding using robotic arm
File:Laser baby.gif|center|User laser sensor to find the CTWD
</gallery>

[[File:aml3d_logo.jpg|center|100px]]
=== Project team ===
==== Project students ====
* Anh Tran
* Nhat Nguyen
==== Supervisors ====
* Dr. Brian Ng
* Dr. Paul Colegrove (AML3D, sponsored company)
=== Objectives ===
The objective of this project is to increase the efficiency of the manufacture process at AML3D. In order to do so, the team will investigate into the possibility of automating and optimising the quality control processes. The two quality control processes that are currently being implemented at AML3D are measuring layer height using laser sensors, and human supervision for detecting defects. These processes add overhead into production time and usage of human resource, which is not desired. To achieve the goal, machine learning methods will be used extensively to analyse the electrical signatures of the weld process.

== Background ==
=== Wire Arc Additive Manufacturing ===
Wire and Arc Additive Manufacturing (WAAM) is a type of additive manufacturing that uses electric arc as the heat source and material wire to feed the manufacture process <ref name="WAAM"> S. W. Williams, F. Martina, A. C. Addison, J. Ding, G. Pardal & P. Colegrove (2016) Wire + Arc Additive Manufacturing, Materials Science and Technology, 32:7, 641-647, DOI: 10.1179/1743284715Y.0000000073 </ref>. WAAM has been investigated since the 1990s <ref name="WAAM" />, but only recently that it received more attention from the manufacture world.

Its significant comes from the ability to manufacture complex model with less time and less material. Figure 1a and 1b show real parts that was manufactured by AML3D. Such custom made parts might takes months before be ready to be shipped, but with WAAM, the production time can reduce down to weeks. Currently, the industries that benefit the most from WAAM are maritime and aerospace.

Similar to other additive manufacture methods, WAAM achieves such results by sliding the models into multiple layers, and then build the model layer by layer. The movement control is normally handled by a robotic arm (AML3D uses ABB's Arc Welder robot), and the welding path is generated by a Computer Aid Manufacture (CAM) software.

[[File:welding_part1.JPG|thumb|Figure 1a: A propeller]]
[[File:welding_part2.jpg|thumb|Figure 1b: Another part of the ship]]
[[File:abb_robot.jpg|thumb|Figure 2: ABB's IRB 1520ID Arc Welder]]

=== Gas metal arc welding ===
<div><ul>
<li style="display: inline-block;"> [[File:gmaw_welding.png|thumb|500px|none|Figure 2: GMAW welding process <ref name="welding_book"> Norrish, J. (2006). Advanced welding processes. Elsevier. </ref>]] </li>
</ul></div>

Gas metal arc welding (GMAW), or also known as metal inert gas (MIG), is the welding process used at AML3D for WAAM. Due to its high deposition rate and economic benefits, GMAW has became more popular. We will explore two variants of GMAW, which are Pulse Multi Control and Cold Metal Transfer. Both processes are developed by Fronius, an Austrian welding company. Note that it is not required to understand the welding physics to follow this wiki. The next two following sections' purpose is to highlight the high variability, high dynamic nature in welding.

[[File:fronius.png|thumb|Figure 3: Fronius logo]]

==== Pulse Multi Control ====
Pulse Multi Control (PMC) welding is Fronius' modified Pulsed GMAW. The advantage of Pulsed GMAW is the ability to control the metal droplet transfer in welding. Note that from figure 4, the metal droplet is characterised by a downward slope of the current pulse. Figure 5 shows the current signal captured from experiments conducted at AML3D (PMC is used).

<div><ul>
<li style="display: inline-block;"> [[File:Pulsed_gmaw_transfer_mode.png|thumb|500px|none|Figure 4: Relationship between pulse current and droplet transfer<ref name="welding_book" />]] </li>
<li style="display: inline-block;"> [[File:Pmc_current_in_time_domain.png|thumb|650px|none|Figure 5: Current signal captured from experiments conducted at AML3D (PMC process)]] </li>
</ul></div>
==== Cold Metal Transfer ====
Cold Metal Transfer (CMT) is a complex welding process, also developed by Fronius. By detecting a short circuit (mode C and D in figure 8), CMT can make adjustment such as retracting the welding material to cool down, and therefore create a smoother, more stable weld. The complexity of CMT can be seen in the current signal of the process (figure 9). Compare to PMC, CMT's signal has more variation during one signal period.

<div><ul>
<li style="display: inline-block;"> [[File:Transfer_mode.png|thumb|800px|none|Figure 8: Mechanism of metal droplet transfer mode]] </li>
<li style="display: inline-block;"> [[File:Cmt1_current_in_time_domain.png|thumb|735px|none|Figure 9: Current signal captured from experiments conducted at AML3D (CMT process)]] </li>
</ul></div>
=== Machine Learning ===
Machine learning (or statistical learning) refer to a set of tools to give data scientists an insight into the data and make better decisions based on those information <ref name="introduction_ml_book"> James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An introduction to statistical learning (Vol. 112, p. 18). New York: Springer. </ref>. In this project, we will explore the capabilities of classical machine learning methods (i.e anything but Deep Learning) due to the limitation of our available data. By the time of writing this wiki, we have only explored Support Vector Machine.

==== Support Vector Machine ====
The main idea of Support Vector Machine (SVM) is to draw hyperplanes that best separate multiple classes of data. The original SVM judges best decision hyperplane based on the separation margin between data classes (figure 10). However, there are multiple variants of SVM, such as hard/soft margin SVM, Nu-SVM, One class SVM (to solve anomaly detection problem). In our project, we choose to use a soft margin implementation of SVM. The benefit of soft margin SVM (as oppose to hard margin SVM) is that by allowing some misclassifications, we get a wider margin and therefore the solution becomes more "generalise" (i.e work better with a new set of dataset that is independent from the training dataset).

<div><ul>
<li style="display: inline-block;"> [[File:svm_margin.png|thumb|400px|none|Figure 10: Separation margin <ref name="bishop_book"> Bishop, C. M. (2006). Pattern recognition and machine learning. Springer. </ref>]] </li>
</ul></div>

==== Kernel tricks ====
Figure 10 is of linearly separable cases for SVM. For dataset that is not easy to separate with a hyperplane in their original features space (such as figure 11), the data often get projected into a higher dimension space where it is easier to separate (figure 14). Such technique is called the Kernel Tricks. It is important to note that Kernel Tricks does not actually "map" the data from '''X''' dimension to '''Z''' dimension, but rather pairwise compare the similarity of the input space in '''Z''' dimension.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_nonseparable.png |thumb|350px|none|Figure 11: Nonseparable dataset <ref name="bishop_book"></ref>]] </li>
<li style="display: inline-block;"> [[File:Svm_kernel_trick.png |thumb|650px|none|Figure 12: Simplified example of the Kernel tricks]] </li>
</ul></div>

== Previous studies ==
There are many researches into defect detection in GMAW process. The three most dominant analysis methods are spectroscopic analysis, acoustic analysis, and electrical signal analysis.

=== Spectroscopic analysis ===
In <ref name="spectroscopic_analysis">D. Bebiano and S. Alfaro, “A weld defects detection system based on a spectrometer,” Sensors, vol. 9, no. 4, pp. 2851–2861, 2009.</ref>, the authors captured the intensity of radiation emission of electric arc. It is obvious from figure 13 that good weld's intensity stays stable while defect one's is characterised by a sudden peak. Therefore, a threshold is set to detect the defect welds.

<div><ul>
<li style="display: inline-block;"> [[File:spectroscopic_analysis.png|thumb|400px|none|Figure 13: Spectroscopic analysis <ref name="spectroscopic_analysis"></ref>]] </li>
</ul></div>

Other research into using spectroscopic data to monitor that can be of interest is <ref>D. Naso, B. Turchiano, and P. Pantaleo, “A fuzzy-logic based optical sensor for online weld defect-detection,” IEEE transactions on Industrial Informatics, vol. 1, no. 4, pp. 259–273, 2005.</ref>.

=== Acoustic analysis ===
With <ref name="acoustic_analysis">E. Cayo and S. C. Alfaro, “A non-intrusive GMA welding process quality monitoring system using acoustic sensing,” Sensors, vol. 9, no. 9, pp. 7150–7166, 2009.</ref>, the author discovered that arc ignition is characterised by a distinct sound and therefore it is possible to capture the frequency of ignition through acoustic sensors. A tolerance band was set and both ignition frequency and sound pressure level were used to monitor the quality of the weld.

<div><ul>
<li style="display: inline-block;"> [[File:acoustic_analysis.png|thumb|400px|none|Figure 14: Acoustic analysis <ref name="acoustic_analysis"></ref>]] </li>
</ul></div>

Other research into using acoustic signal to monitor that can be of interest are <ref>M. Fidali, “Detection of welding process instabilities using acoustic signals,” in International Congress on Technical Diagnostic, pp. 191–201, Springer, 2016.</ref>, <ref>L. Zhang, A. C. Basantes-Defaz, D. Ozevin, and E. Indacochea, “Real-time monitoring of welding process using air-coupled ultrasonics and acoustic emission,” The International Journal of Advanced Manufacturing Technology, vol. 101, no. 5-8, pp. 1623–1634, 2019.</ref>.

=== Electrical signal analysis ===
With <ref name="electrical_analysis_1">A. Sumesh, K. Rameshkumar, A. Raja, K. Mohandas, A. Santhakumari, and R. Shyambabu, “Establishing correlation between current and voltage signatures of the arc and weld defects in GMAW process,” Arabian Journal for Science and Engineering, vol. 42, no. 11, pp. 4649–4665, 2017.</ref>, the author plot the probability density distribution of voltage signal and discovered that the stable weld's signal concentrate in one region while unstable signal spread out. While in the paper, the authors did not developed a monitoring method using their findings, it shows that there is a significant differences between stable and unstable weld's electrical signature.

<div><ul>
<li style="display: inline-block;"> [[File:Electrical analysis pdd.png|thumb|400px|none|Figure 15: Voltage PDD <ref name="electrical_analysis_1"></ref>]] </li>
</ul></div>

In <ref name="electrical_analysis_2">S. Simpson, “Signature images for arc welding fault detection,” Science and Technology of Welding and Joining, vol. 12, no. 6, pp. 481–486, 2007.</ref>, the authors shows that by plotting the signature image of voltage and current scatter plot, both arcing and short-circuiting region's signature image can notify a defects welds. A more visual representation of short circuiting and arcing region can be found in [[#Gas metal arc welding|Gas metal arc welding section]].

<div><ul>
<li style="display: inline-block;"> [[File:Electrical voltage signature.png|thumb|400px|none|Figure 16a: Electrical voltage signature of stable weld <ref name="electrical_analysis_2"></ref>]] </li>
<li style="display: inline-block;"> [[File:Electrical signature 2.png|thumb|400px|none|Figure 16b: Electrical voltage signature of unstable weld <ref name="electrical_analysis_2"></ref>]] </li>
</ul></div>

Other research into using electrical signal to monitor that can be of interest are <ref>R. Madigan, “Arc sensing for defects in constant-voltage gas metal arc welding,” Welding Journal, vol. 78, pp. 322S–328S, 1999.</ref>, <ref>Y. Huang, K. Wang, Z. Zhou, X. Zhou, and J. Fang, “Stability evaluation of short-circuiting gas metal arc welding based on ensemble empirical mode decomposition,” Measurement Science and Technology, vol. 28, no. 3, p. 035006, 2017.</ref>,<ref>Z. Zhang, X. Chen, H. Chen, J. Zhong, and S. Chen, “Online welding quality monitoring based on feature extraction of arc voltage signal,” The International Journal of Advanced Manufacturing Technology, vol. 70, no. 9-12, pp. 1661–1671, 2014</ref>,<ref>X. Li and S. Simpson, “Parametric approach to positional fault detection in short arc welding,” Science and Technology of welding and joining, vol. 14, no. 2, pp. 146–151, 2009.</ref>, and <ref>E. Wei, D. Farson, R. Richardson, and H. Ludewig, “Detection of weld surface porosity by statistical analysis of arc current in gas metal arc welding,” Journal of Manufacturing Processes, vol. 3, no. 1, pp. 50–59, 2001.</ref>.

== Experiment setup ==
The welding system uses TPS 320i C Pulse (figure 15a), produced by Fronius, to supply power, feed wire, and system cooling. For movement control of the welding torch, the system use ABB’s IRB 1520ID (figure 15b), which is a dedicated arc welding robot. The Data Acquisition device is NI's DAQ. Each channel (voltage and current) is sampled at 25kHz while the dominant frequency for voltage and current signal are around 130Hz (for CMT process) and 190Hz (for PMC process).

[[File:Fronius_power_supply.jpg |thumb|Figure 17a: Fronius' Power Supply]]
[[File:Welding_robot2.jpg |thumb|Figure 17b: ABB's welding robot]]

[[File:Ni_daq.jpg|thumb|Figure 16: NI's DAQ device]]

== Statistical Analysis ==

=== Preprocessing ===
As the captured signal coming from Fronius' power supply, the signals also pick up switching power noise. The noise will interfere with the accuracy of the analysis if not filtered. In this project, we applied a 1000Hz digital Low Pass Butterworth filter. The effect of the filter can be seen in figure 18, while the comparation of using different filters can be seen in the gallery.

<div><ul>
<li style="display: inline-block;"> [[File:Filter_effect.png |thumb|750px|none|Figure 18: Effect of 1000Hz low pass filter]] </li>
<li style="display: inline-block;"> [[File:Freq_response.png |thumb|750px|none|Figure 19: Frequency response of the in-use filter]] </li>
</ul></div>

<gallery mode="slideshow" caption="Gallery: Comparison of different filtering method" widths=75px heights=75px>
File:Effect_of_filtering2.png|widths:75px|heights:75px|In use filter vs 300Hz low pass filter
File:Effect_of_filtering3.png|widths:75px|heights:75px|In use filter vs 500Hz low pass filter
File:Effect_of_filtering4.png|widths:75px|heights:75px|In use filter vs 750Hz low pass filter
File:Effect_of_filtering5.png|widths:75px|heights:75px|In use filter vs 2000Hz low pass filter
</gallery>

=== Power Spectral Density ===
Our goal for this project is to detect the defect welds and predict the layer heights (the contact tip to work piece distance, CTWD, to be more accurate) in real time (refer to [[#Relating Contact Tip to Work-piece Distance to layer height|this section]] to understand the relationship between Contact Tip to Work-piece Distance and Layer height). An analysis of the current power spectral density (PSD) shows that there is an inverse proportional relationship between the CTWD of a weld and the frequency where its '''current signal''' PSD reach the peak, and also the magnitude of the peak. Since the peak of PSD can be fairly sensitive, we only consider the frequency of the peak PSD.

Figure 20 shows the PSD plot of voltage, current, and power signal of the weld using PMC process. In this experiment, we created multiple one-line, 8 seconds welds. These eight-second welds then get segmented into smaller two-second chunk to increase the number of data and decrease the processing time.

<div><ul>
<li style="display: inline-block;"> [[File:Psd pmc good bad all lowfreq.png |thumb|750px|none|Figure 20: Power spectral density of PMC process]] </li>
<li style="display: inline-block;"> [[File:One line weld.jpg |thumb|578px|none|Figure 21: Simple one-line weld experiment]] </li>
<li style="display: inline-block;"> [[File:Cmt_psd_stable_allCTWD_zoomed.png |thumb|750px|none|Figure 22: Simple one-line weld experiment]] </li>

</ul></div>

It seems like the inverse proportional relationship holds true for most cases. Other experiments showed that for a short period at the beginning and ending of the weld, the signal can behave differently relative to the rest of the weld. Currently, to migrate with this problem, data will not be collected and analysed (both offline and online) at these periods. However, the above results seem to only hold true for PMC processes. With CMT process, the current PSD plot indicates no relationship between the CTWD and the frequency at peak (figure 24).

For the next stage of the project, our plan is to develop a linear regression model to predict the CTWD for PMC process. With CMT process, we will have to look into different analysis techniques.

=== Support Vector Machine ===
By looking at the average voltage and average current of each segment of weld while making a simple straight path (PMC process), we realised there is a visible separation margin between stable and unstable welds. Using SVM with RBF kernel, we were able to correctly classify the quality of the weld with accuracy of 98%.

However, with CMT process, the separation cannot be observed.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_pmcNoOxide_rbf.png |thumb|750px|none|Figure 23: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:CmtNoOxide_average_voltage_current.png |thumb|750px|none|Figure 24: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

=== CTWD Regression ===
With CTWD prediction, the result will only be useful if it is done on-the-line while making the part. And therefore, as oppose to defect detection, we cannot have a short calibration process every constant time to check for worn contact tip. By analysing the data while building a real part, we found that the average voltage (for PMC process) and average current (for CMT process) seems to have the strongest linear relationship with the CTWD. The acceptable error for CTWD prediction is +/- 1 mm (according to AML3D requirement). For both PMC and CMT process, our root means square error is less than 1 (PMC: 0.68 and CMT: 0.83), which satisfied the requirement.

<div><ul>
<li style="display: inline-block;"> [[File:Linear_regression_pmc_wall.png |thumb|750px|none|Figure 25: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:Linear_regression_cmt_wall.png |thumb|750px|none|Figure 26: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

== Result discussion ==
These results show that there is a strong possibility of defect detection and CTWD prediction in welding. However, these initial findings are still a proof of concept. In the next stage of our research, we will refine the algorithm and integrate into AML's pipeline.

== References ==
<references />

Projects:2019s2-20001 Using Machine Learning to Determine Deposit Height and Defects for Wire + Arc Additive Manufacture (3D printing)

2019-10-15T05:26:52Z

A1713568: /* Statistical Analysis */

[[Category:Projects]]
[[Category:Final Year Projects]]
[[Category:2019s2|24501]]
== Introduction ==
AML3D is one of the leading company in the Wire Arc Additive Manufacturing (WAAM) industry. In short, WAAM is the Metal 3D printing. The technology has the ability to reduce up to 75% of manufacture time, cost, and material. Currently, WAAM is benefiting other industry such as: mining, defence, marine, and aviation.

<gallery mode="slideshow">
File:Welding baby.gif|center|Welding using robotic arm
File:Laser baby.gif|center|User laser sensor to find the CTWD
</gallery>

[[File:aml3d_logo.jpg|center|100px]]
=== Project team ===
==== Project students ====
* Anh Tran
* Nhat Nguyen
==== Supervisors ====
* Dr. Brian Ng
* Dr. Paul Colegrove (AML3D, sponsored company)
=== Objectives ===
The objective of this project is to increase the efficiency of the manufacture process at AML3D. In order to do so, the team will investigate into the possibility of automating and optimising the quality control processes. The two quality control processes that are currently being implemented at AML3D are measuring layer height using laser sensors, and human supervision for detecting defects. These processes add overhead into production time and usage of human resource, which is not desired. To achieve the goal, machine learning methods will be used extensively to analyse the electrical signatures of the weld process.

== Background ==
=== Wire Arc Additive Manufacturing ===
Wire and Arc Additive Manufacturing (WAAM) is a type of additive manufacturing that uses electric arc as the heat source and material wire to feed the manufacture process <ref name="WAAM"> S. W. Williams, F. Martina, A. C. Addison, J. Ding, G. Pardal & P. Colegrove (2016) Wire + Arc Additive Manufacturing, Materials Science and Technology, 32:7, 641-647, DOI: 10.1179/1743284715Y.0000000073 </ref>. WAAM has been investigated since the 1990s <ref name="WAAM" />, but only recently that it received more attention from the manufacture world.

Its significant comes from the ability to manufacture complex model with less time and less material. Figure 1a and 1b show real parts that was manufactured by AML3D. Such custom made parts might takes months before be ready to be shipped, but with WAAM, the production time can reduce down to weeks. Currently, the industries that benefit the most from WAAM are maritime and aerospace.

Similar to other additive manufacture methods, WAAM achieves such results by sliding the models into multiple layers, and then build the model layer by layer. The movement control is normally handled by a robotic arm (AML3D uses ABB's Arc Welder robot), and the welding path is generated by a Computer Aid Manufacture (CAM) software.

[[File:welding_part1.JPG|thumb|Figure 1a: A propeller]]
[[File:welding_part2.jpg|thumb|Figure 1b: Another part of the ship]]
[[File:abb_robot.jpg|thumb|Figure 2: ABB's IRB 1520ID Arc Welder]]

=== Gas metal arc welding ===
<div><ul>
<li style="display: inline-block;"> [[File:gmaw_welding.png|thumb|500px|none|Figure 2: GMAW welding process <ref name="welding_book"> Norrish, J. (2006). Advanced welding processes. Elsevier. </ref>]] </li>
</ul></div>

Gas metal arc welding (GMAW), or also known as metal inert gas (MIG), is the welding process used at AML3D for WAAM. Due to its high deposition rate and economic benefits, GMAW has became more popular. We will explore two variants of GMAW, which are Pulse Multi Control and Cold Metal Transfer. Both processes are developed by Fronius, an Austrian welding company. Note that it is not required to understand the welding physics to follow this wiki. The next two following sections' purpose is to highlight the high variability, high dynamic nature in welding.

[[File:fronius.png|thumb|Figure 3: Fronius logo]]

==== Pulse Multi Control ====
Pulse Multi Control (PMC) welding is Fronius' modified Pulsed GMAW. The advantage of Pulsed GMAW is the ability to control the metal droplet transfer in welding. Note that from figure 4, the metal droplet is characterised by a downward slope of the current pulse. Figure 5 shows the current signal captured from experiments conducted at AML3D (PMC is used).

<div><ul>
<li style="display: inline-block;"> [[File:Pulsed_gmaw_transfer_mode.png|thumb|500px|none|Figure 4: Relationship between pulse current and droplet transfer<ref name="welding_book" />]] </li>
<li style="display: inline-block;"> [[File:Pmc_current_in_time_domain.png|thumb|650px|none|Figure 5: Current signal captured from experiments conducted at AML3D (PMC process)]] </li>
</ul></div>
==== Cold Metal Transfer ====
Cold Metal Transfer (CMT) is a complex welding process, also developed by Fronius. By detecting a short circuit (mode C and D in figure 8), CMT can make adjustment such as retracting the welding material to cool down, and therefore create a smoother, more stable weld. The complexity of CMT can be seen in the current signal of the process (figure 9). Compare to PMC, CMT's signal has more variation during one signal period.

<div><ul>
<li style="display: inline-block;"> [[File:Transfer_mode.png|thumb|800px|none|Figure 8: Mechanism of metal droplet transfer mode]] </li>
<li style="display: inline-block;"> [[File:Cmt1_current_in_time_domain.png|thumb|735px|none|Figure 9: Current signal captured from experiments conducted at AML3D (CMT process)]] </li>
</ul></div>
=== Machine Learning ===
Machine learning (or statistical learning) refer to a set of tools to give data scientists an insight into the data and make better decisions based on those information <ref name="introduction_ml_book"> James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An introduction to statistical learning (Vol. 112, p. 18). New York: Springer. </ref>. In this project, we will explore the capabilities of classical machine learning methods (i.e anything but Deep Learning) due to the limitation of our available data. By the time of writing this wiki, we have only explored Support Vector Machine.

==== Support Vector Machine ====
The main idea of Support Vector Machine (SVM) is to draw hyperplanes that best separate multiple classes of data. The original SVM judges best decision hyperplane based on the separation margin between data classes (figure 10). However, there are multiple variants of SVM, such as hard/soft margin SVM, Nu-SVM, One class SVM (to solve anomaly detection problem). In our project, we choose to use a soft margin implementation of SVM. The benefit of soft margin SVM (as oppose to hard margin SVM) is that by allowing some misclassifications, we get a wider margin and therefore the solution becomes more "generalise" (i.e work better with a new set of dataset that is independent from the training dataset).

<div><ul>
<li style="display: inline-block;"> [[File:svm_margin.png|thumb|400px|none|Figure 10: Separation margin <ref name="bishop_book"> Bishop, C. M. (2006). Pattern recognition and machine learning. Springer. </ref>]] </li>
</ul></div>

==== Kernel tricks ====
Figure 10 is of linearly separable cases for SVM. For dataset that is not easy to separate with a hyperplane in their original features space (such as figure 11), the data often get projected into a higher dimension space where it is easier to separate (figure 14). Such technique is called the Kernel Tricks. It is important to note that Kernel Tricks does not actually "map" the data from '''X''' dimension to '''Z''' dimension, but rather pairwise compare the similarity of the input space in '''Z''' dimension.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_nonseparable.png |thumb|350px|none|Figure 11: Nonseparable dataset <ref name="bishop_book"></ref>]] </li>
<li style="display: inline-block;"> [[File:Svm_kernel_trick.png |thumb|650px|none|Figure 12: Simplified example of the Kernel tricks]] </li>
</ul></div>

== Previous studies ==
There are many researches into defect detection in GMAW process. The three most dominant analysis methods are spectroscopic analysis, acoustic analysis, and electrical signal analysis.

=== Spectroscopic analysis ===
In <ref name="spectroscopic_analysis">D. Bebiano and S. Alfaro, “A weld defects detection system based on a spectrometer,” Sensors, vol. 9, no. 4, pp. 2851–2861, 2009.</ref>, the authors captured the intensity of radiation emission of electric arc. It is obvious from figure 13 that good weld's intensity stays stable while defect one's is characterised by a sudden peak. Therefore, a threshold is set to detect the defect welds.

<div><ul>
<li style="display: inline-block;"> [[File:spectroscopic_analysis.png|thumb|400px|none|Figure 13: Spectroscopic analysis <ref name="spectroscopic_analysis"></ref>]] </li>
</ul></div>

Other research into using spectroscopic data to monitor that can be of interest is <ref>D. Naso, B. Turchiano, and P. Pantaleo, “A fuzzy-logic based optical sensor for online weld defect-detection,” IEEE transactions on Industrial Informatics, vol. 1, no. 4, pp. 259–273, 2005.</ref>.

=== Acoustic analysis ===
With <ref name="acoustic_analysis">E. Cayo and S. C. Alfaro, “A non-intrusive GMA welding process quality monitoring system using acoustic sensing,” Sensors, vol. 9, no. 9, pp. 7150–7166, 2009.</ref>, the author discovered that arc ignition is characterised by a distinct sound and therefore it is possible to capture the frequency of ignition through acoustic sensors. A tolerance band was set and both ignition frequency and sound pressure level were used to monitor the quality of the weld.

<div><ul>
<li style="display: inline-block;"> [[File:acoustic_analysis.png|thumb|400px|none|Figure 14: Acoustic analysis <ref name="acoustic_analysis"></ref>]] </li>
</ul></div>

Other research into using acoustic signal to monitor that can be of interest are <ref>M. Fidali, “Detection of welding process instabilities using acoustic signals,” in International Congress on Technical Diagnostic, pp. 191–201, Springer, 2016.</ref>, <ref>L. Zhang, A. C. Basantes-Defaz, D. Ozevin, and E. Indacochea, “Real-time monitoring of welding process using air-coupled ultrasonics and acoustic emission,” The International Journal of Advanced Manufacturing Technology, vol. 101, no. 5-8, pp. 1623–1634, 2019.</ref>.

=== Electrical signal analysis ===
With <ref name="electrical_analysis_1">A. Sumesh, K. Rameshkumar, A. Raja, K. Mohandas, A. Santhakumari, and R. Shyambabu, “Establishing correlation between current and voltage signatures of the arc and weld defects in GMAW process,” Arabian Journal for Science and Engineering, vol. 42, no. 11, pp. 4649–4665, 2017.</ref>, the author plot the probability density distribution of voltage signal and discovered that the stable weld's signal concentrate in one region while unstable signal spread out. While in the paper, the authors did not developed a monitoring method using their findings, it shows that there is a significant differences between stable and unstable weld's electrical signature.

<div><ul>
<li style="display: inline-block;"> [[File:Electrical analysis pdd.png|thumb|400px|none|Figure 15: Voltage PDD <ref name="electrical_analysis_1"></ref>]] </li>
</ul></div>

In <ref name="electrical_analysis_2">S. Simpson, “Signature images for arc welding fault detection,” Science and Technology of Welding and Joining, vol. 12, no. 6, pp. 481–486, 2007.</ref>, the authors shows that by plotting the signature image of voltage and current scatter plot, both arcing and short-circuiting region's signature image can notify a defects welds. A more visual representation of short circuiting and arcing region can be found in [[#Gas metal arc welding|Gas metal arc welding section]].

<div><ul>
<li style="display: inline-block;"> [[File:Electrical voltage signature.png|thumb|400px|none|Figure 16a: Electrical voltage signature of stable weld <ref name="electrical_analysis_2"></ref>]] </li>
<li style="display: inline-block;"> [[File:Electrical signature 2.png|thumb|400px|none|Figure 16b: Electrical voltage signature of unstable weld <ref name="electrical_analysis_2"></ref>]] </li>
</ul></div>

Other research into using electrical signal to monitor that can be of interest are <ref>R. Madigan, “Arc sensing for defects in constant-voltage gas metal arc welding,” Welding Journal, vol. 78, pp. 322S–328S, 1999.</ref>, <ref>Y. Huang, K. Wang, Z. Zhou, X. Zhou, and J. Fang, “Stability evaluation of short-circuiting gas metal arc welding based on ensemble empirical mode decomposition,” Measurement Science and Technology, vol. 28, no. 3, p. 035006, 2017.</ref>,<ref>Z. Zhang, X. Chen, H. Chen, J. Zhong, and S. Chen, “Online welding quality monitoring based on feature extraction of arc voltage signal,” The International Journal of Advanced Manufacturing Technology, vol. 70, no. 9-12, pp. 1661–1671, 2014</ref>,<ref>X. Li and S. Simpson, “Parametric approach to positional fault detection in short arc welding,” Science and Technology of welding and joining, vol. 14, no. 2, pp. 146–151, 2009.</ref>, and <ref>E. Wei, D. Farson, R. Richardson, and H. Ludewig, “Detection of weld surface porosity by statistical analysis of arc current in gas metal arc welding,” Journal of Manufacturing Processes, vol. 3, no. 1, pp. 50–59, 2001.</ref>.

== Experiment setup ==
The welding system uses TPS 320i C Pulse (figure 15a), produced by Fronius, to supply power, feed wire, and system cooling. For movement control of the welding torch, the system use ABB’s IRB 1520ID (figure 15b), which is a dedicated arc welding robot. The Data Acquisition device is NI's DAQ. Each channel (voltage and current) is sampled at 25kHz while the dominant frequency for voltage and current signal are around 130Hz (for CMT process) and 190Hz (for PMC process).

[[File:Fronius_power_supply.jpg |thumb|Figure 17a: Fronius' Power Supply]]
[[File:Welding_robot2.jpg |thumb|Figure 17b: ABB's welding robot]]

[[File:Ni_daq.jpg|thumb|Figure 16: NI's DAQ device]]

== Statistical Analysis ==

=== Preprocessing ===
As the captured signal coming from Fronius' power supply, the signals also pick up switching power noise. The noise will interfere with the accuracy of the analysis if not filtered. In this project, we applied a 1000Hz digital Low Pass Butterworth filter. The effect of the filter can be seen in figure 18, while the comparation of using different filters can be seen in the gallery.

<div><ul>
<li style="display: inline-block;"> [[File:Filter_effect.png |thumb|750px|none|Figure 18: Effect of 1000Hz low pass filter]] </li>
<li style="display: inline-block;"> [[File:Freq_response.png |thumb|750px|none|Figure 19: Frequency response of the in-use filter]] </li>
</ul></div>

<gallery mode="slideshow" caption="Gallery: Comparison of different filtering method" widths:75px heights:75px>
File:Effect_of_filtering2.png|widths:75px|heights:75px|In use filter vs 300Hz low pass filter
File:Effect_of_filtering3.png|widths:75px|heights:75px|In use filter vs 500Hz low pass filter
File:Effect_of_filtering4.png|widths:75px|heights:75px|In use filter vs 750Hz low pass filter
File:Effect_of_filtering5.png|widths:75px|heights:75px|In use filter vs 2000Hz low pass filter
</gallery>

=== Power Spectral Density ===
Our goal for this project is to detect the defect welds and predict the layer heights (the contact tip to work piece distance, CTWD, to be more accurate) in real time (refer to [[#Relating Contact Tip to Work-piece Distance to layer height|this section]] to understand the relationship between Contact Tip to Work-piece Distance and Layer height). An analysis of the current power spectral density (PSD) shows that there is an inverse proportional relationship between the CTWD of a weld and the frequency where its '''current signal''' PSD reach the peak, and also the magnitude of the peak. Since the peak of PSD can be fairly sensitive, we only consider the frequency of the peak PSD.

Figure 20 shows the PSD plot of voltage, current, and power signal of the weld using PMC process. In this experiment, we created multiple one-line, 8 seconds welds. These eight-second welds then get segmented into smaller two-second chunk to increase the number of data and decrease the processing time.

<div><ul>
<li style="display: inline-block;"> [[File:Psd pmc good bad all lowfreq.png |thumb|750px|none|Figure 20: Power spectral density of PMC process]] </li>
<li style="display: inline-block;"> [[File:One line weld.jpg |thumb|578px|none|Figure 21: Simple one-line weld experiment]] </li>
<li style="display: inline-block;"> [[File:Cmt_psd_stable_allCTWD_zoomed.png |thumb|750px|none|Figure 22: Simple one-line weld experiment]] </li>

</ul></div>

It seems like the inverse proportional relationship holds true for most cases. Other experiments showed that for a short period at the beginning and ending of the weld, the signal can behave differently relative to the rest of the weld. Currently, to migrate with this problem, data will not be collected and analysed (both offline and online) at these periods. However, the above results seem to only hold true for PMC processes. With CMT process, the current PSD plot indicates no relationship between the CTWD and the frequency at peak (figure 24).

For the next stage of the project, our plan is to develop a linear regression model to predict the CTWD for PMC process. With CMT process, we will have to look into different analysis techniques.

=== Support Vector Machine ===
By looking at the average voltage and average current of each segment of weld while making a simple straight path (PMC process), we realised there is a visible separation margin between stable and unstable welds. Using SVM with RBF kernel, we were able to correctly classify the quality of the weld with accuracy of 98%.

However, with CMT process, the separation cannot be observed.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_pmcNoOxide_rbf.png |thumb|750px|none|Figure 23: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:CmtNoOxide_average_voltage_current.png |thumb|750px|none|Figure 24: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

=== CTWD Regression ===
With CTWD prediction, the result will only be useful if it is done on-the-line while making the part. And therefore, as oppose to defect detection, we cannot have a short calibration process every constant time to check for worn contact tip. By analysing the data while building a real part, we found that the average voltage (for PMC process) and average current (for CMT process) seems to have the strongest linear relationship with the CTWD. The acceptable error for CTWD prediction is +/- 1 mm (according to AML3D requirement). For both PMC and CMT process, our root means square error is less than 1 (PMC: 0.68 and CMT: 0.83), which satisfied the requirement.

<div><ul>
<li style="display: inline-block;"> [[File:Linear_regression_pmc_wall.png |thumb|750px|none|Figure 25: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:Linear_regression_cmt_wall.png |thumb|750px|none|Figure 26: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

== Result discussion ==
These results show that there is a strong possibility of defect detection and CTWD prediction in welding. However, these initial findings are still a proof of concept. In the next stage of our research, we will refine the algorithm and integrate into AML's pipeline.

== References ==
<references />

Projects:2019s2-20001 Using Machine Learning to Determine Deposit Height and Defects for Wire + Arc Additive Manufacture (3D printing)

2019-10-15T05:26:20Z

A1713568: /* Preprocessing */

[[Category:Projects]]
[[Category:Final Year Projects]]
[[Category:2019s2|24501]]
== Introduction ==
AML3D is one of the leading company in the Wire Arc Additive Manufacturing (WAAM) industry. In short, WAAM is the Metal 3D printing. The technology has the ability to reduce up to 75% of manufacture time, cost, and material. Currently, WAAM is benefiting other industry such as: mining, defence, marine, and aviation.

<gallery mode="slideshow">
File:Welding baby.gif|center|Welding using robotic arm
File:Laser baby.gif|center|User laser sensor to find the CTWD
</gallery>

[[File:aml3d_logo.jpg|center|100px]]
=== Project team ===
==== Project students ====
* Anh Tran
* Nhat Nguyen
==== Supervisors ====
* Dr. Brian Ng
* Dr. Paul Colegrove (AML3D, sponsored company)
=== Objectives ===
The objective of this project is to increase the efficiency of the manufacture process at AML3D. In order to do so, the team will investigate into the possibility of automating and optimising the quality control processes. The two quality control processes that are currently being implemented at AML3D are measuring layer height using laser sensors, and human supervision for detecting defects. These processes add overhead into production time and usage of human resource, which is not desired. To achieve the goal, machine learning methods will be used extensively to analyse the electrical signatures of the weld process.

== Background ==
=== Wire Arc Additive Manufacturing ===
Wire and Arc Additive Manufacturing (WAAM) is a type of additive manufacturing that uses electric arc as the heat source and material wire to feed the manufacture process <ref name="WAAM"> S. W. Williams, F. Martina, A. C. Addison, J. Ding, G. Pardal & P. Colegrove (2016) Wire + Arc Additive Manufacturing, Materials Science and Technology, 32:7, 641-647, DOI: 10.1179/1743284715Y.0000000073 </ref>. WAAM has been investigated since the 1990s <ref name="WAAM" />, but only recently that it received more attention from the manufacture world.

Its significant comes from the ability to manufacture complex model with less time and less material. Figure 1a and 1b show real parts that was manufactured by AML3D. Such custom made parts might takes months before be ready to be shipped, but with WAAM, the production time can reduce down to weeks. Currently, the industries that benefit the most from WAAM are maritime and aerospace.

Similar to other additive manufacture methods, WAAM achieves such results by sliding the models into multiple layers, and then build the model layer by layer. The movement control is normally handled by a robotic arm (AML3D uses ABB's Arc Welder robot), and the welding path is generated by a Computer Aid Manufacture (CAM) software.

[[File:welding_part1.JPG|thumb|Figure 1a: A propeller]]
[[File:welding_part2.jpg|thumb|Figure 1b: Another part of the ship]]
[[File:abb_robot.jpg|thumb|Figure 2: ABB's IRB 1520ID Arc Welder]]

=== Gas metal arc welding ===
<div><ul>
<li style="display: inline-block;"> [[File:gmaw_welding.png|thumb|500px|none|Figure 2: GMAW welding process <ref name="welding_book"> Norrish, J. (2006). Advanced welding processes. Elsevier. </ref>]] </li>
</ul></div>

Gas metal arc welding (GMAW), or also known as metal inert gas (MIG), is the welding process used at AML3D for WAAM. Due to its high deposition rate and economic benefits, GMAW has became more popular. We will explore two variants of GMAW, which are Pulse Multi Control and Cold Metal Transfer. Both processes are developed by Fronius, an Austrian welding company. Note that it is not required to understand the welding physics to follow this wiki. The next two following sections' purpose is to highlight the high variability, high dynamic nature in welding.

[[File:fronius.png|thumb|Figure 3: Fronius logo]]

==== Pulse Multi Control ====
Pulse Multi Control (PMC) welding is Fronius' modified Pulsed GMAW. The advantage of Pulsed GMAW is the ability to control the metal droplet transfer in welding. Note that from figure 4, the metal droplet is characterised by a downward slope of the current pulse. Figure 5 shows the current signal captured from experiments conducted at AML3D (PMC is used).

<div><ul>
<li style="display: inline-block;"> [[File:Pulsed_gmaw_transfer_mode.png|thumb|500px|none|Figure 4: Relationship between pulse current and droplet transfer<ref name="welding_book" />]] </li>
<li style="display: inline-block;"> [[File:Pmc_current_in_time_domain.png|thumb|650px|none|Figure 5: Current signal captured from experiments conducted at AML3D (PMC process)]] </li>
</ul></div>
==== Cold Metal Transfer ====
Cold Metal Transfer (CMT) is a complex welding process, also developed by Fronius. By detecting a short circuit (mode C and D in figure 8), CMT can make adjustment such as retracting the welding material to cool down, and therefore create a smoother, more stable weld. The complexity of CMT can be seen in the current signal of the process (figure 9). Compare to PMC, CMT's signal has more variation during one signal period.

<div><ul>
<li style="display: inline-block;"> [[File:Transfer_mode.png|thumb|800px|none|Figure 8: Mechanism of metal droplet transfer mode]] </li>
<li style="display: inline-block;"> [[File:Cmt1_current_in_time_domain.png|thumb|735px|none|Figure 9: Current signal captured from experiments conducted at AML3D (CMT process)]] </li>
</ul></div>
=== Machine Learning ===
Machine learning (or statistical learning) refer to a set of tools to give data scientists an insight into the data and make better decisions based on those information <ref name="introduction_ml_book"> James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An introduction to statistical learning (Vol. 112, p. 18). New York: Springer. </ref>. In this project, we will explore the capabilities of classical machine learning methods (i.e anything but Deep Learning) due to the limitation of our available data. By the time of writing this wiki, we have only explored Support Vector Machine.

==== Support Vector Machine ====
The main idea of Support Vector Machine (SVM) is to draw hyperplanes that best separate multiple classes of data. The original SVM judges best decision hyperplane based on the separation margin between data classes (figure 10). However, there are multiple variants of SVM, such as hard/soft margin SVM, Nu-SVM, One class SVM (to solve anomaly detection problem). In our project, we choose to use a soft margin implementation of SVM. The benefit of soft margin SVM (as oppose to hard margin SVM) is that by allowing some misclassifications, we get a wider margin and therefore the solution becomes more "generalise" (i.e work better with a new set of dataset that is independent from the training dataset).

<div><ul>
<li style="display: inline-block;"> [[File:svm_margin.png|thumb|400px|none|Figure 10: Separation margin <ref name="bishop_book"> Bishop, C. M. (2006). Pattern recognition and machine learning. Springer. </ref>]] </li>
</ul></div>

==== Kernel tricks ====
Figure 10 is of linearly separable cases for SVM. For dataset that is not easy to separate with a hyperplane in their original features space (such as figure 11), the data often get projected into a higher dimension space where it is easier to separate (figure 14). Such technique is called the Kernel Tricks. It is important to note that Kernel Tricks does not actually "map" the data from '''X''' dimension to '''Z''' dimension, but rather pairwise compare the similarity of the input space in '''Z''' dimension.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_nonseparable.png |thumb|350px|none|Figure 11: Nonseparable dataset <ref name="bishop_book"></ref>]] </li>
<li style="display: inline-block;"> [[File:Svm_kernel_trick.png |thumb|650px|none|Figure 12: Simplified example of the Kernel tricks]] </li>
</ul></div>

== Previous studies ==
There are many researches into defect detection in GMAW process. The three most dominant analysis methods are spectroscopic analysis, acoustic analysis, and electrical signal analysis.

=== Spectroscopic analysis ===
In <ref name="spectroscopic_analysis">D. Bebiano and S. Alfaro, “A weld defects detection system based on a spectrometer,” Sensors, vol. 9, no. 4, pp. 2851–2861, 2009.</ref>, the authors captured the intensity of radiation emission of electric arc. It is obvious from figure 13 that good weld's intensity stays stable while defect one's is characterised by a sudden peak. Therefore, a threshold is set to detect the defect welds.

<div><ul>
<li style="display: inline-block;"> [[File:spectroscopic_analysis.png|thumb|400px|none|Figure 13: Spectroscopic analysis <ref name="spectroscopic_analysis"></ref>]] </li>
</ul></div>

Other research into using spectroscopic data to monitor that can be of interest is <ref>D. Naso, B. Turchiano, and P. Pantaleo, “A fuzzy-logic based optical sensor for online weld defect-detection,” IEEE transactions on Industrial Informatics, vol. 1, no. 4, pp. 259–273, 2005.</ref>.

=== Acoustic analysis ===
With <ref name="acoustic_analysis">E. Cayo and S. C. Alfaro, “A non-intrusive GMA welding process quality monitoring system using acoustic sensing,” Sensors, vol. 9, no. 9, pp. 7150–7166, 2009.</ref>, the author discovered that arc ignition is characterised by a distinct sound and therefore it is possible to capture the frequency of ignition through acoustic sensors. A tolerance band was set and both ignition frequency and sound pressure level were used to monitor the quality of the weld.

<div><ul>
<li style="display: inline-block;"> [[File:acoustic_analysis.png|thumb|400px|none|Figure 14: Acoustic analysis <ref name="acoustic_analysis"></ref>]] </li>
</ul></div>

Other research into using acoustic signal to monitor that can be of interest are <ref>M. Fidali, “Detection of welding process instabilities using acoustic signals,” in International Congress on Technical Diagnostic, pp. 191–201, Springer, 2016.</ref>, <ref>L. Zhang, A. C. Basantes-Defaz, D. Ozevin, and E. Indacochea, “Real-time monitoring of welding process using air-coupled ultrasonics and acoustic emission,” The International Journal of Advanced Manufacturing Technology, vol. 101, no. 5-8, pp. 1623–1634, 2019.</ref>.

=== Electrical signal analysis ===
With <ref name="electrical_analysis_1">A. Sumesh, K. Rameshkumar, A. Raja, K. Mohandas, A. Santhakumari, and R. Shyambabu, “Establishing correlation between current and voltage signatures of the arc and weld defects in GMAW process,” Arabian Journal for Science and Engineering, vol. 42, no. 11, pp. 4649–4665, 2017.</ref>, the author plot the probability density distribution of voltage signal and discovered that the stable weld's signal concentrate in one region while unstable signal spread out. While in the paper, the authors did not developed a monitoring method using their findings, it shows that there is a significant differences between stable and unstable weld's electrical signature.

<div><ul>
<li style="display: inline-block;"> [[File:Electrical analysis pdd.png|thumb|400px|none|Figure 15: Voltage PDD <ref name="electrical_analysis_1"></ref>]] </li>
</ul></div>

In <ref name="electrical_analysis_2">S. Simpson, “Signature images for arc welding fault detection,” Science and Technology of Welding and Joining, vol. 12, no. 6, pp. 481–486, 2007.</ref>, the authors shows that by plotting the signature image of voltage and current scatter plot, both arcing and short-circuiting region's signature image can notify a defects welds. A more visual representation of short circuiting and arcing region can be found in [[#Gas metal arc welding|Gas metal arc welding section]].

<div><ul>
<li style="display: inline-block;"> [[File:Electrical voltage signature.png|thumb|400px|none|Figure 16a: Electrical voltage signature of stable weld <ref name="electrical_analysis_2"></ref>]] </li>
<li style="display: inline-block;"> [[File:Electrical signature 2.png|thumb|400px|none|Figure 16b: Electrical voltage signature of unstable weld <ref name="electrical_analysis_2"></ref>]] </li>
</ul></div>

Other research into using electrical signal to monitor that can be of interest are <ref>R. Madigan, “Arc sensing for defects in constant-voltage gas metal arc welding,” Welding Journal, vol. 78, pp. 322S–328S, 1999.</ref>, <ref>Y. Huang, K. Wang, Z. Zhou, X. Zhou, and J. Fang, “Stability evaluation of short-circuiting gas metal arc welding based on ensemble empirical mode decomposition,” Measurement Science and Technology, vol. 28, no. 3, p. 035006, 2017.</ref>,<ref>Z. Zhang, X. Chen, H. Chen, J. Zhong, and S. Chen, “Online welding quality monitoring based on feature extraction of arc voltage signal,” The International Journal of Advanced Manufacturing Technology, vol. 70, no. 9-12, pp. 1661–1671, 2014</ref>,<ref>X. Li and S. Simpson, “Parametric approach to positional fault detection in short arc welding,” Science and Technology of welding and joining, vol. 14, no. 2, pp. 146–151, 2009.</ref>, and <ref>E. Wei, D. Farson, R. Richardson, and H. Ludewig, “Detection of weld surface porosity by statistical analysis of arc current in gas metal arc welding,” Journal of Manufacturing Processes, vol. 3, no. 1, pp. 50–59, 2001.</ref>.

== Experiment setup ==
The welding system uses TPS 320i C Pulse (figure 15a), produced by Fronius, to supply power, feed wire, and system cooling. For movement control of the welding torch, the system use ABB’s IRB 1520ID (figure 15b), which is a dedicated arc welding robot. The Data Acquisition device is NI's DAQ. Each channel (voltage and current) is sampled at 25kHz while the dominant frequency for voltage and current signal are around 130Hz (for CMT process) and 190Hz (for PMC process).

[[File:Fronius_power_supply.jpg |thumb|Figure 17a: Fronius' Power Supply]]
[[File:Welding_robot2.jpg |thumb|Figure 17b: ABB's welding robot]]

[[File:Ni_daq.jpg|thumb|Figure 16: NI's DAQ device]]

== Statistical Analysis ==

=== Preprocessing ===
As the captured signal coming from Fronius' power supply, the signals also pick up switching power noise. The noise will interfere with the accuracy of the analysis if not filtered. In this project, we applied a 1000Hz digital Low Pass Butterworth filter. The effect of the filter can be seen in figure 18, while the comparation of using different filters can be seen in the gallery.

<div><ul>
<li style="display: inline-block;"> [[File:Filter_effect.png |thumb|750px|none|Figure 18: Effect of 1000Hz low pass filter]] </li>
<li style="display: inline-block;"> [[File:Freq_response.png |thumb|750px|none|Figure 19: Frequency response of the in-use filter]] </li>
</ul></div>

<gallery mode="slideshow" caption="Gallery: Comparison of different filtering method">
File:Effect_of_filtering2.png|widths:75px|heights:75px|In use filter vs 300Hz low pass filter
File:Effect_of_filtering3.png|widths:75px|heights:75px|In use filter vs 500Hz low pass filter
File:Effect_of_filtering4.png|widths:75px|heights:75px|In use filter vs 750Hz low pass filter
File:Effect_of_filtering5.png|widths:75px|heights:75px|In use filter vs 2000Hz low pass filter
</gallery>

=== Power Spectral Density ===
Our goal for this project is to detect the defect welds and predict the layer heights (the contact tip to work piece distance, CTWD, to be more accurate) in real time (refer to [[#Relating Contact Tip to Work-piece Distance to layer height|this section]] to understand the relationship between Contact Tip to Work-piece Distance and Layer height). An analysis of the current power spectral density (PSD) shows that there is an inverse proportional relationship between the CTWD of a weld and the frequency where its '''current signal''' PSD reach the peak, and also the magnitude of the peak. Since the peak of PSD can be fairly sensitive, we only consider the frequency of the peak PSD.

Figure 20 shows the PSD plot of voltage, current, and power signal of the weld using PMC process. In this experiment, we created multiple one-line, 8 seconds welds. These eight-second welds then get segmented into smaller two-second chunk to increase the number of data and decrease the processing time.

<div><ul>
<li style="display: inline-block;"> [[File:Psd pmc good bad all lowfreq.png |thumb|750px|none|Figure 20: Power spectral density of PMC process]] </li>
<li style="display: inline-block;"> [[File:One line weld.jpg |thumb|578px|none|Figure 21: Simple one-line weld experiment]] </li>
<li style="display: inline-block;"> [[File:Cmt_psd_stable_allCTWD_zoomed.png |thumb|750px|none|Figure 22: Simple one-line weld experiment]] </li>

</ul></div>

It seems like the inverse proportional relationship holds true for most cases. Other experiments showed that for a short period at the beginning and ending of the weld, the signal can behave differently relative to the rest of the weld. Currently, to migrate with this problem, data will not be collected and analysed (both offline and online) at these periods. However, the above results seem to only hold true for PMC processes. With CMT process, the current PSD plot indicates no relationship between the CTWD and the frequency at peak (figure 24).

For the next stage of the project, our plan is to develop a linear regression model to predict the CTWD for PMC process. With CMT process, we will have to look into different analysis techniques.

=== Support Vector Machine ===
By looking at the average voltage and average current of each segment of weld while making a simple straight path (PMC process), we realised there is a visible separation margin between stable and unstable welds. Using SVM with RBF kernel, we were able to correctly classify the quality of the weld with accuracy of 98%.

However, with CMT process, the separation cannot be observed.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_pmcNoOxide_rbf.png |thumb|750px|none|Figure 23: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:CmtNoOxide_average_voltage_current.png |thumb|750px|none|Figure 24: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

=== CTWD Regression ===
With CTWD prediction, the result will only be useful if it is done on-the-line while making the part. And therefore, as oppose to defect detection, we cannot have a short calibration process every constant time to check for worn contact tip. By analysing the data while building a real part, we found that the average voltage (for PMC process) and average current (for CMT process) seems to have the strongest linear relationship with the CTWD. The acceptable error for CTWD prediction is +/- 1 mm (according to AML3D requirement). For both PMC and CMT process, our root means square error is less than 1 (PMC: 0.68 and CMT: 0.83), which satisfied the requirement.

<div><ul>
<li style="display: inline-block;"> [[File:Linear_regression_pmc_wall.png |thumb|750px|none|Figure 25: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:Linear_regression_cmt_wall.png |thumb|750px|none|Figure 26: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

== Result discussion ==
These results show that there is a strong possibility of defect detection and CTWD prediction in welding. However, these initial findings are still a proof of concept. In the next stage of our research, we will refine the algorithm and integrate into AML's pipeline.

== References ==
<references />

Projects:2019s2-20001 Using Machine Learning to Determine Deposit Height and Defects for Wire + Arc Additive Manufacture (3D printing)

2019-10-15T05:24:39Z

A1713568: /* Preprocessing */

[[Category:Projects]]
[[Category:Final Year Projects]]
[[Category:2019s2|24501]]
== Introduction ==
AML3D is one of the leading company in the Wire Arc Additive Manufacturing (WAAM) industry. In short, WAAM is the Metal 3D printing. The technology has the ability to reduce up to 75% of manufacture time, cost, and material. Currently, WAAM is benefiting other industry such as: mining, defence, marine, and aviation.

<gallery mode="slideshow">
File:Welding baby.gif|center|Welding using robotic arm
File:Laser baby.gif|center|User laser sensor to find the CTWD
</gallery>

[[File:aml3d_logo.jpg|center|100px]]
=== Project team ===
==== Project students ====
* Anh Tran
* Nhat Nguyen
==== Supervisors ====
* Dr. Brian Ng
* Dr. Paul Colegrove (AML3D, sponsored company)
=== Objectives ===
The objective of this project is to increase the efficiency of the manufacture process at AML3D. In order to do so, the team will investigate into the possibility of automating and optimising the quality control processes. The two quality control processes that are currently being implemented at AML3D are measuring layer height using laser sensors, and human supervision for detecting defects. These processes add overhead into production time and usage of human resource, which is not desired. To achieve the goal, machine learning methods will be used extensively to analyse the electrical signatures of the weld process.

== Background ==
=== Wire Arc Additive Manufacturing ===
Wire and Arc Additive Manufacturing (WAAM) is a type of additive manufacturing that uses electric arc as the heat source and material wire to feed the manufacture process <ref name="WAAM"> S. W. Williams, F. Martina, A. C. Addison, J. Ding, G. Pardal & P. Colegrove (2016) Wire + Arc Additive Manufacturing, Materials Science and Technology, 32:7, 641-647, DOI: 10.1179/1743284715Y.0000000073 </ref>. WAAM has been investigated since the 1990s <ref name="WAAM" />, but only recently that it received more attention from the manufacture world.

Its significant comes from the ability to manufacture complex model with less time and less material. Figure 1a and 1b show real parts that was manufactured by AML3D. Such custom made parts might takes months before be ready to be shipped, but with WAAM, the production time can reduce down to weeks. Currently, the industries that benefit the most from WAAM are maritime and aerospace.

Similar to other additive manufacture methods, WAAM achieves such results by sliding the models into multiple layers, and then build the model layer by layer. The movement control is normally handled by a robotic arm (AML3D uses ABB's Arc Welder robot), and the welding path is generated by a Computer Aid Manufacture (CAM) software.

[[File:welding_part1.JPG|thumb|Figure 1a: A propeller]]
[[File:welding_part2.jpg|thumb|Figure 1b: Another part of the ship]]
[[File:abb_robot.jpg|thumb|Figure 2: ABB's IRB 1520ID Arc Welder]]

=== Gas metal arc welding ===
<div><ul>
<li style="display: inline-block;"> [[File:gmaw_welding.png|thumb|500px|none|Figure 2: GMAW welding process <ref name="welding_book"> Norrish, J. (2006). Advanced welding processes. Elsevier. </ref>]] </li>
</ul></div>

Gas metal arc welding (GMAW), or also known as metal inert gas (MIG), is the welding process used at AML3D for WAAM. Due to its high deposition rate and economic benefits, GMAW has became more popular. We will explore two variants of GMAW, which are Pulse Multi Control and Cold Metal Transfer. Both processes are developed by Fronius, an Austrian welding company. Note that it is not required to understand the welding physics to follow this wiki. The next two following sections' purpose is to highlight the high variability, high dynamic nature in welding.

[[File:fronius.png|thumb|Figure 3: Fronius logo]]

==== Pulse Multi Control ====
Pulse Multi Control (PMC) welding is Fronius' modified Pulsed GMAW. The advantage of Pulsed GMAW is the ability to control the metal droplet transfer in welding. Note that from figure 4, the metal droplet is characterised by a downward slope of the current pulse. Figure 5 shows the current signal captured from experiments conducted at AML3D (PMC is used).

<div><ul>
<li style="display: inline-block;"> [[File:Pulsed_gmaw_transfer_mode.png|thumb|500px|none|Figure 4: Relationship between pulse current and droplet transfer<ref name="welding_book" />]] </li>
<li style="display: inline-block;"> [[File:Pmc_current_in_time_domain.png|thumb|650px|none|Figure 5: Current signal captured from experiments conducted at AML3D (PMC process)]] </li>
</ul></div>
==== Cold Metal Transfer ====
Cold Metal Transfer (CMT) is a complex welding process, also developed by Fronius. By detecting a short circuit (mode C and D in figure 8), CMT can make adjustment such as retracting the welding material to cool down, and therefore create a smoother, more stable weld. The complexity of CMT can be seen in the current signal of the process (figure 9). Compare to PMC, CMT's signal has more variation during one signal period.

<div><ul>
<li style="display: inline-block;"> [[File:Transfer_mode.png|thumb|800px|none|Figure 8: Mechanism of metal droplet transfer mode]] </li>
<li style="display: inline-block;"> [[File:Cmt1_current_in_time_domain.png|thumb|735px|none|Figure 9: Current signal captured from experiments conducted at AML3D (CMT process)]] </li>
</ul></div>
=== Machine Learning ===
Machine learning (or statistical learning) refer to a set of tools to give data scientists an insight into the data and make better decisions based on those information <ref name="introduction_ml_book"> James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An introduction to statistical learning (Vol. 112, p. 18). New York: Springer. </ref>. In this project, we will explore the capabilities of classical machine learning methods (i.e anything but Deep Learning) due to the limitation of our available data. By the time of writing this wiki, we have only explored Support Vector Machine.

==== Support Vector Machine ====
The main idea of Support Vector Machine (SVM) is to draw hyperplanes that best separate multiple classes of data. The original SVM judges best decision hyperplane based on the separation margin between data classes (figure 10). However, there are multiple variants of SVM, such as hard/soft margin SVM, Nu-SVM, One class SVM (to solve anomaly detection problem). In our project, we choose to use a soft margin implementation of SVM. The benefit of soft margin SVM (as oppose to hard margin SVM) is that by allowing some misclassifications, we get a wider margin and therefore the solution becomes more "generalise" (i.e work better with a new set of dataset that is independent from the training dataset).

<div><ul>
<li style="display: inline-block;"> [[File:svm_margin.png|thumb|400px|none|Figure 10: Separation margin <ref name="bishop_book"> Bishop, C. M. (2006). Pattern recognition and machine learning. Springer. </ref>]] </li>
</ul></div>

==== Kernel tricks ====
Figure 10 is of linearly separable cases for SVM. For dataset that is not easy to separate with a hyperplane in their original features space (such as figure 11), the data often get projected into a higher dimension space where it is easier to separate (figure 14). Such technique is called the Kernel Tricks. It is important to note that Kernel Tricks does not actually "map" the data from '''X''' dimension to '''Z''' dimension, but rather pairwise compare the similarity of the input space in '''Z''' dimension.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_nonseparable.png |thumb|350px|none|Figure 11: Nonseparable dataset <ref name="bishop_book"></ref>]] </li>
<li style="display: inline-block;"> [[File:Svm_kernel_trick.png |thumb|650px|none|Figure 12: Simplified example of the Kernel tricks]] </li>
</ul></div>

== Previous studies ==
There are many researches into defect detection in GMAW process. The three most dominant analysis methods are spectroscopic analysis, acoustic analysis, and electrical signal analysis.

=== Spectroscopic analysis ===
In <ref name="spectroscopic_analysis">D. Bebiano and S. Alfaro, “A weld defects detection system based on a spectrometer,” Sensors, vol. 9, no. 4, pp. 2851–2861, 2009.</ref>, the authors captured the intensity of radiation emission of electric arc. It is obvious from figure 13 that good weld's intensity stays stable while defect one's is characterised by a sudden peak. Therefore, a threshold is set to detect the defect welds.

<div><ul>
<li style="display: inline-block;"> [[File:spectroscopic_analysis.png|thumb|400px|none|Figure 13: Spectroscopic analysis <ref name="spectroscopic_analysis"></ref>]] </li>
</ul></div>

Other research into using spectroscopic data to monitor that can be of interest is <ref>D. Naso, B. Turchiano, and P. Pantaleo, “A fuzzy-logic based optical sensor for online weld defect-detection,” IEEE transactions on Industrial Informatics, vol. 1, no. 4, pp. 259–273, 2005.</ref>.

=== Acoustic analysis ===
With <ref name="acoustic_analysis">E. Cayo and S. C. Alfaro, “A non-intrusive GMA welding process quality monitoring system using acoustic sensing,” Sensors, vol. 9, no. 9, pp. 7150–7166, 2009.</ref>, the author discovered that arc ignition is characterised by a distinct sound and therefore it is possible to capture the frequency of ignition through acoustic sensors. A tolerance band was set and both ignition frequency and sound pressure level were used to monitor the quality of the weld.

<div><ul>
<li style="display: inline-block;"> [[File:acoustic_analysis.png|thumb|400px|none|Figure 14: Acoustic analysis <ref name="acoustic_analysis"></ref>]] </li>
</ul></div>

Other research into using acoustic signal to monitor that can be of interest are <ref>M. Fidali, “Detection of welding process instabilities using acoustic signals,” in International Congress on Technical Diagnostic, pp. 191–201, Springer, 2016.</ref>, <ref>L. Zhang, A. C. Basantes-Defaz, D. Ozevin, and E. Indacochea, “Real-time monitoring of welding process using air-coupled ultrasonics and acoustic emission,” The International Journal of Advanced Manufacturing Technology, vol. 101, no. 5-8, pp. 1623–1634, 2019.</ref>.

=== Electrical signal analysis ===
With <ref name="electrical_analysis_1">A. Sumesh, K. Rameshkumar, A. Raja, K. Mohandas, A. Santhakumari, and R. Shyambabu, “Establishing correlation between current and voltage signatures of the arc and weld defects in GMAW process,” Arabian Journal for Science and Engineering, vol. 42, no. 11, pp. 4649–4665, 2017.</ref>, the author plot the probability density distribution of voltage signal and discovered that the stable weld's signal concentrate in one region while unstable signal spread out. While in the paper, the authors did not developed a monitoring method using their findings, it shows that there is a significant differences between stable and unstable weld's electrical signature.

<div><ul>
<li style="display: inline-block;"> [[File:Electrical analysis pdd.png|thumb|400px|none|Figure 15: Voltage PDD <ref name="electrical_analysis_1"></ref>]] </li>
</ul></div>

In <ref name="electrical_analysis_2">S. Simpson, “Signature images for arc welding fault detection,” Science and Technology of Welding and Joining, vol. 12, no. 6, pp. 481–486, 2007.</ref>, the authors shows that by plotting the signature image of voltage and current scatter plot, both arcing and short-circuiting region's signature image can notify a defects welds. A more visual representation of short circuiting and arcing region can be found in [[#Gas metal arc welding|Gas metal arc welding section]].

<div><ul>
<li style="display: inline-block;"> [[File:Electrical voltage signature.png|thumb|400px|none|Figure 16a: Electrical voltage signature of stable weld <ref name="electrical_analysis_2"></ref>]] </li>
<li style="display: inline-block;"> [[File:Electrical signature 2.png|thumb|400px|none|Figure 16b: Electrical voltage signature of unstable weld <ref name="electrical_analysis_2"></ref>]] </li>
</ul></div>

Other research into using electrical signal to monitor that can be of interest are <ref>R. Madigan, “Arc sensing for defects in constant-voltage gas metal arc welding,” Welding Journal, vol. 78, pp. 322S–328S, 1999.</ref>, <ref>Y. Huang, K. Wang, Z. Zhou, X. Zhou, and J. Fang, “Stability evaluation of short-circuiting gas metal arc welding based on ensemble empirical mode decomposition,” Measurement Science and Technology, vol. 28, no. 3, p. 035006, 2017.</ref>,<ref>Z. Zhang, X. Chen, H. Chen, J. Zhong, and S. Chen, “Online welding quality monitoring based on feature extraction of arc voltage signal,” The International Journal of Advanced Manufacturing Technology, vol. 70, no. 9-12, pp. 1661–1671, 2014</ref>,<ref>X. Li and S. Simpson, “Parametric approach to positional fault detection in short arc welding,” Science and Technology of welding and joining, vol. 14, no. 2, pp. 146–151, 2009.</ref>, and <ref>E. Wei, D. Farson, R. Richardson, and H. Ludewig, “Detection of weld surface porosity by statistical analysis of arc current in gas metal arc welding,” Journal of Manufacturing Processes, vol. 3, no. 1, pp. 50–59, 2001.</ref>.

== Experiment setup ==
The welding system uses TPS 320i C Pulse (figure 15a), produced by Fronius, to supply power, feed wire, and system cooling. For movement control of the welding torch, the system use ABB’s IRB 1520ID (figure 15b), which is a dedicated arc welding robot. The Data Acquisition device is NI's DAQ. Each channel (voltage and current) is sampled at 25kHz while the dominant frequency for voltage and current signal are around 130Hz (for CMT process) and 190Hz (for PMC process).

[[File:Fronius_power_supply.jpg |thumb|Figure 17a: Fronius' Power Supply]]
[[File:Welding_robot2.jpg |thumb|Figure 17b: ABB's welding robot]]

[[File:Ni_daq.jpg|thumb|Figure 16: NI's DAQ device]]

== Statistical Analysis ==

=== Preprocessing ===
As the captured signal coming from Fronius' power supply, the signals also pick up switching power noise. The noise will interfere with the accuracy of the analysis if not filtered. In this project, we applied a 1000Hz digital Low Pass Butterworth filter. The effect of the filter can be seen in figure 18, while the comparation of using different filters can be seen in the gallery.

<div><ul>
<li style="display: inline-block;"> [[File:Filter_effect.png |thumb|750px|none|Figure 18: Effect of 1000Hz low pass filter]] </li>
<li style="display: inline-block;"> [[File:Freq_response.png |thumb|750px|none|Figure 19: Frequency response of the in-use filter]] </li>
</ul></div>

<gallery mode="slideshow" caption="Gallery: Comparison of different filtering method">
File:Effect_of_filtering2.png|50px|In use filter vs 300Hz low pass filter
File:Effect_of_filtering3.png|50px|In use filter vs 500Hz low pass filter
File:Effect_of_filtering4.png|50px|In use filter vs 750Hz low pass filter
File:Effect_of_filtering5.png|50px|In use filter vs 2000Hz low pass filter
</gallery>

=== Power Spectral Density ===
Our goal for this project is to detect the defect welds and predict the layer heights (the contact tip to work piece distance, CTWD, to be more accurate) in real time (refer to [[#Relating Contact Tip to Work-piece Distance to layer height|this section]] to understand the relationship between Contact Tip to Work-piece Distance and Layer height). An analysis of the current power spectral density (PSD) shows that there is an inverse proportional relationship between the CTWD of a weld and the frequency where its '''current signal''' PSD reach the peak, and also the magnitude of the peak. Since the peak of PSD can be fairly sensitive, we only consider the frequency of the peak PSD.

Figure 20 shows the PSD plot of voltage, current, and power signal of the weld using PMC process. In this experiment, we created multiple one-line, 8 seconds welds. These eight-second welds then get segmented into smaller two-second chunk to increase the number of data and decrease the processing time.

<div><ul>
<li style="display: inline-block;"> [[File:Psd pmc good bad all lowfreq.png |thumb|750px|none|Figure 20: Power spectral density of PMC process]] </li>
<li style="display: inline-block;"> [[File:One line weld.jpg |thumb|578px|none|Figure 21: Simple one-line weld experiment]] </li>
<li style="display: inline-block;"> [[File:Cmt_psd_stable_allCTWD_zoomed.png |thumb|750px|none|Figure 22: Simple one-line weld experiment]] </li>

</ul></div>

It seems like the inverse proportional relationship holds true for most cases. Other experiments showed that for a short period at the beginning and ending of the weld, the signal can behave differently relative to the rest of the weld. Currently, to migrate with this problem, data will not be collected and analysed (both offline and online) at these periods. However, the above results seem to only hold true for PMC processes. With CMT process, the current PSD plot indicates no relationship between the CTWD and the frequency at peak (figure 24).

For the next stage of the project, our plan is to develop a linear regression model to predict the CTWD for PMC process. With CMT process, we will have to look into different analysis techniques.

=== Support Vector Machine ===
By looking at the average voltage and average current of each segment of weld while making a simple straight path (PMC process), we realised there is a visible separation margin between stable and unstable welds. Using SVM with RBF kernel, we were able to correctly classify the quality of the weld with accuracy of 98%.

However, with CMT process, the separation cannot be observed.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_pmcNoOxide_rbf.png |thumb|750px|none|Figure 23: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:CmtNoOxide_average_voltage_current.png |thumb|750px|none|Figure 24: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

=== CTWD Regression ===
With CTWD prediction, the result will only be useful if it is done on-the-line while making the part. And therefore, as oppose to defect detection, we cannot have a short calibration process every constant time to check for worn contact tip. By analysing the data while building a real part, we found that the average voltage (for PMC process) and average current (for CMT process) seems to have the strongest linear relationship with the CTWD. The acceptable error for CTWD prediction is +/- 1 mm (according to AML3D requirement). For both PMC and CMT process, our root means square error is less than 1 (PMC: 0.68 and CMT: 0.83), which satisfied the requirement.

<div><ul>
<li style="display: inline-block;"> [[File:Linear_regression_pmc_wall.png |thumb|750px|none|Figure 25: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:Linear_regression_cmt_wall.png |thumb|750px|none|Figure 26: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

== Result discussion ==
These results show that there is a strong possibility of defect detection and CTWD prediction in welding. However, these initial findings are still a proof of concept. In the next stage of our research, we will refine the algorithm and integrate into AML's pipeline.

== References ==
<references />

Projects:2019s2-20001 Using Machine Learning to Determine Deposit Height and Defects for Wire + Arc Additive Manufacture (3D printing)

2019-10-15T05:24:17Z

A1713568: /* Preprocessing */

[[Category:Projects]]
[[Category:Final Year Projects]]
[[Category:2019s2|24501]]
== Introduction ==
AML3D is one of the leading company in the Wire Arc Additive Manufacturing (WAAM) industry. In short, WAAM is the Metal 3D printing. The technology has the ability to reduce up to 75% of manufacture time, cost, and material. Currently, WAAM is benefiting other industry such as: mining, defence, marine, and aviation.

<gallery mode="slideshow">
File:Welding baby.gif|center|Welding using robotic arm
File:Laser baby.gif|center|User laser sensor to find the CTWD
</gallery>

[[File:aml3d_logo.jpg|center|100px]]
=== Project team ===
==== Project students ====
* Anh Tran
* Nhat Nguyen
==== Supervisors ====
* Dr. Brian Ng
* Dr. Paul Colegrove (AML3D, sponsored company)
=== Objectives ===
The objective of this project is to increase the efficiency of the manufacture process at AML3D. In order to do so, the team will investigate into the possibility of automating and optimising the quality control processes. The two quality control processes that are currently being implemented at AML3D are measuring layer height using laser sensors, and human supervision for detecting defects. These processes add overhead into production time and usage of human resource, which is not desired. To achieve the goal, machine learning methods will be used extensively to analyse the electrical signatures of the weld process.

== Background ==
=== Wire Arc Additive Manufacturing ===
Wire and Arc Additive Manufacturing (WAAM) is a type of additive manufacturing that uses electric arc as the heat source and material wire to feed the manufacture process <ref name="WAAM"> S. W. Williams, F. Martina, A. C. Addison, J. Ding, G. Pardal & P. Colegrove (2016) Wire + Arc Additive Manufacturing, Materials Science and Technology, 32:7, 641-647, DOI: 10.1179/1743284715Y.0000000073 </ref>. WAAM has been investigated since the 1990s <ref name="WAAM" />, but only recently that it received more attention from the manufacture world.

Its significant comes from the ability to manufacture complex model with less time and less material. Figure 1a and 1b show real parts that was manufactured by AML3D. Such custom made parts might takes months before be ready to be shipped, but with WAAM, the production time can reduce down to weeks. Currently, the industries that benefit the most from WAAM are maritime and aerospace.

Similar to other additive manufacture methods, WAAM achieves such results by sliding the models into multiple layers, and then build the model layer by layer. The movement control is normally handled by a robotic arm (AML3D uses ABB's Arc Welder robot), and the welding path is generated by a Computer Aid Manufacture (CAM) software.

[[File:welding_part1.JPG|thumb|Figure 1a: A propeller]]
[[File:welding_part2.jpg|thumb|Figure 1b: Another part of the ship]]
[[File:abb_robot.jpg|thumb|Figure 2: ABB's IRB 1520ID Arc Welder]]

=== Gas metal arc welding ===
<div><ul>
<li style="display: inline-block;"> [[File:gmaw_welding.png|thumb|500px|none|Figure 2: GMAW welding process <ref name="welding_book"> Norrish, J. (2006). Advanced welding processes. Elsevier. </ref>]] </li>
</ul></div>

Gas metal arc welding (GMAW), or also known as metal inert gas (MIG), is the welding process used at AML3D for WAAM. Due to its high deposition rate and economic benefits, GMAW has became more popular. We will explore two variants of GMAW, which are Pulse Multi Control and Cold Metal Transfer. Both processes are developed by Fronius, an Austrian welding company. Note that it is not required to understand the welding physics to follow this wiki. The next two following sections' purpose is to highlight the high variability, high dynamic nature in welding.

[[File:fronius.png|thumb|Figure 3: Fronius logo]]

==== Pulse Multi Control ====
Pulse Multi Control (PMC) welding is Fronius' modified Pulsed GMAW. The advantage of Pulsed GMAW is the ability to control the metal droplet transfer in welding. Note that from figure 4, the metal droplet is characterised by a downward slope of the current pulse. Figure 5 shows the current signal captured from experiments conducted at AML3D (PMC is used).

<div><ul>
<li style="display: inline-block;"> [[File:Pulsed_gmaw_transfer_mode.png|thumb|500px|none|Figure 4: Relationship between pulse current and droplet transfer<ref name="welding_book" />]] </li>
<li style="display: inline-block;"> [[File:Pmc_current_in_time_domain.png|thumb|650px|none|Figure 5: Current signal captured from experiments conducted at AML3D (PMC process)]] </li>
</ul></div>
==== Cold Metal Transfer ====
Cold Metal Transfer (CMT) is a complex welding process, also developed by Fronius. By detecting a short circuit (mode C and D in figure 8), CMT can make adjustment such as retracting the welding material to cool down, and therefore create a smoother, more stable weld. The complexity of CMT can be seen in the current signal of the process (figure 9). Compare to PMC, CMT's signal has more variation during one signal period.

<div><ul>
<li style="display: inline-block;"> [[File:Transfer_mode.png|thumb|800px|none|Figure 8: Mechanism of metal droplet transfer mode]] </li>
<li style="display: inline-block;"> [[File:Cmt1_current_in_time_domain.png|thumb|735px|none|Figure 9: Current signal captured from experiments conducted at AML3D (CMT process)]] </li>
</ul></div>
=== Machine Learning ===
Machine learning (or statistical learning) refer to a set of tools to give data scientists an insight into the data and make better decisions based on those information <ref name="introduction_ml_book"> James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An introduction to statistical learning (Vol. 112, p. 18). New York: Springer. </ref>. In this project, we will explore the capabilities of classical machine learning methods (i.e anything but Deep Learning) due to the limitation of our available data. By the time of writing this wiki, we have only explored Support Vector Machine.

==== Support Vector Machine ====
The main idea of Support Vector Machine (SVM) is to draw hyperplanes that best separate multiple classes of data. The original SVM judges best decision hyperplane based on the separation margin between data classes (figure 10). However, there are multiple variants of SVM, such as hard/soft margin SVM, Nu-SVM, One class SVM (to solve anomaly detection problem). In our project, we choose to use a soft margin implementation of SVM. The benefit of soft margin SVM (as oppose to hard margin SVM) is that by allowing some misclassifications, we get a wider margin and therefore the solution becomes more "generalise" (i.e work better with a new set of dataset that is independent from the training dataset).

<div><ul>
<li style="display: inline-block;"> [[File:svm_margin.png|thumb|400px|none|Figure 10: Separation margin <ref name="bishop_book"> Bishop, C. M. (2006). Pattern recognition and machine learning. Springer. </ref>]] </li>
</ul></div>

==== Kernel tricks ====
Figure 10 is of linearly separable cases for SVM. For dataset that is not easy to separate with a hyperplane in their original features space (such as figure 11), the data often get projected into a higher dimension space where it is easier to separate (figure 14). Such technique is called the Kernel Tricks. It is important to note that Kernel Tricks does not actually "map" the data from '''X''' dimension to '''Z''' dimension, but rather pairwise compare the similarity of the input space in '''Z''' dimension.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_nonseparable.png |thumb|350px|none|Figure 11: Nonseparable dataset <ref name="bishop_book"></ref>]] </li>
<li style="display: inline-block;"> [[File:Svm_kernel_trick.png |thumb|650px|none|Figure 12: Simplified example of the Kernel tricks]] </li>
</ul></div>

== Previous studies ==
There are many researches into defect detection in GMAW process. The three most dominant analysis methods are spectroscopic analysis, acoustic analysis, and electrical signal analysis.

=== Spectroscopic analysis ===
In <ref name="spectroscopic_analysis">D. Bebiano and S. Alfaro, “A weld defects detection system based on a spectrometer,” Sensors, vol. 9, no. 4, pp. 2851–2861, 2009.</ref>, the authors captured the intensity of radiation emission of electric arc. It is obvious from figure 13 that good weld's intensity stays stable while defect one's is characterised by a sudden peak. Therefore, a threshold is set to detect the defect welds.

<div><ul>
<li style="display: inline-block;"> [[File:spectroscopic_analysis.png|thumb|400px|none|Figure 13: Spectroscopic analysis <ref name="spectroscopic_analysis"></ref>]] </li>
</ul></div>

Other research into using spectroscopic data to monitor that can be of interest is <ref>D. Naso, B. Turchiano, and P. Pantaleo, “A fuzzy-logic based optical sensor for online weld defect-detection,” IEEE transactions on Industrial Informatics, vol. 1, no. 4, pp. 259–273, 2005.</ref>.

=== Acoustic analysis ===
With <ref name="acoustic_analysis">E. Cayo and S. C. Alfaro, “A non-intrusive GMA welding process quality monitoring system using acoustic sensing,” Sensors, vol. 9, no. 9, pp. 7150–7166, 2009.</ref>, the author discovered that arc ignition is characterised by a distinct sound and therefore it is possible to capture the frequency of ignition through acoustic sensors. A tolerance band was set and both ignition frequency and sound pressure level were used to monitor the quality of the weld.

<div><ul>
<li style="display: inline-block;"> [[File:acoustic_analysis.png|thumb|400px|none|Figure 14: Acoustic analysis <ref name="acoustic_analysis"></ref>]] </li>
</ul></div>

Other research into using acoustic signal to monitor that can be of interest are <ref>M. Fidali, “Detection of welding process instabilities using acoustic signals,” in International Congress on Technical Diagnostic, pp. 191–201, Springer, 2016.</ref>, <ref>L. Zhang, A. C. Basantes-Defaz, D. Ozevin, and E. Indacochea, “Real-time monitoring of welding process using air-coupled ultrasonics and acoustic emission,” The International Journal of Advanced Manufacturing Technology, vol. 101, no. 5-8, pp. 1623–1634, 2019.</ref>.

=== Electrical signal analysis ===
With <ref name="electrical_analysis_1">A. Sumesh, K. Rameshkumar, A. Raja, K. Mohandas, A. Santhakumari, and R. Shyambabu, “Establishing correlation between current and voltage signatures of the arc and weld defects in GMAW process,” Arabian Journal for Science and Engineering, vol. 42, no. 11, pp. 4649–4665, 2017.</ref>, the author plot the probability density distribution of voltage signal and discovered that the stable weld's signal concentrate in one region while unstable signal spread out. While in the paper, the authors did not developed a monitoring method using their findings, it shows that there is a significant differences between stable and unstable weld's electrical signature.

<div><ul>
<li style="display: inline-block;"> [[File:Electrical analysis pdd.png|thumb|400px|none|Figure 15: Voltage PDD <ref name="electrical_analysis_1"></ref>]] </li>
</ul></div>

In <ref name="electrical_analysis_2">S. Simpson, “Signature images for arc welding fault detection,” Science and Technology of Welding and Joining, vol. 12, no. 6, pp. 481–486, 2007.</ref>, the authors shows that by plotting the signature image of voltage and current scatter plot, both arcing and short-circuiting region's signature image can notify a defects welds. A more visual representation of short circuiting and arcing region can be found in [[#Gas metal arc welding|Gas metal arc welding section]].

<div><ul>
<li style="display: inline-block;"> [[File:Electrical voltage signature.png|thumb|400px|none|Figure 16a: Electrical voltage signature of stable weld <ref name="electrical_analysis_2"></ref>]] </li>
<li style="display: inline-block;"> [[File:Electrical signature 2.png|thumb|400px|none|Figure 16b: Electrical voltage signature of unstable weld <ref name="electrical_analysis_2"></ref>]] </li>
</ul></div>

Other research into using electrical signal to monitor that can be of interest are <ref>R. Madigan, “Arc sensing for defects in constant-voltage gas metal arc welding,” Welding Journal, vol. 78, pp. 322S–328S, 1999.</ref>, <ref>Y. Huang, K. Wang, Z. Zhou, X. Zhou, and J. Fang, “Stability evaluation of short-circuiting gas metal arc welding based on ensemble empirical mode decomposition,” Measurement Science and Technology, vol. 28, no. 3, p. 035006, 2017.</ref>,<ref>Z. Zhang, X. Chen, H. Chen, J. Zhong, and S. Chen, “Online welding quality monitoring based on feature extraction of arc voltage signal,” The International Journal of Advanced Manufacturing Technology, vol. 70, no. 9-12, pp. 1661–1671, 2014</ref>,<ref>X. Li and S. Simpson, “Parametric approach to positional fault detection in short arc welding,” Science and Technology of welding and joining, vol. 14, no. 2, pp. 146–151, 2009.</ref>, and <ref>E. Wei, D. Farson, R. Richardson, and H. Ludewig, “Detection of weld surface porosity by statistical analysis of arc current in gas metal arc welding,” Journal of Manufacturing Processes, vol. 3, no. 1, pp. 50–59, 2001.</ref>.

== Experiment setup ==
The welding system uses TPS 320i C Pulse (figure 15a), produced by Fronius, to supply power, feed wire, and system cooling. For movement control of the welding torch, the system use ABB’s IRB 1520ID (figure 15b), which is a dedicated arc welding robot. The Data Acquisition device is NI's DAQ. Each channel (voltage and current) is sampled at 25kHz while the dominant frequency for voltage and current signal are around 130Hz (for CMT process) and 190Hz (for PMC process).

[[File:Fronius_power_supply.jpg |thumb|Figure 17a: Fronius' Power Supply]]
[[File:Welding_robot2.jpg |thumb|Figure 17b: ABB's welding robot]]

[[File:Ni_daq.jpg|thumb|Figure 16: NI's DAQ device]]

== Statistical Analysis ==

=== Preprocessing ===
As the captured signal coming from Fronius' power supply, the signals also pick up switching power noise. The noise will interfere with the accuracy of the analysis if not filtered. In this project, we applied a 1000Hz digital Low Pass Butterworth filter. The effect of the filter can be seen in figure 18, while the comparation of using different filters can be seen in the gallery.

<div><ul>
<li style="display: inline-block;"> [[File:Filter_effect.png |thumb|750px|none|Figure 18: Effect of 1000Hz low pass filter]] </li>
<li style="display: inline-block;"> [[File:Freq_response.png |thumb|750px|none|Figure 19: Frequency response of the in-use filter]] </li>
</ul></div>

<gallery mode="slideshow" caption="Gallery: Comparison of different filtering method">
File:Effect_of_filtering2.png|100px|In use filter vs 300Hz low pass filter
File:Effect_of_filtering3.png|100px|In use filter vs 500Hz low pass filter
File:Effect_of_filtering4.png|100px|In use filter vs 750Hz low pass filter
File:Effect_of_filtering5.png|100px|In use filter vs 2000Hz low pass filter
</gallery>

=== Power Spectral Density ===
Our goal for this project is to detect the defect welds and predict the layer heights (the contact tip to work piece distance, CTWD, to be more accurate) in real time (refer to [[#Relating Contact Tip to Work-piece Distance to layer height|this section]] to understand the relationship between Contact Tip to Work-piece Distance and Layer height). An analysis of the current power spectral density (PSD) shows that there is an inverse proportional relationship between the CTWD of a weld and the frequency where its '''current signal''' PSD reach the peak, and also the magnitude of the peak. Since the peak of PSD can be fairly sensitive, we only consider the frequency of the peak PSD.

Figure 20 shows the PSD plot of voltage, current, and power signal of the weld using PMC process. In this experiment, we created multiple one-line, 8 seconds welds. These eight-second welds then get segmented into smaller two-second chunk to increase the number of data and decrease the processing time.

<div><ul>
<li style="display: inline-block;"> [[File:Psd pmc good bad all lowfreq.png |thumb|750px|none|Figure 20: Power spectral density of PMC process]] </li>
<li style="display: inline-block;"> [[File:One line weld.jpg |thumb|578px|none|Figure 21: Simple one-line weld experiment]] </li>
<li style="display: inline-block;"> [[File:Cmt_psd_stable_allCTWD_zoomed.png |thumb|750px|none|Figure 22: Simple one-line weld experiment]] </li>

</ul></div>

It seems like the inverse proportional relationship holds true for most cases. Other experiments showed that for a short period at the beginning and ending of the weld, the signal can behave differently relative to the rest of the weld. Currently, to migrate with this problem, data will not be collected and analysed (both offline and online) at these periods. However, the above results seem to only hold true for PMC processes. With CMT process, the current PSD plot indicates no relationship between the CTWD and the frequency at peak (figure 24).

For the next stage of the project, our plan is to develop a linear regression model to predict the CTWD for PMC process. With CMT process, we will have to look into different analysis techniques.

=== Support Vector Machine ===
By looking at the average voltage and average current of each segment of weld while making a simple straight path (PMC process), we realised there is a visible separation margin between stable and unstable welds. Using SVM with RBF kernel, we were able to correctly classify the quality of the weld with accuracy of 98%.

However, with CMT process, the separation cannot be observed.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_pmcNoOxide_rbf.png |thumb|750px|none|Figure 23: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:CmtNoOxide_average_voltage_current.png |thumb|750px|none|Figure 24: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

=== CTWD Regression ===
With CTWD prediction, the result will only be useful if it is done on-the-line while making the part. And therefore, as oppose to defect detection, we cannot have a short calibration process every constant time to check for worn contact tip. By analysing the data while building a real part, we found that the average voltage (for PMC process) and average current (for CMT process) seems to have the strongest linear relationship with the CTWD. The acceptable error for CTWD prediction is +/- 1 mm (according to AML3D requirement). For both PMC and CMT process, our root means square error is less than 1 (PMC: 0.68 and CMT: 0.83), which satisfied the requirement.

<div><ul>
<li style="display: inline-block;"> [[File:Linear_regression_pmc_wall.png |thumb|750px|none|Figure 25: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:Linear_regression_cmt_wall.png |thumb|750px|none|Figure 26: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

== Result discussion ==
These results show that there is a strong possibility of defect detection and CTWD prediction in welding. However, these initial findings are still a proof of concept. In the next stage of our research, we will refine the algorithm and integrate into AML's pipeline.

== References ==
<references />

Projects:2019s2-20001 Using Machine Learning to Determine Deposit Height and Defects for Wire + Arc Additive Manufacture (3D printing)

2019-10-15T05:23:09Z

A1713568: /* Preprocessing */

[[Category:Projects]]
[[Category:Final Year Projects]]
[[Category:2019s2|24501]]
== Introduction ==
AML3D is one of the leading company in the Wire Arc Additive Manufacturing (WAAM) industry. In short, WAAM is the Metal 3D printing. The technology has the ability to reduce up to 75% of manufacture time, cost, and material. Currently, WAAM is benefiting other industry such as: mining, defence, marine, and aviation.

<gallery mode="slideshow">
File:Welding baby.gif|center|Welding using robotic arm
File:Laser baby.gif|center|User laser sensor to find the CTWD
</gallery>

[[File:aml3d_logo.jpg|center|100px]]
=== Project team ===
==== Project students ====
* Anh Tran
* Nhat Nguyen
==== Supervisors ====
* Dr. Brian Ng
* Dr. Paul Colegrove (AML3D, sponsored company)
=== Objectives ===
The objective of this project is to increase the efficiency of the manufacture process at AML3D. In order to do so, the team will investigate into the possibility of automating and optimising the quality control processes. The two quality control processes that are currently being implemented at AML3D are measuring layer height using laser sensors, and human supervision for detecting defects. These processes add overhead into production time and usage of human resource, which is not desired. To achieve the goal, machine learning methods will be used extensively to analyse the electrical signatures of the weld process.

== Background ==
=== Wire Arc Additive Manufacturing ===
Wire and Arc Additive Manufacturing (WAAM) is a type of additive manufacturing that uses electric arc as the heat source and material wire to feed the manufacture process <ref name="WAAM"> S. W. Williams, F. Martina, A. C. Addison, J. Ding, G. Pardal & P. Colegrove (2016) Wire + Arc Additive Manufacturing, Materials Science and Technology, 32:7, 641-647, DOI: 10.1179/1743284715Y.0000000073 </ref>. WAAM has been investigated since the 1990s <ref name="WAAM" />, but only recently that it received more attention from the manufacture world.

Its significant comes from the ability to manufacture complex model with less time and less material. Figure 1a and 1b show real parts that was manufactured by AML3D. Such custom made parts might takes months before be ready to be shipped, but with WAAM, the production time can reduce down to weeks. Currently, the industries that benefit the most from WAAM are maritime and aerospace.

Similar to other additive manufacture methods, WAAM achieves such results by sliding the models into multiple layers, and then build the model layer by layer. The movement control is normally handled by a robotic arm (AML3D uses ABB's Arc Welder robot), and the welding path is generated by a Computer Aid Manufacture (CAM) software.

[[File:welding_part1.JPG|thumb|Figure 1a: A propeller]]
[[File:welding_part2.jpg|thumb|Figure 1b: Another part of the ship]]
[[File:abb_robot.jpg|thumb|Figure 2: ABB's IRB 1520ID Arc Welder]]

=== Gas metal arc welding ===
<div><ul>
<li style="display: inline-block;"> [[File:gmaw_welding.png|thumb|500px|none|Figure 2: GMAW welding process <ref name="welding_book"> Norrish, J. (2006). Advanced welding processes. Elsevier. </ref>]] </li>
</ul></div>

Gas metal arc welding (GMAW), or also known as metal inert gas (MIG), is the welding process used at AML3D for WAAM. Due to its high deposition rate and economic benefits, GMAW has became more popular. We will explore two variants of GMAW, which are Pulse Multi Control and Cold Metal Transfer. Both processes are developed by Fronius, an Austrian welding company. Note that it is not required to understand the welding physics to follow this wiki. The next two following sections' purpose is to highlight the high variability, high dynamic nature in welding.

[[File:fronius.png|thumb|Figure 3: Fronius logo]]

==== Pulse Multi Control ====
Pulse Multi Control (PMC) welding is Fronius' modified Pulsed GMAW. The advantage of Pulsed GMAW is the ability to control the metal droplet transfer in welding. Note that from figure 4, the metal droplet is characterised by a downward slope of the current pulse. Figure 5 shows the current signal captured from experiments conducted at AML3D (PMC is used).

<div><ul>
<li style="display: inline-block;"> [[File:Pulsed_gmaw_transfer_mode.png|thumb|500px|none|Figure 4: Relationship between pulse current and droplet transfer<ref name="welding_book" />]] </li>
<li style="display: inline-block;"> [[File:Pmc_current_in_time_domain.png|thumb|650px|none|Figure 5: Current signal captured from experiments conducted at AML3D (PMC process)]] </li>
</ul></div>
==== Cold Metal Transfer ====
Cold Metal Transfer (CMT) is a complex welding process, also developed by Fronius. By detecting a short circuit (mode C and D in figure 8), CMT can make adjustment such as retracting the welding material to cool down, and therefore create a smoother, more stable weld. The complexity of CMT can be seen in the current signal of the process (figure 9). Compare to PMC, CMT's signal has more variation during one signal period.

<div><ul>
<li style="display: inline-block;"> [[File:Transfer_mode.png|thumb|800px|none|Figure 8: Mechanism of metal droplet transfer mode]] </li>
<li style="display: inline-block;"> [[File:Cmt1_current_in_time_domain.png|thumb|735px|none|Figure 9: Current signal captured from experiments conducted at AML3D (CMT process)]] </li>
</ul></div>
=== Machine Learning ===
Machine learning (or statistical learning) refer to a set of tools to give data scientists an insight into the data and make better decisions based on those information <ref name="introduction_ml_book"> James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An introduction to statistical learning (Vol. 112, p. 18). New York: Springer. </ref>. In this project, we will explore the capabilities of classical machine learning methods (i.e anything but Deep Learning) due to the limitation of our available data. By the time of writing this wiki, we have only explored Support Vector Machine.

==== Support Vector Machine ====
The main idea of Support Vector Machine (SVM) is to draw hyperplanes that best separate multiple classes of data. The original SVM judges best decision hyperplane based on the separation margin between data classes (figure 10). However, there are multiple variants of SVM, such as hard/soft margin SVM, Nu-SVM, One class SVM (to solve anomaly detection problem). In our project, we choose to use a soft margin implementation of SVM. The benefit of soft margin SVM (as oppose to hard margin SVM) is that by allowing some misclassifications, we get a wider margin and therefore the solution becomes more "generalise" (i.e work better with a new set of dataset that is independent from the training dataset).

<div><ul>
<li style="display: inline-block;"> [[File:svm_margin.png|thumb|400px|none|Figure 10: Separation margin <ref name="bishop_book"> Bishop, C. M. (2006). Pattern recognition and machine learning. Springer. </ref>]] </li>
</ul></div>

==== Kernel tricks ====
Figure 10 is of linearly separable cases for SVM. For dataset that is not easy to separate with a hyperplane in their original features space (such as figure 11), the data often get projected into a higher dimension space where it is easier to separate (figure 14). Such technique is called the Kernel Tricks. It is important to note that Kernel Tricks does not actually "map" the data from '''X''' dimension to '''Z''' dimension, but rather pairwise compare the similarity of the input space in '''Z''' dimension.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_nonseparable.png |thumb|350px|none|Figure 11: Nonseparable dataset <ref name="bishop_book"></ref>]] </li>
<li style="display: inline-block;"> [[File:Svm_kernel_trick.png |thumb|650px|none|Figure 12: Simplified example of the Kernel tricks]] </li>
</ul></div>

== Previous studies ==
There are many researches into defect detection in GMAW process. The three most dominant analysis methods are spectroscopic analysis, acoustic analysis, and electrical signal analysis.

=== Spectroscopic analysis ===
In <ref name="spectroscopic_analysis">D. Bebiano and S. Alfaro, “A weld defects detection system based on a spectrometer,” Sensors, vol. 9, no. 4, pp. 2851–2861, 2009.</ref>, the authors captured the intensity of radiation emission of electric arc. It is obvious from figure 13 that good weld's intensity stays stable while defect one's is characterised by a sudden peak. Therefore, a threshold is set to detect the defect welds.

<div><ul>
<li style="display: inline-block;"> [[File:spectroscopic_analysis.png|thumb|400px|none|Figure 13: Spectroscopic analysis <ref name="spectroscopic_analysis"></ref>]] </li>
</ul></div>

Other research into using spectroscopic data to monitor that can be of interest is <ref>D. Naso, B. Turchiano, and P. Pantaleo, “A fuzzy-logic based optical sensor for online weld defect-detection,” IEEE transactions on Industrial Informatics, vol. 1, no. 4, pp. 259–273, 2005.</ref>.

=== Acoustic analysis ===
With <ref name="acoustic_analysis">E. Cayo and S. C. Alfaro, “A non-intrusive GMA welding process quality monitoring system using acoustic sensing,” Sensors, vol. 9, no. 9, pp. 7150–7166, 2009.</ref>, the author discovered that arc ignition is characterised by a distinct sound and therefore it is possible to capture the frequency of ignition through acoustic sensors. A tolerance band was set and both ignition frequency and sound pressure level were used to monitor the quality of the weld.

<div><ul>
<li style="display: inline-block;"> [[File:acoustic_analysis.png|thumb|400px|none|Figure 14: Acoustic analysis <ref name="acoustic_analysis"></ref>]] </li>
</ul></div>

Other research into using acoustic signal to monitor that can be of interest are <ref>M. Fidali, “Detection of welding process instabilities using acoustic signals,” in International Congress on Technical Diagnostic, pp. 191–201, Springer, 2016.</ref>, <ref>L. Zhang, A. C. Basantes-Defaz, D. Ozevin, and E. Indacochea, “Real-time monitoring of welding process using air-coupled ultrasonics and acoustic emission,” The International Journal of Advanced Manufacturing Technology, vol. 101, no. 5-8, pp. 1623–1634, 2019.</ref>.

=== Electrical signal analysis ===
With <ref name="electrical_analysis_1">A. Sumesh, K. Rameshkumar, A. Raja, K. Mohandas, A. Santhakumari, and R. Shyambabu, “Establishing correlation between current and voltage signatures of the arc and weld defects in GMAW process,” Arabian Journal for Science and Engineering, vol. 42, no. 11, pp. 4649–4665, 2017.</ref>, the author plot the probability density distribution of voltage signal and discovered that the stable weld's signal concentrate in one region while unstable signal spread out. While in the paper, the authors did not developed a monitoring method using their findings, it shows that there is a significant differences between stable and unstable weld's electrical signature.

<div><ul>
<li style="display: inline-block;"> [[File:Electrical analysis pdd.png|thumb|400px|none|Figure 15: Voltage PDD <ref name="electrical_analysis_1"></ref>]] </li>
</ul></div>

In <ref name="electrical_analysis_2">S. Simpson, “Signature images for arc welding fault detection,” Science and Technology of Welding and Joining, vol. 12, no. 6, pp. 481–486, 2007.</ref>, the authors shows that by plotting the signature image of voltage and current scatter plot, both arcing and short-circuiting region's signature image can notify a defects welds. A more visual representation of short circuiting and arcing region can be found in [[#Gas metal arc welding|Gas metal arc welding section]].

<div><ul>
<li style="display: inline-block;"> [[File:Electrical voltage signature.png|thumb|400px|none|Figure 16a: Electrical voltage signature of stable weld <ref name="electrical_analysis_2"></ref>]] </li>
<li style="display: inline-block;"> [[File:Electrical signature 2.png|thumb|400px|none|Figure 16b: Electrical voltage signature of unstable weld <ref name="electrical_analysis_2"></ref>]] </li>
</ul></div>

Other research into using electrical signal to monitor that can be of interest are <ref>R. Madigan, “Arc sensing for defects in constant-voltage gas metal arc welding,” Welding Journal, vol. 78, pp. 322S–328S, 1999.</ref>, <ref>Y. Huang, K. Wang, Z. Zhou, X. Zhou, and J. Fang, “Stability evaluation of short-circuiting gas metal arc welding based on ensemble empirical mode decomposition,” Measurement Science and Technology, vol. 28, no. 3, p. 035006, 2017.</ref>,<ref>Z. Zhang, X. Chen, H. Chen, J. Zhong, and S. Chen, “Online welding quality monitoring based on feature extraction of arc voltage signal,” The International Journal of Advanced Manufacturing Technology, vol. 70, no. 9-12, pp. 1661–1671, 2014</ref>,<ref>X. Li and S. Simpson, “Parametric approach to positional fault detection in short arc welding,” Science and Technology of welding and joining, vol. 14, no. 2, pp. 146–151, 2009.</ref>, and <ref>E. Wei, D. Farson, R. Richardson, and H. Ludewig, “Detection of weld surface porosity by statistical analysis of arc current in gas metal arc welding,” Journal of Manufacturing Processes, vol. 3, no. 1, pp. 50–59, 2001.</ref>.

== Experiment setup ==
The welding system uses TPS 320i C Pulse (figure 15a), produced by Fronius, to supply power, feed wire, and system cooling. For movement control of the welding torch, the system use ABB’s IRB 1520ID (figure 15b), which is a dedicated arc welding robot. The Data Acquisition device is NI's DAQ. Each channel (voltage and current) is sampled at 25kHz while the dominant frequency for voltage and current signal are around 130Hz (for CMT process) and 190Hz (for PMC process).

[[File:Fronius_power_supply.jpg |thumb|Figure 17a: Fronius' Power Supply]]
[[File:Welding_robot2.jpg |thumb|Figure 17b: ABB's welding robot]]

[[File:Ni_daq.jpg|thumb|Figure 16: NI's DAQ device]]

== Statistical Analysis ==

=== Preprocessing ===
As the captured signal coming from Fronius' power supply, the signals also pick up switching power noise. The noise will interfere with the accuracy of the analysis if not filtered. In this project, we applied a 1000Hz digital Low Pass Butterworth filter. The effect of the filter can be seen in figure 18, while the comparation of using different filters can be seen in the gallery.

<div><ul>
<li style="display: inline-block;"> [[File:Filter_effect.png |thumb|750px|none|Figure 18: Effect of 1000Hz low pass filter]] </li>
<li style="display: inline-block;"> [[File:Freq_response.png |thumb|750px|none|Figure 19: Frequency response of the in-use filter]] </li>
</ul></div>

<gallery mode="slideshow" caption="Gallery: Comparison of different filtering method">
File:Effect_of_filtering2.png|In use filter vs 300Hz low pass filter
File:Effect_of_filtering3.png|In use filter vs 500Hz low pass filter
File:Effect_of_filtering4.png|In use filter vs 750Hz low pass filter
File:Effect_of_filtering5.png|In use filter vs 2000Hz low pass filter
</gallery>

=== Power Spectral Density ===
Our goal for this project is to detect the defect welds and predict the layer heights (the contact tip to work piece distance, CTWD, to be more accurate) in real time (refer to [[#Relating Contact Tip to Work-piece Distance to layer height|this section]] to understand the relationship between Contact Tip to Work-piece Distance and Layer height). An analysis of the current power spectral density (PSD) shows that there is an inverse proportional relationship between the CTWD of a weld and the frequency where its '''current signal''' PSD reach the peak, and also the magnitude of the peak. Since the peak of PSD can be fairly sensitive, we only consider the frequency of the peak PSD.

Figure 20 shows the PSD plot of voltage, current, and power signal of the weld using PMC process. In this experiment, we created multiple one-line, 8 seconds welds. These eight-second welds then get segmented into smaller two-second chunk to increase the number of data and decrease the processing time.

<div><ul>
<li style="display: inline-block;"> [[File:Psd pmc good bad all lowfreq.png |thumb|750px|none|Figure 20: Power spectral density of PMC process]] </li>
<li style="display: inline-block;"> [[File:One line weld.jpg |thumb|578px|none|Figure 21: Simple one-line weld experiment]] </li>
<li style="display: inline-block;"> [[File:Cmt_psd_stable_allCTWD_zoomed.png |thumb|750px|none|Figure 22: Simple one-line weld experiment]] </li>

</ul></div>

It seems like the inverse proportional relationship holds true for most cases. Other experiments showed that for a short period at the beginning and ending of the weld, the signal can behave differently relative to the rest of the weld. Currently, to migrate with this problem, data will not be collected and analysed (both offline and online) at these periods. However, the above results seem to only hold true for PMC processes. With CMT process, the current PSD plot indicates no relationship between the CTWD and the frequency at peak (figure 24).

For the next stage of the project, our plan is to develop a linear regression model to predict the CTWD for PMC process. With CMT process, we will have to look into different analysis techniques.

=== Support Vector Machine ===
By looking at the average voltage and average current of each segment of weld while making a simple straight path (PMC process), we realised there is a visible separation margin between stable and unstable welds. Using SVM with RBF kernel, we were able to correctly classify the quality of the weld with accuracy of 98%.

However, with CMT process, the separation cannot be observed.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_pmcNoOxide_rbf.png |thumb|750px|none|Figure 23: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:CmtNoOxide_average_voltage_current.png |thumb|750px|none|Figure 24: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

=== CTWD Regression ===
With CTWD prediction, the result will only be useful if it is done on-the-line while making the part. And therefore, as oppose to defect detection, we cannot have a short calibration process every constant time to check for worn contact tip. By analysing the data while building a real part, we found that the average voltage (for PMC process) and average current (for CMT process) seems to have the strongest linear relationship with the CTWD. The acceptable error for CTWD prediction is +/- 1 mm (according to AML3D requirement). For both PMC and CMT process, our root means square error is less than 1 (PMC: 0.68 and CMT: 0.83), which satisfied the requirement.

<div><ul>
<li style="display: inline-block;"> [[File:Linear_regression_pmc_wall.png |thumb|750px|none|Figure 25: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:Linear_regression_cmt_wall.png |thumb|750px|none|Figure 26: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

== Result discussion ==
These results show that there is a strong possibility of defect detection and CTWD prediction in welding. However, these initial findings are still a proof of concept. In the next stage of our research, we will refine the algorithm and integrate into AML's pipeline.

== References ==
<references />

Projects:2019s2-20001 Using Machine Learning to Determine Deposit Height and Defects for Wire + Arc Additive Manufacture (3D printing)

2019-10-15T05:21:39Z

A1713568: /* Introduction */

[[Category:Projects]]
[[Category:Final Year Projects]]
[[Category:2019s2|24501]]
== Introduction ==
AML3D is one of the leading company in the Wire Arc Additive Manufacturing (WAAM) industry. In short, WAAM is the Metal 3D printing. The technology has the ability to reduce up to 75% of manufacture time, cost, and material. Currently, WAAM is benefiting other industry such as: mining, defence, marine, and aviation.

<gallery mode="slideshow">
File:Welding baby.gif|center|Welding using robotic arm
File:Laser baby.gif|center|User laser sensor to find the CTWD
</gallery>

[[File:aml3d_logo.jpg|center|100px]]
=== Project team ===
==== Project students ====
* Anh Tran
* Nhat Nguyen
==== Supervisors ====
* Dr. Brian Ng
* Dr. Paul Colegrove (AML3D, sponsored company)
=== Objectives ===
The objective of this project is to increase the efficiency of the manufacture process at AML3D. In order to do so, the team will investigate into the possibility of automating and optimising the quality control processes. The two quality control processes that are currently being implemented at AML3D are measuring layer height using laser sensors, and human supervision for detecting defects. These processes add overhead into production time and usage of human resource, which is not desired. To achieve the goal, machine learning methods will be used extensively to analyse the electrical signatures of the weld process.

== Background ==
=== Wire Arc Additive Manufacturing ===
Wire and Arc Additive Manufacturing (WAAM) is a type of additive manufacturing that uses electric arc as the heat source and material wire to feed the manufacture process <ref name="WAAM"> S. W. Williams, F. Martina, A. C. Addison, J. Ding, G. Pardal & P. Colegrove (2016) Wire + Arc Additive Manufacturing, Materials Science and Technology, 32:7, 641-647, DOI: 10.1179/1743284715Y.0000000073 </ref>. WAAM has been investigated since the 1990s <ref name="WAAM" />, but only recently that it received more attention from the manufacture world.

Its significant comes from the ability to manufacture complex model with less time and less material. Figure 1a and 1b show real parts that was manufactured by AML3D. Such custom made parts might takes months before be ready to be shipped, but with WAAM, the production time can reduce down to weeks. Currently, the industries that benefit the most from WAAM are maritime and aerospace.

Similar to other additive manufacture methods, WAAM achieves such results by sliding the models into multiple layers, and then build the model layer by layer. The movement control is normally handled by a robotic arm (AML3D uses ABB's Arc Welder robot), and the welding path is generated by a Computer Aid Manufacture (CAM) software.

[[File:welding_part1.JPG|thumb|Figure 1a: A propeller]]
[[File:welding_part2.jpg|thumb|Figure 1b: Another part of the ship]]
[[File:abb_robot.jpg|thumb|Figure 2: ABB's IRB 1520ID Arc Welder]]

=== Gas metal arc welding ===
<div><ul>
<li style="display: inline-block;"> [[File:gmaw_welding.png|thumb|500px|none|Figure 2: GMAW welding process <ref name="welding_book"> Norrish, J. (2006). Advanced welding processes. Elsevier. </ref>]] </li>
</ul></div>

Gas metal arc welding (GMAW), or also known as metal inert gas (MIG), is the welding process used at AML3D for WAAM. Due to its high deposition rate and economic benefits, GMAW has became more popular. We will explore two variants of GMAW, which are Pulse Multi Control and Cold Metal Transfer. Both processes are developed by Fronius, an Austrian welding company. Note that it is not required to understand the welding physics to follow this wiki. The next two following sections' purpose is to highlight the high variability, high dynamic nature in welding.

[[File:fronius.png|thumb|Figure 3: Fronius logo]]

==== Pulse Multi Control ====
Pulse Multi Control (PMC) welding is Fronius' modified Pulsed GMAW. The advantage of Pulsed GMAW is the ability to control the metal droplet transfer in welding. Note that from figure 4, the metal droplet is characterised by a downward slope of the current pulse. Figure 5 shows the current signal captured from experiments conducted at AML3D (PMC is used).

<div><ul>
<li style="display: inline-block;"> [[File:Pulsed_gmaw_transfer_mode.png|thumb|500px|none|Figure 4: Relationship between pulse current and droplet transfer<ref name="welding_book" />]] </li>
<li style="display: inline-block;"> [[File:Pmc_current_in_time_domain.png|thumb|650px|none|Figure 5: Current signal captured from experiments conducted at AML3D (PMC process)]] </li>
</ul></div>
==== Cold Metal Transfer ====
Cold Metal Transfer (CMT) is a complex welding process, also developed by Fronius. By detecting a short circuit (mode C and D in figure 8), CMT can make adjustment such as retracting the welding material to cool down, and therefore create a smoother, more stable weld. The complexity of CMT can be seen in the current signal of the process (figure 9). Compare to PMC, CMT's signal has more variation during one signal period.

<div><ul>
<li style="display: inline-block;"> [[File:Transfer_mode.png|thumb|800px|none|Figure 8: Mechanism of metal droplet transfer mode]] </li>
<li style="display: inline-block;"> [[File:Cmt1_current_in_time_domain.png|thumb|735px|none|Figure 9: Current signal captured from experiments conducted at AML3D (CMT process)]] </li>
</ul></div>
=== Machine Learning ===
Machine learning (or statistical learning) refer to a set of tools to give data scientists an insight into the data and make better decisions based on those information <ref name="introduction_ml_book"> James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An introduction to statistical learning (Vol. 112, p. 18). New York: Springer. </ref>. In this project, we will explore the capabilities of classical machine learning methods (i.e anything but Deep Learning) due to the limitation of our available data. By the time of writing this wiki, we have only explored Support Vector Machine.

==== Support Vector Machine ====
The main idea of Support Vector Machine (SVM) is to draw hyperplanes that best separate multiple classes of data. The original SVM judges best decision hyperplane based on the separation margin between data classes (figure 10). However, there are multiple variants of SVM, such as hard/soft margin SVM, Nu-SVM, One class SVM (to solve anomaly detection problem). In our project, we choose to use a soft margin implementation of SVM. The benefit of soft margin SVM (as oppose to hard margin SVM) is that by allowing some misclassifications, we get a wider margin and therefore the solution becomes more "generalise" (i.e work better with a new set of dataset that is independent from the training dataset).

<div><ul>
<li style="display: inline-block;"> [[File:svm_margin.png|thumb|400px|none|Figure 10: Separation margin <ref name="bishop_book"> Bishop, C. M. (2006). Pattern recognition and machine learning. Springer. </ref>]] </li>
</ul></div>

==== Kernel tricks ====
Figure 10 is of linearly separable cases for SVM. For dataset that is not easy to separate with a hyperplane in their original features space (such as figure 11), the data often get projected into a higher dimension space where it is easier to separate (figure 14). Such technique is called the Kernel Tricks. It is important to note that Kernel Tricks does not actually "map" the data from '''X''' dimension to '''Z''' dimension, but rather pairwise compare the similarity of the input space in '''Z''' dimension.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_nonseparable.png |thumb|350px|none|Figure 11: Nonseparable dataset <ref name="bishop_book"></ref>]] </li>
<li style="display: inline-block;"> [[File:Svm_kernel_trick.png |thumb|650px|none|Figure 12: Simplified example of the Kernel tricks]] </li>
</ul></div>

== Previous studies ==
There are many researches into defect detection in GMAW process. The three most dominant analysis methods are spectroscopic analysis, acoustic analysis, and electrical signal analysis.

=== Spectroscopic analysis ===
In <ref name="spectroscopic_analysis">D. Bebiano and S. Alfaro, “A weld defects detection system based on a spectrometer,” Sensors, vol. 9, no. 4, pp. 2851–2861, 2009.</ref>, the authors captured the intensity of radiation emission of electric arc. It is obvious from figure 13 that good weld's intensity stays stable while defect one's is characterised by a sudden peak. Therefore, a threshold is set to detect the defect welds.

<div><ul>
<li style="display: inline-block;"> [[File:spectroscopic_analysis.png|thumb|400px|none|Figure 13: Spectroscopic analysis <ref name="spectroscopic_analysis"></ref>]] </li>
</ul></div>

Other research into using spectroscopic data to monitor that can be of interest is <ref>D. Naso, B. Turchiano, and P. Pantaleo, “A fuzzy-logic based optical sensor for online weld defect-detection,” IEEE transactions on Industrial Informatics, vol. 1, no. 4, pp. 259–273, 2005.</ref>.

=== Acoustic analysis ===
With <ref name="acoustic_analysis">E. Cayo and S. C. Alfaro, “A non-intrusive GMA welding process quality monitoring system using acoustic sensing,” Sensors, vol. 9, no. 9, pp. 7150–7166, 2009.</ref>, the author discovered that arc ignition is characterised by a distinct sound and therefore it is possible to capture the frequency of ignition through acoustic sensors. A tolerance band was set and both ignition frequency and sound pressure level were used to monitor the quality of the weld.

<div><ul>
<li style="display: inline-block;"> [[File:acoustic_analysis.png|thumb|400px|none|Figure 14: Acoustic analysis <ref name="acoustic_analysis"></ref>]] </li>
</ul></div>

Other research into using acoustic signal to monitor that can be of interest are <ref>M. Fidali, “Detection of welding process instabilities using acoustic signals,” in International Congress on Technical Diagnostic, pp. 191–201, Springer, 2016.</ref>, <ref>L. Zhang, A. C. Basantes-Defaz, D. Ozevin, and E. Indacochea, “Real-time monitoring of welding process using air-coupled ultrasonics and acoustic emission,” The International Journal of Advanced Manufacturing Technology, vol. 101, no. 5-8, pp. 1623–1634, 2019.</ref>.

=== Electrical signal analysis ===
With <ref name="electrical_analysis_1">A. Sumesh, K. Rameshkumar, A. Raja, K. Mohandas, A. Santhakumari, and R. Shyambabu, “Establishing correlation between current and voltage signatures of the arc and weld defects in GMAW process,” Arabian Journal for Science and Engineering, vol. 42, no. 11, pp. 4649–4665, 2017.</ref>, the author plot the probability density distribution of voltage signal and discovered that the stable weld's signal concentrate in one region while unstable signal spread out. While in the paper, the authors did not developed a monitoring method using their findings, it shows that there is a significant differences between stable and unstable weld's electrical signature.

<div><ul>
<li style="display: inline-block;"> [[File:Electrical analysis pdd.png|thumb|400px|none|Figure 15: Voltage PDD <ref name="electrical_analysis_1"></ref>]] </li>
</ul></div>

In <ref name="electrical_analysis_2">S. Simpson, “Signature images for arc welding fault detection,” Science and Technology of Welding and Joining, vol. 12, no. 6, pp. 481–486, 2007.</ref>, the authors shows that by plotting the signature image of voltage and current scatter plot, both arcing and short-circuiting region's signature image can notify a defects welds. A more visual representation of short circuiting and arcing region can be found in [[#Gas metal arc welding|Gas metal arc welding section]].

<div><ul>
<li style="display: inline-block;"> [[File:Electrical voltage signature.png|thumb|400px|none|Figure 16a: Electrical voltage signature of stable weld <ref name="electrical_analysis_2"></ref>]] </li>
<li style="display: inline-block;"> [[File:Electrical signature 2.png|thumb|400px|none|Figure 16b: Electrical voltage signature of unstable weld <ref name="electrical_analysis_2"></ref>]] </li>
</ul></div>

Other research into using electrical signal to monitor that can be of interest are <ref>R. Madigan, “Arc sensing for defects in constant-voltage gas metal arc welding,” Welding Journal, vol. 78, pp. 322S–328S, 1999.</ref>, <ref>Y. Huang, K. Wang, Z. Zhou, X. Zhou, and J. Fang, “Stability evaluation of short-circuiting gas metal arc welding based on ensemble empirical mode decomposition,” Measurement Science and Technology, vol. 28, no. 3, p. 035006, 2017.</ref>,<ref>Z. Zhang, X. Chen, H. Chen, J. Zhong, and S. Chen, “Online welding quality monitoring based on feature extraction of arc voltage signal,” The International Journal of Advanced Manufacturing Technology, vol. 70, no. 9-12, pp. 1661–1671, 2014</ref>,<ref>X. Li and S. Simpson, “Parametric approach to positional fault detection in short arc welding,” Science and Technology of welding and joining, vol. 14, no. 2, pp. 146–151, 2009.</ref>, and <ref>E. Wei, D. Farson, R. Richardson, and H. Ludewig, “Detection of weld surface porosity by statistical analysis of arc current in gas metal arc welding,” Journal of Manufacturing Processes, vol. 3, no. 1, pp. 50–59, 2001.</ref>.

== Experiment setup ==
The welding system uses TPS 320i C Pulse (figure 15a), produced by Fronius, to supply power, feed wire, and system cooling. For movement control of the welding torch, the system use ABB’s IRB 1520ID (figure 15b), which is a dedicated arc welding robot. The Data Acquisition device is NI's DAQ. Each channel (voltage and current) is sampled at 25kHz while the dominant frequency for voltage and current signal are around 130Hz (for CMT process) and 190Hz (for PMC process).

[[File:Fronius_power_supply.jpg |thumb|Figure 17a: Fronius' Power Supply]]
[[File:Welding_robot2.jpg |thumb|Figure 17b: ABB's welding robot]]

[[File:Ni_daq.jpg|thumb|Figure 16: NI's DAQ device]]

== Statistical Analysis ==

=== Preprocessing ===
As the captured signal coming from Fronius' power supply, the signals also pick up switching power noise. The noise will interfere with the accuracy of the analysis if not filtered. In this project, we applied a 1000Hz digital Low Pass Butterworth filter. The effect of the filter can be seen in figure 18, while the comparation of using different filters can be seen in the gallery.

<div><ul>
<li style="display: inline-block;"> [[File:Filter_effect.png |thumb|750px|none|Figure 18: Effect of 1000Hz low pass filter]] </li>
<li style="display: inline-block;"> [[File:Freq_response.png |thumb|750px|none|Figure 19: Frequency response of the in-use filter]] </li>
</ul></div>

<gallery mode="packed-hover" caption="Gallery: Comparison of different filtering method" perrow=2>
File:Effect_of_filtering2.png|In use filter vs 300Hz low pass filter
File:Effect_of_filtering3.png|In use filter vs 500Hz low pass filter
File:Effect_of_filtering4.png|In use filter vs 750Hz low pass filter
File:Effect_of_filtering5.png|In use filter vs 2000Hz low pass filter
</gallery>

=== Power Spectral Density ===
Our goal for this project is to detect the defect welds and predict the layer heights (the contact tip to work piece distance, CTWD, to be more accurate) in real time (refer to [[#Relating Contact Tip to Work-piece Distance to layer height|this section]] to understand the relationship between Contact Tip to Work-piece Distance and Layer height). An analysis of the current power spectral density (PSD) shows that there is an inverse proportional relationship between the CTWD of a weld and the frequency where its '''current signal''' PSD reach the peak, and also the magnitude of the peak. Since the peak of PSD can be fairly sensitive, we only consider the frequency of the peak PSD.

Figure 20 shows the PSD plot of voltage, current, and power signal of the weld using PMC process. In this experiment, we created multiple one-line, 8 seconds welds. These eight-second welds then get segmented into smaller two-second chunk to increase the number of data and decrease the processing time.

<div><ul>
<li style="display: inline-block;"> [[File:Psd pmc good bad all lowfreq.png |thumb|750px|none|Figure 20: Power spectral density of PMC process]] </li>
<li style="display: inline-block;"> [[File:One line weld.jpg |thumb|578px|none|Figure 21: Simple one-line weld experiment]] </li>
<li style="display: inline-block;"> [[File:Cmt_psd_stable_allCTWD_zoomed.png |thumb|750px|none|Figure 22: Simple one-line weld experiment]] </li>

</ul></div>

It seems like the inverse proportional relationship holds true for most cases. Other experiments showed that for a short period at the beginning and ending of the weld, the signal can behave differently relative to the rest of the weld. Currently, to migrate with this problem, data will not be collected and analysed (both offline and online) at these periods. However, the above results seem to only hold true for PMC processes. With CMT process, the current PSD plot indicates no relationship between the CTWD and the frequency at peak (figure 24).

For the next stage of the project, our plan is to develop a linear regression model to predict the CTWD for PMC process. With CMT process, we will have to look into different analysis techniques.

=== Support Vector Machine ===
By looking at the average voltage and average current of each segment of weld while making a simple straight path (PMC process), we realised there is a visible separation margin between stable and unstable welds. Using SVM with RBF kernel, we were able to correctly classify the quality of the weld with accuracy of 98%.

However, with CMT process, the separation cannot be observed.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_pmcNoOxide_rbf.png |thumb|750px|none|Figure 23: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:CmtNoOxide_average_voltage_current.png |thumb|750px|none|Figure 24: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

=== CTWD Regression ===
With CTWD prediction, the result will only be useful if it is done on-the-line while making the part. And therefore, as oppose to defect detection, we cannot have a short calibration process every constant time to check for worn contact tip. By analysing the data while building a real part, we found that the average voltage (for PMC process) and average current (for CMT process) seems to have the strongest linear relationship with the CTWD. The acceptable error for CTWD prediction is +/- 1 mm (according to AML3D requirement). For both PMC and CMT process, our root means square error is less than 1 (PMC: 0.68 and CMT: 0.83), which satisfied the requirement.

<div><ul>
<li style="display: inline-block;"> [[File:Linear_regression_pmc_wall.png |thumb|750px|none|Figure 25: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:Linear_regression_cmt_wall.png |thumb|750px|none|Figure 26: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

== Result discussion ==
These results show that there is a strong possibility of defect detection and CTWD prediction in welding. However, these initial findings are still a proof of concept. In the next stage of our research, we will refine the algorithm and integrate into AML's pipeline.

== References ==
<references />

Projects:2019s2-20001 Using Machine Learning to Determine Deposit Height and Defects for Wire + Arc Additive Manufacture (3D printing)

2019-10-15T05:14:46Z

A1713568: /* Introduction */

[[Category:Projects]]
[[Category:Final Year Projects]]
[[Category:2019s2|24501]]
== Introduction ==
AML3D is one of the leading company in the Wire Arc Additive Manufacturing (WAAM) industry. In short, WAAM is the Metal 3D printing. The technology has the ability to reduce up to 75% of manufacture time, cost, and material. Currently, WAAM is benefiting other industry such as: mining, defence, marine, and aviation.

<div><ul>
<li style="display: inline-block;"> [[File:Welding baby.gif|upright=2.0|Welding using robotic arm|none|thumb]] </li>
<li style="display: inline-block;"> [[File:Laser baby.gif|upright=2.0|User laser sensor to find the CTWD|none|thumb]] </li>
</ul></div>

[[File:aml3d_logo.jpg|100px]]
=== Project team ===
==== Project students ====
* Anh Tran
* Nhat Nguyen
==== Supervisors ====
* Dr. Brian Ng
* Dr. Paul Colegrove (AML3D, sponsored company)
=== Objectives ===
The objective of this project is to increase the efficiency of the manufacture process at AML3D. In order to do so, the team will investigate into the possibility of automating and optimising the quality control processes. The two quality control processes that are currently being implemented at AML3D are measuring layer height using laser sensors, and human supervision for detecting defects. These processes add overhead into production time and usage of human resource, which is not desired. To achieve the goal, machine learning methods will be used extensively to analyse the electrical signatures of the weld process.

== Background ==
=== Wire Arc Additive Manufacturing ===
Wire and Arc Additive Manufacturing (WAAM) is a type of additive manufacturing that uses electric arc as the heat source and material wire to feed the manufacture process <ref name="WAAM"> S. W. Williams, F. Martina, A. C. Addison, J. Ding, G. Pardal & P. Colegrove (2016) Wire + Arc Additive Manufacturing, Materials Science and Technology, 32:7, 641-647, DOI: 10.1179/1743284715Y.0000000073 </ref>. WAAM has been investigated since the 1990s <ref name="WAAM" />, but only recently that it received more attention from the manufacture world.

Its significant comes from the ability to manufacture complex model with less time and less material. Figure 1a and 1b show real parts that was manufactured by AML3D. Such custom made parts might takes months before be ready to be shipped, but with WAAM, the production time can reduce down to weeks. Currently, the industries that benefit the most from WAAM are maritime and aerospace.

Similar to other additive manufacture methods, WAAM achieves such results by sliding the models into multiple layers, and then build the model layer by layer. The movement control is normally handled by a robotic arm (AML3D uses ABB's Arc Welder robot), and the welding path is generated by a Computer Aid Manufacture (CAM) software.

[[File:welding_part1.JPG|thumb|Figure 1a: A propeller]]
[[File:welding_part2.jpg|thumb|Figure 1b: Another part of the ship]]
[[File:abb_robot.jpg|thumb|Figure 2: ABB's IRB 1520ID Arc Welder]]

=== Gas metal arc welding ===
<div><ul>
<li style="display: inline-block;"> [[File:gmaw_welding.png|thumb|500px|none|Figure 2: GMAW welding process <ref name="welding_book"> Norrish, J. (2006). Advanced welding processes. Elsevier. </ref>]] </li>
</ul></div>

Gas metal arc welding (GMAW), or also known as metal inert gas (MIG), is the welding process used at AML3D for WAAM. Due to its high deposition rate and economic benefits, GMAW has became more popular. We will explore two variants of GMAW, which are Pulse Multi Control and Cold Metal Transfer. Both processes are developed by Fronius, an Austrian welding company. Note that it is not required to understand the welding physics to follow this wiki. The next two following sections' purpose is to highlight the high variability, high dynamic nature in welding.

[[File:fronius.png|thumb|Figure 3: Fronius logo]]

==== Pulse Multi Control ====
Pulse Multi Control (PMC) welding is Fronius' modified Pulsed GMAW. The advantage of Pulsed GMAW is the ability to control the metal droplet transfer in welding. Note that from figure 4, the metal droplet is characterised by a downward slope of the current pulse. Figure 5 shows the current signal captured from experiments conducted at AML3D (PMC is used).

<div><ul>
<li style="display: inline-block;"> [[File:Pulsed_gmaw_transfer_mode.png|thumb|500px|none|Figure 4: Relationship between pulse current and droplet transfer<ref name="welding_book" />]] </li>
<li style="display: inline-block;"> [[File:Pmc_current_in_time_domain.png|thumb|650px|none|Figure 5: Current signal captured from experiments conducted at AML3D (PMC process)]] </li>
</ul></div>
==== Cold Metal Transfer ====
Cold Metal Transfer (CMT) is a complex welding process, also developed by Fronius. By detecting a short circuit (mode C and D in figure 8), CMT can make adjustment such as retracting the welding material to cool down, and therefore create a smoother, more stable weld. The complexity of CMT can be seen in the current signal of the process (figure 9). Compare to PMC, CMT's signal has more variation during one signal period.

<div><ul>
<li style="display: inline-block;"> [[File:Transfer_mode.png|thumb|800px|none|Figure 8: Mechanism of metal droplet transfer mode]] </li>
<li style="display: inline-block;"> [[File:Cmt1_current_in_time_domain.png|thumb|735px|none|Figure 9: Current signal captured from experiments conducted at AML3D (CMT process)]] </li>
</ul></div>
=== Machine Learning ===
Machine learning (or statistical learning) refer to a set of tools to give data scientists an insight into the data and make better decisions based on those information <ref name="introduction_ml_book"> James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An introduction to statistical learning (Vol. 112, p. 18). New York: Springer. </ref>. In this project, we will explore the capabilities of classical machine learning methods (i.e anything but Deep Learning) due to the limitation of our available data. By the time of writing this wiki, we have only explored Support Vector Machine.

==== Support Vector Machine ====
The main idea of Support Vector Machine (SVM) is to draw hyperplanes that best separate multiple classes of data. The original SVM judges best decision hyperplane based on the separation margin between data classes (figure 10). However, there are multiple variants of SVM, such as hard/soft margin SVM, Nu-SVM, One class SVM (to solve anomaly detection problem). In our project, we choose to use a soft margin implementation of SVM. The benefit of soft margin SVM (as oppose to hard margin SVM) is that by allowing some misclassifications, we get a wider margin and therefore the solution becomes more "generalise" (i.e work better with a new set of dataset that is independent from the training dataset).

<div><ul>
<li style="display: inline-block;"> [[File:svm_margin.png|thumb|400px|none|Figure 10: Separation margin <ref name="bishop_book"> Bishop, C. M. (2006). Pattern recognition and machine learning. Springer. </ref>]] </li>
</ul></div>

==== Kernel tricks ====
Figure 10 is of linearly separable cases for SVM. For dataset that is not easy to separate with a hyperplane in their original features space (such as figure 11), the data often get projected into a higher dimension space where it is easier to separate (figure 14). Such technique is called the Kernel Tricks. It is important to note that Kernel Tricks does not actually "map" the data from '''X''' dimension to '''Z''' dimension, but rather pairwise compare the similarity of the input space in '''Z''' dimension.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_nonseparable.png |thumb|350px|none|Figure 11: Nonseparable dataset <ref name="bishop_book"></ref>]] </li>
<li style="display: inline-block;"> [[File:Svm_kernel_trick.png |thumb|650px|none|Figure 12: Simplified example of the Kernel tricks]] </li>
</ul></div>

== Previous studies ==
There are many researches into defect detection in GMAW process. The three most dominant analysis methods are spectroscopic analysis, acoustic analysis, and electrical signal analysis.

=== Spectroscopic analysis ===
In <ref name="spectroscopic_analysis">D. Bebiano and S. Alfaro, “A weld defects detection system based on a spectrometer,” Sensors, vol. 9, no. 4, pp. 2851–2861, 2009.</ref>, the authors captured the intensity of radiation emission of electric arc. It is obvious from figure 13 that good weld's intensity stays stable while defect one's is characterised by a sudden peak. Therefore, a threshold is set to detect the defect welds.

<div><ul>
<li style="display: inline-block;"> [[File:spectroscopic_analysis.png|thumb|400px|none|Figure 13: Spectroscopic analysis <ref name="spectroscopic_analysis"></ref>]] </li>
</ul></div>

Other research into using spectroscopic data to monitor that can be of interest is <ref>D. Naso, B. Turchiano, and P. Pantaleo, “A fuzzy-logic based optical sensor for online weld defect-detection,” IEEE transactions on Industrial Informatics, vol. 1, no. 4, pp. 259–273, 2005.</ref>.

=== Acoustic analysis ===
With <ref name="acoustic_analysis">E. Cayo and S. C. Alfaro, “A non-intrusive GMA welding process quality monitoring system using acoustic sensing,” Sensors, vol. 9, no. 9, pp. 7150–7166, 2009.</ref>, the author discovered that arc ignition is characterised by a distinct sound and therefore it is possible to capture the frequency of ignition through acoustic sensors. A tolerance band was set and both ignition frequency and sound pressure level were used to monitor the quality of the weld.

<div><ul>
<li style="display: inline-block;"> [[File:acoustic_analysis.png|thumb|400px|none|Figure 14: Acoustic analysis <ref name="acoustic_analysis"></ref>]] </li>
</ul></div>

Other research into using acoustic signal to monitor that can be of interest are <ref>M. Fidali, “Detection of welding process instabilities using acoustic signals,” in International Congress on Technical Diagnostic, pp. 191–201, Springer, 2016.</ref>, <ref>L. Zhang, A. C. Basantes-Defaz, D. Ozevin, and E. Indacochea, “Real-time monitoring of welding process using air-coupled ultrasonics and acoustic emission,” The International Journal of Advanced Manufacturing Technology, vol. 101, no. 5-8, pp. 1623–1634, 2019.</ref>.

=== Electrical signal analysis ===
With <ref name="electrical_analysis_1">A. Sumesh, K. Rameshkumar, A. Raja, K. Mohandas, A. Santhakumari, and R. Shyambabu, “Establishing correlation between current and voltage signatures of the arc and weld defects in GMAW process,” Arabian Journal for Science and Engineering, vol. 42, no. 11, pp. 4649–4665, 2017.</ref>, the author plot the probability density distribution of voltage signal and discovered that the stable weld's signal concentrate in one region while unstable signal spread out. While in the paper, the authors did not developed a monitoring method using their findings, it shows that there is a significant differences between stable and unstable weld's electrical signature.

<div><ul>
<li style="display: inline-block;"> [[File:Electrical analysis pdd.png|thumb|400px|none|Figure 15: Voltage PDD <ref name="electrical_analysis_1"></ref>]] </li>
</ul></div>

In <ref name="electrical_analysis_2">S. Simpson, “Signature images for arc welding fault detection,” Science and Technology of Welding and Joining, vol. 12, no. 6, pp. 481–486, 2007.</ref>, the authors shows that by plotting the signature image of voltage and current scatter plot, both arcing and short-circuiting region's signature image can notify a defects welds. A more visual representation of short circuiting and arcing region can be found in [[#Gas metal arc welding|Gas metal arc welding section]].

<div><ul>
<li style="display: inline-block;"> [[File:Electrical voltage signature.png|thumb|400px|none|Figure 16a: Electrical voltage signature of stable weld <ref name="electrical_analysis_2"></ref>]] </li>
<li style="display: inline-block;"> [[File:Electrical signature 2.png|thumb|400px|none|Figure 16b: Electrical voltage signature of unstable weld <ref name="electrical_analysis_2"></ref>]] </li>
</ul></div>

Other research into using electrical signal to monitor that can be of interest are <ref>R. Madigan, “Arc sensing for defects in constant-voltage gas metal arc welding,” Welding Journal, vol. 78, pp. 322S–328S, 1999.</ref>, <ref>Y. Huang, K. Wang, Z. Zhou, X. Zhou, and J. Fang, “Stability evaluation of short-circuiting gas metal arc welding based on ensemble empirical mode decomposition,” Measurement Science and Technology, vol. 28, no. 3, p. 035006, 2017.</ref>,<ref>Z. Zhang, X. Chen, H. Chen, J. Zhong, and S. Chen, “Online welding quality monitoring based on feature extraction of arc voltage signal,” The International Journal of Advanced Manufacturing Technology, vol. 70, no. 9-12, pp. 1661–1671, 2014</ref>,<ref>X. Li and S. Simpson, “Parametric approach to positional fault detection in short arc welding,” Science and Technology of welding and joining, vol. 14, no. 2, pp. 146–151, 2009.</ref>, and <ref>E. Wei, D. Farson, R. Richardson, and H. Ludewig, “Detection of weld surface porosity by statistical analysis of arc current in gas metal arc welding,” Journal of Manufacturing Processes, vol. 3, no. 1, pp. 50–59, 2001.</ref>.

== Experiment setup ==
The welding system uses TPS 320i C Pulse (figure 15a), produced by Fronius, to supply power, feed wire, and system cooling. For movement control of the welding torch, the system use ABB’s IRB 1520ID (figure 15b), which is a dedicated arc welding robot. The Data Acquisition device is NI's DAQ. Each channel (voltage and current) is sampled at 25kHz while the dominant frequency for voltage and current signal are around 130Hz (for CMT process) and 190Hz (for PMC process).

[[File:Fronius_power_supply.jpg |thumb|Figure 17a: Fronius' Power Supply]]
[[File:Welding_robot2.jpg |thumb|Figure 17b: ABB's welding robot]]

[[File:Ni_daq.jpg|thumb|Figure 16: NI's DAQ device]]

== Statistical Analysis ==

=== Preprocessing ===
As the captured signal coming from Fronius' power supply, the signals also pick up switching power noise. The noise will interfere with the accuracy of the analysis if not filtered. In this project, we applied a 1000Hz digital Low Pass Butterworth filter. The effect of the filter can be seen in figure 18, while the comparation of using different filters can be seen in the gallery.

<div><ul>
<li style="display: inline-block;"> [[File:Filter_effect.png |thumb|750px|none|Figure 18: Effect of 1000Hz low pass filter]] </li>
<li style="display: inline-block;"> [[File:Freq_response.png |thumb|750px|none|Figure 19: Frequency response of the in-use filter]] </li>
</ul></div>

<gallery mode="packed-hover" caption="Gallery: Comparison of different filtering method" perrow=2>
File:Effect_of_filtering2.png|In use filter vs 300Hz low pass filter
File:Effect_of_filtering3.png|In use filter vs 500Hz low pass filter
File:Effect_of_filtering4.png|In use filter vs 750Hz low pass filter
File:Effect_of_filtering5.png|In use filter vs 2000Hz low pass filter
</gallery>

=== Power Spectral Density ===
Our goal for this project is to detect the defect welds and predict the layer heights (the contact tip to work piece distance, CTWD, to be more accurate) in real time (refer to [[#Relating Contact Tip to Work-piece Distance to layer height|this section]] to understand the relationship between Contact Tip to Work-piece Distance and Layer height). An analysis of the current power spectral density (PSD) shows that there is an inverse proportional relationship between the CTWD of a weld and the frequency where its '''current signal''' PSD reach the peak, and also the magnitude of the peak. Since the peak of PSD can be fairly sensitive, we only consider the frequency of the peak PSD.

Figure 20 shows the PSD plot of voltage, current, and power signal of the weld using PMC process. In this experiment, we created multiple one-line, 8 seconds welds. These eight-second welds then get segmented into smaller two-second chunk to increase the number of data and decrease the processing time.

<div><ul>
<li style="display: inline-block;"> [[File:Psd pmc good bad all lowfreq.png |thumb|750px|none|Figure 20: Power spectral density of PMC process]] </li>
<li style="display: inline-block;"> [[File:One line weld.jpg |thumb|578px|none|Figure 21: Simple one-line weld experiment]] </li>
<li style="display: inline-block;"> [[File:Cmt_psd_stable_allCTWD_zoomed.png |thumb|750px|none|Figure 22: Simple one-line weld experiment]] </li>

</ul></div>

It seems like the inverse proportional relationship holds true for most cases. Other experiments showed that for a short period at the beginning and ending of the weld, the signal can behave differently relative to the rest of the weld. Currently, to migrate with this problem, data will not be collected and analysed (both offline and online) at these periods. However, the above results seem to only hold true for PMC processes. With CMT process, the current PSD plot indicates no relationship between the CTWD and the frequency at peak (figure 24).

For the next stage of the project, our plan is to develop a linear regression model to predict the CTWD for PMC process. With CMT process, we will have to look into different analysis techniques.

=== Support Vector Machine ===
By looking at the average voltage and average current of each segment of weld while making a simple straight path (PMC process), we realised there is a visible separation margin between stable and unstable welds. Using SVM with RBF kernel, we were able to correctly classify the quality of the weld with accuracy of 98%.

However, with CMT process, the separation cannot be observed.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_pmcNoOxide_rbf.png |thumb|750px|none|Figure 23: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:CmtNoOxide_average_voltage_current.png |thumb|750px|none|Figure 24: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

=== CTWD Regression ===
With CTWD prediction, the result will only be useful if it is done on-the-line while making the part. And therefore, as oppose to defect detection, we cannot have a short calibration process every constant time to check for worn contact tip. By analysing the data while building a real part, we found that the average voltage (for PMC process) and average current (for CMT process) seems to have the strongest linear relationship with the CTWD. The acceptable error for CTWD prediction is +/- 1 mm (according to AML3D requirement). For both PMC and CMT process, our root means square error is less than 1 (PMC: 0.68 and CMT: 0.83), which satisfied the requirement.

<div><ul>
<li style="display: inline-block;"> [[File:Linear_regression_pmc_wall.png |thumb|750px|none|Figure 25: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:Linear_regression_cmt_wall.png |thumb|750px|none|Figure 26: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

== Result discussion ==
These results show that there is a strong possibility of defect detection and CTWD prediction in welding. However, these initial findings are still a proof of concept. In the next stage of our research, we will refine the algorithm and integrate into AML's pipeline.

== References ==
<references />

Projects:2019s2-20001 Using Machine Learning to Determine Deposit Height and Defects for Wire + Arc Additive Manufacture (3D printing)

2019-10-15T05:13:51Z

A1713568: /* Introduction */

[[Category:Projects]]
[[Category:Final Year Projects]]
[[Category:2019s2|24501]]
== Introduction ==
AML3D is one of the leading company in the Wire Arc Additive Manufacturing (WAAM) industry. In short, WAAM is the Metal 3D printing. The technology has the ability to reduce up to 75% of manufacture time, cost, and material. Currently, WAAM is benefiting other industry such as: mining, defence, marine, and aviation.

[[File:Welding baby.gif|upright=2.0|Welding using robotic arm|none|thumb]]
[[File:Laser baby.gif|upright=2.0|User laser sensor to find the CTWD|none|thumb]]

[[File:aml3d_logo.jpg|100px]]
=== Project team ===
==== Project students ====
* Anh Tran
* Nhat Nguyen
==== Supervisors ====
* Dr. Brian Ng
* Dr. Paul Colegrove (AML3D, sponsored company)
=== Objectives ===
The objective of this project is to increase the efficiency of the manufacture process at AML3D. In order to do so, the team will investigate into the possibility of automating and optimising the quality control processes. The two quality control processes that are currently being implemented at AML3D are measuring layer height using laser sensors, and human supervision for detecting defects. These processes add overhead into production time and usage of human resource, which is not desired. To achieve the goal, machine learning methods will be used extensively to analyse the electrical signatures of the weld process.

== Background ==
=== Wire Arc Additive Manufacturing ===
Wire and Arc Additive Manufacturing (WAAM) is a type of additive manufacturing that uses electric arc as the heat source and material wire to feed the manufacture process <ref name="WAAM"> S. W. Williams, F. Martina, A. C. Addison, J. Ding, G. Pardal & P. Colegrove (2016) Wire + Arc Additive Manufacturing, Materials Science and Technology, 32:7, 641-647, DOI: 10.1179/1743284715Y.0000000073 </ref>. WAAM has been investigated since the 1990s <ref name="WAAM" />, but only recently that it received more attention from the manufacture world.

Its significant comes from the ability to manufacture complex model with less time and less material. Figure 1a and 1b show real parts that was manufactured by AML3D. Such custom made parts might takes months before be ready to be shipped, but with WAAM, the production time can reduce down to weeks. Currently, the industries that benefit the most from WAAM are maritime and aerospace.

Similar to other additive manufacture methods, WAAM achieves such results by sliding the models into multiple layers, and then build the model layer by layer. The movement control is normally handled by a robotic arm (AML3D uses ABB's Arc Welder robot), and the welding path is generated by a Computer Aid Manufacture (CAM) software.

[[File:welding_part1.JPG|thumb|Figure 1a: A propeller]]
[[File:welding_part2.jpg|thumb|Figure 1b: Another part of the ship]]
[[File:abb_robot.jpg|thumb|Figure 2: ABB's IRB 1520ID Arc Welder]]

=== Gas metal arc welding ===
<div><ul>
<li style="display: inline-block;"> [[File:gmaw_welding.png|thumb|500px|none|Figure 2: GMAW welding process <ref name="welding_book"> Norrish, J. (2006). Advanced welding processes. Elsevier. </ref>]] </li>
</ul></div>

Gas metal arc welding (GMAW), or also known as metal inert gas (MIG), is the welding process used at AML3D for WAAM. Due to its high deposition rate and economic benefits, GMAW has became more popular. We will explore two variants of GMAW, which are Pulse Multi Control and Cold Metal Transfer. Both processes are developed by Fronius, an Austrian welding company. Note that it is not required to understand the welding physics to follow this wiki. The next two following sections' purpose is to highlight the high variability, high dynamic nature in welding.

[[File:fronius.png|thumb|Figure 3: Fronius logo]]

==== Pulse Multi Control ====
Pulse Multi Control (PMC) welding is Fronius' modified Pulsed GMAW. The advantage of Pulsed GMAW is the ability to control the metal droplet transfer in welding. Note that from figure 4, the metal droplet is characterised by a downward slope of the current pulse. Figure 5 shows the current signal captured from experiments conducted at AML3D (PMC is used).

<div><ul>
<li style="display: inline-block;"> [[File:Pulsed_gmaw_transfer_mode.png|thumb|500px|none|Figure 4: Relationship between pulse current and droplet transfer<ref name="welding_book" />]] </li>
<li style="display: inline-block;"> [[File:Pmc_current_in_time_domain.png|thumb|650px|none|Figure 5: Current signal captured from experiments conducted at AML3D (PMC process)]] </li>
</ul></div>
==== Cold Metal Transfer ====
Cold Metal Transfer (CMT) is a complex welding process, also developed by Fronius. By detecting a short circuit (mode C and D in figure 8), CMT can make adjustment such as retracting the welding material to cool down, and therefore create a smoother, more stable weld. The complexity of CMT can be seen in the current signal of the process (figure 9). Compare to PMC, CMT's signal has more variation during one signal period.

<div><ul>
<li style="display: inline-block;"> [[File:Transfer_mode.png|thumb|800px|none|Figure 8: Mechanism of metal droplet transfer mode]] </li>
<li style="display: inline-block;"> [[File:Cmt1_current_in_time_domain.png|thumb|735px|none|Figure 9: Current signal captured from experiments conducted at AML3D (CMT process)]] </li>
</ul></div>
=== Machine Learning ===
Machine learning (or statistical learning) refer to a set of tools to give data scientists an insight into the data and make better decisions based on those information <ref name="introduction_ml_book"> James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An introduction to statistical learning (Vol. 112, p. 18). New York: Springer. </ref>. In this project, we will explore the capabilities of classical machine learning methods (i.e anything but Deep Learning) due to the limitation of our available data. By the time of writing this wiki, we have only explored Support Vector Machine.

==== Support Vector Machine ====
The main idea of Support Vector Machine (SVM) is to draw hyperplanes that best separate multiple classes of data. The original SVM judges best decision hyperplane based on the separation margin between data classes (figure 10). However, there are multiple variants of SVM, such as hard/soft margin SVM, Nu-SVM, One class SVM (to solve anomaly detection problem). In our project, we choose to use a soft margin implementation of SVM. The benefit of soft margin SVM (as oppose to hard margin SVM) is that by allowing some misclassifications, we get a wider margin and therefore the solution becomes more "generalise" (i.e work better with a new set of dataset that is independent from the training dataset).

<div><ul>
<li style="display: inline-block;"> [[File:svm_margin.png|thumb|400px|none|Figure 10: Separation margin <ref name="bishop_book"> Bishop, C. M. (2006). Pattern recognition and machine learning. Springer. </ref>]] </li>
</ul></div>

==== Kernel tricks ====
Figure 10 is of linearly separable cases for SVM. For dataset that is not easy to separate with a hyperplane in their original features space (such as figure 11), the data often get projected into a higher dimension space where it is easier to separate (figure 14). Such technique is called the Kernel Tricks. It is important to note that Kernel Tricks does not actually "map" the data from '''X''' dimension to '''Z''' dimension, but rather pairwise compare the similarity of the input space in '''Z''' dimension.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_nonseparable.png |thumb|350px|none|Figure 11: Nonseparable dataset <ref name="bishop_book"></ref>]] </li>
<li style="display: inline-block;"> [[File:Svm_kernel_trick.png |thumb|650px|none|Figure 12: Simplified example of the Kernel tricks]] </li>
</ul></div>

== Previous studies ==
There are many researches into defect detection in GMAW process. The three most dominant analysis methods are spectroscopic analysis, acoustic analysis, and electrical signal analysis.

=== Spectroscopic analysis ===
In <ref name="spectroscopic_analysis">D. Bebiano and S. Alfaro, “A weld defects detection system based on a spectrometer,” Sensors, vol. 9, no. 4, pp. 2851–2861, 2009.</ref>, the authors captured the intensity of radiation emission of electric arc. It is obvious from figure 13 that good weld's intensity stays stable while defect one's is characterised by a sudden peak. Therefore, a threshold is set to detect the defect welds.

<div><ul>
<li style="display: inline-block;"> [[File:spectroscopic_analysis.png|thumb|400px|none|Figure 13: Spectroscopic analysis <ref name="spectroscopic_analysis"></ref>]] </li>
</ul></div>

Other research into using spectroscopic data to monitor that can be of interest is <ref>D. Naso, B. Turchiano, and P. Pantaleo, “A fuzzy-logic based optical sensor for online weld defect-detection,” IEEE transactions on Industrial Informatics, vol. 1, no. 4, pp. 259–273, 2005.</ref>.

=== Acoustic analysis ===
With <ref name="acoustic_analysis">E. Cayo and S. C. Alfaro, “A non-intrusive GMA welding process quality monitoring system using acoustic sensing,” Sensors, vol. 9, no. 9, pp. 7150–7166, 2009.</ref>, the author discovered that arc ignition is characterised by a distinct sound and therefore it is possible to capture the frequency of ignition through acoustic sensors. A tolerance band was set and both ignition frequency and sound pressure level were used to monitor the quality of the weld.

<div><ul>
<li style="display: inline-block;"> [[File:acoustic_analysis.png|thumb|400px|none|Figure 14: Acoustic analysis <ref name="acoustic_analysis"></ref>]] </li>
</ul></div>

Other research into using acoustic signal to monitor that can be of interest are <ref>M. Fidali, “Detection of welding process instabilities using acoustic signals,” in International Congress on Technical Diagnostic, pp. 191–201, Springer, 2016.</ref>, <ref>L. Zhang, A. C. Basantes-Defaz, D. Ozevin, and E. Indacochea, “Real-time monitoring of welding process using air-coupled ultrasonics and acoustic emission,” The International Journal of Advanced Manufacturing Technology, vol. 101, no. 5-8, pp. 1623–1634, 2019.</ref>.

=== Electrical signal analysis ===
With <ref name="electrical_analysis_1">A. Sumesh, K. Rameshkumar, A. Raja, K. Mohandas, A. Santhakumari, and R. Shyambabu, “Establishing correlation between current and voltage signatures of the arc and weld defects in GMAW process,” Arabian Journal for Science and Engineering, vol. 42, no. 11, pp. 4649–4665, 2017.</ref>, the author plot the probability density distribution of voltage signal and discovered that the stable weld's signal concentrate in one region while unstable signal spread out. While in the paper, the authors did not developed a monitoring method using their findings, it shows that there is a significant differences between stable and unstable weld's electrical signature.

<div><ul>
<li style="display: inline-block;"> [[File:Electrical analysis pdd.png|thumb|400px|none|Figure 15: Voltage PDD <ref name="electrical_analysis_1"></ref>]] </li>
</ul></div>

In <ref name="electrical_analysis_2">S. Simpson, “Signature images for arc welding fault detection,” Science and Technology of Welding and Joining, vol. 12, no. 6, pp. 481–486, 2007.</ref>, the authors shows that by plotting the signature image of voltage and current scatter plot, both arcing and short-circuiting region's signature image can notify a defects welds. A more visual representation of short circuiting and arcing region can be found in [[#Gas metal arc welding|Gas metal arc welding section]].

<div><ul>
<li style="display: inline-block;"> [[File:Electrical voltage signature.png|thumb|400px|none|Figure 16a: Electrical voltage signature of stable weld <ref name="electrical_analysis_2"></ref>]] </li>
<li style="display: inline-block;"> [[File:Electrical signature 2.png|thumb|400px|none|Figure 16b: Electrical voltage signature of unstable weld <ref name="electrical_analysis_2"></ref>]] </li>
</ul></div>

Other research into using electrical signal to monitor that can be of interest are <ref>R. Madigan, “Arc sensing for defects in constant-voltage gas metal arc welding,” Welding Journal, vol. 78, pp. 322S–328S, 1999.</ref>, <ref>Y. Huang, K. Wang, Z. Zhou, X. Zhou, and J. Fang, “Stability evaluation of short-circuiting gas metal arc welding based on ensemble empirical mode decomposition,” Measurement Science and Technology, vol. 28, no. 3, p. 035006, 2017.</ref>,<ref>Z. Zhang, X. Chen, H. Chen, J. Zhong, and S. Chen, “Online welding quality monitoring based on feature extraction of arc voltage signal,” The International Journal of Advanced Manufacturing Technology, vol. 70, no. 9-12, pp. 1661–1671, 2014</ref>,<ref>X. Li and S. Simpson, “Parametric approach to positional fault detection in short arc welding,” Science and Technology of welding and joining, vol. 14, no. 2, pp. 146–151, 2009.</ref>, and <ref>E. Wei, D. Farson, R. Richardson, and H. Ludewig, “Detection of weld surface porosity by statistical analysis of arc current in gas metal arc welding,” Journal of Manufacturing Processes, vol. 3, no. 1, pp. 50–59, 2001.</ref>.

== Experiment setup ==
The welding system uses TPS 320i C Pulse (figure 15a), produced by Fronius, to supply power, feed wire, and system cooling. For movement control of the welding torch, the system use ABB’s IRB 1520ID (figure 15b), which is a dedicated arc welding robot. The Data Acquisition device is NI's DAQ. Each channel (voltage and current) is sampled at 25kHz while the dominant frequency for voltage and current signal are around 130Hz (for CMT process) and 190Hz (for PMC process).

[[File:Fronius_power_supply.jpg |thumb|Figure 17a: Fronius' Power Supply]]
[[File:Welding_robot2.jpg |thumb|Figure 17b: ABB's welding robot]]

[[File:Ni_daq.jpg|thumb|Figure 16: NI's DAQ device]]

== Statistical Analysis ==

=== Preprocessing ===
As the captured signal coming from Fronius' power supply, the signals also pick up switching power noise. The noise will interfere with the accuracy of the analysis if not filtered. In this project, we applied a 1000Hz digital Low Pass Butterworth filter. The effect of the filter can be seen in figure 18, while the comparation of using different filters can be seen in the gallery.

<div><ul>
<li style="display: inline-block;"> [[File:Filter_effect.png |thumb|750px|none|Figure 18: Effect of 1000Hz low pass filter]] </li>
<li style="display: inline-block;"> [[File:Freq_response.png |thumb|750px|none|Figure 19: Frequency response of the in-use filter]] </li>
</ul></div>

<gallery mode="packed-hover" caption="Gallery: Comparison of different filtering method" perrow=2>
File:Effect_of_filtering2.png|In use filter vs 300Hz low pass filter
File:Effect_of_filtering3.png|In use filter vs 500Hz low pass filter
File:Effect_of_filtering4.png|In use filter vs 750Hz low pass filter
File:Effect_of_filtering5.png|In use filter vs 2000Hz low pass filter
</gallery>

=== Power Spectral Density ===
Our goal for this project is to detect the defect welds and predict the layer heights (the contact tip to work piece distance, CTWD, to be more accurate) in real time (refer to [[#Relating Contact Tip to Work-piece Distance to layer height|this section]] to understand the relationship between Contact Tip to Work-piece Distance and Layer height). An analysis of the current power spectral density (PSD) shows that there is an inverse proportional relationship between the CTWD of a weld and the frequency where its '''current signal''' PSD reach the peak, and also the magnitude of the peak. Since the peak of PSD can be fairly sensitive, we only consider the frequency of the peak PSD.

Figure 20 shows the PSD plot of voltage, current, and power signal of the weld using PMC process. In this experiment, we created multiple one-line, 8 seconds welds. These eight-second welds then get segmented into smaller two-second chunk to increase the number of data and decrease the processing time.

<div><ul>
<li style="display: inline-block;"> [[File:Psd pmc good bad all lowfreq.png |thumb|750px|none|Figure 20: Power spectral density of PMC process]] </li>
<li style="display: inline-block;"> [[File:One line weld.jpg |thumb|578px|none|Figure 21: Simple one-line weld experiment]] </li>
<li style="display: inline-block;"> [[File:Cmt_psd_stable_allCTWD_zoomed.png |thumb|750px|none|Figure 22: Simple one-line weld experiment]] </li>

</ul></div>

It seems like the inverse proportional relationship holds true for most cases. Other experiments showed that for a short period at the beginning and ending of the weld, the signal can behave differently relative to the rest of the weld. Currently, to migrate with this problem, data will not be collected and analysed (both offline and online) at these periods. However, the above results seem to only hold true for PMC processes. With CMT process, the current PSD plot indicates no relationship between the CTWD and the frequency at peak (figure 24).

For the next stage of the project, our plan is to develop a linear regression model to predict the CTWD for PMC process. With CMT process, we will have to look into different analysis techniques.

=== Support Vector Machine ===
By looking at the average voltage and average current of each segment of weld while making a simple straight path (PMC process), we realised there is a visible separation margin between stable and unstable welds. Using SVM with RBF kernel, we were able to correctly classify the quality of the weld with accuracy of 98%.

However, with CMT process, the separation cannot be observed.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_pmcNoOxide_rbf.png |thumb|750px|none|Figure 23: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:CmtNoOxide_average_voltage_current.png |thumb|750px|none|Figure 24: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

=== CTWD Regression ===
With CTWD prediction, the result will only be useful if it is done on-the-line while making the part. And therefore, as oppose to defect detection, we cannot have a short calibration process every constant time to check for worn contact tip. By analysing the data while building a real part, we found that the average voltage (for PMC process) and average current (for CMT process) seems to have the strongest linear relationship with the CTWD. The acceptable error for CTWD prediction is +/- 1 mm (according to AML3D requirement). For both PMC and CMT process, our root means square error is less than 1 (PMC: 0.68 and CMT: 0.83), which satisfied the requirement.

<div><ul>
<li style="display: inline-block;"> [[File:Linear_regression_pmc_wall.png |thumb|750px|none|Figure 25: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:Linear_regression_cmt_wall.png |thumb|750px|none|Figure 26: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

== Result discussion ==
These results show that there is a strong possibility of defect detection and CTWD prediction in welding. However, these initial findings are still a proof of concept. In the next stage of our research, we will refine the algorithm and integrate into AML's pipeline.

== References ==
<references />

Projects:2019s2-20001 Using Machine Learning to Determine Deposit Height and Defects for Wire + Arc Additive Manufacture (3D printing)

2019-10-15T05:10:28Z

A1713568:

[[Category:Projects]]
[[Category:Final Year Projects]]
[[Category:2019s2|24501]]
== Introduction ==
AML3D is one of the leading company in the Wire Arc Additive Manufacturing (WAAM) industry. In short, WAAM is the Metal 3D printing. The technology has the ability to reduce up to 75% of manufacture time, cost, and material. Currently, WAAM is benefiting other industry such as: mining, defence, marine, and aviation.

<gallery>
Welding baby.gif|Look at this bad boy welding robot
Laser baby.gif|Laser 'em dead baby boy!
</gallery>

[[File:aml3d_logo.jpg|100px]]
=== Project team ===
==== Project students ====
* Anh Tran
* Nhat Nguyen
==== Supervisors ====
* Dr. Brian Ng
* Dr. Paul Colegrove (AML3D, sponsored company)
=== Objectives ===
The objective of this project is to increase the efficiency of the manufacture process at AML3D. In order to do so, the team will investigate into the possibility of automating and optimising the quality control processes. The two quality control processes that are currently being implemented at AML3D are measuring layer height using laser sensors, and human supervision for detecting defects. These processes add overhead into production time and usage of human resource, which is not desired. To achieve the goal, machine learning methods will be used extensively to analyse the electrical signatures of the weld process.

== Background ==
=== Wire Arc Additive Manufacturing ===
Wire and Arc Additive Manufacturing (WAAM) is a type of additive manufacturing that uses electric arc as the heat source and material wire to feed the manufacture process <ref name="WAAM"> S. W. Williams, F. Martina, A. C. Addison, J. Ding, G. Pardal & P. Colegrove (2016) Wire + Arc Additive Manufacturing, Materials Science and Technology, 32:7, 641-647, DOI: 10.1179/1743284715Y.0000000073 </ref>. WAAM has been investigated since the 1990s <ref name="WAAM" />, but only recently that it received more attention from the manufacture world.

Its significant comes from the ability to manufacture complex model with less time and less material. Figure 1a and 1b show real parts that was manufactured by AML3D. Such custom made parts might takes months before be ready to be shipped, but with WAAM, the production time can reduce down to weeks. Currently, the industries that benefit the most from WAAM are maritime and aerospace.

Similar to other additive manufacture methods, WAAM achieves such results by sliding the models into multiple layers, and then build the model layer by layer. The movement control is normally handled by a robotic arm (AML3D uses ABB's Arc Welder robot), and the welding path is generated by a Computer Aid Manufacture (CAM) software.

[[File:welding_part1.JPG|thumb|Figure 1a: A propeller]]
[[File:welding_part2.jpg|thumb|Figure 1b: Another part of the ship]]
[[File:abb_robot.jpg|thumb|Figure 2: ABB's IRB 1520ID Arc Welder]]

=== Gas metal arc welding ===
<div><ul>
<li style="display: inline-block;"> [[File:gmaw_welding.png|thumb|500px|none|Figure 2: GMAW welding process <ref name="welding_book"> Norrish, J. (2006). Advanced welding processes. Elsevier. </ref>]] </li>
</ul></div>

Gas metal arc welding (GMAW), or also known as metal inert gas (MIG), is the welding process used at AML3D for WAAM. Due to its high deposition rate and economic benefits, GMAW has became more popular. We will explore two variants of GMAW, which are Pulse Multi Control and Cold Metal Transfer. Both processes are developed by Fronius, an Austrian welding company. Note that it is not required to understand the welding physics to follow this wiki. The next two following sections' purpose is to highlight the high variability, high dynamic nature in welding.

[[File:fronius.png|thumb|Figure 3: Fronius logo]]

==== Pulse Multi Control ====
Pulse Multi Control (PMC) welding is Fronius' modified Pulsed GMAW. The advantage of Pulsed GMAW is the ability to control the metal droplet transfer in welding. Note that from figure 4, the metal droplet is characterised by a downward slope of the current pulse. Figure 5 shows the current signal captured from experiments conducted at AML3D (PMC is used).

<div><ul>
<li style="display: inline-block;"> [[File:Pulsed_gmaw_transfer_mode.png|thumb|500px|none|Figure 4: Relationship between pulse current and droplet transfer<ref name="welding_book" />]] </li>
<li style="display: inline-block;"> [[File:Pmc_current_in_time_domain.png|thumb|650px|none|Figure 5: Current signal captured from experiments conducted at AML3D (PMC process)]] </li>
</ul></div>
==== Cold Metal Transfer ====
Cold Metal Transfer (CMT) is a complex welding process, also developed by Fronius. By detecting a short circuit (mode C and D in figure 8), CMT can make adjustment such as retracting the welding material to cool down, and therefore create a smoother, more stable weld. The complexity of CMT can be seen in the current signal of the process (figure 9). Compare to PMC, CMT's signal has more variation during one signal period.

<div><ul>
<li style="display: inline-block;"> [[File:Transfer_mode.png|thumb|800px|none|Figure 8: Mechanism of metal droplet transfer mode]] </li>
<li style="display: inline-block;"> [[File:Cmt1_current_in_time_domain.png|thumb|735px|none|Figure 9: Current signal captured from experiments conducted at AML3D (CMT process)]] </li>
</ul></div>
=== Machine Learning ===
Machine learning (or statistical learning) refer to a set of tools to give data scientists an insight into the data and make better decisions based on those information <ref name="introduction_ml_book"> James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An introduction to statistical learning (Vol. 112, p. 18). New York: Springer. </ref>. In this project, we will explore the capabilities of classical machine learning methods (i.e anything but Deep Learning) due to the limitation of our available data. By the time of writing this wiki, we have only explored Support Vector Machine.

==== Support Vector Machine ====
The main idea of Support Vector Machine (SVM) is to draw hyperplanes that best separate multiple classes of data. The original SVM judges best decision hyperplane based on the separation margin between data classes (figure 10). However, there are multiple variants of SVM, such as hard/soft margin SVM, Nu-SVM, One class SVM (to solve anomaly detection problem). In our project, we choose to use a soft margin implementation of SVM. The benefit of soft margin SVM (as oppose to hard margin SVM) is that by allowing some misclassifications, we get a wider margin and therefore the solution becomes more "generalise" (i.e work better with a new set of dataset that is independent from the training dataset).

<div><ul>
<li style="display: inline-block;"> [[File:svm_margin.png|thumb|400px|none|Figure 10: Separation margin <ref name="bishop_book"> Bishop, C. M. (2006). Pattern recognition and machine learning. Springer. </ref>]] </li>
</ul></div>

==== Kernel tricks ====
Figure 10 is of linearly separable cases for SVM. For dataset that is not easy to separate with a hyperplane in their original features space (such as figure 11), the data often get projected into a higher dimension space where it is easier to separate (figure 14). Such technique is called the Kernel Tricks. It is important to note that Kernel Tricks does not actually "map" the data from '''X''' dimension to '''Z''' dimension, but rather pairwise compare the similarity of the input space in '''Z''' dimension.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_nonseparable.png |thumb|350px|none|Figure 11: Nonseparable dataset <ref name="bishop_book"></ref>]] </li>
<li style="display: inline-block;"> [[File:Svm_kernel_trick.png |thumb|650px|none|Figure 12: Simplified example of the Kernel tricks]] </li>
</ul></div>

== Previous studies ==
There are many researches into defect detection in GMAW process. The three most dominant analysis methods are spectroscopic analysis, acoustic analysis, and electrical signal analysis.

=== Spectroscopic analysis ===
In <ref name="spectroscopic_analysis">D. Bebiano and S. Alfaro, “A weld defects detection system based on a spectrometer,” Sensors, vol. 9, no. 4, pp. 2851–2861, 2009.</ref>, the authors captured the intensity of radiation emission of electric arc. It is obvious from figure 13 that good weld's intensity stays stable while defect one's is characterised by a sudden peak. Therefore, a threshold is set to detect the defect welds.

<div><ul>
<li style="display: inline-block;"> [[File:spectroscopic_analysis.png|thumb|400px|none|Figure 13: Spectroscopic analysis <ref name="spectroscopic_analysis"></ref>]] </li>
</ul></div>

Other research into using spectroscopic data to monitor that can be of interest is <ref>D. Naso, B. Turchiano, and P. Pantaleo, “A fuzzy-logic based optical sensor for online weld defect-detection,” IEEE transactions on Industrial Informatics, vol. 1, no. 4, pp. 259–273, 2005.</ref>.

=== Acoustic analysis ===
With <ref name="acoustic_analysis">E. Cayo and S. C. Alfaro, “A non-intrusive GMA welding process quality monitoring system using acoustic sensing,” Sensors, vol. 9, no. 9, pp. 7150–7166, 2009.</ref>, the author discovered that arc ignition is characterised by a distinct sound and therefore it is possible to capture the frequency of ignition through acoustic sensors. A tolerance band was set and both ignition frequency and sound pressure level were used to monitor the quality of the weld.

<div><ul>
<li style="display: inline-block;"> [[File:acoustic_analysis.png|thumb|400px|none|Figure 14: Acoustic analysis <ref name="acoustic_analysis"></ref>]] </li>
</ul></div>

Other research into using acoustic signal to monitor that can be of interest are <ref>M. Fidali, “Detection of welding process instabilities using acoustic signals,” in International Congress on Technical Diagnostic, pp. 191–201, Springer, 2016.</ref>, <ref>L. Zhang, A. C. Basantes-Defaz, D. Ozevin, and E. Indacochea, “Real-time monitoring of welding process using air-coupled ultrasonics and acoustic emission,” The International Journal of Advanced Manufacturing Technology, vol. 101, no. 5-8, pp. 1623–1634, 2019.</ref>.

=== Electrical signal analysis ===
With <ref name="electrical_analysis_1">A. Sumesh, K. Rameshkumar, A. Raja, K. Mohandas, A. Santhakumari, and R. Shyambabu, “Establishing correlation between current and voltage signatures of the arc and weld defects in GMAW process,” Arabian Journal for Science and Engineering, vol. 42, no. 11, pp. 4649–4665, 2017.</ref>, the author plot the probability density distribution of voltage signal and discovered that the stable weld's signal concentrate in one region while unstable signal spread out. While in the paper, the authors did not developed a monitoring method using their findings, it shows that there is a significant differences between stable and unstable weld's electrical signature.

<div><ul>
<li style="display: inline-block;"> [[File:Electrical analysis pdd.png|thumb|400px|none|Figure 15: Voltage PDD <ref name="electrical_analysis_1"></ref>]] </li>
</ul></div>

In <ref name="electrical_analysis_2">S. Simpson, “Signature images for arc welding fault detection,” Science and Technology of Welding and Joining, vol. 12, no. 6, pp. 481–486, 2007.</ref>, the authors shows that by plotting the signature image of voltage and current scatter plot, both arcing and short-circuiting region's signature image can notify a defects welds. A more visual representation of short circuiting and arcing region can be found in [[#Gas metal arc welding|Gas metal arc welding section]].

<div><ul>
<li style="display: inline-block;"> [[File:Electrical voltage signature.png|thumb|400px|none|Figure 16a: Electrical voltage signature of stable weld <ref name="electrical_analysis_2"></ref>]] </li>
<li style="display: inline-block;"> [[File:Electrical signature 2.png|thumb|400px|none|Figure 16b: Electrical voltage signature of unstable weld <ref name="electrical_analysis_2"></ref>]] </li>
</ul></div>

Other research into using electrical signal to monitor that can be of interest are <ref>R. Madigan, “Arc sensing for defects in constant-voltage gas metal arc welding,” Welding Journal, vol. 78, pp. 322S–328S, 1999.</ref>, <ref>Y. Huang, K. Wang, Z. Zhou, X. Zhou, and J. Fang, “Stability evaluation of short-circuiting gas metal arc welding based on ensemble empirical mode decomposition,” Measurement Science and Technology, vol. 28, no. 3, p. 035006, 2017.</ref>,<ref>Z. Zhang, X. Chen, H. Chen, J. Zhong, and S. Chen, “Online welding quality monitoring based on feature extraction of arc voltage signal,” The International Journal of Advanced Manufacturing Technology, vol. 70, no. 9-12, pp. 1661–1671, 2014</ref>,<ref>X. Li and S. Simpson, “Parametric approach to positional fault detection in short arc welding,” Science and Technology of welding and joining, vol. 14, no. 2, pp. 146–151, 2009.</ref>, and <ref>E. Wei, D. Farson, R. Richardson, and H. Ludewig, “Detection of weld surface porosity by statistical analysis of arc current in gas metal arc welding,” Journal of Manufacturing Processes, vol. 3, no. 1, pp. 50–59, 2001.</ref>.

== Experiment setup ==
The welding system uses TPS 320i C Pulse (figure 15a), produced by Fronius, to supply power, feed wire, and system cooling. For movement control of the welding torch, the system use ABB’s IRB 1520ID (figure 15b), which is a dedicated arc welding robot. The Data Acquisition device is NI's DAQ. Each channel (voltage and current) is sampled at 25kHz while the dominant frequency for voltage and current signal are around 130Hz (for CMT process) and 190Hz (for PMC process).

[[File:Fronius_power_supply.jpg |thumb|Figure 17a: Fronius' Power Supply]]
[[File:Welding_robot2.jpg |thumb|Figure 17b: ABB's welding robot]]

[[File:Ni_daq.jpg|thumb|Figure 16: NI's DAQ device]]

== Statistical Analysis ==

=== Preprocessing ===
As the captured signal coming from Fronius' power supply, the signals also pick up switching power noise. The noise will interfere with the accuracy of the analysis if not filtered. In this project, we applied a 1000Hz digital Low Pass Butterworth filter. The effect of the filter can be seen in figure 18, while the comparation of using different filters can be seen in the gallery.

<div><ul>
<li style="display: inline-block;"> [[File:Filter_effect.png |thumb|750px|none|Figure 18: Effect of 1000Hz low pass filter]] </li>
<li style="display: inline-block;"> [[File:Freq_response.png |thumb|750px|none|Figure 19: Frequency response of the in-use filter]] </li>
</ul></div>

<gallery mode="packed-hover" caption="Gallery: Comparison of different filtering method" perrow=2>
File:Effect_of_filtering2.png|In use filter vs 300Hz low pass filter
File:Effect_of_filtering3.png|In use filter vs 500Hz low pass filter
File:Effect_of_filtering4.png|In use filter vs 750Hz low pass filter
File:Effect_of_filtering5.png|In use filter vs 2000Hz low pass filter
</gallery>

=== Power Spectral Density ===
Our goal for this project is to detect the defect welds and predict the layer heights (the contact tip to work piece distance, CTWD, to be more accurate) in real time (refer to [[#Relating Contact Tip to Work-piece Distance to layer height|this section]] to understand the relationship between Contact Tip to Work-piece Distance and Layer height). An analysis of the current power spectral density (PSD) shows that there is an inverse proportional relationship between the CTWD of a weld and the frequency where its '''current signal''' PSD reach the peak, and also the magnitude of the peak. Since the peak of PSD can be fairly sensitive, we only consider the frequency of the peak PSD.

Figure 20 shows the PSD plot of voltage, current, and power signal of the weld using PMC process. In this experiment, we created multiple one-line, 8 seconds welds. These eight-second welds then get segmented into smaller two-second chunk to increase the number of data and decrease the processing time.

<div><ul>
<li style="display: inline-block;"> [[File:Psd pmc good bad all lowfreq.png |thumb|750px|none|Figure 20: Power spectral density of PMC process]] </li>
<li style="display: inline-block;"> [[File:One line weld.jpg |thumb|578px|none|Figure 21: Simple one-line weld experiment]] </li>
<li style="display: inline-block;"> [[File:Cmt_psd_stable_allCTWD_zoomed.png |thumb|750px|none|Figure 22: Simple one-line weld experiment]] </li>

</ul></div>

It seems like the inverse proportional relationship holds true for most cases. Other experiments showed that for a short period at the beginning and ending of the weld, the signal can behave differently relative to the rest of the weld. Currently, to migrate with this problem, data will not be collected and analysed (both offline and online) at these periods. However, the above results seem to only hold true for PMC processes. With CMT process, the current PSD plot indicates no relationship between the CTWD and the frequency at peak (figure 24).

For the next stage of the project, our plan is to develop a linear regression model to predict the CTWD for PMC process. With CMT process, we will have to look into different analysis techniques.

=== Support Vector Machine ===
By looking at the average voltage and average current of each segment of weld while making a simple straight path (PMC process), we realised there is a visible separation margin between stable and unstable welds. Using SVM with RBF kernel, we were able to correctly classify the quality of the weld with accuracy of 98%.

However, with CMT process, the separation cannot be observed.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_pmcNoOxide_rbf.png |thumb|750px|none|Figure 23: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:CmtNoOxide_average_voltage_current.png |thumb|750px|none|Figure 24: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

=== CTWD Regression ===
With CTWD prediction, the result will only be useful if it is done on-the-line while making the part. And therefore, as oppose to defect detection, we cannot have a short calibration process every constant time to check for worn contact tip. By analysing the data while building a real part, we found that the average voltage (for PMC process) and average current (for CMT process) seems to have the strongest linear relationship with the CTWD. The acceptable error for CTWD prediction is +/- 1 mm (according to AML3D requirement). For both PMC and CMT process, our root means square error is less than 1 (PMC: 0.68 and CMT: 0.83), which satisfied the requirement.

<div><ul>
<li style="display: inline-block;"> [[File:Linear_regression_pmc_wall.png |thumb|750px|none|Figure 25: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:Linear_regression_cmt_wall.png |thumb|750px|none|Figure 26: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

== Result discussion ==
These results show that there is a strong possibility of defect detection and CTWD prediction in welding. However, these initial findings are still a proof of concept. In the next stage of our research, we will refine the algorithm and integrate into AML's pipeline.

== References ==
<references />

File:Laser baby.gif

2019-10-15T05:10:01Z

A1713568:

File:Welding baby.gif

2019-10-15T05:05:51Z

A1713568:

Projects:2019s2-20001 Using Machine Learning to Determine Deposit Height and Defects for Wire + Arc Additive Manufacture (3D printing)

2019-10-15T04:47:41Z

A1713568: /* Experiment setup */

[[Category:Projects]]
[[Category:Final Year Projects]]
[[Category:2019s2|24501]]
== Introduction ==
AML3D is one of the leading company in the Wire Arc Additive Manufacturing (WAAM) industry. In short, WAAM is the Metal 3D printing. The technology has the ability to reduce up to 75% of manufacture time, cost, and material. Currently, WAAM is benefiting other industry such as: mining, defence, marine, and aviation.

[[File:aml3d_logo.jpg|100px]]
=== Project team ===
==== Project students ====
* Anh Tran
* Nhat Nguyen
==== Supervisors ====
* Dr. Brian Ng
* Dr. Paul Colegrove (AML3D, sponsored company)
=== Objectives ===
The objective of this project is to increase the efficiency of the manufacture process at AML3D. In order to do so, the team will investigate into the possibility of automating and optimising the quality control processes. The two quality control processes that are currently being implemented at AML3D are measuring layer height using laser sensors, and human supervision for detecting defects. These processes add overhead into production time and usage of human resource, which is not desired. To achieve the goal, machine learning methods will be used extensively to analyse the electrical signatures of the weld process.

== Background ==
=== Wire Arc Additive Manufacturing ===
Wire and Arc Additive Manufacturing (WAAM) is a type of additive manufacturing that uses electric arc as the heat source and material wire to feed the manufacture process <ref name="WAAM"> S. W. Williams, F. Martina, A. C. Addison, J. Ding, G. Pardal & P. Colegrove (2016) Wire + Arc Additive Manufacturing, Materials Science and Technology, 32:7, 641-647, DOI: 10.1179/1743284715Y.0000000073 </ref>. WAAM has been investigated since the 1990s <ref name="WAAM" />, but only recently that it received more attention from the manufacture world.

Its significant comes from the ability to manufacture complex model with less time and less material. Figure 1a and 1b show real parts that was manufactured by AML3D. Such custom made parts might takes months before be ready to be shipped, but with WAAM, the production time can reduce down to weeks. Currently, the industries that benefit the most from WAAM are maritime and aerospace.

Similar to other additive manufacture methods, WAAM achieves such results by sliding the models into multiple layers, and then build the model layer by layer. The movement control is normally handled by a robotic arm (AML3D uses ABB's Arc Welder robot), and the welding path is generated by a Computer Aid Manufacture (CAM) software.

[[File:welding_part1.JPG|thumb|Figure 1a: A propeller]]
[[File:welding_part2.jpg|thumb|Figure 1b: Another part of the ship]]
[[File:abb_robot.jpg|thumb|Figure 2: ABB's IRB 1520ID Arc Welder]]

=== Gas metal arc welding ===
<div><ul>
<li style="display: inline-block;"> [[File:gmaw_welding.png|thumb|500px|none|Figure 2: GMAW welding process <ref name="welding_book"> Norrish, J. (2006). Advanced welding processes. Elsevier. </ref>]] </li>
</ul></div>

Gas metal arc welding (GMAW), or also known as metal inert gas (MIG), is the welding process used at AML3D for WAAM. Due to its high deposition rate and economic benefits, GMAW has became more popular. We will explore two variants of GMAW, which are Pulse Multi Control and Cold Metal Transfer. Both processes are developed by Fronius, an Austrian welding company. Note that it is not required to understand the welding physics to follow this wiki. The next two following sections' purpose is to highlight the high variability, high dynamic nature in welding.

[[File:fronius.png|thumb|Figure 3: Fronius logo]]

==== Pulse Multi Control ====
Pulse Multi Control (PMC) welding is Fronius' modified Pulsed GMAW. The advantage of Pulsed GMAW is the ability to control the metal droplet transfer in welding. Note that from figure 4, the metal droplet is characterised by a downward slope of the current pulse. Figure 5 shows the current signal captured from experiments conducted at AML3D (PMC is used).

<div><ul>
<li style="display: inline-block;"> [[File:Pulsed_gmaw_transfer_mode.png|thumb|500px|none|Figure 4: Relationship between pulse current and droplet transfer<ref name="welding_book" />]] </li>
<li style="display: inline-block;"> [[File:Pmc_current_in_time_domain.png|thumb|650px|none|Figure 5: Current signal captured from experiments conducted at AML3D (PMC process)]] </li>
</ul></div>
==== Cold Metal Transfer ====
Cold Metal Transfer (CMT) is a complex welding process, also developed by Fronius. By detecting a short circuit (mode C and D in figure 8), CMT can make adjustment such as retracting the welding material to cool down, and therefore create a smoother, more stable weld. The complexity of CMT can be seen in the current signal of the process (figure 9). Compare to PMC, CMT's signal has more variation during one signal period.

<div><ul>
<li style="display: inline-block;"> [[File:Transfer_mode.png|thumb|800px|none|Figure 8: Mechanism of metal droplet transfer mode]] </li>
<li style="display: inline-block;"> [[File:Cmt1_current_in_time_domain.png|thumb|735px|none|Figure 9: Current signal captured from experiments conducted at AML3D (CMT process)]] </li>
</ul></div>
=== Machine Learning ===
Machine learning (or statistical learning) refer to a set of tools to give data scientists an insight into the data and make better decisions based on those information <ref name="introduction_ml_book"> James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An introduction to statistical learning (Vol. 112, p. 18). New York: Springer. </ref>. In this project, we will explore the capabilities of classical machine learning methods (i.e anything but Deep Learning) due to the limitation of our available data. By the time of writing this wiki, we have only explored Support Vector Machine.

==== Support Vector Machine ====
The main idea of Support Vector Machine (SVM) is to draw hyperplanes that best separate multiple classes of data. The original SVM judges best decision hyperplane based on the separation margin between data classes (figure 10). However, there are multiple variants of SVM, such as hard/soft margin SVM, Nu-SVM, One class SVM (to solve anomaly detection problem). In our project, we choose to use a soft margin implementation of SVM. The benefit of soft margin SVM (as oppose to hard margin SVM) is that by allowing some misclassifications, we get a wider margin and therefore the solution becomes more "generalise" (i.e work better with a new set of dataset that is independent from the training dataset).

<div><ul>
<li style="display: inline-block;"> [[File:svm_margin.png|thumb|400px|none|Figure 10: Separation margin <ref name="bishop_book"> Bishop, C. M. (2006). Pattern recognition and machine learning. Springer. </ref>]] </li>
</ul></div>

==== Kernel tricks ====
Figure 10 is of linearly separable cases for SVM. For dataset that is not easy to separate with a hyperplane in their original features space (such as figure 11), the data often get projected into a higher dimension space where it is easier to separate (figure 14). Such technique is called the Kernel Tricks. It is important to note that Kernel Tricks does not actually "map" the data from '''X''' dimension to '''Z''' dimension, but rather pairwise compare the similarity of the input space in '''Z''' dimension.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_nonseparable.png |thumb|350px|none|Figure 11: Nonseparable dataset <ref name="bishop_book"></ref>]] </li>
<li style="display: inline-block;"> [[File:Svm_kernel_trick.png |thumb|650px|none|Figure 12: Simplified example of the Kernel tricks]] </li>
</ul></div>

== Previous studies ==
There are many researches into defect detection in GMAW process. The three most dominant analysis methods are spectroscopic analysis, acoustic analysis, and electrical signal analysis.

=== Spectroscopic analysis ===
In <ref name="spectroscopic_analysis">D. Bebiano and S. Alfaro, “A weld defects detection system based on a spectrometer,” Sensors, vol. 9, no. 4, pp. 2851–2861, 2009.</ref>, the authors captured the intensity of radiation emission of electric arc. It is obvious from figure 13 that good weld's intensity stays stable while defect one's is characterised by a sudden peak. Therefore, a threshold is set to detect the defect welds.

<div><ul>
<li style="display: inline-block;"> [[File:spectroscopic_analysis.png|thumb|400px|none|Figure 13: Spectroscopic analysis <ref name="spectroscopic_analysis"></ref>]] </li>
</ul></div>

Other research into using spectroscopic data to monitor that can be of interest is <ref>D. Naso, B. Turchiano, and P. Pantaleo, “A fuzzy-logic based optical sensor for online weld defect-detection,” IEEE transactions on Industrial Informatics, vol. 1, no. 4, pp. 259–273, 2005.</ref>.

=== Acoustic analysis ===
With <ref name="acoustic_analysis">E. Cayo and S. C. Alfaro, “A non-intrusive GMA welding process quality monitoring system using acoustic sensing,” Sensors, vol. 9, no. 9, pp. 7150–7166, 2009.</ref>, the author discovered that arc ignition is characterised by a distinct sound and therefore it is possible to capture the frequency of ignition through acoustic sensors. A tolerance band was set and both ignition frequency and sound pressure level were used to monitor the quality of the weld.

<div><ul>
<li style="display: inline-block;"> [[File:acoustic_analysis.png|thumb|400px|none|Figure 14: Acoustic analysis <ref name="acoustic_analysis"></ref>]] </li>
</ul></div>

Other research into using acoustic signal to monitor that can be of interest are <ref>M. Fidali, “Detection of welding process instabilities using acoustic signals,” in International Congress on Technical Diagnostic, pp. 191–201, Springer, 2016.</ref>, <ref>L. Zhang, A. C. Basantes-Defaz, D. Ozevin, and E. Indacochea, “Real-time monitoring of welding process using air-coupled ultrasonics and acoustic emission,” The International Journal of Advanced Manufacturing Technology, vol. 101, no. 5-8, pp. 1623–1634, 2019.</ref>.

=== Electrical signal analysis ===
With <ref name="electrical_analysis_1">A. Sumesh, K. Rameshkumar, A. Raja, K. Mohandas, A. Santhakumari, and R. Shyambabu, “Establishing correlation between current and voltage signatures of the arc and weld defects in GMAW process,” Arabian Journal for Science and Engineering, vol. 42, no. 11, pp. 4649–4665, 2017.</ref>, the author plot the probability density distribution of voltage signal and discovered that the stable weld's signal concentrate in one region while unstable signal spread out. While in the paper, the authors did not developed a monitoring method using their findings, it shows that there is a significant differences between stable and unstable weld's electrical signature.

<div><ul>
<li style="display: inline-block;"> [[File:Electrical analysis pdd.png|thumb|400px|none|Figure 15: Voltage PDD <ref name="electrical_analysis_1"></ref>]] </li>
</ul></div>

In <ref name="electrical_analysis_2">S. Simpson, “Signature images for arc welding fault detection,” Science and Technology of Welding and Joining, vol. 12, no. 6, pp. 481–486, 2007.</ref>, the authors shows that by plotting the signature image of voltage and current scatter plot, both arcing and short-circuiting region's signature image can notify a defects welds. A more visual representation of short circuiting and arcing region can be found in [[#Gas metal arc welding|Gas metal arc welding section]].

<div><ul>
<li style="display: inline-block;"> [[File:Electrical voltage signature.png|thumb|400px|none|Figure 16a: Electrical voltage signature of stable weld <ref name="electrical_analysis_2"></ref>]] </li>
<li style="display: inline-block;"> [[File:Electrical signature 2.png|thumb|400px|none|Figure 16b: Electrical voltage signature of unstable weld <ref name="electrical_analysis_2"></ref>]] </li>
</ul></div>

Other research into using electrical signal to monitor that can be of interest are <ref>R. Madigan, “Arc sensing for defects in constant-voltage gas metal arc welding,” Welding Journal, vol. 78, pp. 322S–328S, 1999.</ref>, <ref>Y. Huang, K. Wang, Z. Zhou, X. Zhou, and J. Fang, “Stability evaluation of short-circuiting gas metal arc welding based on ensemble empirical mode decomposition,” Measurement Science and Technology, vol. 28, no. 3, p. 035006, 2017.</ref>,<ref>Z. Zhang, X. Chen, H. Chen, J. Zhong, and S. Chen, “Online welding quality monitoring based on feature extraction of arc voltage signal,” The International Journal of Advanced Manufacturing Technology, vol. 70, no. 9-12, pp. 1661–1671, 2014</ref>,<ref>X. Li and S. Simpson, “Parametric approach to positional fault detection in short arc welding,” Science and Technology of welding and joining, vol. 14, no. 2, pp. 146–151, 2009.</ref>, and <ref>E. Wei, D. Farson, R. Richardson, and H. Ludewig, “Detection of weld surface porosity by statistical analysis of arc current in gas metal arc welding,” Journal of Manufacturing Processes, vol. 3, no. 1, pp. 50–59, 2001.</ref>.

== Experiment setup ==
The welding system uses TPS 320i C Pulse (figure 15a), produced by Fronius, to supply power, feed wire, and system cooling. For movement control of the welding torch, the system use ABB’s IRB 1520ID (figure 15b), which is a dedicated arc welding robot. The Data Acquisition device is NI's DAQ. Each channel (voltage and current) is sampled at 25kHz while the dominant frequency for voltage and current signal are around 130Hz (for CMT process) and 190Hz (for PMC process).

[[File:Fronius_power_supply.jpg |thumb|Figure 17a: Fronius' Power Supply]]
[[File:Welding_robot2.jpg |thumb|Figure 17b: ABB's welding robot]]

[[File:Ni_daq.jpg|thumb|Figure 16: NI's DAQ device]]

== Statistical Analysis ==

=== Preprocessing ===
As the captured signal coming from Fronius' power supply, the signals also pick up switching power noise. The noise will interfere with the accuracy of the analysis if not filtered. In this project, we applied a 1000Hz digital Low Pass Butterworth filter. The effect of the filter can be seen in figure 18, while the comparation of using different filters can be seen in the gallery.

<div><ul>
<li style="display: inline-block;"> [[File:Filter_effect.png |thumb|750px|none|Figure 18: Effect of 1000Hz low pass filter]] </li>
<li style="display: inline-block;"> [[File:Freq_response.png |thumb|750px|none|Figure 19: Frequency response of the in-use filter]] </li>
</ul></div>

<gallery mode="packed-hover" caption="Gallery: Comparison of different filtering method" perrow=2>
File:Effect_of_filtering2.png|In use filter vs 300Hz low pass filter
File:Effect_of_filtering3.png|In use filter vs 500Hz low pass filter
File:Effect_of_filtering4.png|In use filter vs 750Hz low pass filter
File:Effect_of_filtering5.png|In use filter vs 2000Hz low pass filter
</gallery>

=== Power Spectral Density ===
Our goal for this project is to detect the defect welds and predict the layer heights (the contact tip to work piece distance, CTWD, to be more accurate) in real time (refer to [[#Relating Contact Tip to Work-piece Distance to layer height|this section]] to understand the relationship between Contact Tip to Work-piece Distance and Layer height). An analysis of the current power spectral density (PSD) shows that there is an inverse proportional relationship between the CTWD of a weld and the frequency where its '''current signal''' PSD reach the peak, and also the magnitude of the peak. Since the peak of PSD can be fairly sensitive, we only consider the frequency of the peak PSD.

Figure 20 shows the PSD plot of voltage, current, and power signal of the weld using PMC process. In this experiment, we created multiple one-line, 8 seconds welds. These eight-second welds then get segmented into smaller two-second chunk to increase the number of data and decrease the processing time.

<div><ul>
<li style="display: inline-block;"> [[File:Psd pmc good bad all lowfreq.png |thumb|750px|none|Figure 20: Power spectral density of PMC process]] </li>
<li style="display: inline-block;"> [[File:One line weld.jpg |thumb|578px|none|Figure 21: Simple one-line weld experiment]] </li>
<li style="display: inline-block;"> [[File:Cmt_psd_stable_allCTWD_zoomed.png |thumb|750px|none|Figure 22: Simple one-line weld experiment]] </li>

</ul></div>

It seems like the inverse proportional relationship holds true for most cases. Other experiments showed that for a short period at the beginning and ending of the weld, the signal can behave differently relative to the rest of the weld. Currently, to migrate with this problem, data will not be collected and analysed (both offline and online) at these periods. However, the above results seem to only hold true for PMC processes. With CMT process, the current PSD plot indicates no relationship between the CTWD and the frequency at peak (figure 24).

For the next stage of the project, our plan is to develop a linear regression model to predict the CTWD for PMC process. With CMT process, we will have to look into different analysis techniques.

=== Support Vector Machine ===
By looking at the average voltage and average current of each segment of weld while making a simple straight path (PMC process), we realised there is a visible separation margin between stable and unstable welds. Using SVM with RBF kernel, we were able to correctly classify the quality of the weld with accuracy of 98%.

However, with CMT process, the separation cannot be observed.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_pmcNoOxide_rbf.png |thumb|750px|none|Figure 23: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:CmtNoOxide_average_voltage_current.png |thumb|750px|none|Figure 24: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

=== CTWD Regression ===
With CTWD prediction, the result will only be useful if it is done on-the-line while making the part. And therefore, as oppose to defect detection, we cannot have a short calibration process every constant time to check for worn contact tip. By analysing the data while building a real part, we found that the average voltage (for PMC process) and average current (for CMT process) seems to have the strongest linear relationship with the CTWD. The acceptable error for CTWD prediction is +/- 1 mm (according to AML3D requirement). For both PMC and CMT process, our root means square error is less than 1 (PMC: 0.68 and CMT: 0.83), which satisfied the requirement.

<div><ul>
<li style="display: inline-block;"> [[File:Linear_regression_pmc_wall.png |thumb|750px|none|Figure 25: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:Linear_regression_cmt_wall.png |thumb|750px|none|Figure 26: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

== Result discussion ==
These results show that there is a strong possibility of defect detection and CTWD prediction in welding. However, these initial findings are still a proof of concept. In the next stage of our research, we will refine the algorithm and integrate into AML's pipeline.

== References ==
<references />

File:Welding robot2.jpg

2019-10-15T04:47:16Z

A1713568:

File:Fronius power supply2.jpg

2019-10-15T04:46:08Z

A1713568: flipped

== Summary ==
flipped

File:Fronius power supply.jpg

2019-10-15T04:44:52Z

A1713568: A1713568 uploaded a new version of File:Fronius power supply.jpg

Projects:2019s2-20001 Using Machine Learning to Determine Deposit Height and Defects for Wire + Arc Additive Manufacture (3D printing)

2019-10-15T04:40:11Z

A1713568: /* Experiment setup */

[[Category:Projects]]
[[Category:Final Year Projects]]
[[Category:2019s2|24501]]
== Introduction ==
AML3D is one of the leading company in the Wire Arc Additive Manufacturing (WAAM) industry. In short, WAAM is the Metal 3D printing. The technology has the ability to reduce up to 75% of manufacture time, cost, and material. Currently, WAAM is benefiting other industry such as: mining, defence, marine, and aviation.

[[File:aml3d_logo.jpg|100px]]
=== Project team ===
==== Project students ====
* Anh Tran
* Nhat Nguyen
==== Supervisors ====
* Dr. Brian Ng
* Dr. Paul Colegrove (AML3D, sponsored company)
=== Objectives ===
The objective of this project is to increase the efficiency of the manufacture process at AML3D. In order to do so, the team will investigate into the possibility of automating and optimising the quality control processes. The two quality control processes that are currently being implemented at AML3D are measuring layer height using laser sensors, and human supervision for detecting defects. These processes add overhead into production time and usage of human resource, which is not desired. To achieve the goal, machine learning methods will be used extensively to analyse the electrical signatures of the weld process.

== Background ==
=== Wire Arc Additive Manufacturing ===
Wire and Arc Additive Manufacturing (WAAM) is a type of additive manufacturing that uses electric arc as the heat source and material wire to feed the manufacture process <ref name="WAAM"> S. W. Williams, F. Martina, A. C. Addison, J. Ding, G. Pardal & P. Colegrove (2016) Wire + Arc Additive Manufacturing, Materials Science and Technology, 32:7, 641-647, DOI: 10.1179/1743284715Y.0000000073 </ref>. WAAM has been investigated since the 1990s <ref name="WAAM" />, but only recently that it received more attention from the manufacture world.

Its significant comes from the ability to manufacture complex model with less time and less material. Figure 1a and 1b show real parts that was manufactured by AML3D. Such custom made parts might takes months before be ready to be shipped, but with WAAM, the production time can reduce down to weeks. Currently, the industries that benefit the most from WAAM are maritime and aerospace.

Similar to other additive manufacture methods, WAAM achieves such results by sliding the models into multiple layers, and then build the model layer by layer. The movement control is normally handled by a robotic arm (AML3D uses ABB's Arc Welder robot), and the welding path is generated by a Computer Aid Manufacture (CAM) software.

[[File:welding_part1.JPG|thumb|Figure 1a: A propeller]]
[[File:welding_part2.jpg|thumb|Figure 1b: Another part of the ship]]
[[File:abb_robot.jpg|thumb|Figure 2: ABB's IRB 1520ID Arc Welder]]

=== Gas metal arc welding ===
<div><ul>
<li style="display: inline-block;"> [[File:gmaw_welding.png|thumb|500px|none|Figure 2: GMAW welding process <ref name="welding_book"> Norrish, J. (2006). Advanced welding processes. Elsevier. </ref>]] </li>
</ul></div>

Gas metal arc welding (GMAW), or also known as metal inert gas (MIG), is the welding process used at AML3D for WAAM. Due to its high deposition rate and economic benefits, GMAW has became more popular. We will explore two variants of GMAW, which are Pulse Multi Control and Cold Metal Transfer. Both processes are developed by Fronius, an Austrian welding company. Note that it is not required to understand the welding physics to follow this wiki. The next two following sections' purpose is to highlight the high variability, high dynamic nature in welding.

[[File:fronius.png|thumb|Figure 3: Fronius logo]]

==== Pulse Multi Control ====
Pulse Multi Control (PMC) welding is Fronius' modified Pulsed GMAW. The advantage of Pulsed GMAW is the ability to control the metal droplet transfer in welding. Note that from figure 4, the metal droplet is characterised by a downward slope of the current pulse. Figure 5 shows the current signal captured from experiments conducted at AML3D (PMC is used).

<div><ul>
<li style="display: inline-block;"> [[File:Pulsed_gmaw_transfer_mode.png|thumb|500px|none|Figure 4: Relationship between pulse current and droplet transfer<ref name="welding_book" />]] </li>
<li style="display: inline-block;"> [[File:Pmc_current_in_time_domain.png|thumb|650px|none|Figure 5: Current signal captured from experiments conducted at AML3D (PMC process)]] </li>
</ul></div>
==== Cold Metal Transfer ====
Cold Metal Transfer (CMT) is a complex welding process, also developed by Fronius. By detecting a short circuit (mode C and D in figure 8), CMT can make adjustment such as retracting the welding material to cool down, and therefore create a smoother, more stable weld. The complexity of CMT can be seen in the current signal of the process (figure 9). Compare to PMC, CMT's signal has more variation during one signal period.

<div><ul>
<li style="display: inline-block;"> [[File:Transfer_mode.png|thumb|800px|none|Figure 8: Mechanism of metal droplet transfer mode]] </li>
<li style="display: inline-block;"> [[File:Cmt1_current_in_time_domain.png|thumb|735px|none|Figure 9: Current signal captured from experiments conducted at AML3D (CMT process)]] </li>
</ul></div>
=== Machine Learning ===
Machine learning (or statistical learning) refer to a set of tools to give data scientists an insight into the data and make better decisions based on those information <ref name="introduction_ml_book"> James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An introduction to statistical learning (Vol. 112, p. 18). New York: Springer. </ref>. In this project, we will explore the capabilities of classical machine learning methods (i.e anything but Deep Learning) due to the limitation of our available data. By the time of writing this wiki, we have only explored Support Vector Machine.

==== Support Vector Machine ====
The main idea of Support Vector Machine (SVM) is to draw hyperplanes that best separate multiple classes of data. The original SVM judges best decision hyperplane based on the separation margin between data classes (figure 10). However, there are multiple variants of SVM, such as hard/soft margin SVM, Nu-SVM, One class SVM (to solve anomaly detection problem). In our project, we choose to use a soft margin implementation of SVM. The benefit of soft margin SVM (as oppose to hard margin SVM) is that by allowing some misclassifications, we get a wider margin and therefore the solution becomes more "generalise" (i.e work better with a new set of dataset that is independent from the training dataset).

<div><ul>
<li style="display: inline-block;"> [[File:svm_margin.png|thumb|400px|none|Figure 10: Separation margin <ref name="bishop_book"> Bishop, C. M. (2006). Pattern recognition and machine learning. Springer. </ref>]] </li>
</ul></div>

==== Kernel tricks ====
Figure 10 is of linearly separable cases for SVM. For dataset that is not easy to separate with a hyperplane in their original features space (such as figure 11), the data often get projected into a higher dimension space where it is easier to separate (figure 14). Such technique is called the Kernel Tricks. It is important to note that Kernel Tricks does not actually "map" the data from '''X''' dimension to '''Z''' dimension, but rather pairwise compare the similarity of the input space in '''Z''' dimension.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_nonseparable.png |thumb|350px|none|Figure 11: Nonseparable dataset <ref name="bishop_book"></ref>]] </li>
<li style="display: inline-block;"> [[File:Svm_kernel_trick.png |thumb|650px|none|Figure 12: Simplified example of the Kernel tricks]] </li>
</ul></div>

== Previous studies ==
There are many researches into defect detection in GMAW process. The three most dominant analysis methods are spectroscopic analysis, acoustic analysis, and electrical signal analysis.

=== Spectroscopic analysis ===
In <ref name="spectroscopic_analysis">D. Bebiano and S. Alfaro, “A weld defects detection system based on a spectrometer,” Sensors, vol. 9, no. 4, pp. 2851–2861, 2009.</ref>, the authors captured the intensity of radiation emission of electric arc. It is obvious from figure 13 that good weld's intensity stays stable while defect one's is characterised by a sudden peak. Therefore, a threshold is set to detect the defect welds.

<div><ul>
<li style="display: inline-block;"> [[File:spectroscopic_analysis.png|thumb|400px|none|Figure 13: Spectroscopic analysis <ref name="spectroscopic_analysis"></ref>]] </li>
</ul></div>

Other research into using spectroscopic data to monitor that can be of interest is <ref>D. Naso, B. Turchiano, and P. Pantaleo, “A fuzzy-logic based optical sensor for online weld defect-detection,” IEEE transactions on Industrial Informatics, vol. 1, no. 4, pp. 259–273, 2005.</ref>.

=== Acoustic analysis ===
With <ref name="acoustic_analysis">E. Cayo and S. C. Alfaro, “A non-intrusive GMA welding process quality monitoring system using acoustic sensing,” Sensors, vol. 9, no. 9, pp. 7150–7166, 2009.</ref>, the author discovered that arc ignition is characterised by a distinct sound and therefore it is possible to capture the frequency of ignition through acoustic sensors. A tolerance band was set and both ignition frequency and sound pressure level were used to monitor the quality of the weld.

<div><ul>
<li style="display: inline-block;"> [[File:acoustic_analysis.png|thumb|400px|none|Figure 14: Acoustic analysis <ref name="acoustic_analysis"></ref>]] </li>
</ul></div>

Other research into using acoustic signal to monitor that can be of interest are <ref>M. Fidali, “Detection of welding process instabilities using acoustic signals,” in International Congress on Technical Diagnostic, pp. 191–201, Springer, 2016.</ref>, <ref>L. Zhang, A. C. Basantes-Defaz, D. Ozevin, and E. Indacochea, “Real-time monitoring of welding process using air-coupled ultrasonics and acoustic emission,” The International Journal of Advanced Manufacturing Technology, vol. 101, no. 5-8, pp. 1623–1634, 2019.</ref>.

=== Electrical signal analysis ===
With <ref name="electrical_analysis_1">A. Sumesh, K. Rameshkumar, A. Raja, K. Mohandas, A. Santhakumari, and R. Shyambabu, “Establishing correlation between current and voltage signatures of the arc and weld defects in GMAW process,” Arabian Journal for Science and Engineering, vol. 42, no. 11, pp. 4649–4665, 2017.</ref>, the author plot the probability density distribution of voltage signal and discovered that the stable weld's signal concentrate in one region while unstable signal spread out. While in the paper, the authors did not developed a monitoring method using their findings, it shows that there is a significant differences between stable and unstable weld's electrical signature.

<div><ul>
<li style="display: inline-block;"> [[File:Electrical analysis pdd.png|thumb|400px|none|Figure 15: Voltage PDD <ref name="electrical_analysis_1"></ref>]] </li>
</ul></div>

In <ref name="electrical_analysis_2">S. Simpson, “Signature images for arc welding fault detection,” Science and Technology of Welding and Joining, vol. 12, no. 6, pp. 481–486, 2007.</ref>, the authors shows that by plotting the signature image of voltage and current scatter plot, both arcing and short-circuiting region's signature image can notify a defects welds. A more visual representation of short circuiting and arcing region can be found in [[#Gas metal arc welding|Gas metal arc welding section]].

<div><ul>
<li style="display: inline-block;"> [[File:Electrical voltage signature.png|thumb|400px|none|Figure 16a: Electrical voltage signature of stable weld <ref name="electrical_analysis_2"></ref>]] </li>
<li style="display: inline-block;"> [[File:Electrical signature 2.png|thumb|400px|none|Figure 16b: Electrical voltage signature of unstable weld <ref name="electrical_analysis_2"></ref>]] </li>
</ul></div>

Other research into using electrical signal to monitor that can be of interest are <ref>R. Madigan, “Arc sensing for defects in constant-voltage gas metal arc welding,” Welding Journal, vol. 78, pp. 322S–328S, 1999.</ref>, <ref>Y. Huang, K. Wang, Z. Zhou, X. Zhou, and J. Fang, “Stability evaluation of short-circuiting gas metal arc welding based on ensemble empirical mode decomposition,” Measurement Science and Technology, vol. 28, no. 3, p. 035006, 2017.</ref>,<ref>Z. Zhang, X. Chen, H. Chen, J. Zhong, and S. Chen, “Online welding quality monitoring based on feature extraction of arc voltage signal,” The International Journal of Advanced Manufacturing Technology, vol. 70, no. 9-12, pp. 1661–1671, 2014</ref>,<ref>X. Li and S. Simpson, “Parametric approach to positional fault detection in short arc welding,” Science and Technology of welding and joining, vol. 14, no. 2, pp. 146–151, 2009.</ref>, and <ref>E. Wei, D. Farson, R. Richardson, and H. Ludewig, “Detection of weld surface porosity by statistical analysis of arc current in gas metal arc welding,” Journal of Manufacturing Processes, vol. 3, no. 1, pp. 50–59, 2001.</ref>.

== Experiment setup ==
The welding system uses TPS 320i C Pulse (figure 15a), produced by Fronius, to supply power, feed wire, and system cooling. For movement control of the welding torch, the system use ABB’s IRB 1520ID (figure 15b), which is a dedicated arc welding robot. The Data Acquisition device is NI's DAQ. Each channel (voltage and current) is sampled at 25kHz while the dominant frequency for voltage and current signal are around 130Hz (for CMT process) and 190Hz (for PMC process).

[[File:Fronius_power_supply.jpg |thumb|Figure 17a: Fronius' Power Supply]]
[[File:Welding_robot.jpg |thumb|Figure 17b: ABB's welding robot]]

[[File:Ni_daq.jpg|thumb|Figure 16: NI's DAQ device]]

== Statistical Analysis ==

=== Preprocessing ===
As the captured signal coming from Fronius' power supply, the signals also pick up switching power noise. The noise will interfere with the accuracy of the analysis if not filtered. In this project, we applied a 1000Hz digital Low Pass Butterworth filter. The effect of the filter can be seen in figure 18, while the comparation of using different filters can be seen in the gallery.

<div><ul>
<li style="display: inline-block;"> [[File:Filter_effect.png |thumb|750px|none|Figure 18: Effect of 1000Hz low pass filter]] </li>
<li style="display: inline-block;"> [[File:Freq_response.png |thumb|750px|none|Figure 19: Frequency response of the in-use filter]] </li>
</ul></div>

<gallery mode="packed-hover" caption="Gallery: Comparison of different filtering method" perrow=2>
File:Effect_of_filtering2.png|In use filter vs 300Hz low pass filter
File:Effect_of_filtering3.png|In use filter vs 500Hz low pass filter
File:Effect_of_filtering4.png|In use filter vs 750Hz low pass filter
File:Effect_of_filtering5.png|In use filter vs 2000Hz low pass filter
</gallery>

=== Power Spectral Density ===
Our goal for this project is to detect the defect welds and predict the layer heights (the contact tip to work piece distance, CTWD, to be more accurate) in real time (refer to [[#Relating Contact Tip to Work-piece Distance to layer height|this section]] to understand the relationship between Contact Tip to Work-piece Distance and Layer height). An analysis of the current power spectral density (PSD) shows that there is an inverse proportional relationship between the CTWD of a weld and the frequency where its '''current signal''' PSD reach the peak, and also the magnitude of the peak. Since the peak of PSD can be fairly sensitive, we only consider the frequency of the peak PSD.

Figure 20 shows the PSD plot of voltage, current, and power signal of the weld using PMC process. In this experiment, we created multiple one-line, 8 seconds welds. These eight-second welds then get segmented into smaller two-second chunk to increase the number of data and decrease the processing time.

<div><ul>
<li style="display: inline-block;"> [[File:Psd pmc good bad all lowfreq.png |thumb|750px|none|Figure 20: Power spectral density of PMC process]] </li>
<li style="display: inline-block;"> [[File:One line weld.jpg |thumb|578px|none|Figure 21: Simple one-line weld experiment]] </li>
<li style="display: inline-block;"> [[File:Cmt_psd_stable_allCTWD_zoomed.png |thumb|750px|none|Figure 22: Simple one-line weld experiment]] </li>

</ul></div>

It seems like the inverse proportional relationship holds true for most cases. Other experiments showed that for a short period at the beginning and ending of the weld, the signal can behave differently relative to the rest of the weld. Currently, to migrate with this problem, data will not be collected and analysed (both offline and online) at these periods. However, the above results seem to only hold true for PMC processes. With CMT process, the current PSD plot indicates no relationship between the CTWD and the frequency at peak (figure 24).

For the next stage of the project, our plan is to develop a linear regression model to predict the CTWD for PMC process. With CMT process, we will have to look into different analysis techniques.

=== Support Vector Machine ===
By looking at the average voltage and average current of each segment of weld while making a simple straight path (PMC process), we realised there is a visible separation margin between stable and unstable welds. Using SVM with RBF kernel, we were able to correctly classify the quality of the weld with accuracy of 98%.

However, with CMT process, the separation cannot be observed.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_pmcNoOxide_rbf.png |thumb|750px|none|Figure 23: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:CmtNoOxide_average_voltage_current.png |thumb|750px|none|Figure 24: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

=== CTWD Regression ===
With CTWD prediction, the result will only be useful if it is done on-the-line while making the part. And therefore, as oppose to defect detection, we cannot have a short calibration process every constant time to check for worn contact tip. By analysing the data while building a real part, we found that the average voltage (for PMC process) and average current (for CMT process) seems to have the strongest linear relationship with the CTWD. The acceptable error for CTWD prediction is +/- 1 mm (according to AML3D requirement). For both PMC and CMT process, our root means square error is less than 1 (PMC: 0.68 and CMT: 0.83), which satisfied the requirement.

<div><ul>
<li style="display: inline-block;"> [[File:Linear_regression_pmc_wall.png |thumb|750px|none|Figure 25: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:Linear_regression_cmt_wall.png |thumb|750px|none|Figure 26: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

== Result discussion ==
These results show that there is a strong possibility of defect detection and CTWD prediction in welding. However, these initial findings are still a proof of concept. In the next stage of our research, we will refine the algorithm and integrate into AML's pipeline.

== References ==
<references />

Projects:2019s2-20001 Using Machine Learning to Determine Deposit Height and Defects for Wire + Arc Additive Manufacture (3D printing)

2019-10-15T04:39:54Z

A1713568: /* Experiment setup */

[[Category:Projects]]
[[Category:Final Year Projects]]
[[Category:2019s2|24501]]
== Introduction ==
AML3D is one of the leading company in the Wire Arc Additive Manufacturing (WAAM) industry. In short, WAAM is the Metal 3D printing. The technology has the ability to reduce up to 75% of manufacture time, cost, and material. Currently, WAAM is benefiting other industry such as: mining, defence, marine, and aviation.

[[File:aml3d_logo.jpg|100px]]
=== Project team ===
==== Project students ====
* Anh Tran
* Nhat Nguyen
==== Supervisors ====
* Dr. Brian Ng
* Dr. Paul Colegrove (AML3D, sponsored company)
=== Objectives ===
The objective of this project is to increase the efficiency of the manufacture process at AML3D. In order to do so, the team will investigate into the possibility of automating and optimising the quality control processes. The two quality control processes that are currently being implemented at AML3D are measuring layer height using laser sensors, and human supervision for detecting defects. These processes add overhead into production time and usage of human resource, which is not desired. To achieve the goal, machine learning methods will be used extensively to analyse the electrical signatures of the weld process.

== Background ==
=== Wire Arc Additive Manufacturing ===
Wire and Arc Additive Manufacturing (WAAM) is a type of additive manufacturing that uses electric arc as the heat source and material wire to feed the manufacture process <ref name="WAAM"> S. W. Williams, F. Martina, A. C. Addison, J. Ding, G. Pardal & P. Colegrove (2016) Wire + Arc Additive Manufacturing, Materials Science and Technology, 32:7, 641-647, DOI: 10.1179/1743284715Y.0000000073 </ref>. WAAM has been investigated since the 1990s <ref name="WAAM" />, but only recently that it received more attention from the manufacture world.

Its significant comes from the ability to manufacture complex model with less time and less material. Figure 1a and 1b show real parts that was manufactured by AML3D. Such custom made parts might takes months before be ready to be shipped, but with WAAM, the production time can reduce down to weeks. Currently, the industries that benefit the most from WAAM are maritime and aerospace.

Similar to other additive manufacture methods, WAAM achieves such results by sliding the models into multiple layers, and then build the model layer by layer. The movement control is normally handled by a robotic arm (AML3D uses ABB's Arc Welder robot), and the welding path is generated by a Computer Aid Manufacture (CAM) software.

[[File:welding_part1.JPG|thumb|Figure 1a: A propeller]]
[[File:welding_part2.jpg|thumb|Figure 1b: Another part of the ship]]
[[File:abb_robot.jpg|thumb|Figure 2: ABB's IRB 1520ID Arc Welder]]

=== Gas metal arc welding ===
<div><ul>
<li style="display: inline-block;"> [[File:gmaw_welding.png|thumb|500px|none|Figure 2: GMAW welding process <ref name="welding_book"> Norrish, J. (2006). Advanced welding processes. Elsevier. </ref>]] </li>
</ul></div>

Gas metal arc welding (GMAW), or also known as metal inert gas (MIG), is the welding process used at AML3D for WAAM. Due to its high deposition rate and economic benefits, GMAW has became more popular. We will explore two variants of GMAW, which are Pulse Multi Control and Cold Metal Transfer. Both processes are developed by Fronius, an Austrian welding company. Note that it is not required to understand the welding physics to follow this wiki. The next two following sections' purpose is to highlight the high variability, high dynamic nature in welding.

[[File:fronius.png|thumb|Figure 3: Fronius logo]]

==== Pulse Multi Control ====
Pulse Multi Control (PMC) welding is Fronius' modified Pulsed GMAW. The advantage of Pulsed GMAW is the ability to control the metal droplet transfer in welding. Note that from figure 4, the metal droplet is characterised by a downward slope of the current pulse. Figure 5 shows the current signal captured from experiments conducted at AML3D (PMC is used).

<div><ul>
<li style="display: inline-block;"> [[File:Pulsed_gmaw_transfer_mode.png|thumb|500px|none|Figure 4: Relationship between pulse current and droplet transfer<ref name="welding_book" />]] </li>
<li style="display: inline-block;"> [[File:Pmc_current_in_time_domain.png|thumb|650px|none|Figure 5: Current signal captured from experiments conducted at AML3D (PMC process)]] </li>
</ul></div>
==== Cold Metal Transfer ====
Cold Metal Transfer (CMT) is a complex welding process, also developed by Fronius. By detecting a short circuit (mode C and D in figure 8), CMT can make adjustment such as retracting the welding material to cool down, and therefore create a smoother, more stable weld. The complexity of CMT can be seen in the current signal of the process (figure 9). Compare to PMC, CMT's signal has more variation during one signal period.

<div><ul>
<li style="display: inline-block;"> [[File:Transfer_mode.png|thumb|800px|none|Figure 8: Mechanism of metal droplet transfer mode]] </li>
<li style="display: inline-block;"> [[File:Cmt1_current_in_time_domain.png|thumb|735px|none|Figure 9: Current signal captured from experiments conducted at AML3D (CMT process)]] </li>
</ul></div>
=== Machine Learning ===
Machine learning (or statistical learning) refer to a set of tools to give data scientists an insight into the data and make better decisions based on those information <ref name="introduction_ml_book"> James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An introduction to statistical learning (Vol. 112, p. 18). New York: Springer. </ref>. In this project, we will explore the capabilities of classical machine learning methods (i.e anything but Deep Learning) due to the limitation of our available data. By the time of writing this wiki, we have only explored Support Vector Machine.

==== Support Vector Machine ====
The main idea of Support Vector Machine (SVM) is to draw hyperplanes that best separate multiple classes of data. The original SVM judges best decision hyperplane based on the separation margin between data classes (figure 10). However, there are multiple variants of SVM, such as hard/soft margin SVM, Nu-SVM, One class SVM (to solve anomaly detection problem). In our project, we choose to use a soft margin implementation of SVM. The benefit of soft margin SVM (as oppose to hard margin SVM) is that by allowing some misclassifications, we get a wider margin and therefore the solution becomes more "generalise" (i.e work better with a new set of dataset that is independent from the training dataset).

<div><ul>
<li style="display: inline-block;"> [[File:svm_margin.png|thumb|400px|none|Figure 10: Separation margin <ref name="bishop_book"> Bishop, C. M. (2006). Pattern recognition and machine learning. Springer. </ref>]] </li>
</ul></div>

==== Kernel tricks ====
Figure 10 is of linearly separable cases for SVM. For dataset that is not easy to separate with a hyperplane in their original features space (such as figure 11), the data often get projected into a higher dimension space where it is easier to separate (figure 14). Such technique is called the Kernel Tricks. It is important to note that Kernel Tricks does not actually "map" the data from '''X''' dimension to '''Z''' dimension, but rather pairwise compare the similarity of the input space in '''Z''' dimension.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_nonseparable.png |thumb|350px|none|Figure 11: Nonseparable dataset <ref name="bishop_book"></ref>]] </li>
<li style="display: inline-block;"> [[File:Svm_kernel_trick.png |thumb|650px|none|Figure 12: Simplified example of the Kernel tricks]] </li>
</ul></div>

== Previous studies ==
There are many researches into defect detection in GMAW process. The three most dominant analysis methods are spectroscopic analysis, acoustic analysis, and electrical signal analysis.

=== Spectroscopic analysis ===
In <ref name="spectroscopic_analysis">D. Bebiano and S. Alfaro, “A weld defects detection system based on a spectrometer,” Sensors, vol. 9, no. 4, pp. 2851–2861, 2009.</ref>, the authors captured the intensity of radiation emission of electric arc. It is obvious from figure 13 that good weld's intensity stays stable while defect one's is characterised by a sudden peak. Therefore, a threshold is set to detect the defect welds.

<div><ul>
<li style="display: inline-block;"> [[File:spectroscopic_analysis.png|thumb|400px|none|Figure 13: Spectroscopic analysis <ref name="spectroscopic_analysis"></ref>]] </li>
</ul></div>

Other research into using spectroscopic data to monitor that can be of interest is <ref>D. Naso, B. Turchiano, and P. Pantaleo, “A fuzzy-logic based optical sensor for online weld defect-detection,” IEEE transactions on Industrial Informatics, vol. 1, no. 4, pp. 259–273, 2005.</ref>.

=== Acoustic analysis ===
With <ref name="acoustic_analysis">E. Cayo and S. C. Alfaro, “A non-intrusive GMA welding process quality monitoring system using acoustic sensing,” Sensors, vol. 9, no. 9, pp. 7150–7166, 2009.</ref>, the author discovered that arc ignition is characterised by a distinct sound and therefore it is possible to capture the frequency of ignition through acoustic sensors. A tolerance band was set and both ignition frequency and sound pressure level were used to monitor the quality of the weld.

<div><ul>
<li style="display: inline-block;"> [[File:acoustic_analysis.png|thumb|400px|none|Figure 14: Acoustic analysis <ref name="acoustic_analysis"></ref>]] </li>
</ul></div>

Other research into using acoustic signal to monitor that can be of interest are <ref>M. Fidali, “Detection of welding process instabilities using acoustic signals,” in International Congress on Technical Diagnostic, pp. 191–201, Springer, 2016.</ref>, <ref>L. Zhang, A. C. Basantes-Defaz, D. Ozevin, and E. Indacochea, “Real-time monitoring of welding process using air-coupled ultrasonics and acoustic emission,” The International Journal of Advanced Manufacturing Technology, vol. 101, no. 5-8, pp. 1623–1634, 2019.</ref>.

=== Electrical signal analysis ===
With <ref name="electrical_analysis_1">A. Sumesh, K. Rameshkumar, A. Raja, K. Mohandas, A. Santhakumari, and R. Shyambabu, “Establishing correlation between current and voltage signatures of the arc and weld defects in GMAW process,” Arabian Journal for Science and Engineering, vol. 42, no. 11, pp. 4649–4665, 2017.</ref>, the author plot the probability density distribution of voltage signal and discovered that the stable weld's signal concentrate in one region while unstable signal spread out. While in the paper, the authors did not developed a monitoring method using their findings, it shows that there is a significant differences between stable and unstable weld's electrical signature.

<div><ul>
<li style="display: inline-block;"> [[File:Electrical analysis pdd.png|thumb|400px|none|Figure 15: Voltage PDD <ref name="electrical_analysis_1"></ref>]] </li>
</ul></div>

In <ref name="electrical_analysis_2">S. Simpson, “Signature images for arc welding fault detection,” Science and Technology of Welding and Joining, vol. 12, no. 6, pp. 481–486, 2007.</ref>, the authors shows that by plotting the signature image of voltage and current scatter plot, both arcing and short-circuiting region's signature image can notify a defects welds. A more visual representation of short circuiting and arcing region can be found in [[#Gas metal arc welding|Gas metal arc welding section]].

<div><ul>
<li style="display: inline-block;"> [[File:Electrical voltage signature.png|thumb|400px|none|Figure 16a: Electrical voltage signature of stable weld <ref name="electrical_analysis_2"></ref>]] </li>
<li style="display: inline-block;"> [[File:Electrical signature 2.png|thumb|400px|none|Figure 16b: Electrical voltage signature of unstable weld <ref name="electrical_analysis_2"></ref>]] </li>
</ul></div>

Other research into using electrical signal to monitor that can be of interest are <ref>R. Madigan, “Arc sensing for defects in constant-voltage gas metal arc welding,” Welding Journal, vol. 78, pp. 322S–328S, 1999.</ref>, <ref>Y. Huang, K. Wang, Z. Zhou, X. Zhou, and J. Fang, “Stability evaluation of short-circuiting gas metal arc welding based on ensemble empirical mode decomposition,” Measurement Science and Technology, vol. 28, no. 3, p. 035006, 2017.</ref>,<ref>Z. Zhang, X. Chen, H. Chen, J. Zhong, and S. Chen, “Online welding quality monitoring based on feature extraction of arc voltage signal,” The International Journal of Advanced Manufacturing Technology, vol. 70, no. 9-12, pp. 1661–1671, 2014</ref>,<ref>X. Li and S. Simpson, “Parametric approach to positional fault detection in short arc welding,” Science and Technology of welding and joining, vol. 14, no. 2, pp. 146–151, 2009.</ref>, and <ref>E. Wei, D. Farson, R. Richardson, and H. Ludewig, “Detection of weld surface porosity by statistical analysis of arc current in gas metal arc welding,” Journal of Manufacturing Processes, vol. 3, no. 1, pp. 50–59, 2001.</ref>.

== Experiment setup ==
The welding system uses TPS 320i C Pulse (figure 15a), produced by Fronius, to supply power, feed wire, and system cooling. For movement control of the welding torch, the system use ABB’s IRB 1520ID (figure 15b), which is a dedicated arc welding robot. The Data Acquisition device is NI's DAQ. Each channel (voltage and current) is sampled at 25kHz while the dominant frequency for voltage and current signal are around 130Hz (for CMT process) and 190Hz (for PMC process).

[[File:Fronius_power_supply.jpg |thumb|none|Figure 17a: Fronius' Power Supply]]
[[File:Welding_robot.jpg |thumb|none|Figure 17b: ABB's welding robot]]

[[File:Ni_daq.jpg|thumb|Figure 16: NI's DAQ device]]

== Statistical Analysis ==

=== Preprocessing ===
As the captured signal coming from Fronius' power supply, the signals also pick up switching power noise. The noise will interfere with the accuracy of the analysis if not filtered. In this project, we applied a 1000Hz digital Low Pass Butterworth filter. The effect of the filter can be seen in figure 18, while the comparation of using different filters can be seen in the gallery.

<div><ul>
<li style="display: inline-block;"> [[File:Filter_effect.png |thumb|750px|none|Figure 18: Effect of 1000Hz low pass filter]] </li>
<li style="display: inline-block;"> [[File:Freq_response.png |thumb|750px|none|Figure 19: Frequency response of the in-use filter]] </li>
</ul></div>

<gallery mode="packed-hover" caption="Gallery: Comparison of different filtering method" perrow=2>
File:Effect_of_filtering2.png|In use filter vs 300Hz low pass filter
File:Effect_of_filtering3.png|In use filter vs 500Hz low pass filter
File:Effect_of_filtering4.png|In use filter vs 750Hz low pass filter
File:Effect_of_filtering5.png|In use filter vs 2000Hz low pass filter
</gallery>

=== Power Spectral Density ===
Our goal for this project is to detect the defect welds and predict the layer heights (the contact tip to work piece distance, CTWD, to be more accurate) in real time (refer to [[#Relating Contact Tip to Work-piece Distance to layer height|this section]] to understand the relationship between Contact Tip to Work-piece Distance and Layer height). An analysis of the current power spectral density (PSD) shows that there is an inverse proportional relationship between the CTWD of a weld and the frequency where its '''current signal''' PSD reach the peak, and also the magnitude of the peak. Since the peak of PSD can be fairly sensitive, we only consider the frequency of the peak PSD.

Figure 20 shows the PSD plot of voltage, current, and power signal of the weld using PMC process. In this experiment, we created multiple one-line, 8 seconds welds. These eight-second welds then get segmented into smaller two-second chunk to increase the number of data and decrease the processing time.

<div><ul>
<li style="display: inline-block;"> [[File:Psd pmc good bad all lowfreq.png |thumb|750px|none|Figure 20: Power spectral density of PMC process]] </li>
<li style="display: inline-block;"> [[File:One line weld.jpg |thumb|578px|none|Figure 21: Simple one-line weld experiment]] </li>
<li style="display: inline-block;"> [[File:Cmt_psd_stable_allCTWD_zoomed.png |thumb|750px|none|Figure 22: Simple one-line weld experiment]] </li>

</ul></div>

It seems like the inverse proportional relationship holds true for most cases. Other experiments showed that for a short period at the beginning and ending of the weld, the signal can behave differently relative to the rest of the weld. Currently, to migrate with this problem, data will not be collected and analysed (both offline and online) at these periods. However, the above results seem to only hold true for PMC processes. With CMT process, the current PSD plot indicates no relationship between the CTWD and the frequency at peak (figure 24).

For the next stage of the project, our plan is to develop a linear regression model to predict the CTWD for PMC process. With CMT process, we will have to look into different analysis techniques.

=== Support Vector Machine ===
By looking at the average voltage and average current of each segment of weld while making a simple straight path (PMC process), we realised there is a visible separation margin between stable and unstable welds. Using SVM with RBF kernel, we were able to correctly classify the quality of the weld with accuracy of 98%.

However, with CMT process, the separation cannot be observed.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_pmcNoOxide_rbf.png |thumb|750px|none|Figure 23: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:CmtNoOxide_average_voltage_current.png |thumb|750px|none|Figure 24: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

=== CTWD Regression ===
With CTWD prediction, the result will only be useful if it is done on-the-line while making the part. And therefore, as oppose to defect detection, we cannot have a short calibration process every constant time to check for worn contact tip. By analysing the data while building a real part, we found that the average voltage (for PMC process) and average current (for CMT process) seems to have the strongest linear relationship with the CTWD. The acceptable error for CTWD prediction is +/- 1 mm (according to AML3D requirement). For both PMC and CMT process, our root means square error is less than 1 (PMC: 0.68 and CMT: 0.83), which satisfied the requirement.

<div><ul>
<li style="display: inline-block;"> [[File:Linear_regression_pmc_wall.png |thumb|750px|none|Figure 25: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:Linear_regression_cmt_wall.png |thumb|750px|none|Figure 26: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

== Result discussion ==
These results show that there is a strong possibility of defect detection and CTWD prediction in welding. However, these initial findings are still a proof of concept. In the next stage of our research, we will refine the algorithm and integrate into AML's pipeline.

== References ==
<references />

Projects:2019s2-20001 Using Machine Learning to Determine Deposit Height and Defects for Wire + Arc Additive Manufacture (3D printing)

2019-10-15T04:38:41Z

A1713568:

[[Category:Projects]]
[[Category:Final Year Projects]]
[[Category:2019s2|24501]]
== Introduction ==
AML3D is one of the leading company in the Wire Arc Additive Manufacturing (WAAM) industry. In short, WAAM is the Metal 3D printing. The technology has the ability to reduce up to 75% of manufacture time, cost, and material. Currently, WAAM is benefiting other industry such as: mining, defence, marine, and aviation.

[[File:aml3d_logo.jpg|100px]]
=== Project team ===
==== Project students ====
* Anh Tran
* Nhat Nguyen
==== Supervisors ====
* Dr. Brian Ng
* Dr. Paul Colegrove (AML3D, sponsored company)
=== Objectives ===
The objective of this project is to increase the efficiency of the manufacture process at AML3D. In order to do so, the team will investigate into the possibility of automating and optimising the quality control processes. The two quality control processes that are currently being implemented at AML3D are measuring layer height using laser sensors, and human supervision for detecting defects. These processes add overhead into production time and usage of human resource, which is not desired. To achieve the goal, machine learning methods will be used extensively to analyse the electrical signatures of the weld process.

== Background ==
=== Wire Arc Additive Manufacturing ===
Wire and Arc Additive Manufacturing (WAAM) is a type of additive manufacturing that uses electric arc as the heat source and material wire to feed the manufacture process <ref name="WAAM"> S. W. Williams, F. Martina, A. C. Addison, J. Ding, G. Pardal & P. Colegrove (2016) Wire + Arc Additive Manufacturing, Materials Science and Technology, 32:7, 641-647, DOI: 10.1179/1743284715Y.0000000073 </ref>. WAAM has been investigated since the 1990s <ref name="WAAM" />, but only recently that it received more attention from the manufacture world.

Its significant comes from the ability to manufacture complex model with less time and less material. Figure 1a and 1b show real parts that was manufactured by AML3D. Such custom made parts might takes months before be ready to be shipped, but with WAAM, the production time can reduce down to weeks. Currently, the industries that benefit the most from WAAM are maritime and aerospace.

Similar to other additive manufacture methods, WAAM achieves such results by sliding the models into multiple layers, and then build the model layer by layer. The movement control is normally handled by a robotic arm (AML3D uses ABB's Arc Welder robot), and the welding path is generated by a Computer Aid Manufacture (CAM) software.

[[File:welding_part1.JPG|thumb|Figure 1a: A propeller]]
[[File:welding_part2.jpg|thumb|Figure 1b: Another part of the ship]]
[[File:abb_robot.jpg|thumb|Figure 2: ABB's IRB 1520ID Arc Welder]]

=== Gas metal arc welding ===
<div><ul>
<li style="display: inline-block;"> [[File:gmaw_welding.png|thumb|500px|none|Figure 2: GMAW welding process <ref name="welding_book"> Norrish, J. (2006). Advanced welding processes. Elsevier. </ref>]] </li>
</ul></div>

Gas metal arc welding (GMAW), or also known as metal inert gas (MIG), is the welding process used at AML3D for WAAM. Due to its high deposition rate and economic benefits, GMAW has became more popular. We will explore two variants of GMAW, which are Pulse Multi Control and Cold Metal Transfer. Both processes are developed by Fronius, an Austrian welding company. Note that it is not required to understand the welding physics to follow this wiki. The next two following sections' purpose is to highlight the high variability, high dynamic nature in welding.

[[File:fronius.png|thumb|Figure 3: Fronius logo]]

==== Pulse Multi Control ====
Pulse Multi Control (PMC) welding is Fronius' modified Pulsed GMAW. The advantage of Pulsed GMAW is the ability to control the metal droplet transfer in welding. Note that from figure 4, the metal droplet is characterised by a downward slope of the current pulse. Figure 5 shows the current signal captured from experiments conducted at AML3D (PMC is used).

<div><ul>
<li style="display: inline-block;"> [[File:Pulsed_gmaw_transfer_mode.png|thumb|500px|none|Figure 4: Relationship between pulse current and droplet transfer<ref name="welding_book" />]] </li>
<li style="display: inline-block;"> [[File:Pmc_current_in_time_domain.png|thumb|650px|none|Figure 5: Current signal captured from experiments conducted at AML3D (PMC process)]] </li>
</ul></div>
==== Cold Metal Transfer ====
Cold Metal Transfer (CMT) is a complex welding process, also developed by Fronius. By detecting a short circuit (mode C and D in figure 8), CMT can make adjustment such as retracting the welding material to cool down, and therefore create a smoother, more stable weld. The complexity of CMT can be seen in the current signal of the process (figure 9). Compare to PMC, CMT's signal has more variation during one signal period.

<div><ul>
<li style="display: inline-block;"> [[File:Transfer_mode.png|thumb|800px|none|Figure 8: Mechanism of metal droplet transfer mode]] </li>
<li style="display: inline-block;"> [[File:Cmt1_current_in_time_domain.png|thumb|735px|none|Figure 9: Current signal captured from experiments conducted at AML3D (CMT process)]] </li>
</ul></div>
=== Machine Learning ===
Machine learning (or statistical learning) refer to a set of tools to give data scientists an insight into the data and make better decisions based on those information <ref name="introduction_ml_book"> James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An introduction to statistical learning (Vol. 112, p. 18). New York: Springer. </ref>. In this project, we will explore the capabilities of classical machine learning methods (i.e anything but Deep Learning) due to the limitation of our available data. By the time of writing this wiki, we have only explored Support Vector Machine.

==== Support Vector Machine ====
The main idea of Support Vector Machine (SVM) is to draw hyperplanes that best separate multiple classes of data. The original SVM judges best decision hyperplane based on the separation margin between data classes (figure 10). However, there are multiple variants of SVM, such as hard/soft margin SVM, Nu-SVM, One class SVM (to solve anomaly detection problem). In our project, we choose to use a soft margin implementation of SVM. The benefit of soft margin SVM (as oppose to hard margin SVM) is that by allowing some misclassifications, we get a wider margin and therefore the solution becomes more "generalise" (i.e work better with a new set of dataset that is independent from the training dataset).

<div><ul>
<li style="display: inline-block;"> [[File:svm_margin.png|thumb|400px|none|Figure 10: Separation margin <ref name="bishop_book"> Bishop, C. M. (2006). Pattern recognition and machine learning. Springer. </ref>]] </li>
</ul></div>

==== Kernel tricks ====
Figure 10 is of linearly separable cases for SVM. For dataset that is not easy to separate with a hyperplane in their original features space (such as figure 11), the data often get projected into a higher dimension space where it is easier to separate (figure 14). Such technique is called the Kernel Tricks. It is important to note that Kernel Tricks does not actually "map" the data from '''X''' dimension to '''Z''' dimension, but rather pairwise compare the similarity of the input space in '''Z''' dimension.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_nonseparable.png |thumb|350px|none|Figure 11: Nonseparable dataset <ref name="bishop_book"></ref>]] </li>
<li style="display: inline-block;"> [[File:Svm_kernel_trick.png |thumb|650px|none|Figure 12: Simplified example of the Kernel tricks]] </li>
</ul></div>

== Previous studies ==
There are many researches into defect detection in GMAW process. The three most dominant analysis methods are spectroscopic analysis, acoustic analysis, and electrical signal analysis.

=== Spectroscopic analysis ===
In <ref name="spectroscopic_analysis">D. Bebiano and S. Alfaro, “A weld defects detection system based on a spectrometer,” Sensors, vol. 9, no. 4, pp. 2851–2861, 2009.</ref>, the authors captured the intensity of radiation emission of electric arc. It is obvious from figure 13 that good weld's intensity stays stable while defect one's is characterised by a sudden peak. Therefore, a threshold is set to detect the defect welds.

<div><ul>
<li style="display: inline-block;"> [[File:spectroscopic_analysis.png|thumb|400px|none|Figure 13: Spectroscopic analysis <ref name="spectroscopic_analysis"></ref>]] </li>
</ul></div>

Other research into using spectroscopic data to monitor that can be of interest is <ref>D. Naso, B. Turchiano, and P. Pantaleo, “A fuzzy-logic based optical sensor for online weld defect-detection,” IEEE transactions on Industrial Informatics, vol. 1, no. 4, pp. 259–273, 2005.</ref>.

=== Acoustic analysis ===
With <ref name="acoustic_analysis">E. Cayo and S. C. Alfaro, “A non-intrusive GMA welding process quality monitoring system using acoustic sensing,” Sensors, vol. 9, no. 9, pp. 7150–7166, 2009.</ref>, the author discovered that arc ignition is characterised by a distinct sound and therefore it is possible to capture the frequency of ignition through acoustic sensors. A tolerance band was set and both ignition frequency and sound pressure level were used to monitor the quality of the weld.

<div><ul>
<li style="display: inline-block;"> [[File:acoustic_analysis.png|thumb|400px|none|Figure 14: Acoustic analysis <ref name="acoustic_analysis"></ref>]] </li>
</ul></div>

Other research into using acoustic signal to monitor that can be of interest are <ref>M. Fidali, “Detection of welding process instabilities using acoustic signals,” in International Congress on Technical Diagnostic, pp. 191–201, Springer, 2016.</ref>, <ref>L. Zhang, A. C. Basantes-Defaz, D. Ozevin, and E. Indacochea, “Real-time monitoring of welding process using air-coupled ultrasonics and acoustic emission,” The International Journal of Advanced Manufacturing Technology, vol. 101, no. 5-8, pp. 1623–1634, 2019.</ref>.

=== Electrical signal analysis ===
With <ref name="electrical_analysis_1">A. Sumesh, K. Rameshkumar, A. Raja, K. Mohandas, A. Santhakumari, and R. Shyambabu, “Establishing correlation between current and voltage signatures of the arc and weld defects in GMAW process,” Arabian Journal for Science and Engineering, vol. 42, no. 11, pp. 4649–4665, 2017.</ref>, the author plot the probability density distribution of voltage signal and discovered that the stable weld's signal concentrate in one region while unstable signal spread out. While in the paper, the authors did not developed a monitoring method using their findings, it shows that there is a significant differences between stable and unstable weld's electrical signature.

<div><ul>
<li style="display: inline-block;"> [[File:Electrical analysis pdd.png|thumb|400px|none|Figure 15: Voltage PDD <ref name="electrical_analysis_1"></ref>]] </li>
</ul></div>

In <ref name="electrical_analysis_2">S. Simpson, “Signature images for arc welding fault detection,” Science and Technology of Welding and Joining, vol. 12, no. 6, pp. 481–486, 2007.</ref>, the authors shows that by plotting the signature image of voltage and current scatter plot, both arcing and short-circuiting region's signature image can notify a defects welds. A more visual representation of short circuiting and arcing region can be found in [[#Gas metal arc welding|Gas metal arc welding section]].

<div><ul>
<li style="display: inline-block;"> [[File:Electrical voltage signature.png|thumb|400px|none|Figure 16a: Electrical voltage signature of stable weld <ref name="electrical_analysis_2"></ref>]] </li>
<li style="display: inline-block;"> [[File:Electrical signature 2.png|thumb|400px|none|Figure 16b: Electrical voltage signature of unstable weld <ref name="electrical_analysis_2"></ref>]] </li>
</ul></div>

Other research into using electrical signal to monitor that can be of interest are <ref>R. Madigan, “Arc sensing for defects in constant-voltage gas metal arc welding,” Welding Journal, vol. 78, pp. 322S–328S, 1999.</ref>, <ref>Y. Huang, K. Wang, Z. Zhou, X. Zhou, and J. Fang, “Stability evaluation of short-circuiting gas metal arc welding based on ensemble empirical mode decomposition,” Measurement Science and Technology, vol. 28, no. 3, p. 035006, 2017.</ref>,<ref>Z. Zhang, X. Chen, H. Chen, J. Zhong, and S. Chen, “Online welding quality monitoring based on feature extraction of arc voltage signal,” The International Journal of Advanced Manufacturing Technology, vol. 70, no. 9-12, pp. 1661–1671, 2014</ref>,<ref>X. Li and S. Simpson, “Parametric approach to positional fault detection in short arc welding,” Science and Technology of welding and joining, vol. 14, no. 2, pp. 146–151, 2009.</ref>, and <ref>E. Wei, D. Farson, R. Richardson, and H. Ludewig, “Detection of weld surface porosity by statistical analysis of arc current in gas metal arc welding,” Journal of Manufacturing Processes, vol. 3, no. 1, pp. 50–59, 2001.</ref>.

== Experiment setup ==
The welding system uses TPS 320i C Pulse (figure 15a), produced by Fronius, to supply power, feed wire, and system cooling. For movement control of the welding torch, the system use ABB’s IRB 1520ID (figure 15b), which is a dedicated arc welding robot. The Data Acquisition device is NI's DAQ. Each channel (voltage and current) is sampled at 25kHz while the dominant frequency for voltage and current signal are around 130Hz (for CMT process) and 190Hz (for PMC process).

<div><ul>
<li style="display: inline-block;"> [[File:Fronius_power_supply.jpg |thumb|250px|none|Figure 17a: Fronius' Power Supply]] </li>
<li style="display: inline-block;"> [[File:Welding_robot.jpg |thumb|250px|none|Figure 17b: ABB's welding robot]] </li>
</ul></div>

[[File:Ni_daq.jpg|thumb|Figure 16: NI's DAQ device]]

== Statistical Analysis ==

=== Preprocessing ===
As the captured signal coming from Fronius' power supply, the signals also pick up switching power noise. The noise will interfere with the accuracy of the analysis if not filtered. In this project, we applied a 1000Hz digital Low Pass Butterworth filter. The effect of the filter can be seen in figure 18, while the comparation of using different filters can be seen in the gallery.

<div><ul>
<li style="display: inline-block;"> [[File:Filter_effect.png |thumb|750px|none|Figure 18: Effect of 1000Hz low pass filter]] </li>
<li style="display: inline-block;"> [[File:Freq_response.png |thumb|750px|none|Figure 19: Frequency response of the in-use filter]] </li>
</ul></div>

<gallery mode="packed-hover" caption="Gallery: Comparison of different filtering method" perrow=2>
File:Effect_of_filtering2.png|In use filter vs 300Hz low pass filter
File:Effect_of_filtering3.png|In use filter vs 500Hz low pass filter
File:Effect_of_filtering4.png|In use filter vs 750Hz low pass filter
File:Effect_of_filtering5.png|In use filter vs 2000Hz low pass filter
</gallery>

=== Power Spectral Density ===
Our goal for this project is to detect the defect welds and predict the layer heights (the contact tip to work piece distance, CTWD, to be more accurate) in real time (refer to [[#Relating Contact Tip to Work-piece Distance to layer height|this section]] to understand the relationship between Contact Tip to Work-piece Distance and Layer height). An analysis of the current power spectral density (PSD) shows that there is an inverse proportional relationship between the CTWD of a weld and the frequency where its '''current signal''' PSD reach the peak, and also the magnitude of the peak. Since the peak of PSD can be fairly sensitive, we only consider the frequency of the peak PSD.

Figure 20 shows the PSD plot of voltage, current, and power signal of the weld using PMC process. In this experiment, we created multiple one-line, 8 seconds welds. These eight-second welds then get segmented into smaller two-second chunk to increase the number of data and decrease the processing time.

<div><ul>
<li style="display: inline-block;"> [[File:Psd pmc good bad all lowfreq.png |thumb|750px|none|Figure 20: Power spectral density of PMC process]] </li>
<li style="display: inline-block;"> [[File:One line weld.jpg |thumb|578px|none|Figure 21: Simple one-line weld experiment]] </li>
<li style="display: inline-block;"> [[File:Cmt_psd_stable_allCTWD_zoomed.png |thumb|750px|none|Figure 22: Simple one-line weld experiment]] </li>

</ul></div>

It seems like the inverse proportional relationship holds true for most cases. Other experiments showed that for a short period at the beginning and ending of the weld, the signal can behave differently relative to the rest of the weld. Currently, to migrate with this problem, data will not be collected and analysed (both offline and online) at these periods. However, the above results seem to only hold true for PMC processes. With CMT process, the current PSD plot indicates no relationship between the CTWD and the frequency at peak (figure 24).

For the next stage of the project, our plan is to develop a linear regression model to predict the CTWD for PMC process. With CMT process, we will have to look into different analysis techniques.

=== Support Vector Machine ===
By looking at the average voltage and average current of each segment of weld while making a simple straight path (PMC process), we realised there is a visible separation margin between stable and unstable welds. Using SVM with RBF kernel, we were able to correctly classify the quality of the weld with accuracy of 98%.

However, with CMT process, the separation cannot be observed.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_pmcNoOxide_rbf.png |thumb|750px|none|Figure 23: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:CmtNoOxide_average_voltage_current.png |thumb|750px|none|Figure 24: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

=== CTWD Regression ===
With CTWD prediction, the result will only be useful if it is done on-the-line while making the part. And therefore, as oppose to defect detection, we cannot have a short calibration process every constant time to check for worn contact tip. By analysing the data while building a real part, we found that the average voltage (for PMC process) and average current (for CMT process) seems to have the strongest linear relationship with the CTWD. The acceptable error for CTWD prediction is +/- 1 mm (according to AML3D requirement). For both PMC and CMT process, our root means square error is less than 1 (PMC: 0.68 and CMT: 0.83), which satisfied the requirement.

<div><ul>
<li style="display: inline-block;"> [[File:Linear_regression_pmc_wall.png |thumb|750px|none|Figure 25: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:Linear_regression_cmt_wall.png |thumb|750px|none|Figure 26: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

== Result discussion ==
These results show that there is a strong possibility of defect detection and CTWD prediction in welding. However, these initial findings are still a proof of concept. In the next stage of our research, we will refine the algorithm and integrate into AML's pipeline.

== References ==
<references />

Projects:2019s2-20001 Using Machine Learning to Determine Deposit Height and Defects for Wire + Arc Additive Manufacture (3D printing)

2019-10-15T04:31:10Z

A1713568:

[[Category:Projects]]
[[Category:Final Year Projects]]
[[Category:2019s2|24501]]
== Introduction ==
AML3D is one of the leading company in the Wire Arc Additive Manufacturing (WAAM) industry. In short, WAAM is the Metal 3D printing. The technology has the ability to reduce up to 75% of manufacture time, cost, and material. Currently, WAAM is benefiting other industry such as: mining, defence, marine, and aviation.

[[File:aml3d_logo.jpg|100px]]
=== Project team ===
==== Project students ====
* Anh Tran
* Nhat Nguyen
==== Supervisors ====
* Dr. Brian Ng
* Dr. Paul Colegrove (AML3D, sponsored company)
=== Objectives ===
The objective of this project is to increase the efficiency of the manufacture process at AML3D. In order to do so, the team will investigate into the possibility of automating and optimising the quality control processes. The two quality control processes that are currently being implemented at AML3D are measuring layer height using laser sensors, and human supervision for detecting defects. These processes add overhead into production time and usage of human resource, which is not desired. To achieve the goal, machine learning methods will be used extensively to analyse the electrical signatures of the weld process.

== Background ==
=== Wire Arc Additive Manufacturing ===
Wire and Arc Additive Manufacturing (WAAM) is a type of additive manufacturing that uses electric arc as the heat source and material wire to feed the manufacture process <ref name="WAAM"> S. W. Williams, F. Martina, A. C. Addison, J. Ding, G. Pardal & P. Colegrove (2016) Wire + Arc Additive Manufacturing, Materials Science and Technology, 32:7, 641-647, DOI: 10.1179/1743284715Y.0000000073 </ref>. WAAM has been investigated since the 1990s <ref name="WAAM" />, but only recently that it received more attention from the manufacture world.

Its significant comes from the ability to manufacture complex model with less time and less material. Figure 1a and 1b show real parts that was manufactured by AML3D. Such custom made parts might takes months before be ready to be shipped, but with WAAM, the production time can reduce down to weeks. Currently, the industries that benefit the most from WAAM are maritime and aerospace.

Similar to other additive manufacture methods, WAAM achieves such results by sliding the models into multiple layers, and then build the model layer by layer. The movement control is normally handled by a robotic arm (AML3D uses ABB's Arc Welder robot), and the welding path is generated by a Computer Aid Manufacture (CAM) software.

[[File:welding_part1.JPG|thumb|Figure 1a: A propeller]]
[[File:welding_part2.jpg|thumb|Figure 1b: Another part of the ship]]
[[File:abb_robot.jpg|thumb|Figure 2: ABB's IRB 1520ID Arc Welder]]

=== Gas metal arc welding ===
<div><ul>
<li style="display: inline-block;"> [[File:gmaw_welding.png|thumb|500px|none|Figure 5: GMAW welding process <ref name="welding_book"> Norrish, J. (2006). Advanced welding processes. Elsevier. </ref>]] </li>
</ul></div>

Gas metal arc welding (GMAW), or also known as metal inert gas (MIG), is the welding process used at AML3D for WAAM. Due to its high deposition rate and economic benefits, GMAW has became more popular. We will explore two variants of GMAW, which are Pulse Multi Control and Cold Metal Transfer. Both processes are developed by Fronius, an Austrian welding company. Note that it is not required to understand the welding physics to follow this wiki. The next two following sections' purpose is to highlight the high variability, high dynamic nature in welding.

[[File:fronius.png|thumb|Figure 6: Fronius logo]]

==== Pulse Multi Control ====
Pulse Multi Control (PMC) welding is Fronius' modified Pulsed GMAW. The advantage of Pulsed GMAW is the ability to control the metal droplet transfer in welding. Note that from figure 7, the metal droplet is characterised by a downward slope of the current pulse. Figure 8 shows the current signal captured from experiments conducted at AML3D (PMC is used).

<div><ul>
<li style="display: inline-block;"> [[File:Pulsed_gmaw_transfer_mode.png|thumb|500px|none|Figure 7: Relationship between pulse current and droplet transfer<ref name="welding_book" />]] </li>
<li style="display: inline-block;"> [[File:Pmc_current_in_time_domain.png|thumb|650px|none|Figure 8: Current signal captured from experiments conducted at AML3D (PMC process)]] </li>
</ul></div>
==== Cold Metal Transfer ====
Cold Metal Transfer (CMT) is a complex welding process, also developed by Fronius. By detecting a short circuit (mode C and D in figure 9), CMT can make adjustment such as retracting the welding material to cool down, and therefore create a smoother, more stable weld. The complexity of CMT can be seen in the current signal of the process (figure 10). Compare to PMC, CMT's signal has more variation during one signal period.

<div><ul>
<li style="display: inline-block;"> [[File:Transfer_mode.png|thumb|800px|none|Figure 9: Mechanism of metal droplet transfer mode]] </li>
<li style="display: inline-block;"> [[File:Cmt1_current_in_time_domain.png|thumb|735px|none|Figure 10: Current signal captured from experiments conducted at AML3D (CMT process)]] </li>
</ul></div>
=== Machine Learning ===
Machine learning (or statistical learning) refer to a set of tools to give data scientists an insight into the data and make better decisions based on those information <ref name="introduction_ml_book"> James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An introduction to statistical learning (Vol. 112, p. 18). New York: Springer. </ref>. In this project, we will explore the capabilities of classical machine learning methods (i.e anything but Deep Learning) due to the limitation of our available data. By the time of writing this wiki, we have only explored Support Vector Machine.

==== Support Vector Machine ====
The main idea of Support Vector Machine (SVM) is to draw hyperplanes that best separate multiple classes of data. The original SVM judges best decision hyperplane based on the separation margin between data classes (figure 11). However, there are multiple variants of SVM, such as hard/soft margin SVM, Nu-SVM, One class SVM (to solve anomaly detection problem). In our project, we choose to use a soft margin implementation of SVM. The benefit of soft margin SVM (as oppose to hard margin SVM) is that by allowing some misclassifications, we get a wider margin and therefore the solution becomes more "generalise" (i.e work better with a new set of dataset that is independent from the training dataset).

<div><ul>
<li style="display: inline-block;"> [[File:svm_margin.png|thumb|400px|none|Figure 11: Separation margin <ref name="bishop_book"> Bishop, C. M. (2006). Pattern recognition and machine learning. Springer. </ref>]] </li>
</ul></div>

==== Kernel tricks ====
Both figure 11 and 12 are of linearly separable cases for SVM. For dataset that is not easy to separate with a hyperplane in their original features space (such as figure 13), the data often get projected into a higher dimension space where it is easier to separate (figure 14). Such technique is called the Kernel Tricks. It is important to note that Kernel Tricks does not actually "map" the data from '''X''' dimension to '''Z''' dimension, but rather pairwise compare the similarity of the input space in '''Z''' dimension.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_nonseparable.png |thumb|350px|none|Figure 13: Nonseparable dataset <ref name="bishop_book"></ref>]] </li>
<li style="display: inline-block;"> [[File:Svm_kernel_trick.png |thumb|650px|none|Figure 14: Simplified example of the Kernel tricks]] </li>
</ul></div>

== Previous studies ==
There are many researches into defect detection in GMAW process. The three most dominant analysis methods are spectroscopic analysis, acoustic analysis, and electrical signal analysis.

=== Spectroscopic analysis ===
In <ref name="spectroscopic_analysis">D. Bebiano and S. Alfaro, “A weld defects detection system based on a spectrometer,” Sensors, vol. 9, no. 4, pp. 2851–2861, 2009.</ref>, the authors captured the intensity of radiation emission of electric arc. It is obvious from figure 15 that good weld's intensity stays stable while defect one's is characterised by a sudden peak. Therefore, a threshold is set to detect the defect welds.

<div><ul>
<li style="display: inline-block;"> [[File:spectroscopic_analysis.png|thumb|400px|none|Figure 15: Spectroscopic analysis <ref name="spectroscopic_analysis"></ref>]] </li>
</ul></div>

Other research into using spectroscopic data to monitor that can be of interest is <ref>D. Naso, B. Turchiano, and P. Pantaleo, “A fuzzy-logic based optical sensor for online weld defect-detection,” IEEE transactions on Industrial Informatics, vol. 1, no. 4, pp. 259–273, 2005.</ref>.

=== Acoustic analysis ===
With <ref name="acoustic_analysis">E. Cayo and S. C. Alfaro, “A non-intrusive GMA welding process quality monitoring system using acoustic sensing,” Sensors, vol. 9, no. 9, pp. 7150–7166, 2009.</ref>, the author discovered that arc ignition is characterised by a distinct sound and therefore it is possible to capture the frequency of ignition through acoustic sensors. A tolerance band was set and both ignition frequency and sound pressure level were used to monitor the quality of the weld.

<div><ul>
<li style="display: inline-block;"> [[File:acoustic_analysis.png|thumb|400px|none|Figure 16: Acoustic analysis <ref name="acoustic_analysis"></ref>]] </li>
</ul></div>

Other research into using acoustic signal to monitor that can be of interest are <ref>M. Fidali, “Detection of welding process instabilities using acoustic signals,” in International Congress on Technical Diagnostic, pp. 191–201, Springer, 2016.</ref>, <ref>L. Zhang, A. C. Basantes-Defaz, D. Ozevin, and E. Indacochea, “Real-time monitoring of welding process using air-coupled ultrasonics and acoustic emission,” The International Journal of Advanced Manufacturing Technology, vol. 101, no. 5-8, pp. 1623–1634, 2019.</ref>.

=== Electrical signal analysis ===
With <ref name="electrical_analysis_1">A. Sumesh, K. Rameshkumar, A. Raja, K. Mohandas, A. Santhakumari, and R. Shyambabu, “Establishing correlation between current and voltage signatures of the arc and weld defects in GMAW process,” Arabian Journal for Science and Engineering, vol. 42, no. 11, pp. 4649–4665, 2017.</ref>, the author plot the probability density distribution of voltage signal and discovered that the stable weld's signal concentrate in one region while unstable signal spread out. While in the paper, the authors did not developed a monitoring method using their findings, it shows that there is a significant differences between stable and unstable weld's electrical signature.

<div><ul>
<li style="display: inline-block;"> [[File:Electrical analysis pdd.png|thumb|400px|none|Figure 17: Voltage PDD <ref name="electrical_analysis_1"></ref>]] </li>
</ul></div>

In <ref name="electrical_analysis_2">S. Simpson, “Signature images for arc welding fault detection,” Science and Technology of Welding and Joining, vol. 12, no. 6, pp. 481–486, 2007.</ref>, the authors shows that by plotting the signature image of voltage and current scatter plot, both arcing and short-circuiting region's signature image can notify a defects welds. A more visual representation of short circuiting and arcing region can be found in [[#Gas metal arc welding|Gas metal arc welding section]].

<div><ul>
<li style="display: inline-block;"> [[File:Electrical voltage signature.png|thumb|400px|none|Figure 18a: Electrical voltage signature of stable weld <ref name="electrical_analysis_2"></ref>]] </li>
<li style="display: inline-block;"> [[File:Electrical signature 2.png|thumb|400px|none|Figure 18b: Electrical voltage signature of unstable weld <ref name="electrical_analysis_2"></ref>]] </li>
</ul></div>

Other research into using electrical signal to monitor that can be of interest are <ref>R. Madigan, “Arc sensing for defects in constant-voltage gas metal arc welding,” Welding Journal, vol. 78, pp. 322S–328S, 1999.</ref>, <ref>Y. Huang, K. Wang, Z. Zhou, X. Zhou, and J. Fang, “Stability evaluation of short-circuiting gas metal arc welding based on ensemble empirical mode decomposition,” Measurement Science and Technology, vol. 28, no. 3, p. 035006, 2017.</ref>,<ref>Z. Zhang, X. Chen, H. Chen, J. Zhong, and S. Chen, “Online welding quality monitoring based on feature extraction of arc voltage signal,” The International Journal of Advanced Manufacturing Technology, vol. 70, no. 9-12, pp. 1661–1671, 2014</ref>,<ref>X. Li and S. Simpson, “Parametric approach to positional fault detection in short arc welding,” Science and Technology of welding and joining, vol. 14, no. 2, pp. 146–151, 2009.</ref>, and <ref>E. Wei, D. Farson, R. Richardson, and H. Ludewig, “Detection of weld surface porosity by statistical analysis of arc current in gas metal arc welding,” Journal of Manufacturing Processes, vol. 3, no. 1, pp. 50–59, 2001.</ref>.

== Experiment setup ==
The welding system uses TPS 320i C Pulse (figure 15a), produced by Fronius, to supply power, feed wire, and system cooling. For movement control of the welding torch, the system use ABB’s IRB 1520ID (figure 15b), which is a dedicated arc welding robot. The Data Acquisition device is NI's DAQ. Each channel (voltage and current) is sampled at 25kHz while the dominant frequency for voltage and current signal are around 130Hz (for CMT process) and 190Hz (for PMC process).

<div><ul>
<li style="display: inline-block;"> [[File:Fronius_power_supply.jpg |thumb|250px|none|Figure 19a: Fronius' Power Supply]] </li>
<li style="display: inline-block;"> [[File:Welding_robot.jpg |thumb|250px|none|Figure 19b: ABB's welding robot]] </li>
</ul></div>

[[File:Ni_daq.jpg|thumb|Figure 16: NI's DAQ device]]

== Statistical Analysis ==

=== Preprocessing ===
As the captured signal coming from Fronius' power supply, the signals also pick up switching power noise. The noise will interfere with the accuracy of the analysis if not filtered. In this project, we applied a 1000Hz digital Low Pass Butterworth filter. The effect of the filter can be seen in figure 20, while the comparation of using different filters can be seen in the gallery.

<div><ul>
<li style="display: inline-block;"> [[File:Filter_effect.png |thumb|750px|none|Figure 20: Effect of 1000Hz low pass filter]] </li>
<li style="display: inline-block;"> [[File:Freq_response.png |thumb|750px|none|Figure 21: Frequency response of the in-use filter]] </li>
</ul></div>

<gallery mode="packed-hover" caption="Gallery: Comparison of different filtering method" perrow=2>
File:Effect_of_filtering2.png|In use filter vs 300Hz low pass filter
File:Effect_of_filtering3.png|In use filter vs 500Hz low pass filter
File:Effect_of_filtering4.png|In use filter vs 750Hz low pass filter
File:Effect_of_filtering5.png|In use filter vs 2000Hz low pass filter
</gallery>

=== Power Spectral Density ===
Our goal for this project is to detect the defect welds and predict the layer heights (the contact tip to work piece distance, CTWD, to be more accurate) in real time (refer to [[#Relating Contact Tip to Work-piece Distance to layer height|this section]] to understand the relationship between Contact Tip to Work-piece Distance and Layer height). An analysis of the current power spectral density (PSD) shows that there is an inverse proportional relationship between the CTWD of a weld and the frequency where its '''current signal''' PSD reach the peak, and also the magnitude of the peak. Since the peak of PSD can be fairly sensitive, we only consider the frequency of the peak PSD.

Figure 22 shows the PSD plot of voltage, current, and power signal of the weld using PMC process. In this experiment, we created multiple one-line, 8 seconds welds. These eight-second welds then get segmented into smaller two-second chunk to increase the number of data and decrease the processing time.

<div><ul>
<li style="display: inline-block;"> [[File:Psd pmc good bad all lowfreq.png |thumb|750px|none|Figure 22: Power spectral density of PMC process]] </li>
<li style="display: inline-block;"> [[File:One line weld.jpg |thumb|578px|none|Figure 23: Simple one-line weld experiment]] </li>
<li style="display: inline-block;"> [[File:Cmt_psd_stable_allCTWD_zoomed.png |thumb|750px|none|Figure 24: Simple one-line weld experiment]] </li>

</ul></div>

It seems like the inverse proportional relationship holds true for most cases. Other experiments showed that for a short period at the beginning and ending of the weld, the signal can behave differently relative to the rest of the weld. Currently, to migrate with this problem, data will not be collected and analysed (both offline and online) at these periods. However, the above results seem to only hold true for PMC processes. With CMT process, the current PSD plot indicates no relationship between the CTWD and the frequency at peak (figure 24).

For the next stage of the project, our plan is to develop a linear regression model to predict the CTWD for PMC process. With CMT process, we will have to look into different analysis techniques.

=== Support Vector Machine ===
By looking at the average voltage and average current of each segment of weld while making a simple straight path (PMC process), we realised there is a visible separation margin between stable and unstable welds. Using SVM with RBF kernel, we were able to correctly classify the quality of the weld with accuracy of 98%.

However, with CMT process, the separation cannot be observed.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_pmcNoOxide_rbf.png |thumb|750px|none|Figure 20: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:CmtNoOxide_average_voltage_current.png |thumb|750px|none|Figure 20: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

=== CTWD Regression ===
With CTWD prediction, the result will only be useful if it is done on-the-line while making the part. And therefore, as oppose to defect detection, we cannot have a short calibration process every constant time to check for worn contact tip. By analysing the data while building a real part, we found that the average voltage (for PMC process) and average current (for CMT process) seems to have the strongest linear relationship with the CTWD. The acceptable error for CTWD prediction is +/- 1 mm (according to AML3D requirement). For both PMC and CMT process, our root means square error is less than 1 (PMC: 0.68 and CMT: 0.83), which satisfied the requirement.

<div><ul>
<li style="display: inline-block;"> [[File:Linear_regression_pmc_wall.png |thumb|750px|none|Figure 20: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:Linear_regression_cmt_wall.png |thumb|750px|none|Figure 20: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

== Result discussion ==
These results show that there is a strong possibility of defect detection and CTWD prediction in welding. However, these initial findings are still a proof of concept. In the next stage of our research, we will refine the algorithm and integrate into AML's pipeline.

== References ==
<references />

Projects:2019s2-20001 Using Machine Learning to Determine Deposit Height and Defects for Wire + Arc Additive Manufacture (3D printing)

2019-10-15T04:30:46Z

A1713568: /* Result discussion */

[[Category:Projects]]
[[Category:Final Year Projects]]
[[Category:2019s2|24501]]
== Introduction ==
AML3D is one of the leading company in the Wire Arc Additive Manufacturing (WAAM) industry. In short, WAAM is the Metal 3D printing. The technology has the ability to reduce up to 75% of manufacture time, cost, and material. Currently, WAAM is benefiting other industry such as: mining, defence, marine, and aviation.

[[File:aml3d_logo.jpg|100px]]
=== Project team ===
==== Project students ====
* Anh Tran
* Nhat Nguyen
==== Supervisors ====
* Dr. Brian Ng
* Dr. Paul Colegrove (AML3D, sponsored company)
=== Objectives ===
The objective of this project is to increase the efficiency of the manufacture process at AML3D. In order to do so, the team will investigate into the possibility of automating and optimising the quality control processes. The two quality control processes that are currently being implemented at AML3D are measuring layer height using laser sensors, and human supervision for detecting defects. These processes add overhead into production time and usage of human resource, which is not desired. To achieve the goal, machine learning methods will be used extensively to analyse the electrical signatures of the weld process.

== Background ==
=== Wire Arc Additive Manufacturing ===
Wire and Arc Additive Manufacturing (WAAM) is a type of additive manufacturing that uses electric arc as the heat source and material wire to feed the manufacture process <ref name="WAAM"> S. W. Williams, F. Martina, A. C. Addison, J. Ding, G. Pardal & P. Colegrove (2016) Wire + Arc Additive Manufacturing, Materials Science and Technology, 32:7, 641-647, DOI: 10.1179/1743284715Y.0000000073 </ref>. WAAM has been investigated since the 1990s <ref name="WAAM" />, but only recently that it received more attention from the manufacture world.

Its significant comes from the ability to manufacture complex model with less time and less material. Figure 1a and 1b show real parts that was manufactured by AML3D. Such custom made parts might takes months before be ready to be shipped, but with WAAM, the production time can reduce down to weeks. Currently, the industries that benefit the most from WAAM are maritime and aerospace.

Similar to other additive manufacture methods, WAAM achieves such results by sliding the models into multiple layers, and then build the model layer by layer. The movement control is normally handled by a robotic arm (AML3D uses ABB's Arc Welder robot), and the welding path is generated by a Computer Aid Manufacture (CAM) software.

[[File:welding_part1.JPG|thumb|Figure 1a: A propeller]]
[[File:welding_part2.jpg|thumb|Figure 1b: Another part of the ship]]
[[File:abb_robot.jpg|thumb|Figure 2: ABB's IRB 1520ID Arc Welder]]

=== Gas metal arc welding ===
<div><ul>
<li style="display: inline-block;"> [[File:gmaw_welding.png|thumb|500px|none|Figure 5: GMAW welding process <ref name="welding_book"> Norrish, J. (2006). Advanced welding processes. Elsevier. </ref>]] </li>
</ul></div>

Gas metal arc welding (GMAW), or also known as metal inert gas (MIG), is the welding process used at AML3D for WAAM. Due to its high deposition rate and economic benefits, GMAW has became more popular. We will explore two variants of GMAW, which are Pulse Multi Control and Cold Metal Transfer. Both processes are developed by Fronius, an Austrian welding company. Note that it is not required to understand the welding physics to follow this wiki. The next two following sections' purpose is to highlight the high variability, high dynamic nature in welding.

[[File:fronius.png|thumb|Figure 6: Fronius logo]]

==== Pulse Multi Control ====
Pulse Multi Control (PMC) welding is Fronius' modified Pulsed GMAW. The advantage of Pulsed GMAW is the ability to control the metal droplet transfer in welding. Note that from figure 7, the metal droplet is characterised by a downward slope of the current pulse. Figure 8 shows the current signal captured from experiments conducted at AML3D (PMC is used).

<div><ul>
<li style="display: inline-block;"> [[File:Pulsed_gmaw_transfer_mode.png|thumb|500px|none|Figure 7: Relationship between pulse current and droplet transfer<ref name="welding_book" />]] </li>
<li style="display: inline-block;"> [[File:Pmc_current_in_time_domain.png|thumb|650px|none|Figure 8: Current signal captured from experiments conducted at AML3D (PMC process)]] </li>
</ul></div>
==== Cold Metal Transfer ====
Cold Metal Transfer (CMT) is a complex welding process, also developed by Fronius. By detecting a short circuit (mode C and D in figure 9), CMT can make adjustment such as retracting the welding material to cool down, and therefore create a smoother, more stable weld. The complexity of CMT can be seen in the current signal of the process (figure 10). Compare to PMC, CMT's signal has more variation during one signal period.

<div><ul>
<li style="display: inline-block;"> [[File:Transfer_mode.png|thumb|800px|none|Figure 9: Mechanism of metal droplet transfer mode]] </li>
<li style="display: inline-block;"> [[File:Cmt1_current_in_time_domain.png|thumb|735px|none|Figure 10: Current signal captured from experiments conducted at AML3D (CMT process)]] </li>
</ul></div>
=== Machine Learning ===
Machine learning (or statistical learning) refer to a set of tools to give data scientists an insight into the data and make better decisions based on those information <ref name="introduction_ml_book"> James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An introduction to statistical learning (Vol. 112, p. 18). New York: Springer. </ref>. In this project, we will explore the capabilities of classical machine learning methods (i.e anything but Deep Learning) due to the limitation of our available data. By the time of writing this wiki, we have only explored Support Vector Machine.

==== Support Vector Machine ====
The main idea of Support Vector Machine (SVM) is to draw hyperplanes that best separate multiple classes of data. The original SVM judges best decision hyperplane based on the separation margin between data classes (figure 11). However, there are multiple variants of SVM, such as hard/soft margin SVM, Nu-SVM, One class SVM (to solve anomaly detection problem). In our project, we choose to use a soft margin implementation of SVM. The benefit of soft margin SVM (as oppose to hard margin SVM) is that by allowing some misclassifications, we get a wider margin and therefore the solution becomes more "generalise" (i.e work better with a new set of dataset that is independent from the training dataset).

<div><ul>
<li style="display: inline-block;"> [[File:svm_margin.png|thumb|400px|none|Figure 11: Separation margin <ref name="bishop_book"> Bishop, C. M. (2006). Pattern recognition and machine learning. Springer. </ref>]] </li>
</ul></div>

==== Kernel tricks ====
Both figure 11 and 12 are of linearly separable cases for SVM. For dataset that is not easy to separate with a hyperplane in their original features space (such as figure 13), the data often get projected into a higher dimension space where it is easier to separate (figure 14). Such technique is called the Kernel Tricks. It is important to note that Kernel Tricks does not actually "map" the data from '''X''' dimension to '''Z''' dimension, but rather pairwise compare the similarity of the input space in '''Z''' dimension.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_nonseparable.png |thumb|350px|none|Figure 13: Nonseparable dataset <ref name="bishop_book"></ref>]] </li>
<li style="display: inline-block;"> [[File:Svm_kernel_trick.png |thumb|650px|none|Figure 14: Simplified example of the Kernel tricks]] </li>
</ul></div>

== Previous studies ==
There are many researches into defect detection in GMAW process. The three most dominant analysis methods are spectroscopic analysis, acoustic analysis, and electrical signal analysis.

=== Spectroscopic analysis ===
In <ref name="spectroscopic_analysis">D. Bebiano and S. Alfaro, “A weld defects detection system based on a spectrometer,” Sensors, vol. 9, no. 4, pp. 2851–2861, 2009.</ref>, the authors captured the intensity of radiation emission of electric arc. It is obvious from figure 15 that good weld's intensity stays stable while defect one's is characterised by a sudden peak. Therefore, a threshold is set to detect the defect welds.

<div><ul>
<li style="display: inline-block;"> [[File:spectroscopic_analysis.png|thumb|400px|none|Figure 15: Spectroscopic analysis <ref name="spectroscopic_analysis"></ref>]] </li>
</ul></div>

Other research into using spectroscopic data to monitor that can be of interest is <ref>D. Naso, B. Turchiano, and P. Pantaleo, “A fuzzy-logic based optical sensor for online weld defect-detection,” IEEE transactions on Industrial Informatics, vol. 1, no. 4, pp. 259–273, 2005.</ref>.

=== Acoustic analysis ===
With <ref name="acoustic_analysis">E. Cayo and S. C. Alfaro, “A non-intrusive GMA welding process quality monitoring system using acoustic sensing,” Sensors, vol. 9, no. 9, pp. 7150–7166, 2009.</ref>, the author discovered that arc ignition is characterised by a distinct sound and therefore it is possible to capture the frequency of ignition through acoustic sensors. A tolerance band was set and both ignition frequency and sound pressure level were used to monitor the quality of the weld.

<div><ul>
<li style="display: inline-block;"> [[File:acoustic_analysis.png|thumb|400px|none|Figure 16: Acoustic analysis <ref name="acoustic_analysis"></ref>]] </li>
</ul></div>

Other research into using acoustic signal to monitor that can be of interest are <ref>M. Fidali, “Detection of welding process instabilities using acoustic signals,” in International Congress on Technical Diagnostic, pp. 191–201, Springer, 2016.</ref>, <ref>L. Zhang, A. C. Basantes-Defaz, D. Ozevin, and E. Indacochea, “Real-time monitoring of welding process using air-coupled ultrasonics and acoustic emission,” The International Journal of Advanced Manufacturing Technology, vol. 101, no. 5-8, pp. 1623–1634, 2019.</ref>.

=== Electrical signal analysis ===
With <ref name="electrical_analysis_1">A. Sumesh, K. Rameshkumar, A. Raja, K. Mohandas, A. Santhakumari, and R. Shyambabu, “Establishing correlation between current and voltage signatures of the arc and weld defects in GMAW process,” Arabian Journal for Science and Engineering, vol. 42, no. 11, pp. 4649–4665, 2017.</ref>, the author plot the probability density distribution of voltage signal and discovered that the stable weld's signal concentrate in one region while unstable signal spread out. While in the paper, the authors did not developed a monitoring method using their findings, it shows that there is a significant differences between stable and unstable weld's electrical signature.

<div><ul>
<li style="display: inline-block;"> [[File:Electrical analysis pdd.png|thumb|400px|none|Figure 17: Voltage PDD <ref name="electrical_analysis_1"></ref>]] </li>
</ul></div>

In <ref name="electrical_analysis_2">S. Simpson, “Signature images for arc welding fault detection,” Science and Technology of Welding and Joining, vol. 12, no. 6, pp. 481–486, 2007.</ref>, the authors shows that by plotting the signature image of voltage and current scatter plot, both arcing and short-circuiting region's signature image can notify a defects welds. A more visual representation of short circuiting and arcing region can be found in [[#Gas metal arc welding|Gas metal arc welding section]].

<div><ul>
<li style="display: inline-block;"> [[File:Electrical voltage signature.png|thumb|400px|none|Figure 18a: Electrical voltage signature of stable weld <ref name="electrical_analysis_2"></ref>]] </li>
<li style="display: inline-block;"> [[File:Electrical signature 2.png|thumb|400px|none|Figure 18b: Electrical voltage signature of unstable weld <ref name="electrical_analysis_2"></ref>]] </li>
</ul></div>

Other research into using electrical signal to monitor that can be of interest are <ref>R. Madigan, “Arc sensing for defects in constant-voltage gas metal arc welding,” Welding Journal, vol. 78, pp. 322S–328S, 1999.</ref>, <ref>Y. Huang, K. Wang, Z. Zhou, X. Zhou, and J. Fang, “Stability evaluation of short-circuiting gas metal arc welding based on ensemble empirical mode decomposition,” Measurement Science and Technology, vol. 28, no. 3, p. 035006, 2017.</ref>,<ref>Z. Zhang, X. Chen, H. Chen, J. Zhong, and S. Chen, “Online welding quality monitoring based on feature extraction of arc voltage signal,” The International Journal of Advanced Manufacturing Technology, vol. 70, no. 9-12, pp. 1661–1671, 2014</ref>,<ref>X. Li and S. Simpson, “Parametric approach to positional fault detection in short arc welding,” Science and Technology of welding and joining, vol. 14, no. 2, pp. 146–151, 2009.</ref>, and <ref>E. Wei, D. Farson, R. Richardson, and H. Ludewig, “Detection of weld surface porosity by statistical analysis of arc current in gas metal arc welding,” Journal of Manufacturing Processes, vol. 3, no. 1, pp. 50–59, 2001.</ref>.

== Experiment setup ==
The welding system uses TPS 320i C Pulse (figure 15a), produced by Fronius, to supply power, feed wire, and system cooling. For movement control of the welding torch, the system use ABB’s IRB 1520ID (figure 15b), which is a dedicated arc welding robot. The Data Acquisition device is NI's DAQ. Each channel (voltage and current) is sampled at 25kHz while the dominant frequency for voltage and current signal are around 130Hz (for CMT process) and 190Hz (for PMC process).

<div><ul>
<li style="display: inline-block;"> [[File:Fronius_power_supply.jpg |thumb|250px|none|Figure 19a: Fronius' Power Supply]] </li>
<li style="display: inline-block;"> [[File:Welding_robot.jpg |thumb|250px|none|Figure 19b: ABB's welding robot]] </li>
</ul></div>

[[File:Ni_daq.jpg|thumb|Figure 16: NI's DAQ device]]

== Statistical Analysis ==

=== Preprocessing ===
As the captured signal coming from Fronius' power supply, the signals also pick up switching power noise. The noise will interfere with the accuracy of the analysis if not filtered. In this project, we applied a 1000Hz digital Low Pass Butterworth filter. The effect of the filter can be seen in figure 20, while the comparation of using different filters can be seen in the gallery.

<div><ul>
<li style="display: inline-block;"> [[File:Filter_effect.png |thumb|750px|none|Figure 20: Effect of 1000Hz low pass filter]] </li>
<li style="display: inline-block;"> [[File:Freq_response.png |thumb|750px|none|Figure 21: Frequency response of the in-use filter]] </li>
</ul></div>

<gallery mode="packed-hover" caption="Gallery: Comparison of different filtering method" perrow=2>
File:Effect_of_filtering2.png|In use filter vs 300Hz low pass filter
File:Effect_of_filtering3.png|In use filter vs 500Hz low pass filter
File:Effect_of_filtering4.png|In use filter vs 750Hz low pass filter
File:Effect_of_filtering5.png|In use filter vs 2000Hz low pass filter
</gallery>

=== Power Spectral Density ===
Our goal for this project is to detect the defect welds and predict the layer heights (the contact tip to work piece distance, CTWD, to be more accurate) in real time (refer to [[#Relating Contact Tip to Work-piece Distance to layer height|this section]] to understand the relationship between Contact Tip to Work-piece Distance and Layer height). An analysis of the current power spectral density (PSD) shows that there is an inverse proportional relationship between the CTWD of a weld and the frequency where its '''current signal''' PSD reach the peak, and also the magnitude of the peak. Since the peak of PSD can be fairly sensitive, we only consider the frequency of the peak PSD.

Figure 22 shows the PSD plot of voltage, current, and power signal of the weld using PMC process. In this experiment, we created multiple one-line, 8 seconds welds. These eight-second welds then get segmented into smaller two-second chunk to increase the number of data and decrease the processing time.

<div><ul>
<li style="display: inline-block;"> [[File:Psd pmc good bad all lowfreq.png |thumb|750px|none|Figure 22: Power spectral density of PMC process]] </li>
<li style="display: inline-block;"> [[File:One line weld.jpg |thumb|578px|none|Figure 23: Simple one-line weld experiment]] </li>
<li style="display: inline-block;"> [[File:Cmt_psd_stable_allCTWD_zoomed.png |thumb|750px|none|Figure 24: Simple one-line weld experiment]] </li>

</ul></div>

It seems like the inverse proportional relationship holds true for most cases. Other experiments showed that for a short period at the beginning and ending of the weld, the signal can behave differently relative to the rest of the weld. Currently, to migrate with this problem, data will not be collected and analysed (both offline and online) at these periods. However, the above results seem to only hold true for PMC processes. With CMT process, the current PSD plot indicates no relationship between the CTWD and the frequency at peak (figure 24).

For the next stage of the project, our plan is to develop a linear regression model to predict the CTWD for PMC process. With CMT process, we will have to look into different analysis techniques.

=== Support Vector Machine ===
By looking at the average voltage and average current of each segment of weld while making a simple straight path (PMC process), we realised there is a visible separation margin between stable and unstable welds. Using SVM with RBF kernel, we were able to correctly classify the quality of the weld with accuracy of 98%.

However, with CMT process, the separation cannot be observed.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_pmcNoOxide_rbf.png |thumb|750px|none|Figure 20: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:CmtNoOxide_average_voltage_current.png |thumb|750px|none|Figure 20: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

=== CTWD Regression ===
With CTWD prediction, the result will only be useful if it is done on-the-line while making the part. And therefore, as oppose to defect detection, we cannot have a short calibration process every constant time to check for worn contact tip. By analysing the data while building a real part, we found that the average voltage (for PMC process) and average current (for CMT process) seems to have the strongest linear relationship with the CTWD. The acceptable error for CTWD prediction is +/- 1 mm (according to AML3D requirement). For both PMC and CMT process, our root means square error is less than 1 (PMC: 0.68 and CMT: 0.83), which satisfied the requirement.

<div><ul>
<li style="display: inline-block;"> [[File:Linear_regression_pmc_wall.png |thumb|750px|none|Figure 20: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:Linear_regression_cmt_wall.png |thumb|750px|none|Figure 20: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

== Result discussion ==
These results show that there is a strong possibility of defect detection and CTWD prediction in welding. However, these initial findings are still a proof of concept. In the next stage of our research, we will refine the algorithm and integrate into AML's pipeline.

== Conclusion ==

== Future work ==

== References ==
<references />

Projects:2019s2-20001 Using Machine Learning to Determine Deposit Height and Defects for Wire + Arc Additive Manufacture (3D printing)

2019-10-15T04:12:23Z

A1713568: /* CTWD Regression */

[[Category:Projects]]
[[Category:Final Year Projects]]
[[Category:2019s2|24501]]
== Introduction ==
AML3D is one of the leading company in the Wire Arc Additive Manufacturing (WAAM) industry. In short, WAAM is the Metal 3D printing. The technology has the ability to reduce up to 75% of manufacture time, cost, and material. Currently, WAAM is benefiting other industry such as: mining, defence, marine, and aviation.

[[File:aml3d_logo.jpg|100px]]
=== Project team ===
==== Project students ====
* Anh Tran
* Nhat Nguyen
==== Supervisors ====
* Dr. Brian Ng
* Dr. Paul Colegrove (AML3D, sponsored company)
=== Objectives ===
The objective of this project is to increase the efficiency of the manufacture process at AML3D. In order to do so, the team will investigate into the possibility of automating and optimising the quality control processes. The two quality control processes that are currently being implemented at AML3D are measuring layer height using laser sensors, and human supervision for detecting defects. These processes add overhead into production time and usage of human resource, which is not desired. To achieve the goal, machine learning methods will be used extensively to analyse the electrical signatures of the weld process.

== Background ==
=== Wire Arc Additive Manufacturing ===
Wire and Arc Additive Manufacturing (WAAM) is a type of additive manufacturing that uses electric arc as the heat source and material wire to feed the manufacture process <ref name="WAAM"> S. W. Williams, F. Martina, A. C. Addison, J. Ding, G. Pardal & P. Colegrove (2016) Wire + Arc Additive Manufacturing, Materials Science and Technology, 32:7, 641-647, DOI: 10.1179/1743284715Y.0000000073 </ref>. WAAM has been investigated since the 1990s <ref name="WAAM" />, but only recently that it received more attention from the manufacture world.

Its significant comes from the ability to manufacture complex model with less time and less material. Figure 1a and 1b show real parts that was manufactured by AML3D. Such custom made parts might takes months before be ready to be shipped, but with WAAM, the production time can reduce down to weeks. Currently, the industries that benefit the most from WAAM are maritime and aerospace.

Similar to other additive manufacture methods, WAAM achieves such results by sliding the models into multiple layers, and then build the model layer by layer. The movement control is normally handled by a robotic arm (AML3D uses ABB's Arc Welder robot), and the welding path is generated by a Computer Aid Manufacture (CAM) software.

[[File:welding_part1.JPG|thumb|Figure 1a: A propeller]]
[[File:welding_part2.jpg|thumb|Figure 1b: Another part of the ship]]
[[File:abb_robot.jpg|thumb|Figure 2: ABB's IRB 1520ID Arc Welder]]

=== Gas metal arc welding ===
<div><ul>
<li style="display: inline-block;"> [[File:gmaw_welding.png|thumb|500px|none|Figure 5: GMAW welding process <ref name="welding_book"> Norrish, J. (2006). Advanced welding processes. Elsevier. </ref>]] </li>
</ul></div>

Gas metal arc welding (GMAW), or also known as metal inert gas (MIG), is the welding process used at AML3D for WAAM. Due to its high deposition rate and economic benefits, GMAW has became more popular. We will explore two variants of GMAW, which are Pulse Multi Control and Cold Metal Transfer. Both processes are developed by Fronius, an Austrian welding company. Note that it is not required to understand the welding physics to follow this wiki. The next two following sections' purpose is to highlight the high variability, high dynamic nature in welding.

[[File:fronius.png|thumb|Figure 6: Fronius logo]]

==== Pulse Multi Control ====
Pulse Multi Control (PMC) welding is Fronius' modified Pulsed GMAW. The advantage of Pulsed GMAW is the ability to control the metal droplet transfer in welding. Note that from figure 7, the metal droplet is characterised by a downward slope of the current pulse. Figure 8 shows the current signal captured from experiments conducted at AML3D (PMC is used).

<div><ul>
<li style="display: inline-block;"> [[File:Pulsed_gmaw_transfer_mode.png|thumb|500px|none|Figure 7: Relationship between pulse current and droplet transfer<ref name="welding_book" />]] </li>
<li style="display: inline-block;"> [[File:Pmc_current_in_time_domain.png|thumb|650px|none|Figure 8: Current signal captured from experiments conducted at AML3D (PMC process)]] </li>
</ul></div>
==== Cold Metal Transfer ====
Cold Metal Transfer (CMT) is a complex welding process, also developed by Fronius. By detecting a short circuit (mode C and D in figure 9), CMT can make adjustment such as retracting the welding material to cool down, and therefore create a smoother, more stable weld. The complexity of CMT can be seen in the current signal of the process (figure 10). Compare to PMC, CMT's signal has more variation during one signal period.

<div><ul>
<li style="display: inline-block;"> [[File:Transfer_mode.png|thumb|800px|none|Figure 9: Mechanism of metal droplet transfer mode]] </li>
<li style="display: inline-block;"> [[File:Cmt1_current_in_time_domain.png|thumb|735px|none|Figure 10: Current signal captured from experiments conducted at AML3D (CMT process)]] </li>
</ul></div>
=== Machine Learning ===
Machine learning (or statistical learning) refer to a set of tools to give data scientists an insight into the data and make better decisions based on those information <ref name="introduction_ml_book"> James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An introduction to statistical learning (Vol. 112, p. 18). New York: Springer. </ref>. In this project, we will explore the capabilities of classical machine learning methods (i.e anything but Deep Learning) due to the limitation of our available data. By the time of writing this wiki, we have only explored Support Vector Machine.

==== Support Vector Machine ====
The main idea of Support Vector Machine (SVM) is to draw hyperplanes that best separate multiple classes of data. The original SVM judges best decision hyperplane based on the separation margin between data classes (figure 11). However, there are multiple variants of SVM, such as hard/soft margin SVM, Nu-SVM, One class SVM (to solve anomaly detection problem). In our project, we choose to use a soft margin implementation of SVM. The benefit of soft margin SVM (as oppose to hard margin SVM) is that by allowing some misclassifications, we get a wider margin and therefore the solution becomes more "generalise" (i.e work better with a new set of dataset that is independent from the training dataset).

<div><ul>
<li style="display: inline-block;"> [[File:svm_margin.png|thumb|400px|none|Figure 11: Separation margin <ref name="bishop_book"> Bishop, C. M. (2006). Pattern recognition and machine learning. Springer. </ref>]] </li>
</ul></div>

==== Kernel tricks ====
Both figure 11 and 12 are of linearly separable cases for SVM. For dataset that is not easy to separate with a hyperplane in their original features space (such as figure 13), the data often get projected into a higher dimension space where it is easier to separate (figure 14). Such technique is called the Kernel Tricks. It is important to note that Kernel Tricks does not actually "map" the data from '''X''' dimension to '''Z''' dimension, but rather pairwise compare the similarity of the input space in '''Z''' dimension.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_nonseparable.png |thumb|350px|none|Figure 13: Nonseparable dataset <ref name="bishop_book"></ref>]] </li>
<li style="display: inline-block;"> [[File:Svm_kernel_trick.png |thumb|650px|none|Figure 14: Simplified example of the Kernel tricks]] </li>
</ul></div>

== Previous studies ==
There are many researches into defect detection in GMAW process. The three most dominant analysis methods are spectroscopic analysis, acoustic analysis, and electrical signal analysis.

=== Spectroscopic analysis ===
In <ref name="spectroscopic_analysis">D. Bebiano and S. Alfaro, “A weld defects detection system based on a spectrometer,” Sensors, vol. 9, no. 4, pp. 2851–2861, 2009.</ref>, the authors captured the intensity of radiation emission of electric arc. It is obvious from figure 15 that good weld's intensity stays stable while defect one's is characterised by a sudden peak. Therefore, a threshold is set to detect the defect welds.

<div><ul>
<li style="display: inline-block;"> [[File:spectroscopic_analysis.png|thumb|400px|none|Figure 15: Spectroscopic analysis <ref name="spectroscopic_analysis"></ref>]] </li>
</ul></div>

Other research into using spectroscopic data to monitor that can be of interest is <ref>D. Naso, B. Turchiano, and P. Pantaleo, “A fuzzy-logic based optical sensor for online weld defect-detection,” IEEE transactions on Industrial Informatics, vol. 1, no. 4, pp. 259–273, 2005.</ref>.

=== Acoustic analysis ===
With <ref name="acoustic_analysis">E. Cayo and S. C. Alfaro, “A non-intrusive GMA welding process quality monitoring system using acoustic sensing,” Sensors, vol. 9, no. 9, pp. 7150–7166, 2009.</ref>, the author discovered that arc ignition is characterised by a distinct sound and therefore it is possible to capture the frequency of ignition through acoustic sensors. A tolerance band was set and both ignition frequency and sound pressure level were used to monitor the quality of the weld.

<div><ul>
<li style="display: inline-block;"> [[File:acoustic_analysis.png|thumb|400px|none|Figure 16: Acoustic analysis <ref name="acoustic_analysis"></ref>]] </li>
</ul></div>

Other research into using acoustic signal to monitor that can be of interest are <ref>M. Fidali, “Detection of welding process instabilities using acoustic signals,” in International Congress on Technical Diagnostic, pp. 191–201, Springer, 2016.</ref>, <ref>L. Zhang, A. C. Basantes-Defaz, D. Ozevin, and E. Indacochea, “Real-time monitoring of welding process using air-coupled ultrasonics and acoustic emission,” The International Journal of Advanced Manufacturing Technology, vol. 101, no. 5-8, pp. 1623–1634, 2019.</ref>.

=== Electrical signal analysis ===
With <ref name="electrical_analysis_1">A. Sumesh, K. Rameshkumar, A. Raja, K. Mohandas, A. Santhakumari, and R. Shyambabu, “Establishing correlation between current and voltage signatures of the arc and weld defects in GMAW process,” Arabian Journal for Science and Engineering, vol. 42, no. 11, pp. 4649–4665, 2017.</ref>, the author plot the probability density distribution of voltage signal and discovered that the stable weld's signal concentrate in one region while unstable signal spread out. While in the paper, the authors did not developed a monitoring method using their findings, it shows that there is a significant differences between stable and unstable weld's electrical signature.

<div><ul>
<li style="display: inline-block;"> [[File:Electrical analysis pdd.png|thumb|400px|none|Figure 17: Voltage PDD <ref name="electrical_analysis_1"></ref>]] </li>
</ul></div>

In <ref name="electrical_analysis_2">S. Simpson, “Signature images for arc welding fault detection,” Science and Technology of Welding and Joining, vol. 12, no. 6, pp. 481–486, 2007.</ref>, the authors shows that by plotting the signature image of voltage and current scatter plot, both arcing and short-circuiting region's signature image can notify a defects welds. A more visual representation of short circuiting and arcing region can be found in [[#Gas metal arc welding|Gas metal arc welding section]].

<div><ul>
<li style="display: inline-block;"> [[File:Electrical voltage signature.png|thumb|400px|none|Figure 18a: Electrical voltage signature of stable weld <ref name="electrical_analysis_2"></ref>]] </li>
<li style="display: inline-block;"> [[File:Electrical signature 2.png|thumb|400px|none|Figure 18b: Electrical voltage signature of unstable weld <ref name="electrical_analysis_2"></ref>]] </li>
</ul></div>

Other research into using electrical signal to monitor that can be of interest are <ref>R. Madigan, “Arc sensing for defects in constant-voltage gas metal arc welding,” Welding Journal, vol. 78, pp. 322S–328S, 1999.</ref>, <ref>Y. Huang, K. Wang, Z. Zhou, X. Zhou, and J. Fang, “Stability evaluation of short-circuiting gas metal arc welding based on ensemble empirical mode decomposition,” Measurement Science and Technology, vol. 28, no. 3, p. 035006, 2017.</ref>,<ref>Z. Zhang, X. Chen, H. Chen, J. Zhong, and S. Chen, “Online welding quality monitoring based on feature extraction of arc voltage signal,” The International Journal of Advanced Manufacturing Technology, vol. 70, no. 9-12, pp. 1661–1671, 2014</ref>,<ref>X. Li and S. Simpson, “Parametric approach to positional fault detection in short arc welding,” Science and Technology of welding and joining, vol. 14, no. 2, pp. 146–151, 2009.</ref>, and <ref>E. Wei, D. Farson, R. Richardson, and H. Ludewig, “Detection of weld surface porosity by statistical analysis of arc current in gas metal arc welding,” Journal of Manufacturing Processes, vol. 3, no. 1, pp. 50–59, 2001.</ref>.

== Experiment setup ==
The welding system uses TPS 320i C Pulse (figure 15a), produced by Fronius, to supply power, feed wire, and system cooling. For movement control of the welding torch, the system use ABB’s IRB 1520ID (figure 15b), which is a dedicated arc welding robot. The Data Acquisition device is NI's DAQ. Each channel (voltage and current) is sampled at 25kHz while the dominant frequency for voltage and current signal are around 130Hz (for CMT process) and 190Hz (for PMC process).

<div><ul>
<li style="display: inline-block;"> [[File:Fronius_power_supply.jpg |thumb|250px|none|Figure 19a: Fronius' Power Supply]] </li>
<li style="display: inline-block;"> [[File:Welding_robot.jpg |thumb|250px|none|Figure 19b: ABB's welding robot]] </li>
</ul></div>

[[File:Ni_daq.jpg|thumb|Figure 16: NI's DAQ device]]

== Statistical Analysis ==

=== Preprocessing ===
As the captured signal coming from Fronius' power supply, the signals also pick up switching power noise. The noise will interfere with the accuracy of the analysis if not filtered. In this project, we applied a 1000Hz digital Low Pass Butterworth filter. The effect of the filter can be seen in figure 20, while the comparation of using different filters can be seen in the gallery.

<div><ul>
<li style="display: inline-block;"> [[File:Filter_effect.png |thumb|750px|none|Figure 20: Effect of 1000Hz low pass filter]] </li>
<li style="display: inline-block;"> [[File:Freq_response.png |thumb|750px|none|Figure 21: Frequency response of the in-use filter]] </li>
</ul></div>

<gallery mode="packed-hover" caption="Gallery: Comparison of different filtering method" perrow=2>
File:Effect_of_filtering2.png|In use filter vs 300Hz low pass filter
File:Effect_of_filtering3.png|In use filter vs 500Hz low pass filter
File:Effect_of_filtering4.png|In use filter vs 750Hz low pass filter
File:Effect_of_filtering5.png|In use filter vs 2000Hz low pass filter
</gallery>

=== Power Spectral Density ===
Our goal for this project is to detect the defect welds and predict the layer heights (the contact tip to work piece distance, CTWD, to be more accurate) in real time (refer to [[#Relating Contact Tip to Work-piece Distance to layer height|this section]] to understand the relationship between Contact Tip to Work-piece Distance and Layer height). An analysis of the current power spectral density (PSD) shows that there is an inverse proportional relationship between the CTWD of a weld and the frequency where its '''current signal''' PSD reach the peak, and also the magnitude of the peak. Since the peak of PSD can be fairly sensitive, we only consider the frequency of the peak PSD.

Figure 22 shows the PSD plot of voltage, current, and power signal of the weld using PMC process. In this experiment, we created multiple one-line, 8 seconds welds. These eight-second welds then get segmented into smaller two-second chunk to increase the number of data and decrease the processing time.

<div><ul>
<li style="display: inline-block;"> [[File:Psd pmc good bad all lowfreq.png |thumb|750px|none|Figure 22: Power spectral density of PMC process]] </li>
<li style="display: inline-block;"> [[File:One line weld.jpg |thumb|578px|none|Figure 23: Simple one-line weld experiment]] </li>
<li style="display: inline-block;"> [[File:Cmt_psd_stable_allCTWD_zoomed.png |thumb|750px|none|Figure 24: Simple one-line weld experiment]] </li>

</ul></div>

It seems like the inverse proportional relationship holds true for most cases. Other experiments showed that for a short period at the beginning and ending of the weld, the signal can behave differently relative to the rest of the weld. Currently, to migrate with this problem, data will not be collected and analysed (both offline and online) at these periods. However, the above results seem to only hold true for PMC processes. With CMT process, the current PSD plot indicates no relationship between the CTWD and the frequency at peak (figure 24).

For the next stage of the project, our plan is to develop a linear regression model to predict the CTWD for PMC process. With CMT process, we will have to look into different analysis techniques.

=== Support Vector Machine ===
By looking at the average voltage and average current of each segment of weld while making a simple straight path (PMC process), we realised there is a visible separation margin between stable and unstable welds. Using SVM with RBF kernel, we were able to correctly classify the quality of the weld with accuracy of 98%.

However, with CMT process, the separation cannot be observed.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_pmcNoOxide_rbf.png |thumb|750px|none|Figure 20: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:CmtNoOxide_average_voltage_current.png |thumb|750px|none|Figure 20: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

=== CTWD Regression ===
With CTWD prediction, the result will only be useful if it is done on-the-line while making the part. And therefore, as oppose to defect detection, we cannot have a short calibration process every constant time to check for worn contact tip. By analysing the data while building a real part, we found that the average voltage (for PMC process) and average current (for CMT process) seems to have the strongest linear relationship with the CTWD. The acceptable error for CTWD prediction is +/- 1 mm (according to AML3D requirement). For both PMC and CMT process, our root means square error is less than 1 (PMC: 0.68 and CMT: 0.83), which satisfied the requirement.

<div><ul>
<li style="display: inline-block;"> [[File:Linear_regression_pmc_wall.png |thumb|750px|none|Figure 20: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:Linear_regression_cmt_wall.png |thumb|750px|none|Figure 20: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

== Result discussion ==

== Conclusion ==

== Future work ==

== References ==
<references />

Projects:2019s2-20001 Using Machine Learning to Determine Deposit Height and Defects for Wire + Arc Additive Manufacture (3D printing)

2019-10-15T04:09:16Z

A1713568: /* Support Vector Machine */

[[Category:Projects]]
[[Category:Final Year Projects]]
[[Category:2019s2|24501]]
== Introduction ==
AML3D is one of the leading company in the Wire Arc Additive Manufacturing (WAAM) industry. In short, WAAM is the Metal 3D printing. The technology has the ability to reduce up to 75% of manufacture time, cost, and material. Currently, WAAM is benefiting other industry such as: mining, defence, marine, and aviation.

[[File:aml3d_logo.jpg|100px]]
=== Project team ===
==== Project students ====
* Anh Tran
* Nhat Nguyen
==== Supervisors ====
* Dr. Brian Ng
* Dr. Paul Colegrove (AML3D, sponsored company)
=== Objectives ===
The objective of this project is to increase the efficiency of the manufacture process at AML3D. In order to do so, the team will investigate into the possibility of automating and optimising the quality control processes. The two quality control processes that are currently being implemented at AML3D are measuring layer height using laser sensors, and human supervision for detecting defects. These processes add overhead into production time and usage of human resource, which is not desired. To achieve the goal, machine learning methods will be used extensively to analyse the electrical signatures of the weld process.

== Background ==
=== Wire Arc Additive Manufacturing ===
Wire and Arc Additive Manufacturing (WAAM) is a type of additive manufacturing that uses electric arc as the heat source and material wire to feed the manufacture process <ref name="WAAM"> S. W. Williams, F. Martina, A. C. Addison, J. Ding, G. Pardal & P. Colegrove (2016) Wire + Arc Additive Manufacturing, Materials Science and Technology, 32:7, 641-647, DOI: 10.1179/1743284715Y.0000000073 </ref>. WAAM has been investigated since the 1990s <ref name="WAAM" />, but only recently that it received more attention from the manufacture world.

Its significant comes from the ability to manufacture complex model with less time and less material. Figure 1a and 1b show real parts that was manufactured by AML3D. Such custom made parts might takes months before be ready to be shipped, but with WAAM, the production time can reduce down to weeks. Currently, the industries that benefit the most from WAAM are maritime and aerospace.

Similar to other additive manufacture methods, WAAM achieves such results by sliding the models into multiple layers, and then build the model layer by layer. The movement control is normally handled by a robotic arm (AML3D uses ABB's Arc Welder robot), and the welding path is generated by a Computer Aid Manufacture (CAM) software.

[[File:welding_part1.JPG|thumb|Figure 1a: A propeller]]
[[File:welding_part2.jpg|thumb|Figure 1b: Another part of the ship]]
[[File:abb_robot.jpg|thumb|Figure 2: ABB's IRB 1520ID Arc Welder]]

=== Gas metal arc welding ===
<div><ul>
<li style="display: inline-block;"> [[File:gmaw_welding.png|thumb|500px|none|Figure 5: GMAW welding process <ref name="welding_book"> Norrish, J. (2006). Advanced welding processes. Elsevier. </ref>]] </li>
</ul></div>

Gas metal arc welding (GMAW), or also known as metal inert gas (MIG), is the welding process used at AML3D for WAAM. Due to its high deposition rate and economic benefits, GMAW has became more popular. We will explore two variants of GMAW, which are Pulse Multi Control and Cold Metal Transfer. Both processes are developed by Fronius, an Austrian welding company. Note that it is not required to understand the welding physics to follow this wiki. The next two following sections' purpose is to highlight the high variability, high dynamic nature in welding.

[[File:fronius.png|thumb|Figure 6: Fronius logo]]

==== Pulse Multi Control ====
Pulse Multi Control (PMC) welding is Fronius' modified Pulsed GMAW. The advantage of Pulsed GMAW is the ability to control the metal droplet transfer in welding. Note that from figure 7, the metal droplet is characterised by a downward slope of the current pulse. Figure 8 shows the current signal captured from experiments conducted at AML3D (PMC is used).

<div><ul>
<li style="display: inline-block;"> [[File:Pulsed_gmaw_transfer_mode.png|thumb|500px|none|Figure 7: Relationship between pulse current and droplet transfer<ref name="welding_book" />]] </li>
<li style="display: inline-block;"> [[File:Pmc_current_in_time_domain.png|thumb|650px|none|Figure 8: Current signal captured from experiments conducted at AML3D (PMC process)]] </li>
</ul></div>
==== Cold Metal Transfer ====
Cold Metal Transfer (CMT) is a complex welding process, also developed by Fronius. By detecting a short circuit (mode C and D in figure 9), CMT can make adjustment such as retracting the welding material to cool down, and therefore create a smoother, more stable weld. The complexity of CMT can be seen in the current signal of the process (figure 10). Compare to PMC, CMT's signal has more variation during one signal period.

<div><ul>
<li style="display: inline-block;"> [[File:Transfer_mode.png|thumb|800px|none|Figure 9: Mechanism of metal droplet transfer mode]] </li>
<li style="display: inline-block;"> [[File:Cmt1_current_in_time_domain.png|thumb|735px|none|Figure 10: Current signal captured from experiments conducted at AML3D (CMT process)]] </li>
</ul></div>
=== Machine Learning ===
Machine learning (or statistical learning) refer to a set of tools to give data scientists an insight into the data and make better decisions based on those information <ref name="introduction_ml_book"> James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An introduction to statistical learning (Vol. 112, p. 18). New York: Springer. </ref>. In this project, we will explore the capabilities of classical machine learning methods (i.e anything but Deep Learning) due to the limitation of our available data. By the time of writing this wiki, we have only explored Support Vector Machine.

==== Support Vector Machine ====
The main idea of Support Vector Machine (SVM) is to draw hyperplanes that best separate multiple classes of data. The original SVM judges best decision hyperplane based on the separation margin between data classes (figure 11). However, there are multiple variants of SVM, such as hard/soft margin SVM, Nu-SVM, One class SVM (to solve anomaly detection problem). In our project, we choose to use a soft margin implementation of SVM. The benefit of soft margin SVM (as oppose to hard margin SVM) is that by allowing some misclassifications, we get a wider margin and therefore the solution becomes more "generalise" (i.e work better with a new set of dataset that is independent from the training dataset).

<div><ul>
<li style="display: inline-block;"> [[File:svm_margin.png|thumb|400px|none|Figure 11: Separation margin <ref name="bishop_book"> Bishop, C. M. (2006). Pattern recognition and machine learning. Springer. </ref>]] </li>
</ul></div>

==== Kernel tricks ====
Both figure 11 and 12 are of linearly separable cases for SVM. For dataset that is not easy to separate with a hyperplane in their original features space (such as figure 13), the data often get projected into a higher dimension space where it is easier to separate (figure 14). Such technique is called the Kernel Tricks. It is important to note that Kernel Tricks does not actually "map" the data from '''X''' dimension to '''Z''' dimension, but rather pairwise compare the similarity of the input space in '''Z''' dimension.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_nonseparable.png |thumb|350px|none|Figure 13: Nonseparable dataset <ref name="bishop_book"></ref>]] </li>
<li style="display: inline-block;"> [[File:Svm_kernel_trick.png |thumb|650px|none|Figure 14: Simplified example of the Kernel tricks]] </li>
</ul></div>

== Previous studies ==
There are many researches into defect detection in GMAW process. The three most dominant analysis methods are spectroscopic analysis, acoustic analysis, and electrical signal analysis.

=== Spectroscopic analysis ===
In <ref name="spectroscopic_analysis">D. Bebiano and S. Alfaro, “A weld defects detection system based on a spectrometer,” Sensors, vol. 9, no. 4, pp. 2851–2861, 2009.</ref>, the authors captured the intensity of radiation emission of electric arc. It is obvious from figure 15 that good weld's intensity stays stable while defect one's is characterised by a sudden peak. Therefore, a threshold is set to detect the defect welds.

<div><ul>
<li style="display: inline-block;"> [[File:spectroscopic_analysis.png|thumb|400px|none|Figure 15: Spectroscopic analysis <ref name="spectroscopic_analysis"></ref>]] </li>
</ul></div>

Other research into using spectroscopic data to monitor that can be of interest is <ref>D. Naso, B. Turchiano, and P. Pantaleo, “A fuzzy-logic based optical sensor for online weld defect-detection,” IEEE transactions on Industrial Informatics, vol. 1, no. 4, pp. 259–273, 2005.</ref>.

=== Acoustic analysis ===
With <ref name="acoustic_analysis">E. Cayo and S. C. Alfaro, “A non-intrusive GMA welding process quality monitoring system using acoustic sensing,” Sensors, vol. 9, no. 9, pp. 7150–7166, 2009.</ref>, the author discovered that arc ignition is characterised by a distinct sound and therefore it is possible to capture the frequency of ignition through acoustic sensors. A tolerance band was set and both ignition frequency and sound pressure level were used to monitor the quality of the weld.

<div><ul>
<li style="display: inline-block;"> [[File:acoustic_analysis.png|thumb|400px|none|Figure 16: Acoustic analysis <ref name="acoustic_analysis"></ref>]] </li>
</ul></div>

Other research into using acoustic signal to monitor that can be of interest are <ref>M. Fidali, “Detection of welding process instabilities using acoustic signals,” in International Congress on Technical Diagnostic, pp. 191–201, Springer, 2016.</ref>, <ref>L. Zhang, A. C. Basantes-Defaz, D. Ozevin, and E. Indacochea, “Real-time monitoring of welding process using air-coupled ultrasonics and acoustic emission,” The International Journal of Advanced Manufacturing Technology, vol. 101, no. 5-8, pp. 1623–1634, 2019.</ref>.

=== Electrical signal analysis ===
With <ref name="electrical_analysis_1">A. Sumesh, K. Rameshkumar, A. Raja, K. Mohandas, A. Santhakumari, and R. Shyambabu, “Establishing correlation between current and voltage signatures of the arc and weld defects in GMAW process,” Arabian Journal for Science and Engineering, vol. 42, no. 11, pp. 4649–4665, 2017.</ref>, the author plot the probability density distribution of voltage signal and discovered that the stable weld's signal concentrate in one region while unstable signal spread out. While in the paper, the authors did not developed a monitoring method using their findings, it shows that there is a significant differences between stable and unstable weld's electrical signature.

<div><ul>
<li style="display: inline-block;"> [[File:Electrical analysis pdd.png|thumb|400px|none|Figure 17: Voltage PDD <ref name="electrical_analysis_1"></ref>]] </li>
</ul></div>

In <ref name="electrical_analysis_2">S. Simpson, “Signature images for arc welding fault detection,” Science and Technology of Welding and Joining, vol. 12, no. 6, pp. 481–486, 2007.</ref>, the authors shows that by plotting the signature image of voltage and current scatter plot, both arcing and short-circuiting region's signature image can notify a defects welds. A more visual representation of short circuiting and arcing region can be found in [[#Gas metal arc welding|Gas metal arc welding section]].

<div><ul>
<li style="display: inline-block;"> [[File:Electrical voltage signature.png|thumb|400px|none|Figure 18a: Electrical voltage signature of stable weld <ref name="electrical_analysis_2"></ref>]] </li>
<li style="display: inline-block;"> [[File:Electrical signature 2.png|thumb|400px|none|Figure 18b: Electrical voltage signature of unstable weld <ref name="electrical_analysis_2"></ref>]] </li>
</ul></div>

Other research into using electrical signal to monitor that can be of interest are <ref>R. Madigan, “Arc sensing for defects in constant-voltage gas metal arc welding,” Welding Journal, vol. 78, pp. 322S–328S, 1999.</ref>, <ref>Y. Huang, K. Wang, Z. Zhou, X. Zhou, and J. Fang, “Stability evaluation of short-circuiting gas metal arc welding based on ensemble empirical mode decomposition,” Measurement Science and Technology, vol. 28, no. 3, p. 035006, 2017.</ref>,<ref>Z. Zhang, X. Chen, H. Chen, J. Zhong, and S. Chen, “Online welding quality monitoring based on feature extraction of arc voltage signal,” The International Journal of Advanced Manufacturing Technology, vol. 70, no. 9-12, pp. 1661–1671, 2014</ref>,<ref>X. Li and S. Simpson, “Parametric approach to positional fault detection in short arc welding,” Science and Technology of welding and joining, vol. 14, no. 2, pp. 146–151, 2009.</ref>, and <ref>E. Wei, D. Farson, R. Richardson, and H. Ludewig, “Detection of weld surface porosity by statistical analysis of arc current in gas metal arc welding,” Journal of Manufacturing Processes, vol. 3, no. 1, pp. 50–59, 2001.</ref>.

== Experiment setup ==
The welding system uses TPS 320i C Pulse (figure 15a), produced by Fronius, to supply power, feed wire, and system cooling. For movement control of the welding torch, the system use ABB’s IRB 1520ID (figure 15b), which is a dedicated arc welding robot. The Data Acquisition device is NI's DAQ. Each channel (voltage and current) is sampled at 25kHz while the dominant frequency for voltage and current signal are around 130Hz (for CMT process) and 190Hz (for PMC process).

<div><ul>
<li style="display: inline-block;"> [[File:Fronius_power_supply.jpg |thumb|250px|none|Figure 19a: Fronius' Power Supply]] </li>
<li style="display: inline-block;"> [[File:Welding_robot.jpg |thumb|250px|none|Figure 19b: ABB's welding robot]] </li>
</ul></div>

[[File:Ni_daq.jpg|thumb|Figure 16: NI's DAQ device]]

== Statistical Analysis ==

=== Preprocessing ===
As the captured signal coming from Fronius' power supply, the signals also pick up switching power noise. The noise will interfere with the accuracy of the analysis if not filtered. In this project, we applied a 1000Hz digital Low Pass Butterworth filter. The effect of the filter can be seen in figure 20, while the comparation of using different filters can be seen in the gallery.

<div><ul>
<li style="display: inline-block;"> [[File:Filter_effect.png |thumb|750px|none|Figure 20: Effect of 1000Hz low pass filter]] </li>
<li style="display: inline-block;"> [[File:Freq_response.png |thumb|750px|none|Figure 21: Frequency response of the in-use filter]] </li>
</ul></div>

<gallery mode="packed-hover" caption="Gallery: Comparison of different filtering method" perrow=2>
File:Effect_of_filtering2.png|In use filter vs 300Hz low pass filter
File:Effect_of_filtering3.png|In use filter vs 500Hz low pass filter
File:Effect_of_filtering4.png|In use filter vs 750Hz low pass filter
File:Effect_of_filtering5.png|In use filter vs 2000Hz low pass filter
</gallery>

=== Power Spectral Density ===
Our goal for this project is to detect the defect welds and predict the layer heights (the contact tip to work piece distance, CTWD, to be more accurate) in real time (refer to [[#Relating Contact Tip to Work-piece Distance to layer height|this section]] to understand the relationship between Contact Tip to Work-piece Distance and Layer height). An analysis of the current power spectral density (PSD) shows that there is an inverse proportional relationship between the CTWD of a weld and the frequency where its '''current signal''' PSD reach the peak, and also the magnitude of the peak. Since the peak of PSD can be fairly sensitive, we only consider the frequency of the peak PSD.

Figure 22 shows the PSD plot of voltage, current, and power signal of the weld using PMC process. In this experiment, we created multiple one-line, 8 seconds welds. These eight-second welds then get segmented into smaller two-second chunk to increase the number of data and decrease the processing time.

<div><ul>
<li style="display: inline-block;"> [[File:Psd pmc good bad all lowfreq.png |thumb|750px|none|Figure 22: Power spectral density of PMC process]] </li>
<li style="display: inline-block;"> [[File:One line weld.jpg |thumb|578px|none|Figure 23: Simple one-line weld experiment]] </li>
<li style="display: inline-block;"> [[File:Cmt_psd_stable_allCTWD_zoomed.png |thumb|750px|none|Figure 24: Simple one-line weld experiment]] </li>

</ul></div>

It seems like the inverse proportional relationship holds true for most cases. Other experiments showed that for a short period at the beginning and ending of the weld, the signal can behave differently relative to the rest of the weld. Currently, to migrate with this problem, data will not be collected and analysed (both offline and online) at these periods. However, the above results seem to only hold true for PMC processes. With CMT process, the current PSD plot indicates no relationship between the CTWD and the frequency at peak (figure 24).

For the next stage of the project, our plan is to develop a linear regression model to predict the CTWD for PMC process. With CMT process, we will have to look into different analysis techniques.

=== Support Vector Machine ===
By looking at the average voltage and average current of each segment of weld while making a simple straight path (PMC process), we realised there is a visible separation margin between stable and unstable welds. Using SVM with RBF kernel, we were able to correctly classify the quality of the weld with accuracy of 98%.

However, with CMT process, the separation cannot be observed.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_pmcNoOxide_rbf.png |thumb|750px|none|Figure 20: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:CmtNoOxide_average_voltage_current.png |thumb|750px|none|Figure 20: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

=== CTWD Regression ===
With CTWD prediction, we found that the average voltage (for PMC process) and average current (for CMT process) seems to have the strongest linear relationship with the CTWD. The acceptable error for CTWD prediction is +/- 1 mm (according to AML3D requirement). For both PMC and CMT process, our root means square error is less than 1 (PMC: 0.68 and CMT: 0.83).

<div><ul>
<li style="display: inline-block;"> [[File:Linear_regression_pmc_wall.png |thumb|750px|none|Figure 20: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:Linear_regression_cmt_wall.png |thumb|750px|none|Figure 20: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

== Result discussion ==

== Conclusion ==

== Future work ==

== References ==
<references />

Projects:2019s2-20001 Using Machine Learning to Determine Deposit Height and Defects for Wire + Arc Additive Manufacture (3D printing)

2019-10-15T04:08:32Z

A1713568:

[[Category:Projects]]
[[Category:Final Year Projects]]
[[Category:2019s2|24501]]
== Introduction ==
AML3D is one of the leading company in the Wire Arc Additive Manufacturing (WAAM) industry. In short, WAAM is the Metal 3D printing. The technology has the ability to reduce up to 75% of manufacture time, cost, and material. Currently, WAAM is benefiting other industry such as: mining, defence, marine, and aviation.

[[File:aml3d_logo.jpg|100px]]
=== Project team ===
==== Project students ====
* Anh Tran
* Nhat Nguyen
==== Supervisors ====
* Dr. Brian Ng
* Dr. Paul Colegrove (AML3D, sponsored company)
=== Objectives ===
The objective of this project is to increase the efficiency of the manufacture process at AML3D. In order to do so, the team will investigate into the possibility of automating and optimising the quality control processes. The two quality control processes that are currently being implemented at AML3D are measuring layer height using laser sensors, and human supervision for detecting defects. These processes add overhead into production time and usage of human resource, which is not desired. To achieve the goal, machine learning methods will be used extensively to analyse the electrical signatures of the weld process.

== Background ==
=== Wire Arc Additive Manufacturing ===
Wire and Arc Additive Manufacturing (WAAM) is a type of additive manufacturing that uses electric arc as the heat source and material wire to feed the manufacture process <ref name="WAAM"> S. W. Williams, F. Martina, A. C. Addison, J. Ding, G. Pardal & P. Colegrove (2016) Wire + Arc Additive Manufacturing, Materials Science and Technology, 32:7, 641-647, DOI: 10.1179/1743284715Y.0000000073 </ref>. WAAM has been investigated since the 1990s <ref name="WAAM" />, but only recently that it received more attention from the manufacture world.

Its significant comes from the ability to manufacture complex model with less time and less material. Figure 1a and 1b show real parts that was manufactured by AML3D. Such custom made parts might takes months before be ready to be shipped, but with WAAM, the production time can reduce down to weeks. Currently, the industries that benefit the most from WAAM are maritime and aerospace.

Similar to other additive manufacture methods, WAAM achieves such results by sliding the models into multiple layers, and then build the model layer by layer. The movement control is normally handled by a robotic arm (AML3D uses ABB's Arc Welder robot), and the welding path is generated by a Computer Aid Manufacture (CAM) software.

[[File:welding_part1.JPG|thumb|Figure 1a: A propeller]]
[[File:welding_part2.jpg|thumb|Figure 1b: Another part of the ship]]
[[File:abb_robot.jpg|thumb|Figure 2: ABB's IRB 1520ID Arc Welder]]

=== Gas metal arc welding ===
<div><ul>
<li style="display: inline-block;"> [[File:gmaw_welding.png|thumb|500px|none|Figure 5: GMAW welding process <ref name="welding_book"> Norrish, J. (2006). Advanced welding processes. Elsevier. </ref>]] </li>
</ul></div>

Gas metal arc welding (GMAW), or also known as metal inert gas (MIG), is the welding process used at AML3D for WAAM. Due to its high deposition rate and economic benefits, GMAW has became more popular. We will explore two variants of GMAW, which are Pulse Multi Control and Cold Metal Transfer. Both processes are developed by Fronius, an Austrian welding company. Note that it is not required to understand the welding physics to follow this wiki. The next two following sections' purpose is to highlight the high variability, high dynamic nature in welding.

[[File:fronius.png|thumb|Figure 6: Fronius logo]]

==== Pulse Multi Control ====
Pulse Multi Control (PMC) welding is Fronius' modified Pulsed GMAW. The advantage of Pulsed GMAW is the ability to control the metal droplet transfer in welding. Note that from figure 7, the metal droplet is characterised by a downward slope of the current pulse. Figure 8 shows the current signal captured from experiments conducted at AML3D (PMC is used).

<div><ul>
<li style="display: inline-block;"> [[File:Pulsed_gmaw_transfer_mode.png|thumb|500px|none|Figure 7: Relationship between pulse current and droplet transfer<ref name="welding_book" />]] </li>
<li style="display: inline-block;"> [[File:Pmc_current_in_time_domain.png|thumb|650px|none|Figure 8: Current signal captured from experiments conducted at AML3D (PMC process)]] </li>
</ul></div>
==== Cold Metal Transfer ====
Cold Metal Transfer (CMT) is a complex welding process, also developed by Fronius. By detecting a short circuit (mode C and D in figure 9), CMT can make adjustment such as retracting the welding material to cool down, and therefore create a smoother, more stable weld. The complexity of CMT can be seen in the current signal of the process (figure 10). Compare to PMC, CMT's signal has more variation during one signal period.

<div><ul>
<li style="display: inline-block;"> [[File:Transfer_mode.png|thumb|800px|none|Figure 9: Mechanism of metal droplet transfer mode]] </li>
<li style="display: inline-block;"> [[File:Cmt1_current_in_time_domain.png|thumb|735px|none|Figure 10: Current signal captured from experiments conducted at AML3D (CMT process)]] </li>
</ul></div>
=== Machine Learning ===
Machine learning (or statistical learning) refer to a set of tools to give data scientists an insight into the data and make better decisions based on those information <ref name="introduction_ml_book"> James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An introduction to statistical learning (Vol. 112, p. 18). New York: Springer. </ref>. In this project, we will explore the capabilities of classical machine learning methods (i.e anything but Deep Learning) due to the limitation of our available data. By the time of writing this wiki, we have only explored Support Vector Machine.

==== Support Vector Machine ====
The main idea of Support Vector Machine (SVM) is to draw hyperplanes that best separate multiple classes of data. The original SVM judges best decision hyperplane based on the separation margin between data classes (figure 11). However, there are multiple variants of SVM, such as hard/soft margin SVM, Nu-SVM, One class SVM (to solve anomaly detection problem). In our project, we choose to use a soft margin implementation of SVM. The benefit of soft margin SVM (as oppose to hard margin SVM) is that by allowing some misclassifications, we get a wider margin and therefore the solution becomes more "generalise" (i.e work better with a new set of dataset that is independent from the training dataset).

<div><ul>
<li style="display: inline-block;"> [[File:svm_margin.png|thumb|400px|none|Figure 11: Separation margin <ref name="bishop_book"> Bishop, C. M. (2006). Pattern recognition and machine learning. Springer. </ref>]] </li>
</ul></div>

==== Kernel tricks ====
Both figure 11 and 12 are of linearly separable cases for SVM. For dataset that is not easy to separate with a hyperplane in their original features space (such as figure 13), the data often get projected into a higher dimension space where it is easier to separate (figure 14). Such technique is called the Kernel Tricks. It is important to note that Kernel Tricks does not actually "map" the data from '''X''' dimension to '''Z''' dimension, but rather pairwise compare the similarity of the input space in '''Z''' dimension.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_nonseparable.png |thumb|350px|none|Figure 13: Nonseparable dataset <ref name="bishop_book"></ref>]] </li>
<li style="display: inline-block;"> [[File:Svm_kernel_trick.png |thumb|650px|none|Figure 14: Simplified example of the Kernel tricks]] </li>
</ul></div>

== Previous studies ==
There are many researches into defect detection in GMAW process. The three most dominant analysis methods are spectroscopic analysis, acoustic analysis, and electrical signal analysis.

=== Spectroscopic analysis ===
In <ref name="spectroscopic_analysis">D. Bebiano and S. Alfaro, “A weld defects detection system based on a spectrometer,” Sensors, vol. 9, no. 4, pp. 2851–2861, 2009.</ref>, the authors captured the intensity of radiation emission of electric arc. It is obvious from figure 15 that good weld's intensity stays stable while defect one's is characterised by a sudden peak. Therefore, a threshold is set to detect the defect welds.

<div><ul>
<li style="display: inline-block;"> [[File:spectroscopic_analysis.png|thumb|400px|none|Figure 15: Spectroscopic analysis <ref name="spectroscopic_analysis"></ref>]] </li>
</ul></div>

Other research into using spectroscopic data to monitor that can be of interest is <ref>D. Naso, B. Turchiano, and P. Pantaleo, “A fuzzy-logic based optical sensor for online weld defect-detection,” IEEE transactions on Industrial Informatics, vol. 1, no. 4, pp. 259–273, 2005.</ref>.

=== Acoustic analysis ===
With <ref name="acoustic_analysis">E. Cayo and S. C. Alfaro, “A non-intrusive GMA welding process quality monitoring system using acoustic sensing,” Sensors, vol. 9, no. 9, pp. 7150–7166, 2009.</ref>, the author discovered that arc ignition is characterised by a distinct sound and therefore it is possible to capture the frequency of ignition through acoustic sensors. A tolerance band was set and both ignition frequency and sound pressure level were used to monitor the quality of the weld.

<div><ul>
<li style="display: inline-block;"> [[File:acoustic_analysis.png|thumb|400px|none|Figure 16: Acoustic analysis <ref name="acoustic_analysis"></ref>]] </li>
</ul></div>

Other research into using acoustic signal to monitor that can be of interest are <ref>M. Fidali, “Detection of welding process instabilities using acoustic signals,” in International Congress on Technical Diagnostic, pp. 191–201, Springer, 2016.</ref>, <ref>L. Zhang, A. C. Basantes-Defaz, D. Ozevin, and E. Indacochea, “Real-time monitoring of welding process using air-coupled ultrasonics and acoustic emission,” The International Journal of Advanced Manufacturing Technology, vol. 101, no. 5-8, pp. 1623–1634, 2019.</ref>.

=== Electrical signal analysis ===
With <ref name="electrical_analysis_1">A. Sumesh, K. Rameshkumar, A. Raja, K. Mohandas, A. Santhakumari, and R. Shyambabu, “Establishing correlation between current and voltage signatures of the arc and weld defects in GMAW process,” Arabian Journal for Science and Engineering, vol. 42, no. 11, pp. 4649–4665, 2017.</ref>, the author plot the probability density distribution of voltage signal and discovered that the stable weld's signal concentrate in one region while unstable signal spread out. While in the paper, the authors did not developed a monitoring method using their findings, it shows that there is a significant differences between stable and unstable weld's electrical signature.

<div><ul>
<li style="display: inline-block;"> [[File:Electrical analysis pdd.png|thumb|400px|none|Figure 17: Voltage PDD <ref name="electrical_analysis_1"></ref>]] </li>
</ul></div>

In <ref name="electrical_analysis_2">S. Simpson, “Signature images for arc welding fault detection,” Science and Technology of Welding and Joining, vol. 12, no. 6, pp. 481–486, 2007.</ref>, the authors shows that by plotting the signature image of voltage and current scatter plot, both arcing and short-circuiting region's signature image can notify a defects welds. A more visual representation of short circuiting and arcing region can be found in [[#Gas metal arc welding|Gas metal arc welding section]].

<div><ul>
<li style="display: inline-block;"> [[File:Electrical voltage signature.png|thumb|400px|none|Figure 18a: Electrical voltage signature of stable weld <ref name="electrical_analysis_2"></ref>]] </li>
<li style="display: inline-block;"> [[File:Electrical signature 2.png|thumb|400px|none|Figure 18b: Electrical voltage signature of unstable weld <ref name="electrical_analysis_2"></ref>]] </li>
</ul></div>

Other research into using electrical signal to monitor that can be of interest are <ref>R. Madigan, “Arc sensing for defects in constant-voltage gas metal arc welding,” Welding Journal, vol. 78, pp. 322S–328S, 1999.</ref>, <ref>Y. Huang, K. Wang, Z. Zhou, X. Zhou, and J. Fang, “Stability evaluation of short-circuiting gas metal arc welding based on ensemble empirical mode decomposition,” Measurement Science and Technology, vol. 28, no. 3, p. 035006, 2017.</ref>,<ref>Z. Zhang, X. Chen, H. Chen, J. Zhong, and S. Chen, “Online welding quality monitoring based on feature extraction of arc voltage signal,” The International Journal of Advanced Manufacturing Technology, vol. 70, no. 9-12, pp. 1661–1671, 2014</ref>,<ref>X. Li and S. Simpson, “Parametric approach to positional fault detection in short arc welding,” Science and Technology of welding and joining, vol. 14, no. 2, pp. 146–151, 2009.</ref>, and <ref>E. Wei, D. Farson, R. Richardson, and H. Ludewig, “Detection of weld surface porosity by statistical analysis of arc current in gas metal arc welding,” Journal of Manufacturing Processes, vol. 3, no. 1, pp. 50–59, 2001.</ref>.

== Experiment setup ==
The welding system uses TPS 320i C Pulse (figure 15a), produced by Fronius, to supply power, feed wire, and system cooling. For movement control of the welding torch, the system use ABB’s IRB 1520ID (figure 15b), which is a dedicated arc welding robot. The Data Acquisition device is NI's DAQ. Each channel (voltage and current) is sampled at 25kHz while the dominant frequency for voltage and current signal are around 130Hz (for CMT process) and 190Hz (for PMC process).

<div><ul>
<li style="display: inline-block;"> [[File:Fronius_power_supply.jpg |thumb|250px|none|Figure 19a: Fronius' Power Supply]] </li>
<li style="display: inline-block;"> [[File:Welding_robot.jpg |thumb|250px|none|Figure 19b: ABB's welding robot]] </li>
</ul></div>

[[File:Ni_daq.jpg|thumb|Figure 16: NI's DAQ device]]

== Statistical Analysis ==

=== Preprocessing ===
As the captured signal coming from Fronius' power supply, the signals also pick up switching power noise. The noise will interfere with the accuracy of the analysis if not filtered. In this project, we applied a 1000Hz digital Low Pass Butterworth filter. The effect of the filter can be seen in figure 20, while the comparation of using different filters can be seen in the gallery.

<div><ul>
<li style="display: inline-block;"> [[File:Filter_effect.png |thumb|750px|none|Figure 20: Effect of 1000Hz low pass filter]] </li>
<li style="display: inline-block;"> [[File:Freq_response.png |thumb|750px|none|Figure 21: Frequency response of the in-use filter]] </li>
</ul></div>

<gallery mode="packed-hover" caption="Gallery: Comparison of different filtering method" perrow=2>
File:Effect_of_filtering2.png|In use filter vs 300Hz low pass filter
File:Effect_of_filtering3.png|In use filter vs 500Hz low pass filter
File:Effect_of_filtering4.png|In use filter vs 750Hz low pass filter
File:Effect_of_filtering5.png|In use filter vs 2000Hz low pass filter
</gallery>

=== Power Spectral Density ===
Our goal for this project is to detect the defect welds and predict the layer heights (the contact tip to work piece distance, CTWD, to be more accurate) in real time (refer to [[#Relating Contact Tip to Work-piece Distance to layer height|this section]] to understand the relationship between Contact Tip to Work-piece Distance and Layer height). An analysis of the current power spectral density (PSD) shows that there is an inverse proportional relationship between the CTWD of a weld and the frequency where its '''current signal''' PSD reach the peak, and also the magnitude of the peak. Since the peak of PSD can be fairly sensitive, we only consider the frequency of the peak PSD.

Figure 22 shows the PSD plot of voltage, current, and power signal of the weld using PMC process. In this experiment, we created multiple one-line, 8 seconds welds. These eight-second welds then get segmented into smaller two-second chunk to increase the number of data and decrease the processing time.

<div><ul>
<li style="display: inline-block;"> [[File:Psd pmc good bad all lowfreq.png |thumb|750px|none|Figure 22: Power spectral density of PMC process]] </li>
<li style="display: inline-block;"> [[File:One line weld.jpg |thumb|578px|none|Figure 23: Simple one-line weld experiment]] </li>
<li style="display: inline-block;"> [[File:Cmt_psd_stable_allCTWD_zoomed.png |thumb|750px|none|Figure 24: Simple one-line weld experiment]] </li>

</ul></div>

It seems like the inverse proportional relationship holds true for most cases. Other experiments showed that for a short period at the beginning and ending of the weld, the signal can behave differently relative to the rest of the weld. Currently, to migrate with this problem, data will not be collected and analysed (both offline and online) at these periods. However, the above results seem to only hold true for PMC processes. With CMT process, the current PSD plot indicates no relationship between the CTWD and the frequency at peak (figure 24).

For the next stage of the project, our plan is to develop a linear regression model to predict the CTWD for PMC process. With CMT process, we will have to look into different analysis techniques.

=== Support Vector Machine ===
By looking at the average voltage and average current of each segment of weld (PMC process), we realised there is a visible separation margin between stable and unstable welds. Using SVM with RBF kernel, we were able to correctly classify the quality of the weld with accuracy of 98%.

However, with CMT process, the separation cannot be observed.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_pmcNoOxide_rbf.png |thumb|750px|none|Figure 20: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:CmtNoOxide_average_voltage_current.png |thumb|750px|none|Figure 20: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

=== CTWD Regression ===
With CTWD prediction, we found that the average voltage (for PMC process) and average current (for CMT process) seems to have the strongest linear relationship with the CTWD. The acceptable error for CTWD prediction is +/- 1 mm (according to AML3D requirement). For both PMC and CMT process, our root means square error is less than 1 (PMC: 0.68 and CMT: 0.83).

<div><ul>
<li style="display: inline-block;"> [[File:Linear_regression_pmc_wall.png |thumb|750px|none|Figure 20: Support Vector Machine classifying unstable data (PMC process)]] </li>
<li style="display: inline-block;"> [[File:Linear_regression_cmt_wall.png |thumb|750px|none|Figure 20: Average voltage vs Average current (CMT process)]] </li>
</ul></div>

== Result discussion ==

== Conclusion ==

== Future work ==

== References ==
<references />

File:Linear regression cmt wall.png

2019-10-15T04:05:30Z

A1713568:

File:Linear regression pmc wall.png

2019-10-15T04:04:56Z

A1713568:

File:CmtNoOxide average voltage current.png

2019-10-15T03:59:41Z

A1713568:

File:Svm pmcNoOxide rbf.png

2019-10-15T03:42:15Z

A1713568:

Projects:2019s2-20001 Using Machine Learning to Determine Deposit Height and Defects for Wire + Arc Additive Manufacture (3D printing)

2019-10-08T01:23:30Z

A1713568: /* Abstract */

[[Category:Projects]]
[[Category:Final Year Projects]]
[[Category:2019s2|24501]]
== Introduction ==
3D printing is an emerging technology that has the potential to significantly reduce material usage through the production of near net-shape parts. Many of the systems for 3D printing are based on lasers and powders; however the deposition rate with such systems is very low making the production of large-scale parts difficult. AML Technologies specialises in the use of Wire + Arc Additive Manufacture (WAAM) where deposition is based on arc welding processes and the deposition rates are an order of magnitude greater. When building 3D printed parts, even a relatively small layer height error of only 0.1 mm can produce large build height errors when multiplied across the many layers of a typical build. This can make path planning difficult, so in-process layer height measurement is an essential building block of any production 3D printing system. A variety of techniques can be used for monitoring the layer height including laser scanners, and arc monitoring. It is the latter technique that will be explored in this project due to its robustness, and low cost of implementation – it only requires the measurement of arc current and voltage. Furthermore, it can potentially be used to detect defects by identifying waveform irregularities.
=== Project team ===
==== Project students ====
* Anh Tran
* Nhat Nguyen
==== Supervisors ====
* Dr. Brian Ng
* Dr. Paul Colegrove (AML3D, sponsored company)

[[File:aml3d_logo.jpg|100px]]
=== Objectives ===
The objective of this project is to increase the efficiency of the manufacture process at AML3D. In order to do so, the team will investigate into the possibility of automating and optimising the quality control processes. The two quality control processes that are currently being implemented at AML3D are measuring layer height using laser sensors, and human supervision for detecting defects. These processes add overhead into production time and usage of human resource, which is not desired. To achieve the goal, it is expected that machine learning methods will be used extensively to analyse the electrical signatures of the weld process.

== Background ==
=== Wire Arc Additive Manufacturing ===
Wire and Arc Additive Manufacturing (WAAM) is a type of additive manufacturing that uses electric arc as the heat source and material wire to feed the manufacture process <ref name="WAAM"> S. W. Williams, F. Martina, A. C. Addison, J. Ding, G. Pardal & P. Colegrove (2016) Wire + Arc Additive Manufacturing, Materials Science and Technology, 32:7, 641-647, DOI: 10.1179/1743284715Y.0000000073 </ref>. WAAM has been investigated since the 1990s <ref name="WAAM" />, but only recently that it received more attention from the manufacture world.

Its significant comes from the ability to manufacture complex model with less time and less material. Figure 1a and 1b show real parts that was manufactured by AML3D. Such custom made parts might takes months before be ready to be shipped, but with WAAM, the production time can reduce down to weeks. Currently, the industries that benefit the most from WAAM are maritime and aerospace.

Similar to other additive manufacture methods, WAAM achieves such results by sliding the models into multiple layers, and then build the model layer by layer. The movement control is normally handled by a robotic arm (AML3D uses ABB's Arc Welder robot), and the welding path is generated by a Computer Aid Manufacture (CAM) software.

[[File:welding_part1.JPG|thumb|Figure 1a: A propeller]]
[[File:welding_part2.jpg|thumb|Figure 1b: Another part of the ship]]
[[File:abb_robot.jpg|thumb|Figure 2: ABB's IRB 1520ID Arc Welder]]

Currently at AML3D, the manufacture process for a part can take anywhere from days to weeks. However, the contact tip (where the welding gun deposits the wire) needs replacement every couple of hours. A worn contact tip such as figure 3 will lead to unstable weld and the this effect will propagate and accumulate through multiple layers. Such defects in the manufacture process will no doubt be financially taxing for AML3D. Apart from constant human supervision, another quality control process implemented at AML3D is height measurement of each layer, after the layer is built. This process is implemented to ensure the height of each layer is in the acceptable range (1-2mm). As mentioned before, these quality control processes are costing AML3D in term of time and finance. Our project will focus on replace them with a less time consuming, more automating process.

<div><ul>
<li style="display: inline-block;"> [[File:contact_tip.jpg|thumb|none|400px|Figure 3a: A worn contact tip]] </li>
<li style="display: inline-block;"> [[File:layers.jpeg|thumb|none|530px|Figure 3b: Close up photo of layers]] </li>
</ul></div>

==== Relating Contact Tip to Work-piece Distance to layer height ====
At AML3D, the sensor captures the Contact Tip to Work-piece Distance (CTWD), and from there, we can deduce the layer height. A more detail explanation is of follow:

<div><ul>
<li style="display: inline-block;"> [[File:Example_of_layer_height_measurement.jpg|thumb|none|400px|Figure 4: From CTWD to layer height]] </li>
</ul></div>

The CTWD can be set at the beginning of the weld, however, as the welding process running, the actual CTWD fluctuate around the set CTWD. For example, after finish constructing one layer, the laser sensor capture the model height at 3 different position: A, B and C. Let the 3 model height be m1, m2 and m3. Calculate n_mean and for argument sake, if the set CTWD is 15 mm, the actual CTWD at A is [[File:Ctwd_at_a.png|180px]]. The height of one layer can be calculated as [[File:Layer_height_at_a.png|210px]]. A worked example can be found at table 1.

{| class="wikitable" name="Table 1"
|+Table 1: Worked example
|-
! CTWD
! m1
! m2
! m3
! m_mean
! c1
! c2
! c3
|-
| 15
| 20
| 22
| 25
| 22.3
| 17.3
| 15.3
| 17.7
|}

=== Gas metal arc welding ===
<div><ul>
<li style="display: inline-block;"> [[File:gmaw_welding.png|thumb|500px|none|Figure 5: GMAW welding process <ref name="welding_book"> Norrish, J. (2006). Advanced welding processes. Elsevier. </ref>]] </li>
</ul></div>

Gas metal arc welding (GMAW), or also known as metal inert gas (MIG), is the welding process used at AML3D for WAAM. Due to its high deposition rate and economic benefits, GMAW has became more popular. We will explore two variants of GMAW, which are Pulse Multi Control and Cold Metal Transfer. Both processes are developed by Fronius, an Austrian welding company. Note that it is not required to understand the welding physics to follow this wiki. The next two following sections' purpose is to highlight the high variability, high dynamic nature in welding.

[[File:fronius.png|thumb|Figure 6: Fronius logo]]

==== Pulse Multi Control ====
Pulse Multi Control (PMC) welding is Fronius' modified Pulsed GMAW. The advantage of Pulsed GMAW is the ability to control the metal droplet transfer in welding. Note that from figure 7, the metal droplet is characterised by a downward slope of the current pulse. Figure 8 shows the current signal captured from experiments conducted at AML3D (PMC is used).

<div><ul>
<li style="display: inline-block;"> [[File:Pulsed_gmaw_transfer_mode.png|thumb|500px|none|Figure 7: Relationship between pulse current and droplet transfer<ref name="welding_book" />]] </li>
<li style="display: inline-block;"> [[File:Pmc_current_in_time_domain.png|thumb|650px|none|Figure 8: Current signal captured from experiments conducted at AML3D (PMC process)]] </li>
</ul></div>
==== Cold Metal Transfer ====
Cold Metal Transfer (CMT) is a complex welding process, also developed by Fronius. By detecting a short circuit (mode C and D in figure 9), CMT can make adjustment such as retracting the welding material to cool down, and therefore create a smoother, more stable weld. The complexity of CMT can be seen in the current signal of the process (figure 10). Compare to PMC, CMT's signal has more variation during one signal period.

<div><ul>
<li style="display: inline-block;"> [[File:Transfer_mode.png|thumb|800px|none|Figure 9: Mechanism of metal droplet transfer mode]] </li>
<li style="display: inline-block;"> [[File:Cmt1_current_in_time_domain.png|thumb|735px|none|Figure 10: Current signal captured from experiments conducted at AML3D (CMT process)]] </li>
</ul></div>
=== Machine Learning ===
Machine learning (or statistical learning) refer to a set of tools to give data scientists an insight into the data and make better decisions based on those information <ref name="introduction_ml_book"> James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An introduction to statistical learning (Vol. 112, p. 18). New York: Springer. </ref>. In this project, we will explore the capabilities of classical machine learning methods (i.e anything but Deep Learning) due to the limitation of our available data. By the time of writing this wiki, we have only explored Support Vector Machine.

==== Support Vector Machine ====
The main idea of Support Vector Machine (SVM) is to draw hyperplanes that best separate multiple classes of data. The original SVM judges best decision hyperplane based on the separation margin between data classes (figure 11). However, there are multiple variants of SVM, such as hard/soft margin SVM, Nu-SVM, One class SVM (to solve anomaly detection problem). In our project, we choose to use a soft margin implementation of SVM. The benefit of soft margin SVM (as oppose to hard margin SVM) is that by allowing some misclassifications, we get a wider margin and therefore the solution becomes more "generalise" (i.e work better with a new set of dataset that is independent from the training dataset).

<div><ul>
<li style="display: inline-block;"> [[File:svm_margin.png|thumb|400px|none|Figure 11: Separation margin <ref name="bishop_book"> Bishop, C. M. (2006). Pattern recognition and machine learning. Springer. </ref>]] </li>
</ul></div>

SVM algorithm makes use of [https://en.wikipedia.org/wiki/Lagrange_multiplier Lagrange multiplier]. Let's the linear model of the dataset be:

[[File:Linear_model.png|150px]]

where [[File:Phi_math.png|10px]] denotes the feature space transformation, '''w''' denotes the weights and ''b'' denotes bias constant.

Let [[File:vector_x_n.png|17px]] be the input vector (where each index in the vector denotes a feature), and [[File:target_n.png|15px]] be the target (the class of input vector). It is showed that in order to maximise the (soft) margin, we will need to minimise the following:

[[File:Soft margin minimise equation.png|150px]]

with subject to:

[[File:Soft margin minimise condition.png|150px]]

where N is the total number of training data, C is the regularisation parameter to control the effect of slack variable [[File:slack_variable.png|17px]]. A more visual presentation of slack variables can be found in figure 12.

<div><ul>
<li style="display: inline-block;"> [[File:svm_slack_variable.png|thumb|400px|none|Figure 12: Slack variable <ref name="bishop_book"></ref>]] </li>
</ul></div>

If we were to apply the Lagrangian method to solve the above minimisation problem, we can find the decision hyperplane for the SVM algorithm. However, we shall not discuss the solution here as it is not the main focus of the project.

==== Kernel tricks ====
Both figure 11 and 12 are of linearly separable cases for SVM. For dataset that is not easy to separate with a hyperplane in their original features space (such as figure 13), the data often get projected into a higher dimension space where it is easier to separate (figure 14). Such technique is called the Kernel Tricks. It is important to note that Kernel Tricks does not actually "map" the data from '''X''' dimension to '''Z''' dimension, but rather pairwise compare the similarity of the input space in '''Z''' dimension. However, we shall not dive too deep into this as it is not the main focus of the project.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_nonseparable.png |thumb|350px|none|Figure 13: Nonseparable dataset <ref name="bishop_book"></ref>]] </li>
<li style="display: inline-block;"> [[File:Svm_kernel_trick.png |thumb|650px|none|Figure 14: Simplified example of the Kernel tricks]] </li>
</ul></div>

== Previous studies ==
There are many researches into defect detection in GMAW process. The three most dominant analysis methods are spectroscopic analysis, acoustic analysis, and electrical signal analysis.

=== Spectroscopic analysis ===
In <ref name="spectroscopic_analysis">D. Bebiano and S. Alfaro, “A weld defects detection system based on a spectrometer,” Sensors, vol. 9, no. 4, pp. 2851–2861, 2009.</ref>, the authors captured the intensity of radiation emission of electric arc. It is obvious from figure 15 that good weld's intensity stays stable while defect one's is characterised by a sudden peak. Therefore, a threshold is set to detect the defect welds.

<div><ul>
<li style="display: inline-block;"> [[File:spectroscopic_analysis.png|thumb|400px|none|Figure 15: Spectroscopic analysis <ref name="spectroscopic_analysis"></ref>]] </li>
</ul></div>

Other research into using spectroscopic data to monitor that can be of interest is <ref>D. Naso, B. Turchiano, and P. Pantaleo, “A fuzzy-logic based optical sensor for online weld defect-detection,” IEEE transactions on Industrial Informatics, vol. 1, no. 4, pp. 259–273, 2005.</ref>.

=== Acoustic analysis ===
With <ref name="acoustic_analysis">E. Cayo and S. C. Alfaro, “A non-intrusive GMA welding process quality monitoring system using acoustic sensing,” Sensors, vol. 9, no. 9, pp. 7150–7166, 2009.</ref>, the author discovered that arc ignition is characterised by a distinct sound and therefore it is possible to capture the frequency of ignition through acoustic sensors. A tolerance band was set and both ignition frequency and sound pressure level were used to monitor the quality of the weld.

<div><ul>
<li style="display: inline-block;"> [[File:acoustic_analysis.png|thumb|400px|none|Figure 16: Acoustic analysis <ref name="acoustic_analysis"></ref>]] </li>
</ul></div>

Other research into using acoustic signal to monitor that can be of interest are <ref>M. Fidali, “Detection of welding process instabilities using acoustic signals,” in International Congress on Technical Diagnostic, pp. 191–201, Springer, 2016.</ref>, <ref>L. Zhang, A. C. Basantes-Defaz, D. Ozevin, and E. Indacochea, “Real-time monitoring of welding process using air-coupled ultrasonics and acoustic emission,” The International Journal of Advanced Manufacturing Technology, vol. 101, no. 5-8, pp. 1623–1634, 2019.</ref>.

=== Electrical signal analysis ===
With <ref name="electrical_analysis_1">A. Sumesh, K. Rameshkumar, A. Raja, K. Mohandas, A. Santhakumari, and R. Shyambabu, “Establishing correlation between current and voltage signatures of the arc and weld defects in GMAW process,” Arabian Journal for Science and Engineering, vol. 42, no. 11, pp. 4649–4665, 2017.</ref>, the author plot the probability density distribution of voltage signal and discovered that the stable weld's signal concentrate in one region while unstable signal spread out. While in the paper, the authors did not developed a monitoring method using their findings, it shows that there is a significant differences between stable and unstable weld's electrical signature.

<div><ul>
<li style="display: inline-block;"> [[File:Electrical analysis pdd.png|thumb|400px|none|Figure 17: Voltage PDD <ref name="electrical_analysis_1"></ref>]] </li>
</ul></div>

In <ref name="electrical_analysis_2">S. Simpson, “Signature images for arc welding fault detection,” Science and Technology of Welding and Joining, vol. 12, no. 6, pp. 481–486, 2007.</ref>, the authors shows that by plotting the signature image of voltage and current scatter plot, both arcing and short-circuiting region's signature image can notify a defects welds. A more visual representation of short circuiting and arcing region can be found in [[#Gas metal arc welding|Gas metal arc welding section]].

<div><ul>
<li style="display: inline-block;"> [[File:Electrical voltage signature.png|thumb|400px|none|Figure 18a: Electrical voltage signature of stable weld <ref name="electrical_analysis_2"></ref>]] </li>
<li style="display: inline-block;"> [[File:Electrical signature 2.png|thumb|400px|none|Figure 18b: Electrical voltage signature of unstable weld <ref name="electrical_analysis_2"></ref>]] </li>
</ul></div>

Other research into using electrical signal to monitor that can be of interest are <ref>R. Madigan, “Arc sensing for defects in constant-voltage gas metal arc welding,” Welding Journal, vol. 78, pp. 322S–328S, 1999.</ref>, <ref>Y. Huang, K. Wang, Z. Zhou, X. Zhou, and J. Fang, “Stability evaluation of short-circuiting gas metal arc welding based on ensemble empirical mode decomposition,” Measurement Science and Technology, vol. 28, no. 3, p. 035006, 2017.</ref>,<ref>Z. Zhang, X. Chen, H. Chen, J. Zhong, and S. Chen, “Online welding quality monitoring based on feature extraction of arc voltage signal,” The International Journal of Advanced Manufacturing Technology, vol. 70, no. 9-12, pp. 1661–1671, 2014</ref>,<ref>X. Li and S. Simpson, “Parametric approach to positional fault detection in short arc welding,” Science and Technology of welding and joining, vol. 14, no. 2, pp. 146–151, 2009.</ref>, and <ref>E. Wei, D. Farson, R. Richardson, and H. Ludewig, “Detection of weld surface porosity by statistical analysis of arc current in gas metal arc welding,” Journal of Manufacturing Processes, vol. 3, no. 1, pp. 50–59, 2001.</ref>.

== Experiment setup ==
The welding system uses TPS 320i C Pulse (figure 15a), produced by Fronius, to supply power, feed wire, and system cooling. For movement control of the welding torch, the system use ABB’s IRB 1520ID (figure 15b), which is a dedicated arc welding robot. The Data Acquisition device is NI's DAQ. Each channel (voltage and current) is sampled at 25kHz while the dominant frequency for voltage and current signal are around 130Hz (for CMT process) and 190Hz (for PMC process).

<div><ul>
<li style="display: inline-block;"> [[File:Fronius_power_supply.jpg |thumb|250px|none|Figure 19a: Fronius' Power Supply]] </li>
<li style="display: inline-block;"> [[File:Welding_robot.jpg |thumb|250px|none|Figure 19b: ABB's welding robot]] </li>
</ul></div>

[[File:Ni_daq.jpg|thumb|Figure 16: NI's DAQ device]]

== Statistical Analysis ==

=== Preprocessing ===
As the captured signal coming from Fronius' power supply, the signals also pick up switching power noise. The noise will interfere with the accuracy of the analysis if not filtered. In this project, we applied a 1000Hz digital Low Pass Butterworth filter. The effect of the filter can be seen in figure 20, while the comparation of using different filters can be seen in the gallery.

<div><ul>
<li style="display: inline-block;"> [[File:Filter_effect.png |thumb|750px|none|Figure 20: Effect of 1000Hz low pass filter]] </li>
<li style="display: inline-block;"> [[File:Freq_response.png |thumb|750px|none|Figure 21: Frequency response of the in-use filter]] </li>
</ul></div>

<gallery mode="packed-hover" caption="Gallery: Comparison of different filtering method" perrow=2>
File:Effect_of_filtering2.png|In use filter vs 300Hz low pass filter
File:Effect_of_filtering3.png|In use filter vs 500Hz low pass filter
File:Effect_of_filtering4.png|In use filter vs 750Hz low pass filter
File:Effect_of_filtering5.png|In use filter vs 2000Hz low pass filter
</gallery>

=== Power Spectral Density ===
==== Prerequisite knowledge ====
Looking at the spectrum is another way to analyse the fundamental information of the signal, in frequency domain <ref name="signal_processing_book">Prandoni, P., & Vetterli, M. (2008). Signal processing for communications. EPFL press.</ref>. Note that the term "Power" in Power Spectral Density does not refer to the convention power definition. In signal processing, energy of a discrete time signal is defined as follow:

[[File:Signal_energy_formular.png|150px]]

The way to interpret this definition is if we consider the signal to be of "Voltage" unit, ''E'' is the total energy (in joules) dissipated over a 1 Ohm resistor <ref name="signal_processing_book"></ref>. The power spectral density of a signal over time interval [-M,M] is defined as follow:

[[File:Psd_formular.png|300px]]

where [[File:Fourier series symbol.png|67px]] is the truncated Fourier transform:

[[File:Fourier_series_formular.png|250px]]

==== Analysis ====
Our goal for this project is to detect the defect welds and predict the layer heights (the contact tip to work piece distance, CTWD, to be more accurate) in real time (refer to [[#Relating Contact Tip to Work-piece Distance to layer height|this section]] to understand the relationship between Contact Tip to Work-piece Distance and Layer height). An analysis of the current power spectral density (PSD) shows that there is an inverse proportional relationship between the CTWD of a weld and the frequency where its '''current signal''' PSD reach the peak, and also the magnitude of the peak. Since the peak of PSD can be fairly sensitive, we only consider the frequency of the peak PSD.

Figure 22 shows the PSD plot of voltage, current, and power signal of the weld using PMC process. In this experiment, we created multiple one-line, 8 seconds welds. These eight-second welds then get segmented into smaller two-second chunk to increase the number of data and decrease the processing time.

<div><ul>
<li style="display: inline-block;"> [[File:Psd pmc good bad all lowfreq.png |thumb|750px|none|Figure 22: Power spectral density of PMC process]] </li>
<li style="display: inline-block;"> [[File:One line weld.jpg |thumb|578px|none|Figure 23: Simple one-line weld experiment]] </li>
<li style="display: inline-block;"> [[File:Cmt_psd_stable_allCTWD_zoomed.png |thumb|750px|none|Figure 24: Simple one-line weld experiment]] </li>

</ul></div>

It seems like the inverse proportional relationship holds true for most cases. Other experiments showed that for a short period at the beginning and ending of the weld, the signal can behave differently relative to the rest of the weld. Currently, to migrate with this problem, data will not be collected and analysed (both offline and online) at these periods. However, the above results seem to only hold true for PMC processes. With CMT process, the current PSD plot indicates no relationship between the CTWD and the frequency at peak (figure 24).

For the next stage of the project, our plan is to develop a linear regression model to predict the CTWD for PMC process. With CMT process, we will have to look into different analysis techniques.

=== Support Vector Machine ===
==== Prerequisite knowledge ====
Refer to [[#Support Vector Machine|Support Vector Machine]] section for more detail.

As mentioned before, we model the dataset as a linear model: [[File:Linear model.png|150px]]. In the analysis, we try 2 different kernel tricks: linear and radial basis function kernel.

Linear kernel: [[File:Linear_kernel_Trick.png|170px]]

Radial Basis Kernel: [[File:Rbf kernel trick.png|240px]]

Let '''x''' be the training dataset in [[File:Traing_dataset_space.png|45px]] domain that contain N data points and for simplicity, each of the training data only has 1 features. The new kernel computed from these dataset will now have a dimension of N. The key takeaway from this is the new kernel's dimension is proportional to the original training dataset and thus, it is difficult to visualised the dataset after using the kernel trick (such as in figure 14).

In the analysis, the two features we used are ('''average voltage''', '''average current''') as these are the two features best separate data.

==== Analysis ====

=== Features extraction from grayscale image ===
==== Prerequisite knowledge ====

==== Analysis ====

== Result discussion ==

== Conclusion ==

== Future work ==

== References ==
<references />

Projects:2019s2-23101 AI Enabled Information Technologies for Multi-Robot Coordination

2019-10-08T01:23:03Z

A1713568: /* Project students */

[[Category:Projects]]
[[Category:Final Year Projects]]
[[Category:2018s1|106]]
Abstract here
== Introduction ==
Robots nowadays are responsible for various complex tasks. For example, consider the case in which robots are tasking with exploring a nuclear power plant to carry out a rescue mission. Our project is motivated by the need of coordinated operation where a task is efficiently carried out faster by multiple robots.

=== Project team ===
==== Project students ====
* Nha Nam Nguyen Nguyen
* Zeping Zhao

==== Supervisors ====
* Professor Cheng-Chew Lim

==== Advisors ====
* DST Group

=== Objectives ===
In the context of our project, coordination can be demonstrated at two levels. First is at the physical level, which allows each robot to navigate obstacles within a dynamic environment to satisfy team goals. The second is at the organisational level, where coordination distributes information include cognition, cooperative behaviours and trust. Consider a rapidly-changing environment, a team of robot must navigate to a destination based on the given priori knowledge, for example, a map. Each robot will need to collect and process new surrounding information to coordinate information within the team and determine the optimal approach to the team goal.

== Background ==
=== Topic 1 ===
As the robot has to carry out a task under uncertainties due to the changing of environment, these project aims lead to several challenges. The first challenge is navigation problem, where a robot must know where it is in the map, plan the optimal path and avoid any obstacles along the way to the destination. We human use our perfect perception system to navigate ourselves. However, robots use sensors to support with navigation, and no such sensors are perfect. The presence of noise in the sensor system will add uncertainties to the carrying tasks. Another challenge is selecting the robot information exchange mechanism within the team. What would the team react in case a robot did not function properly or the network connection was lost? Under such case, it is required to have an information technology to dynamically exchange information based on the operating condition to satisfy the team goal. Finally, is the challenge of how to avoid other robots during navigation task. Generally, we can not consider a robot as a moving obstacle because in the worst case, both robots will try to avoid by moving to the same side, and then to another side. By the way, it is called the reciprocal dance in robotics.

== Method ==
For a robot to know where it is in a map, Adaptive Monte Carlo Localisation also known as AMCL is an improved version of Monte Carlo Localisation is used. Monte Carlo Localisation generates a large number of samples to represent the robot position likelihood. Those samples later will converge to the most likely robot position based on the sensor data. AMCL can decrease the number of samples needed once the robot is localised, which saves a lot of computational resources.
To navigate to a destination, the combination of A* and Dynamic Windows Approach is used. A* is a global path planning algorithm to find the shortest path to a destination using heuristic, which allows a generally faster execution time compared with the exact path planning technique such as Dijkstra algorithm. Dynamic Window Approach is a local path planning technique for calculating the optimal velocity reach the destination follow the global path while avoiding obstacles that can possibly collide with the robot in the velocity search space.
The information exchange mechanism within a team of robots can be divided into centralised and distributed coordination. Centralised coordination allows all robots to operate through a single leader, which can not guarantee the robustness even though it is easy to implement. For example, the robot team could fail to execute tasks if the leader was not functioned properly. The distributed coordination provides a more flexible approach to multi-robot communication, which allows each robot to operate on its own without a leader.
To be able to avoid other robots while moving, Collision Avoidance with Localisation Uncertainty or CALU, is a relatively new technique based on Velocity Obstacles that provides a fully distributed communication among robots. Velocity Obstacles maintains a set of velocities that potentially lead to collision. CALU will choose velocities outside of the collision velocity range to avoid obstacles and use bounded AMCL to provide each robot with the localisation information of others.

== Results ==
To achieve our proposed approach, we will present our project plan for 2 semesters. The first stage is to get ourselves familiar with ROS concepts and install a ROS working environment. During this stage, we also conduct a literature review to identify the potential approaches to our projects. This also is the stage we design a base environment that the whole project will be tested on. The second stage will be about navigate a single robot to the destination. Zeping will be working on building maps of the environment and I will be responsible for localisation and navigation. After each unit is completed, we will perform system integration and conduct testing on the physical environment. It marks the end of the project plan for this semester. Next semester, we will work on the robot coordination. It is planned that Zeping will be in charge of establishing a distributed network among robots, while I will be working on the coordinated navigation. The deliverables for semester 2 are the completed prototype with all project aims satisfied and a thesis.

== Conclusion ==
In conclusion, we would like to report on the progress of our project up till now. We have built several environment maps with different parameters to provide a basis for quantifying the mapping performance. We also have done developing the navigation function of a single robot. A robot now has the ability to localise itself, and move to a specified location within the map. We are on the way of optimising the algorithm parameters to choose the best fit for our project. What has been done are tested in both physical and simulation environment.

== References ==
[1] Pyo, Y., Cho, H., Jung. R. & Lim, T. 2017. ROS Robot Programming, ROBOTIS Co.,Ltd. pp.10-314.

[2] Wang, H., Huang, M. & Wu, D. 2019 A Quantitative Analysis on Gmapping Algorithm Parameters Based on Lidar in Small Area Environment, Springer Nature Singapore Pte Ltd. pp.482.

[3]KOUBAA, A. (2019). Robot Path Planning and Cooperation. [S.l.]: SPRINGER, p.4.

[4]CLAES, D., HENNES, D., TUYLS, K. & MEEUSSEN, W. 2012. Collision Avoidance under Bounded Localization Uncertainty. 2012 Ieee/Rsj International Conference on Intelligent Robots and Systems (Iros), 1192-1198.

[5]Thrun, S., Burgard, W. and Fox, D. (2010). Probabilistic robotics. Cambridge, Mass.: MIT Press, pp.263-265.

[6]SARIFF, N. & BUNIYAMIN, N. 2006. An Overview of Autonomous Mobile Robot Path Planning Algorithms. 2006 4th Student Conference on Research and Development.

[7]FOX, D., BURGARD, W. & THRUN, S. 1997. The dynamic window approach to collision avoidance. IEEE Robotics & Automation Magazine, 4, 23-33.

[8]KOUBAA, A. 2017. Robot Operating System (ROS) : The Complete Reference (Volume 2), Cham, Cham: Springer International Publishing AG.

[9]CLAES, D., HENNES, D., TUYLS, K. & MEEUSSEN, W. 2012. Collision avoidance under bounded localization uncertainty.

Projects:2019s2-23101 AI Enabled Information Technologies for Multi-Robot Coordination

2019-10-08T01:22:52Z

A1713568:

[[Category:Projects]]
[[Category:Final Year Projects]]
[[Category:2018s1|106]]
Abstract here
== Introduction ==
Robots nowadays are responsible for various complex tasks. For example, consider the case in which robots are tasking with exploring a nuclear power plant to carry out a rescue mission. Our project is motivated by the need of coordinated operation where a task is efficiently carried out faster by multiple robots.

=== Project team ===
==== Project students ====
hi
* Nha Nam Nguyen Nguyen
* Zeping Zhao
==== Supervisors ====
* Professor Cheng-Chew Lim

==== Advisors ====
* DST Group

=== Objectives ===
In the context of our project, coordination can be demonstrated at two levels. First is at the physical level, which allows each robot to navigate obstacles within a dynamic environment to satisfy team goals. The second is at the organisational level, where coordination distributes information include cognition, cooperative behaviours and trust. Consider a rapidly-changing environment, a team of robot must navigate to a destination based on the given priori knowledge, for example, a map. Each robot will need to collect and process new surrounding information to coordinate information within the team and determine the optimal approach to the team goal.

== Background ==
=== Topic 1 ===
As the robot has to carry out a task under uncertainties due to the changing of environment, these project aims lead to several challenges. The first challenge is navigation problem, where a robot must know where it is in the map, plan the optimal path and avoid any obstacles along the way to the destination. We human use our perfect perception system to navigate ourselves. However, robots use sensors to support with navigation, and no such sensors are perfect. The presence of noise in the sensor system will add uncertainties to the carrying tasks. Another challenge is selecting the robot information exchange mechanism within the team. What would the team react in case a robot did not function properly or the network connection was lost? Under such case, it is required to have an information technology to dynamically exchange information based on the operating condition to satisfy the team goal. Finally, is the challenge of how to avoid other robots during navigation task. Generally, we can not consider a robot as a moving obstacle because in the worst case, both robots will try to avoid by moving to the same side, and then to another side. By the way, it is called the reciprocal dance in robotics.

== Method ==
For a robot to know where it is in a map, Adaptive Monte Carlo Localisation also known as AMCL is an improved version of Monte Carlo Localisation is used. Monte Carlo Localisation generates a large number of samples to represent the robot position likelihood. Those samples later will converge to the most likely robot position based on the sensor data. AMCL can decrease the number of samples needed once the robot is localised, which saves a lot of computational resources.
To navigate to a destination, the combination of A* and Dynamic Windows Approach is used. A* is a global path planning algorithm to find the shortest path to a destination using heuristic, which allows a generally faster execution time compared with the exact path planning technique such as Dijkstra algorithm. Dynamic Window Approach is a local path planning technique for calculating the optimal velocity reach the destination follow the global path while avoiding obstacles that can possibly collide with the robot in the velocity search space.
The information exchange mechanism within a team of robots can be divided into centralised and distributed coordination. Centralised coordination allows all robots to operate through a single leader, which can not guarantee the robustness even though it is easy to implement. For example, the robot team could fail to execute tasks if the leader was not functioned properly. The distributed coordination provides a more flexible approach to multi-robot communication, which allows each robot to operate on its own without a leader.
To be able to avoid other robots while moving, Collision Avoidance with Localisation Uncertainty or CALU, is a relatively new technique based on Velocity Obstacles that provides a fully distributed communication among robots. Velocity Obstacles maintains a set of velocities that potentially lead to collision. CALU will choose velocities outside of the collision velocity range to avoid obstacles and use bounded AMCL to provide each robot with the localisation information of others.

== Results ==
To achieve our proposed approach, we will present our project plan for 2 semesters. The first stage is to get ourselves familiar with ROS concepts and install a ROS working environment. During this stage, we also conduct a literature review to identify the potential approaches to our projects. This also is the stage we design a base environment that the whole project will be tested on. The second stage will be about navigate a single robot to the destination. Zeping will be working on building maps of the environment and I will be responsible for localisation and navigation. After each unit is completed, we will perform system integration and conduct testing on the physical environment. It marks the end of the project plan for this semester. Next semester, we will work on the robot coordination. It is planned that Zeping will be in charge of establishing a distributed network among robots, while I will be working on the coordinated navigation. The deliverables for semester 2 are the completed prototype with all project aims satisfied and a thesis.

== Conclusion ==
In conclusion, we would like to report on the progress of our project up till now. We have built several environment maps with different parameters to provide a basis for quantifying the mapping performance. We also have done developing the navigation function of a single robot. A robot now has the ability to localise itself, and move to a specified location within the map. We are on the way of optimising the algorithm parameters to choose the best fit for our project. What has been done are tested in both physical and simulation environment.

== References ==
[1] Pyo, Y., Cho, H., Jung. R. & Lim, T. 2017. ROS Robot Programming, ROBOTIS Co.,Ltd. pp.10-314.

[2] Wang, H., Huang, M. & Wu, D. 2019 A Quantitative Analysis on Gmapping Algorithm Parameters Based on Lidar in Small Area Environment, Springer Nature Singapore Pte Ltd. pp.482.

[3]KOUBAA, A. (2019). Robot Path Planning and Cooperation. [S.l.]: SPRINGER, p.4.

[4]CLAES, D., HENNES, D., TUYLS, K. & MEEUSSEN, W. 2012. Collision Avoidance under Bounded Localization Uncertainty. 2012 Ieee/Rsj International Conference on Intelligent Robots and Systems (Iros), 1192-1198.

[5]Thrun, S., Burgard, W. and Fox, D. (2010). Probabilistic robotics. Cambridge, Mass.: MIT Press, pp.263-265.

[6]SARIFF, N. & BUNIYAMIN, N. 2006. An Overview of Autonomous Mobile Robot Path Planning Algorithms. 2006 4th Student Conference on Research and Development.

[7]FOX, D., BURGARD, W. & THRUN, S. 1997. The dynamic window approach to collision avoidance. IEEE Robotics & Automation Magazine, 4, 23-33.

[8]KOUBAA, A. 2017. Robot Operating System (ROS) : The Complete Reference (Volume 2), Cham, Cham: Springer International Publishing AG.

[9]CLAES, D., HENNES, D., TUYLS, K. & MEEUSSEN, W. 2012. Collision avoidance under bounded localization uncertainty.

Projects:General disclaimer

2019-10-08T01:22:00Z

A1713568: Created page with "hi"

Projects:2019s2-20001 Using Machine Learning to Determine Deposit Height and Defects for Wire + Arc Additive Manufacture (3D printing)

2019-10-07T23:43:31Z

A1713568: /* Prerequisite knowledge */

[[Category:Projects]]
[[Category:Final Year Projects]]
[[Category:2019s2|24501]]
== Abstract ==
Additive manufacturing is an emerging technology that is complementary to subtracting manufacturing. AM is able to create complex model at the cost of production time. An example of AM is Wire + Arc Additive Manufacture. After slicing the manufacturing process of one model to multiple layers, WAAM uses electric arc as a source of heat to feed metal wire and create each layer. However, a slight height inaccuracy in this process will result in an expensive faulty, as error can be multiplied across many layers. Thus, in-process height measurement and faulty detection are critical to the production process. In this thesis, we will look into a machine learning approach to develop an on-the-line algorithm that automatically detects faulty processes and make height measurement by analysing electrical signal (current and voltage) of a Gas Metal Arc Welding machine. This project will introduce the
approaches to tackle the same problem by other researchers, describe the design of our experiment, development of our machine learning model and discuss the final result.

'''Keywords''': Additive manufacturing, Gas metal arc welding, Metal inert gas, Machine learning, Fault detection, Signal analysis
== Introduction ==
3D printing is an emerging technology that has the potential to significantly reduce material usage through the production of near net-shape parts. Many of the systems for 3D printing are based on lasers and powders; however the deposition rate with such systems is very low making the production of large-scale parts difficult. AML Technologies specialises in the use of Wire + Arc Additive Manufacture (WAAM) where deposition is based on arc welding processes and the deposition rates are an order of magnitude greater. When building 3D printed parts, even a relatively small layer height error of only 0.1 mm can produce large build height errors when multiplied across the many layers of a typical build. This can make path planning difficult, so in-process layer height measurement is an essential building block of any production 3D printing system. A variety of techniques can be used for monitoring the layer height including laser scanners, and arc monitoring. It is the latter technique that will be explored in this project due to its robustness, and low cost of implementation – it only requires the measurement of arc current and voltage. Furthermore, it can potentially be used to detect defects by identifying waveform irregularities.
=== Project team ===
==== Project students ====
* Anh Tran
* Nhat Nguyen
==== Supervisors ====
* Dr. Brian Ng
* Dr. Paul Colegrove (AML3D, sponsored company)

[[File:aml3d_logo.jpg|100px]]
=== Objectives ===
The objective of this project is to increase the efficiency of the manufacture process at AML3D. In order to do so, the team will investigate into the possibility of automating and optimising the quality control processes. The two quality control processes that are currently being implemented at AML3D are measuring layer height using laser sensors, and human supervision for detecting defects. These processes add overhead into production time and usage of human resource, which is not desired. To achieve the goal, it is expected that machine learning methods will be used extensively to analyse the electrical signatures of the weld process.

== Background ==
=== Wire Arc Additive Manufacturing ===
Wire and Arc Additive Manufacturing (WAAM) is a type of additive manufacturing that uses electric arc as the heat source and material wire to feed the manufacture process <ref name="WAAM"> S. W. Williams, F. Martina, A. C. Addison, J. Ding, G. Pardal & P. Colegrove (2016) Wire + Arc Additive Manufacturing, Materials Science and Technology, 32:7, 641-647, DOI: 10.1179/1743284715Y.0000000073 </ref>. WAAM has been investigated since the 1990s <ref name="WAAM" />, but only recently that it received more attention from the manufacture world.

Its significant comes from the ability to manufacture complex model with less time and less material. Figure 1a and 1b show real parts that was manufactured by AML3D. Such custom made parts might takes months before be ready to be shipped, but with WAAM, the production time can reduce down to weeks. Currently, the industries that benefit the most from WAAM are maritime and aerospace.

Similar to other additive manufacture methods, WAAM achieves such results by sliding the models into multiple layers, and then build the model layer by layer. The movement control is normally handled by a robotic arm (AML3D uses ABB's Arc Welder robot), and the welding path is generated by a Computer Aid Manufacture (CAM) software.

[[File:welding_part1.JPG|thumb|Figure 1a: A propeller]]
[[File:welding_part2.jpg|thumb|Figure 1b: Another part of the ship]]
[[File:abb_robot.jpg|thumb|Figure 2: ABB's IRB 1520ID Arc Welder]]

Currently at AML3D, the manufacture process for a part can take anywhere from days to weeks. However, the contact tip (where the welding gun deposits the wire) needs replacement every couple of hours. A worn contact tip such as figure 3 will lead to unstable weld and the this effect will propagate and accumulate through multiple layers. Such defects in the manufacture process will no doubt be financially taxing for AML3D. Apart from constant human supervision, another quality control process implemented at AML3D is height measurement of each layer, after the layer is built. This process is implemented to ensure the height of each layer is in the acceptable range (1-2mm). As mentioned before, these quality control processes are costing AML3D in term of time and finance. Our project will focus on replace them with a less time consuming, more automating process.

<div><ul>
<li style="display: inline-block;"> [[File:contact_tip.jpg|thumb|none|400px|Figure 3a: A worn contact tip]] </li>
<li style="display: inline-block;"> [[File:layers.jpeg|thumb|none|530px|Figure 3b: Close up photo of layers]] </li>
</ul></div>

==== Relating Contact Tip to Work-piece Distance to layer height ====
At AML3D, the sensor captures the Contact Tip to Work-piece Distance (CTWD), and from there, we can deduce the layer height. A more detail explanation is of follow:

<div><ul>
<li style="display: inline-block;"> [[File:Example_of_layer_height_measurement.jpg|thumb|none|400px|Figure 4: From CTWD to layer height]] </li>
</ul></div>

The CTWD can be set at the beginning of the weld, however, as the welding process running, the actual CTWD fluctuate around the set CTWD. For example, after finish constructing one layer, the laser sensor capture the model height at 3 different position: A, B and C. Let the 3 model height be m1, m2 and m3. Calculate n_mean and for argument sake, if the set CTWD is 15 mm, the actual CTWD at A is [[File:Ctwd_at_a.png|180px]]. The height of one layer can be calculated as [[File:Layer_height_at_a.png|210px]]. A worked example can be found at table 1.

{| class="wikitable" name="Table 1"
|+Table 1: Worked example
|-
! CTWD
! m1
! m2
! m3
! m_mean
! c1
! c2
! c3
|-
| 15
| 20
| 22
| 25
| 22.3
| 17.3
| 15.3
| 17.7
|}

=== Gas metal arc welding ===
<div><ul>
<li style="display: inline-block;"> [[File:gmaw_welding.png|thumb|500px|none|Figure 5: GMAW welding process <ref name="welding_book"> Norrish, J. (2006). Advanced welding processes. Elsevier. </ref>]] </li>
</ul></div>

Gas metal arc welding (GMAW), or also known as metal inert gas (MIG), is the welding process used at AML3D for WAAM. Due to its high deposition rate and economic benefits, GMAW has became more popular. We will explore two variants of GMAW, which are Pulse Multi Control and Cold Metal Transfer. Both processes are developed by Fronius, an Austrian welding company. Note that it is not required to understand the welding physics to follow this wiki. The next two following sections' purpose is to highlight the high variability, high dynamic nature in welding.

[[File:fronius.png|thumb|Figure 6: Fronius logo]]

==== Pulse Multi Control ====
Pulse Multi Control (PMC) welding is Fronius' modified Pulsed GMAW. The advantage of Pulsed GMAW is the ability to control the metal droplet transfer in welding. Note that from figure 7, the metal droplet is characterised by a downward slope of the current pulse. Figure 8 shows the current signal captured from experiments conducted at AML3D (PMC is used).

<div><ul>
<li style="display: inline-block;"> [[File:Pulsed_gmaw_transfer_mode.png|thumb|500px|none|Figure 7: Relationship between pulse current and droplet transfer<ref name="welding_book" />]] </li>
<li style="display: inline-block;"> [[File:Pmc_current_in_time_domain.png|thumb|650px|none|Figure 8: Current signal captured from experiments conducted at AML3D (PMC process)]] </li>
</ul></div>
==== Cold Metal Transfer ====
Cold Metal Transfer (CMT) is a complex welding process, also developed by Fronius. By detecting a short circuit (mode C and D in figure 9), CMT can make adjustment such as retracting the welding material to cool down, and therefore create a smoother, more stable weld. The complexity of CMT can be seen in the current signal of the process (figure 10). Compare to PMC, CMT's signal has more variation during one signal period.

<div><ul>
<li style="display: inline-block;"> [[File:Transfer_mode.png|thumb|800px|none|Figure 9: Mechanism of metal droplet transfer mode]] </li>
<li style="display: inline-block;"> [[File:Cmt1_current_in_time_domain.png|thumb|735px|none|Figure 10: Current signal captured from experiments conducted at AML3D (CMT process)]] </li>
</ul></div>
=== Machine Learning ===
Machine learning (or statistical learning) refer to a set of tools to give data scientists an insight into the data and make better decisions based on those information <ref name="introduction_ml_book"> James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An introduction to statistical learning (Vol. 112, p. 18). New York: Springer. </ref>. In this project, we will explore the capabilities of classical machine learning methods (i.e anything but Deep Learning) due to the limitation of our available data. By the time of writing this wiki, we have only explored Support Vector Machine.

==== Support Vector Machine ====
The main idea of Support Vector Machine (SVM) is to draw hyperplanes that best separate multiple classes of data. The original SVM judges best decision hyperplane based on the separation margin between data classes (figure 11). However, there are multiple variants of SVM, such as hard/soft margin SVM, Nu-SVM, One class SVM (to solve anomaly detection problem). In our project, we choose to use a soft margin implementation of SVM. The benefit of soft margin SVM (as oppose to hard margin SVM) is that by allowing some misclassifications, we get a wider margin and therefore the solution becomes more "generalise" (i.e work better with a new set of dataset that is independent from the training dataset).

<div><ul>
<li style="display: inline-block;"> [[File:svm_margin.png|thumb|400px|none|Figure 11: Separation margin <ref name="bishop_book"> Bishop, C. M. (2006). Pattern recognition and machine learning. Springer. </ref>]] </li>
</ul></div>

SVM algorithm makes use of [https://en.wikipedia.org/wiki/Lagrange_multiplier Lagrange multiplier]. Let's the linear model of the dataset be:

[[File:Linear_model.png|150px]]

where [[File:Phi_math.png|10px]] denotes the feature space transformation, '''w''' denotes the weights and ''b'' denotes bias constant.

Let [[File:vector_x_n.png|17px]] be the input vector (where each index in the vector denotes a feature), and [[File:target_n.png|15px]] be the target (the class of input vector). It is showed that in order to maximise the (soft) margin, we will need to minimise the following:

[[File:Soft margin minimise equation.png|150px]]

with subject to:

[[File:Soft margin minimise condition.png|150px]]

where N is the total number of training data, C is the regularisation parameter to control the effect of slack variable [[File:slack_variable.png|17px]]. A more visual presentation of slack variables can be found in figure 12.

<div><ul>
<li style="display: inline-block;"> [[File:svm_slack_variable.png|thumb|400px|none|Figure 12: Slack variable <ref name="bishop_book"></ref>]] </li>
</ul></div>

If we were to apply the Lagrangian method to solve the above minimisation problem, we can find the decision hyperplane for the SVM algorithm. However, we shall not discuss the solution here as it is not the main focus of the project.

==== Kernel tricks ====
Both figure 11 and 12 are of linearly separable cases for SVM. For dataset that is not easy to separate with a hyperplane in their original features space (such as figure 13), the data often get projected into a higher dimension space where it is easier to separate (figure 14). Such technique is called the Kernel Tricks. It is important to note that Kernel Tricks does not actually "map" the data from '''X''' dimension to '''Z''' dimension, but rather pairwise compare the similarity of the input space in '''Z''' dimension. However, we shall not dive too deep into this as it is not the main focus of the project.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_nonseparable.png |thumb|350px|none|Figure 13: Nonseparable dataset <ref name="bishop_book"></ref>]] </li>
<li style="display: inline-block;"> [[File:Svm_kernel_trick.png |thumb|650px|none|Figure 14: Simplified example of the Kernel tricks]] </li>
</ul></div>

== Previous studies ==
There are many researches into defect detection in GMAW process. The three most dominant analysis methods are spectroscopic analysis, acoustic analysis, and electrical signal analysis.

=== Spectroscopic analysis ===
In <ref name="spectroscopic_analysis">D. Bebiano and S. Alfaro, “A weld defects detection system based on a spectrometer,” Sensors, vol. 9, no. 4, pp. 2851–2861, 2009.</ref>, the authors captured the intensity of radiation emission of electric arc. It is obvious from figure 15 that good weld's intensity stays stable while defect one's is characterised by a sudden peak. Therefore, a threshold is set to detect the defect welds.

<div><ul>
<li style="display: inline-block;"> [[File:spectroscopic_analysis.png|thumb|400px|none|Figure 15: Spectroscopic analysis <ref name="spectroscopic_analysis"></ref>]] </li>
</ul></div>

Other research into using spectroscopic data to monitor that can be of interest is <ref>D. Naso, B. Turchiano, and P. Pantaleo, “A fuzzy-logic based optical sensor for online weld defect-detection,” IEEE transactions on Industrial Informatics, vol. 1, no. 4, pp. 259–273, 2005.</ref>.

=== Acoustic analysis ===
With <ref name="acoustic_analysis">E. Cayo and S. C. Alfaro, “A non-intrusive GMA welding process quality monitoring system using acoustic sensing,” Sensors, vol. 9, no. 9, pp. 7150–7166, 2009.</ref>, the author discovered that arc ignition is characterised by a distinct sound and therefore it is possible to capture the frequency of ignition through acoustic sensors. A tolerance band was set and both ignition frequency and sound pressure level were used to monitor the quality of the weld.

<div><ul>
<li style="display: inline-block;"> [[File:acoustic_analysis.png|thumb|400px|none|Figure 16: Acoustic analysis <ref name="acoustic_analysis"></ref>]] </li>
</ul></div>

Other research into using acoustic signal to monitor that can be of interest are <ref>M. Fidali, “Detection of welding process instabilities using acoustic signals,” in International Congress on Technical Diagnostic, pp. 191–201, Springer, 2016.</ref>, <ref>L. Zhang, A. C. Basantes-Defaz, D. Ozevin, and E. Indacochea, “Real-time monitoring of welding process using air-coupled ultrasonics and acoustic emission,” The International Journal of Advanced Manufacturing Technology, vol. 101, no. 5-8, pp. 1623–1634, 2019.</ref>.

=== Electrical signal analysis ===
With <ref name="electrical_analysis_1">A. Sumesh, K. Rameshkumar, A. Raja, K. Mohandas, A. Santhakumari, and R. Shyambabu, “Establishing correlation between current and voltage signatures of the arc and weld defects in GMAW process,” Arabian Journal for Science and Engineering, vol. 42, no. 11, pp. 4649–4665, 2017.</ref>, the author plot the probability density distribution of voltage signal and discovered that the stable weld's signal concentrate in one region while unstable signal spread out. While in the paper, the authors did not developed a monitoring method using their findings, it shows that there is a significant differences between stable and unstable weld's electrical signature.

<div><ul>
<li style="display: inline-block;"> [[File:Electrical analysis pdd.png|thumb|400px|none|Figure 17: Voltage PDD <ref name="electrical_analysis_1"></ref>]] </li>
</ul></div>

In <ref name="electrical_analysis_2">S. Simpson, “Signature images for arc welding fault detection,” Science and Technology of Welding and Joining, vol. 12, no. 6, pp. 481–486, 2007.</ref>, the authors shows that by plotting the signature image of voltage and current scatter plot, both arcing and short-circuiting region's signature image can notify a defects welds. A more visual representation of short circuiting and arcing region can be found in [[#Gas metal arc welding|Gas metal arc welding section]].

<div><ul>
<li style="display: inline-block;"> [[File:Electrical voltage signature.png|thumb|400px|none|Figure 18a: Electrical voltage signature of stable weld <ref name="electrical_analysis_2"></ref>]] </li>
<li style="display: inline-block;"> [[File:Electrical signature 2.png|thumb|400px|none|Figure 18b: Electrical voltage signature of unstable weld <ref name="electrical_analysis_2"></ref>]] </li>
</ul></div>

Other research into using electrical signal to monitor that can be of interest are <ref>R. Madigan, “Arc sensing for defects in constant-voltage gas metal arc welding,” Welding Journal, vol. 78, pp. 322S–328S, 1999.</ref>, <ref>Y. Huang, K. Wang, Z. Zhou, X. Zhou, and J. Fang, “Stability evaluation of short-circuiting gas metal arc welding based on ensemble empirical mode decomposition,” Measurement Science and Technology, vol. 28, no. 3, p. 035006, 2017.</ref>,<ref>Z. Zhang, X. Chen, H. Chen, J. Zhong, and S. Chen, “Online welding quality monitoring based on feature extraction of arc voltage signal,” The International Journal of Advanced Manufacturing Technology, vol. 70, no. 9-12, pp. 1661–1671, 2014</ref>,<ref>X. Li and S. Simpson, “Parametric approach to positional fault detection in short arc welding,” Science and Technology of welding and joining, vol. 14, no. 2, pp. 146–151, 2009.</ref>, and <ref>E. Wei, D. Farson, R. Richardson, and H. Ludewig, “Detection of weld surface porosity by statistical analysis of arc current in gas metal arc welding,” Journal of Manufacturing Processes, vol. 3, no. 1, pp. 50–59, 2001.</ref>.

== Experiment setup ==
The welding system uses TPS 320i C Pulse (figure 15a), produced by Fronius, to supply power, feed wire, and system cooling. For movement control of the welding torch, the system use ABB’s IRB 1520ID (figure 15b), which is a dedicated arc welding robot. The Data Acquisition device is NI's DAQ. Each channel (voltage and current) is sampled at 25kHz while the dominant frequency for voltage and current signal are around 130Hz (for CMT process) and 190Hz (for PMC process).

<div><ul>
<li style="display: inline-block;"> [[File:Fronius_power_supply.jpg |thumb|250px|none|Figure 19a: Fronius' Power Supply]] </li>
<li style="display: inline-block;"> [[File:Welding_robot.jpg |thumb|250px|none|Figure 19b: ABB's welding robot]] </li>
</ul></div>

[[File:Ni_daq.jpg|thumb|Figure 16: NI's DAQ device]]

== Statistical Analysis ==

=== Preprocessing ===
As the captured signal coming from Fronius' power supply, the signals also pick up switching power noise. The noise will interfere with the accuracy of the analysis if not filtered. In this project, we applied a 1000Hz digital Low Pass Butterworth filter. The effect of the filter can be seen in figure 20, while the comparation of using different filters can be seen in the gallery.

<div><ul>
<li style="display: inline-block;"> [[File:Filter_effect.png |thumb|750px|none|Figure 20: Effect of 1000Hz low pass filter]] </li>
<li style="display: inline-block;"> [[File:Freq_response.png |thumb|750px|none|Figure 21: Frequency response of the in-use filter]] </li>
</ul></div>

<gallery mode="packed-hover" caption="Gallery: Comparison of different filtering method" perrow=2>
File:Effect_of_filtering2.png|In use filter vs 300Hz low pass filter
File:Effect_of_filtering3.png|In use filter vs 500Hz low pass filter
File:Effect_of_filtering4.png|In use filter vs 750Hz low pass filter
File:Effect_of_filtering5.png|In use filter vs 2000Hz low pass filter
</gallery>

=== Power Spectral Density ===
==== Prerequisite knowledge ====
Looking at the spectrum is another way to analyse the fundamental information of the signal, in frequency domain <ref name="signal_processing_book">Prandoni, P., & Vetterli, M. (2008). Signal processing for communications. EPFL press.</ref>. Note that the term "Power" in Power Spectral Density does not refer to the convention power definition. In signal processing, energy of a discrete time signal is defined as follow:

[[File:Signal_energy_formular.png|150px]]

The way to interpret this definition is if we consider the signal to be of "Voltage" unit, ''E'' is the total energy (in joules) dissipated over a 1 Ohm resistor <ref name="signal_processing_book"></ref>. The power spectral density of a signal over time interval [-M,M] is defined as follow:

[[File:Psd_formular.png|300px]]

where [[File:Fourier series symbol.png|67px]] is the truncated Fourier transform:

[[File:Fourier_series_formular.png|250px]]

==== Analysis ====
Our goal for this project is to detect the defect welds and predict the layer heights (the contact tip to work piece distance, CTWD, to be more accurate) in real time (refer to [[#Relating Contact Tip to Work-piece Distance to layer height|this section]] to understand the relationship between Contact Tip to Work-piece Distance and Layer height). An analysis of the current power spectral density (PSD) shows that there is an inverse proportional relationship between the CTWD of a weld and the frequency where its '''current signal''' PSD reach the peak, and also the magnitude of the peak. Since the peak of PSD can be fairly sensitive, we only consider the frequency of the peak PSD.

Figure 22 shows the PSD plot of voltage, current, and power signal of the weld using PMC process. In this experiment, we created multiple one-line, 8 seconds welds. These eight-second welds then get segmented into smaller two-second chunk to increase the number of data and decrease the processing time.

<div><ul>
<li style="display: inline-block;"> [[File:Psd pmc good bad all lowfreq.png |thumb|750px|none|Figure 22: Power spectral density of PMC process]] </li>
<li style="display: inline-block;"> [[File:One line weld.jpg |thumb|578px|none|Figure 23: Simple one-line weld experiment]] </li>
<li style="display: inline-block;"> [[File:Cmt_psd_stable_allCTWD_zoomed.png |thumb|750px|none|Figure 24: Simple one-line weld experiment]] </li>

</ul></div>

It seems like the inverse proportional relationship holds true for most cases. Other experiments showed that for a short period at the beginning and ending of the weld, the signal can behave differently relative to the rest of the weld. Currently, to migrate with this problem, data will not be collected and analysed (both offline and online) at these periods. However, the above results seem to only hold true for PMC processes. With CMT process, the current PSD plot indicates no relationship between the CTWD and the frequency at peak (figure 24).

For the next stage of the project, our plan is to develop a linear regression model to predict the CTWD for PMC process. With CMT process, we will have to look into different analysis techniques.

=== Support Vector Machine ===
==== Prerequisite knowledge ====
Refer to [[#Support Vector Machine|Support Vector Machine]] section for more detail.

As mentioned before, we model the dataset as a linear model: [[File:Linear model.png|150px]]. In the analysis, we try 2 different kernel tricks: linear and radial basis function kernel.

Linear kernel: [[File:Linear_kernel_Trick.png|170px]]

Radial Basis Kernel: [[File:Rbf kernel trick.png|240px]]

Let '''x''' be the training dataset in [[File:Traing_dataset_space.png|45px]] domain that contain N data points and for simplicity, each of the training data only has 1 features. The new kernel computed from these dataset will now have a dimension of N. The key takeaway from this is the new kernel's dimension is proportional to the original training dataset and thus, it is difficult to visualised the dataset after using the kernel trick (such as in figure 14).

In the analysis, the two features we used are ('''average voltage''', '''average current''') as these are the two features best separate data.

==== Analysis ====

=== Features extraction from grayscale image ===
==== Prerequisite knowledge ====

==== Analysis ====

== Result discussion ==

== Conclusion ==

== Future work ==

== References ==
<references />

Projects:2019s2-20001 Using Machine Learning to Determine Deposit Height and Defects for Wire + Arc Additive Manufacture (3D printing)

2019-10-07T23:42:16Z

A1713568: /* Prerequisite knowledge */

[[Category:Projects]]
[[Category:Final Year Projects]]
[[Category:2019s2|24501]]
== Abstract ==
Additive manufacturing is an emerging technology that is complementary to subtracting manufacturing. AM is able to create complex model at the cost of production time. An example of AM is Wire + Arc Additive Manufacture. After slicing the manufacturing process of one model to multiple layers, WAAM uses electric arc as a source of heat to feed metal wire and create each layer. However, a slight height inaccuracy in this process will result in an expensive faulty, as error can be multiplied across many layers. Thus, in-process height measurement and faulty detection are critical to the production process. In this thesis, we will look into a machine learning approach to develop an on-the-line algorithm that automatically detects faulty processes and make height measurement by analysing electrical signal (current and voltage) of a Gas Metal Arc Welding machine. This project will introduce the
approaches to tackle the same problem by other researchers, describe the design of our experiment, development of our machine learning model and discuss the final result.

'''Keywords''': Additive manufacturing, Gas metal arc welding, Metal inert gas, Machine learning, Fault detection, Signal analysis
== Introduction ==
3D printing is an emerging technology that has the potential to significantly reduce material usage through the production of near net-shape parts. Many of the systems for 3D printing are based on lasers and powders; however the deposition rate with such systems is very low making the production of large-scale parts difficult. AML Technologies specialises in the use of Wire + Arc Additive Manufacture (WAAM) where deposition is based on arc welding processes and the deposition rates are an order of magnitude greater. When building 3D printed parts, even a relatively small layer height error of only 0.1 mm can produce large build height errors when multiplied across the many layers of a typical build. This can make path planning difficult, so in-process layer height measurement is an essential building block of any production 3D printing system. A variety of techniques can be used for monitoring the layer height including laser scanners, and arc monitoring. It is the latter technique that will be explored in this project due to its robustness, and low cost of implementation – it only requires the measurement of arc current and voltage. Furthermore, it can potentially be used to detect defects by identifying waveform irregularities.
=== Project team ===
==== Project students ====
* Anh Tran
* Nhat Nguyen
==== Supervisors ====
* Dr. Brian Ng
* Dr. Paul Colegrove (AML3D, sponsored company)

[[File:aml3d_logo.jpg|100px]]
=== Objectives ===
The objective of this project is to increase the efficiency of the manufacture process at AML3D. In order to do so, the team will investigate into the possibility of automating and optimising the quality control processes. The two quality control processes that are currently being implemented at AML3D are measuring layer height using laser sensors, and human supervision for detecting defects. These processes add overhead into production time and usage of human resource, which is not desired. To achieve the goal, it is expected that machine learning methods will be used extensively to analyse the electrical signatures of the weld process.

== Background ==
=== Wire Arc Additive Manufacturing ===
Wire and Arc Additive Manufacturing (WAAM) is a type of additive manufacturing that uses electric arc as the heat source and material wire to feed the manufacture process <ref name="WAAM"> S. W. Williams, F. Martina, A. C. Addison, J. Ding, G. Pardal & P. Colegrove (2016) Wire + Arc Additive Manufacturing, Materials Science and Technology, 32:7, 641-647, DOI: 10.1179/1743284715Y.0000000073 </ref>. WAAM has been investigated since the 1990s <ref name="WAAM" />, but only recently that it received more attention from the manufacture world.

Its significant comes from the ability to manufacture complex model with less time and less material. Figure 1a and 1b show real parts that was manufactured by AML3D. Such custom made parts might takes months before be ready to be shipped, but with WAAM, the production time can reduce down to weeks. Currently, the industries that benefit the most from WAAM are maritime and aerospace.

Similar to other additive manufacture methods, WAAM achieves such results by sliding the models into multiple layers, and then build the model layer by layer. The movement control is normally handled by a robotic arm (AML3D uses ABB's Arc Welder robot), and the welding path is generated by a Computer Aid Manufacture (CAM) software.

[[File:welding_part1.JPG|thumb|Figure 1a: A propeller]]
[[File:welding_part2.jpg|thumb|Figure 1b: Another part of the ship]]
[[File:abb_robot.jpg|thumb|Figure 2: ABB's IRB 1520ID Arc Welder]]

Currently at AML3D, the manufacture process for a part can take anywhere from days to weeks. However, the contact tip (where the welding gun deposits the wire) needs replacement every couple of hours. A worn contact tip such as figure 3 will lead to unstable weld and the this effect will propagate and accumulate through multiple layers. Such defects in the manufacture process will no doubt be financially taxing for AML3D. Apart from constant human supervision, another quality control process implemented at AML3D is height measurement of each layer, after the layer is built. This process is implemented to ensure the height of each layer is in the acceptable range (1-2mm). As mentioned before, these quality control processes are costing AML3D in term of time and finance. Our project will focus on replace them with a less time consuming, more automating process.

<div><ul>
<li style="display: inline-block;"> [[File:contact_tip.jpg|thumb|none|400px|Figure 3a: A worn contact tip]] </li>
<li style="display: inline-block;"> [[File:layers.jpeg|thumb|none|530px|Figure 3b: Close up photo of layers]] </li>
</ul></div>

==== Relating Contact Tip to Work-piece Distance to layer height ====
At AML3D, the sensor captures the Contact Tip to Work-piece Distance (CTWD), and from there, we can deduce the layer height. A more detail explanation is of follow:

<div><ul>
<li style="display: inline-block;"> [[File:Example_of_layer_height_measurement.jpg|thumb|none|400px|Figure 4: From CTWD to layer height]] </li>
</ul></div>

The CTWD can be set at the beginning of the weld, however, as the welding process running, the actual CTWD fluctuate around the set CTWD. For example, after finish constructing one layer, the laser sensor capture the model height at 3 different position: A, B and C. Let the 3 model height be m1, m2 and m3. Calculate n_mean and for argument sake, if the set CTWD is 15 mm, the actual CTWD at A is [[File:Ctwd_at_a.png|180px]]. The height of one layer can be calculated as [[File:Layer_height_at_a.png|210px]]. A worked example can be found at table 1.

{| class="wikitable" name="Table 1"
|+Table 1: Worked example
|-
! CTWD
! m1
! m2
! m3
! m_mean
! c1
! c2
! c3
|-
| 15
| 20
| 22
| 25
| 22.3
| 17.3
| 15.3
| 17.7
|}

=== Gas metal arc welding ===
<div><ul>
<li style="display: inline-block;"> [[File:gmaw_welding.png|thumb|500px|none|Figure 5: GMAW welding process <ref name="welding_book"> Norrish, J. (2006). Advanced welding processes. Elsevier. </ref>]] </li>
</ul></div>

Gas metal arc welding (GMAW), or also known as metal inert gas (MIG), is the welding process used at AML3D for WAAM. Due to its high deposition rate and economic benefits, GMAW has became more popular. We will explore two variants of GMAW, which are Pulse Multi Control and Cold Metal Transfer. Both processes are developed by Fronius, an Austrian welding company. Note that it is not required to understand the welding physics to follow this wiki. The next two following sections' purpose is to highlight the high variability, high dynamic nature in welding.

[[File:fronius.png|thumb|Figure 6: Fronius logo]]

==== Pulse Multi Control ====
Pulse Multi Control (PMC) welding is Fronius' modified Pulsed GMAW. The advantage of Pulsed GMAW is the ability to control the metal droplet transfer in welding. Note that from figure 7, the metal droplet is characterised by a downward slope of the current pulse. Figure 8 shows the current signal captured from experiments conducted at AML3D (PMC is used).

<div><ul>
<li style="display: inline-block;"> [[File:Pulsed_gmaw_transfer_mode.png|thumb|500px|none|Figure 7: Relationship between pulse current and droplet transfer<ref name="welding_book" />]] </li>
<li style="display: inline-block;"> [[File:Pmc_current_in_time_domain.png|thumb|650px|none|Figure 8: Current signal captured from experiments conducted at AML3D (PMC process)]] </li>
</ul></div>
==== Cold Metal Transfer ====
Cold Metal Transfer (CMT) is a complex welding process, also developed by Fronius. By detecting a short circuit (mode C and D in figure 9), CMT can make adjustment such as retracting the welding material to cool down, and therefore create a smoother, more stable weld. The complexity of CMT can be seen in the current signal of the process (figure 10). Compare to PMC, CMT's signal has more variation during one signal period.

<div><ul>
<li style="display: inline-block;"> [[File:Transfer_mode.png|thumb|800px|none|Figure 9: Mechanism of metal droplet transfer mode]] </li>
<li style="display: inline-block;"> [[File:Cmt1_current_in_time_domain.png|thumb|735px|none|Figure 10: Current signal captured from experiments conducted at AML3D (CMT process)]] </li>
</ul></div>
=== Machine Learning ===
Machine learning (or statistical learning) refer to a set of tools to give data scientists an insight into the data and make better decisions based on those information <ref name="introduction_ml_book"> James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An introduction to statistical learning (Vol. 112, p. 18). New York: Springer. </ref>. In this project, we will explore the capabilities of classical machine learning methods (i.e anything but Deep Learning) due to the limitation of our available data. By the time of writing this wiki, we have only explored Support Vector Machine.

==== Support Vector Machine ====
The main idea of Support Vector Machine (SVM) is to draw hyperplanes that best separate multiple classes of data. The original SVM judges best decision hyperplane based on the separation margin between data classes (figure 11). However, there are multiple variants of SVM, such as hard/soft margin SVM, Nu-SVM, One class SVM (to solve anomaly detection problem). In our project, we choose to use a soft margin implementation of SVM. The benefit of soft margin SVM (as oppose to hard margin SVM) is that by allowing some misclassifications, we get a wider margin and therefore the solution becomes more "generalise" (i.e work better with a new set of dataset that is independent from the training dataset).

<div><ul>
<li style="display: inline-block;"> [[File:svm_margin.png|thumb|400px|none|Figure 11: Separation margin <ref name="bishop_book"> Bishop, C. M. (2006). Pattern recognition and machine learning. Springer. </ref>]] </li>
</ul></div>

SVM algorithm makes use of [https://en.wikipedia.org/wiki/Lagrange_multiplier Lagrange multiplier]. Let's the linear model of the dataset be:

[[File:Linear_model.png|150px]]

where [[File:Phi_math.png|10px]] denotes the feature space transformation, '''w''' denotes the weights and ''b'' denotes bias constant.

Let [[File:vector_x_n.png|17px]] be the input vector (where each index in the vector denotes a feature), and [[File:target_n.png|15px]] be the target (the class of input vector). It is showed that in order to maximise the (soft) margin, we will need to minimise the following:

[[File:Soft margin minimise equation.png|150px]]

with subject to:

[[File:Soft margin minimise condition.png|150px]]

where N is the total number of training data, C is the regularisation parameter to control the effect of slack variable [[File:slack_variable.png|17px]]. A more visual presentation of slack variables can be found in figure 12.

<div><ul>
<li style="display: inline-block;"> [[File:svm_slack_variable.png|thumb|400px|none|Figure 12: Slack variable <ref name="bishop_book"></ref>]] </li>
</ul></div>

If we were to apply the Lagrangian method to solve the above minimisation problem, we can find the decision hyperplane for the SVM algorithm. However, we shall not discuss the solution here as it is not the main focus of the project.

==== Kernel tricks ====
Both figure 11 and 12 are of linearly separable cases for SVM. For dataset that is not easy to separate with a hyperplane in their original features space (such as figure 13), the data often get projected into a higher dimension space where it is easier to separate (figure 14). Such technique is called the Kernel Tricks. It is important to note that Kernel Tricks does not actually "map" the data from '''X''' dimension to '''Z''' dimension, but rather pairwise compare the similarity of the input space in '''Z''' dimension. However, we shall not dive too deep into this as it is not the main focus of the project.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_nonseparable.png |thumb|350px|none|Figure 13: Nonseparable dataset <ref name="bishop_book"></ref>]] </li>
<li style="display: inline-block;"> [[File:Svm_kernel_trick.png |thumb|650px|none|Figure 14: Simplified example of the Kernel tricks]] </li>
</ul></div>

== Previous studies ==
There are many researches into defect detection in GMAW process. The three most dominant analysis methods are spectroscopic analysis, acoustic analysis, and electrical signal analysis.

=== Spectroscopic analysis ===
In <ref name="spectroscopic_analysis">D. Bebiano and S. Alfaro, “A weld defects detection system based on a spectrometer,” Sensors, vol. 9, no. 4, pp. 2851–2861, 2009.</ref>, the authors captured the intensity of radiation emission of electric arc. It is obvious from figure 15 that good weld's intensity stays stable while defect one's is characterised by a sudden peak. Therefore, a threshold is set to detect the defect welds.

<div><ul>
<li style="display: inline-block;"> [[File:spectroscopic_analysis.png|thumb|400px|none|Figure 15: Spectroscopic analysis <ref name="spectroscopic_analysis"></ref>]] </li>
</ul></div>

Other research into using spectroscopic data to monitor that can be of interest is <ref>D. Naso, B. Turchiano, and P. Pantaleo, “A fuzzy-logic based optical sensor for online weld defect-detection,” IEEE transactions on Industrial Informatics, vol. 1, no. 4, pp. 259–273, 2005.</ref>.

=== Acoustic analysis ===
With <ref name="acoustic_analysis">E. Cayo and S. C. Alfaro, “A non-intrusive GMA welding process quality monitoring system using acoustic sensing,” Sensors, vol. 9, no. 9, pp. 7150–7166, 2009.</ref>, the author discovered that arc ignition is characterised by a distinct sound and therefore it is possible to capture the frequency of ignition through acoustic sensors. A tolerance band was set and both ignition frequency and sound pressure level were used to monitor the quality of the weld.

<div><ul>
<li style="display: inline-block;"> [[File:acoustic_analysis.png|thumb|400px|none|Figure 16: Acoustic analysis <ref name="acoustic_analysis"></ref>]] </li>
</ul></div>

Other research into using acoustic signal to monitor that can be of interest are <ref>M. Fidali, “Detection of welding process instabilities using acoustic signals,” in International Congress on Technical Diagnostic, pp. 191–201, Springer, 2016.</ref>, <ref>L. Zhang, A. C. Basantes-Defaz, D. Ozevin, and E. Indacochea, “Real-time monitoring of welding process using air-coupled ultrasonics and acoustic emission,” The International Journal of Advanced Manufacturing Technology, vol. 101, no. 5-8, pp. 1623–1634, 2019.</ref>.

=== Electrical signal analysis ===
With <ref name="electrical_analysis_1">A. Sumesh, K. Rameshkumar, A. Raja, K. Mohandas, A. Santhakumari, and R. Shyambabu, “Establishing correlation between current and voltage signatures of the arc and weld defects in GMAW process,” Arabian Journal for Science and Engineering, vol. 42, no. 11, pp. 4649–4665, 2017.</ref>, the author plot the probability density distribution of voltage signal and discovered that the stable weld's signal concentrate in one region while unstable signal spread out. While in the paper, the authors did not developed a monitoring method using their findings, it shows that there is a significant differences between stable and unstable weld's electrical signature.

<div><ul>
<li style="display: inline-block;"> [[File:Electrical analysis pdd.png|thumb|400px|none|Figure 17: Voltage PDD <ref name="electrical_analysis_1"></ref>]] </li>
</ul></div>

In <ref name="electrical_analysis_2">S. Simpson, “Signature images for arc welding fault detection,” Science and Technology of Welding and Joining, vol. 12, no. 6, pp. 481–486, 2007.</ref>, the authors shows that by plotting the signature image of voltage and current scatter plot, both arcing and short-circuiting region's signature image can notify a defects welds. A more visual representation of short circuiting and arcing region can be found in [[#Gas metal arc welding|Gas metal arc welding section]].

<div><ul>
<li style="display: inline-block;"> [[File:Electrical voltage signature.png|thumb|400px|none|Figure 18a: Electrical voltage signature of stable weld <ref name="electrical_analysis_2"></ref>]] </li>
<li style="display: inline-block;"> [[File:Electrical signature 2.png|thumb|400px|none|Figure 18b: Electrical voltage signature of unstable weld <ref name="electrical_analysis_2"></ref>]] </li>
</ul></div>

Other research into using electrical signal to monitor that can be of interest are <ref>R. Madigan, “Arc sensing for defects in constant-voltage gas metal arc welding,” Welding Journal, vol. 78, pp. 322S–328S, 1999.</ref>, <ref>Y. Huang, K. Wang, Z. Zhou, X. Zhou, and J. Fang, “Stability evaluation of short-circuiting gas metal arc welding based on ensemble empirical mode decomposition,” Measurement Science and Technology, vol. 28, no. 3, p. 035006, 2017.</ref>,<ref>Z. Zhang, X. Chen, H. Chen, J. Zhong, and S. Chen, “Online welding quality monitoring based on feature extraction of arc voltage signal,” The International Journal of Advanced Manufacturing Technology, vol. 70, no. 9-12, pp. 1661–1671, 2014</ref>,<ref>X. Li and S. Simpson, “Parametric approach to positional fault detection in short arc welding,” Science and Technology of welding and joining, vol. 14, no. 2, pp. 146–151, 2009.</ref>, and <ref>E. Wei, D. Farson, R. Richardson, and H. Ludewig, “Detection of weld surface porosity by statistical analysis of arc current in gas metal arc welding,” Journal of Manufacturing Processes, vol. 3, no. 1, pp. 50–59, 2001.</ref>.

== Experiment setup ==
The welding system uses TPS 320i C Pulse (figure 15a), produced by Fronius, to supply power, feed wire, and system cooling. For movement control of the welding torch, the system use ABB’s IRB 1520ID (figure 15b), which is a dedicated arc welding robot. The Data Acquisition device is NI's DAQ. Each channel (voltage and current) is sampled at 25kHz while the dominant frequency for voltage and current signal are around 130Hz (for CMT process) and 190Hz (for PMC process).

<div><ul>
<li style="display: inline-block;"> [[File:Fronius_power_supply.jpg |thumb|250px|none|Figure 19a: Fronius' Power Supply]] </li>
<li style="display: inline-block;"> [[File:Welding_robot.jpg |thumb|250px|none|Figure 19b: ABB's welding robot]] </li>
</ul></div>

[[File:Ni_daq.jpg|thumb|Figure 16: NI's DAQ device]]

== Statistical Analysis ==

=== Preprocessing ===
As the captured signal coming from Fronius' power supply, the signals also pick up switching power noise. The noise will interfere with the accuracy of the analysis if not filtered. In this project, we applied a 1000Hz digital Low Pass Butterworth filter. The effect of the filter can be seen in figure 20, while the comparation of using different filters can be seen in the gallery.

<div><ul>
<li style="display: inline-block;"> [[File:Filter_effect.png |thumb|750px|none|Figure 20: Effect of 1000Hz low pass filter]] </li>
<li style="display: inline-block;"> [[File:Freq_response.png |thumb|750px|none|Figure 21: Frequency response of the in-use filter]] </li>
</ul></div>

<gallery mode="packed-hover" caption="Gallery: Comparison of different filtering method" perrow=2>
File:Effect_of_filtering2.png|In use filter vs 300Hz low pass filter
File:Effect_of_filtering3.png|In use filter vs 500Hz low pass filter
File:Effect_of_filtering4.png|In use filter vs 750Hz low pass filter
File:Effect_of_filtering5.png|In use filter vs 2000Hz low pass filter
</gallery>

=== Power Spectral Density ===
==== Prerequisite knowledge ====
Looking at the spectrum is another way to analyse the fundamental information of the signal, in frequency domain <ref name="signal_processing_book">Prandoni, P., & Vetterli, M. (2008). Signal processing for communications. EPFL press.</ref>. Note that the term "Power" in Power Spectral Density does not refer to the convention power definition. In signal processing, energy of a discrete time signal is defined as follow:

[[File:Signal_energy_formular.png|150px]]

The way to interpret this definition is if we consider the signal to be of "Voltage" unit, ''E'' is the total energy (in joules) dissipated over a 1 Ohm resistor <ref name="signal_processing_book"></ref>. The power spectral density of a signal over time interval [-M,M] is defined as follow:

[[File:Psd_formular.png|300px]]

where [[File:Fourier series symbol.png|67px]] is the truncated Fourier transform:

[[File:Fourier_series_formular.png|250px]]

==== Analysis ====
Our goal for this project is to detect the defect welds and predict the layer heights (the contact tip to work piece distance, CTWD, to be more accurate) in real time (refer to [[#Relating Contact Tip to Work-piece Distance to layer height|this section]] to understand the relationship between Contact Tip to Work-piece Distance and Layer height). An analysis of the current power spectral density (PSD) shows that there is an inverse proportional relationship between the CTWD of a weld and the frequency where its '''current signal''' PSD reach the peak, and also the magnitude of the peak. Since the peak of PSD can be fairly sensitive, we only consider the frequency of the peak PSD.

Figure 22 shows the PSD plot of voltage, current, and power signal of the weld using PMC process. In this experiment, we created multiple one-line, 8 seconds welds. These eight-second welds then get segmented into smaller two-second chunk to increase the number of data and decrease the processing time.

<div><ul>
<li style="display: inline-block;"> [[File:Psd pmc good bad all lowfreq.png |thumb|750px|none|Figure 22: Power spectral density of PMC process]] </li>
<li style="display: inline-block;"> [[File:One line weld.jpg |thumb|578px|none|Figure 23: Simple one-line weld experiment]] </li>
<li style="display: inline-block;"> [[File:Cmt_psd_stable_allCTWD_zoomed.png |thumb|750px|none|Figure 24: Simple one-line weld experiment]] </li>

</ul></div>

It seems like the inverse proportional relationship holds true for most cases. Other experiments showed that for a short period at the beginning and ending of the weld, the signal can behave differently relative to the rest of the weld. Currently, to migrate with this problem, data will not be collected and analysed (both offline and online) at these periods. However, the above results seem to only hold true for PMC processes. With CMT process, the current PSD plot indicates no relationship between the CTWD and the frequency at peak (figure 24).

For the next stage of the project, our plan is to develop a linear regression model to predict the CTWD for PMC process. With CMT process, we will have to look into different analysis techniques.

=== Support Vector Machine ===
==== Prerequisite knowledge ====
Refer to [[#Support Vector Machine|Support Vector Machine]] section for more detail.

As mentioned before, we model the dataset as a linear model: [[File:Linear model.png|150px]]. In the analysis, we try 2 different kernel tricks: linear and radial basis function kernel.

Linear kernel: [[File:Linear_kernel_Trick.png|170px]]

Radial Basis Kernel: [[File:Rbf kernel trick.png|240px]]

Let '''x''' be the training dataset in [[File:Traing_dataset_space.png|45px]] domain that contain N data points and for simplicity, each of the training data only has 1 features. The new kernel computed from these dataset will now have a dimension of N. The key takeaway from this is the new kernel's dimension is proportional to the original training dataset and thus, it is difficult to visualised the dataset after using the kernel trick (such as in figure 14).

In the analysis, the two features we used are (average voltage, average current).

==== Analysis ====

=== Features extraction from grayscale image ===
==== Prerequisite knowledge ====

==== Analysis ====

== Result discussion ==

== Conclusion ==

== Future work ==

== References ==
<references />

Projects:2019s2-20001 Using Machine Learning to Determine Deposit Height and Defects for Wire + Arc Additive Manufacture (3D printing)

2019-10-07T23:40:28Z

A1713568: /* Prerequisite knowledge */

[[Category:Projects]]
[[Category:Final Year Projects]]
[[Category:2019s2|24501]]
== Abstract ==
Additive manufacturing is an emerging technology that is complementary to subtracting manufacturing. AM is able to create complex model at the cost of production time. An example of AM is Wire + Arc Additive Manufacture. After slicing the manufacturing process of one model to multiple layers, WAAM uses electric arc as a source of heat to feed metal wire and create each layer. However, a slight height inaccuracy in this process will result in an expensive faulty, as error can be multiplied across many layers. Thus, in-process height measurement and faulty detection are critical to the production process. In this thesis, we will look into a machine learning approach to develop an on-the-line algorithm that automatically detects faulty processes and make height measurement by analysing electrical signal (current and voltage) of a Gas Metal Arc Welding machine. This project will introduce the
approaches to tackle the same problem by other researchers, describe the design of our experiment, development of our machine learning model and discuss the final result.

'''Keywords''': Additive manufacturing, Gas metal arc welding, Metal inert gas, Machine learning, Fault detection, Signal analysis
== Introduction ==
3D printing is an emerging technology that has the potential to significantly reduce material usage through the production of near net-shape parts. Many of the systems for 3D printing are based on lasers and powders; however the deposition rate with such systems is very low making the production of large-scale parts difficult. AML Technologies specialises in the use of Wire + Arc Additive Manufacture (WAAM) where deposition is based on arc welding processes and the deposition rates are an order of magnitude greater. When building 3D printed parts, even a relatively small layer height error of only 0.1 mm can produce large build height errors when multiplied across the many layers of a typical build. This can make path planning difficult, so in-process layer height measurement is an essential building block of any production 3D printing system. A variety of techniques can be used for monitoring the layer height including laser scanners, and arc monitoring. It is the latter technique that will be explored in this project due to its robustness, and low cost of implementation – it only requires the measurement of arc current and voltage. Furthermore, it can potentially be used to detect defects by identifying waveform irregularities.
=== Project team ===
==== Project students ====
* Anh Tran
* Nhat Nguyen
==== Supervisors ====
* Dr. Brian Ng
* Dr. Paul Colegrove (AML3D, sponsored company)

[[File:aml3d_logo.jpg|100px]]
=== Objectives ===
The objective of this project is to increase the efficiency of the manufacture process at AML3D. In order to do so, the team will investigate into the possibility of automating and optimising the quality control processes. The two quality control processes that are currently being implemented at AML3D are measuring layer height using laser sensors, and human supervision for detecting defects. These processes add overhead into production time and usage of human resource, which is not desired. To achieve the goal, it is expected that machine learning methods will be used extensively to analyse the electrical signatures of the weld process.

== Background ==
=== Wire Arc Additive Manufacturing ===
Wire and Arc Additive Manufacturing (WAAM) is a type of additive manufacturing that uses electric arc as the heat source and material wire to feed the manufacture process <ref name="WAAM"> S. W. Williams, F. Martina, A. C. Addison, J. Ding, G. Pardal & P. Colegrove (2016) Wire + Arc Additive Manufacturing, Materials Science and Technology, 32:7, 641-647, DOI: 10.1179/1743284715Y.0000000073 </ref>. WAAM has been investigated since the 1990s <ref name="WAAM" />, but only recently that it received more attention from the manufacture world.

Its significant comes from the ability to manufacture complex model with less time and less material. Figure 1a and 1b show real parts that was manufactured by AML3D. Such custom made parts might takes months before be ready to be shipped, but with WAAM, the production time can reduce down to weeks. Currently, the industries that benefit the most from WAAM are maritime and aerospace.

Similar to other additive manufacture methods, WAAM achieves such results by sliding the models into multiple layers, and then build the model layer by layer. The movement control is normally handled by a robotic arm (AML3D uses ABB's Arc Welder robot), and the welding path is generated by a Computer Aid Manufacture (CAM) software.

[[File:welding_part1.JPG|thumb|Figure 1a: A propeller]]
[[File:welding_part2.jpg|thumb|Figure 1b: Another part of the ship]]
[[File:abb_robot.jpg|thumb|Figure 2: ABB's IRB 1520ID Arc Welder]]

Currently at AML3D, the manufacture process for a part can take anywhere from days to weeks. However, the contact tip (where the welding gun deposits the wire) needs replacement every couple of hours. A worn contact tip such as figure 3 will lead to unstable weld and the this effect will propagate and accumulate through multiple layers. Such defects in the manufacture process will no doubt be financially taxing for AML3D. Apart from constant human supervision, another quality control process implemented at AML3D is height measurement of each layer, after the layer is built. This process is implemented to ensure the height of each layer is in the acceptable range (1-2mm). As mentioned before, these quality control processes are costing AML3D in term of time and finance. Our project will focus on replace them with a less time consuming, more automating process.

<div><ul>
<li style="display: inline-block;"> [[File:contact_tip.jpg|thumb|none|400px|Figure 3a: A worn contact tip]] </li>
<li style="display: inline-block;"> [[File:layers.jpeg|thumb|none|530px|Figure 3b: Close up photo of layers]] </li>
</ul></div>

==== Relating Contact Tip to Work-piece Distance to layer height ====
At AML3D, the sensor captures the Contact Tip to Work-piece Distance (CTWD), and from there, we can deduce the layer height. A more detail explanation is of follow:

<div><ul>
<li style="display: inline-block;"> [[File:Example_of_layer_height_measurement.jpg|thumb|none|400px|Figure 4: From CTWD to layer height]] </li>
</ul></div>

The CTWD can be set at the beginning of the weld, however, as the welding process running, the actual CTWD fluctuate around the set CTWD. For example, after finish constructing one layer, the laser sensor capture the model height at 3 different position: A, B and C. Let the 3 model height be m1, m2 and m3. Calculate n_mean and for argument sake, if the set CTWD is 15 mm, the actual CTWD at A is [[File:Ctwd_at_a.png|180px]]. The height of one layer can be calculated as [[File:Layer_height_at_a.png|210px]]. A worked example can be found at table 1.

{| class="wikitable" name="Table 1"
|+Table 1: Worked example
|-
! CTWD
! m1
! m2
! m3
! m_mean
! c1
! c2
! c3
|-
| 15
| 20
| 22
| 25
| 22.3
| 17.3
| 15.3
| 17.7
|}

=== Gas metal arc welding ===
<div><ul>
<li style="display: inline-block;"> [[File:gmaw_welding.png|thumb|500px|none|Figure 5: GMAW welding process <ref name="welding_book"> Norrish, J. (2006). Advanced welding processes. Elsevier. </ref>]] </li>
</ul></div>

Gas metal arc welding (GMAW), or also known as metal inert gas (MIG), is the welding process used at AML3D for WAAM. Due to its high deposition rate and economic benefits, GMAW has became more popular. We will explore two variants of GMAW, which are Pulse Multi Control and Cold Metal Transfer. Both processes are developed by Fronius, an Austrian welding company. Note that it is not required to understand the welding physics to follow this wiki. The next two following sections' purpose is to highlight the high variability, high dynamic nature in welding.

[[File:fronius.png|thumb|Figure 6: Fronius logo]]

==== Pulse Multi Control ====
Pulse Multi Control (PMC) welding is Fronius' modified Pulsed GMAW. The advantage of Pulsed GMAW is the ability to control the metal droplet transfer in welding. Note that from figure 7, the metal droplet is characterised by a downward slope of the current pulse. Figure 8 shows the current signal captured from experiments conducted at AML3D (PMC is used).

<div><ul>
<li style="display: inline-block;"> [[File:Pulsed_gmaw_transfer_mode.png|thumb|500px|none|Figure 7: Relationship between pulse current and droplet transfer<ref name="welding_book" />]] </li>
<li style="display: inline-block;"> [[File:Pmc_current_in_time_domain.png|thumb|650px|none|Figure 8: Current signal captured from experiments conducted at AML3D (PMC process)]] </li>
</ul></div>
==== Cold Metal Transfer ====
Cold Metal Transfer (CMT) is a complex welding process, also developed by Fronius. By detecting a short circuit (mode C and D in figure 9), CMT can make adjustment such as retracting the welding material to cool down, and therefore create a smoother, more stable weld. The complexity of CMT can be seen in the current signal of the process (figure 10). Compare to PMC, CMT's signal has more variation during one signal period.

<div><ul>
<li style="display: inline-block;"> [[File:Transfer_mode.png|thumb|800px|none|Figure 9: Mechanism of metal droplet transfer mode]] </li>
<li style="display: inline-block;"> [[File:Cmt1_current_in_time_domain.png|thumb|735px|none|Figure 10: Current signal captured from experiments conducted at AML3D (CMT process)]] </li>
</ul></div>
=== Machine Learning ===
Machine learning (or statistical learning) refer to a set of tools to give data scientists an insight into the data and make better decisions based on those information <ref name="introduction_ml_book"> James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An introduction to statistical learning (Vol. 112, p. 18). New York: Springer. </ref>. In this project, we will explore the capabilities of classical machine learning methods (i.e anything but Deep Learning) due to the limitation of our available data. By the time of writing this wiki, we have only explored Support Vector Machine.

==== Support Vector Machine ====
The main idea of Support Vector Machine (SVM) is to draw hyperplanes that best separate multiple classes of data. The original SVM judges best decision hyperplane based on the separation margin between data classes (figure 11). However, there are multiple variants of SVM, such as hard/soft margin SVM, Nu-SVM, One class SVM (to solve anomaly detection problem). In our project, we choose to use a soft margin implementation of SVM. The benefit of soft margin SVM (as oppose to hard margin SVM) is that by allowing some misclassifications, we get a wider margin and therefore the solution becomes more "generalise" (i.e work better with a new set of dataset that is independent from the training dataset).

<div><ul>
<li style="display: inline-block;"> [[File:svm_margin.png|thumb|400px|none|Figure 11: Separation margin <ref name="bishop_book"> Bishop, C. M. (2006). Pattern recognition and machine learning. Springer. </ref>]] </li>
</ul></div>

SVM algorithm makes use of [https://en.wikipedia.org/wiki/Lagrange_multiplier Lagrange multiplier]. Let's the linear model of the dataset be:

[[File:Linear_model.png|150px]]

where [[File:Phi_math.png|10px]] denotes the feature space transformation, '''w''' denotes the weights and ''b'' denotes bias constant.

Let [[File:vector_x_n.png|17px]] be the input vector (where each index in the vector denotes a feature), and [[File:target_n.png|15px]] be the target (the class of input vector). It is showed that in order to maximise the (soft) margin, we will need to minimise the following:

[[File:Soft margin minimise equation.png|150px]]

with subject to:

[[File:Soft margin minimise condition.png|150px]]

where N is the total number of training data, C is the regularisation parameter to control the effect of slack variable [[File:slack_variable.png|17px]]. A more visual presentation of slack variables can be found in figure 12.

<div><ul>
<li style="display: inline-block;"> [[File:svm_slack_variable.png|thumb|400px|none|Figure 12: Slack variable <ref name="bishop_book"></ref>]] </li>
</ul></div>

If we were to apply the Lagrangian method to solve the above minimisation problem, we can find the decision hyperplane for the SVM algorithm. However, we shall not discuss the solution here as it is not the main focus of the project.

==== Kernel tricks ====
Both figure 11 and 12 are of linearly separable cases for SVM. For dataset that is not easy to separate with a hyperplane in their original features space (such as figure 13), the data often get projected into a higher dimension space where it is easier to separate (figure 14). Such technique is called the Kernel Tricks. It is important to note that Kernel Tricks does not actually "map" the data from '''X''' dimension to '''Z''' dimension, but rather pairwise compare the similarity of the input space in '''Z''' dimension. However, we shall not dive too deep into this as it is not the main focus of the project.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_nonseparable.png |thumb|350px|none|Figure 13: Nonseparable dataset <ref name="bishop_book"></ref>]] </li>
<li style="display: inline-block;"> [[File:Svm_kernel_trick.png |thumb|650px|none|Figure 14: Simplified example of the Kernel tricks]] </li>
</ul></div>

== Previous studies ==
There are many researches into defect detection in GMAW process. The three most dominant analysis methods are spectroscopic analysis, acoustic analysis, and electrical signal analysis.

=== Spectroscopic analysis ===
In <ref name="spectroscopic_analysis">D. Bebiano and S. Alfaro, “A weld defects detection system based on a spectrometer,” Sensors, vol. 9, no. 4, pp. 2851–2861, 2009.</ref>, the authors captured the intensity of radiation emission of electric arc. It is obvious from figure 15 that good weld's intensity stays stable while defect one's is characterised by a sudden peak. Therefore, a threshold is set to detect the defect welds.

<div><ul>
<li style="display: inline-block;"> [[File:spectroscopic_analysis.png|thumb|400px|none|Figure 15: Spectroscopic analysis <ref name="spectroscopic_analysis"></ref>]] </li>
</ul></div>

Other research into using spectroscopic data to monitor that can be of interest is <ref>D. Naso, B. Turchiano, and P. Pantaleo, “A fuzzy-logic based optical sensor for online weld defect-detection,” IEEE transactions on Industrial Informatics, vol. 1, no. 4, pp. 259–273, 2005.</ref>.

=== Acoustic analysis ===
With <ref name="acoustic_analysis">E. Cayo and S. C. Alfaro, “A non-intrusive GMA welding process quality monitoring system using acoustic sensing,” Sensors, vol. 9, no. 9, pp. 7150–7166, 2009.</ref>, the author discovered that arc ignition is characterised by a distinct sound and therefore it is possible to capture the frequency of ignition through acoustic sensors. A tolerance band was set and both ignition frequency and sound pressure level were used to monitor the quality of the weld.

<div><ul>
<li style="display: inline-block;"> [[File:acoustic_analysis.png|thumb|400px|none|Figure 16: Acoustic analysis <ref name="acoustic_analysis"></ref>]] </li>
</ul></div>

Other research into using acoustic signal to monitor that can be of interest are <ref>M. Fidali, “Detection of welding process instabilities using acoustic signals,” in International Congress on Technical Diagnostic, pp. 191–201, Springer, 2016.</ref>, <ref>L. Zhang, A. C. Basantes-Defaz, D. Ozevin, and E. Indacochea, “Real-time monitoring of welding process using air-coupled ultrasonics and acoustic emission,” The International Journal of Advanced Manufacturing Technology, vol. 101, no. 5-8, pp. 1623–1634, 2019.</ref>.

=== Electrical signal analysis ===
With <ref name="electrical_analysis_1">A. Sumesh, K. Rameshkumar, A. Raja, K. Mohandas, A. Santhakumari, and R. Shyambabu, “Establishing correlation between current and voltage signatures of the arc and weld defects in GMAW process,” Arabian Journal for Science and Engineering, vol. 42, no. 11, pp. 4649–4665, 2017.</ref>, the author plot the probability density distribution of voltage signal and discovered that the stable weld's signal concentrate in one region while unstable signal spread out. While in the paper, the authors did not developed a monitoring method using their findings, it shows that there is a significant differences between stable and unstable weld's electrical signature.

<div><ul>
<li style="display: inline-block;"> [[File:Electrical analysis pdd.png|thumb|400px|none|Figure 17: Voltage PDD <ref name="electrical_analysis_1"></ref>]] </li>
</ul></div>

In <ref name="electrical_analysis_2">S. Simpson, “Signature images for arc welding fault detection,” Science and Technology of Welding and Joining, vol. 12, no. 6, pp. 481–486, 2007.</ref>, the authors shows that by plotting the signature image of voltage and current scatter plot, both arcing and short-circuiting region's signature image can notify a defects welds. A more visual representation of short circuiting and arcing region can be found in [[#Gas metal arc welding|Gas metal arc welding section]].

<div><ul>
<li style="display: inline-block;"> [[File:Electrical voltage signature.png|thumb|400px|none|Figure 18a: Electrical voltage signature of stable weld <ref name="electrical_analysis_2"></ref>]] </li>
<li style="display: inline-block;"> [[File:Electrical signature 2.png|thumb|400px|none|Figure 18b: Electrical voltage signature of unstable weld <ref name="electrical_analysis_2"></ref>]] </li>
</ul></div>

Other research into using electrical signal to monitor that can be of interest are <ref>R. Madigan, “Arc sensing for defects in constant-voltage gas metal arc welding,” Welding Journal, vol. 78, pp. 322S–328S, 1999.</ref>, <ref>Y. Huang, K. Wang, Z. Zhou, X. Zhou, and J. Fang, “Stability evaluation of short-circuiting gas metal arc welding based on ensemble empirical mode decomposition,” Measurement Science and Technology, vol. 28, no. 3, p. 035006, 2017.</ref>,<ref>Z. Zhang, X. Chen, H. Chen, J. Zhong, and S. Chen, “Online welding quality monitoring based on feature extraction of arc voltage signal,” The International Journal of Advanced Manufacturing Technology, vol. 70, no. 9-12, pp. 1661–1671, 2014</ref>,<ref>X. Li and S. Simpson, “Parametric approach to positional fault detection in short arc welding,” Science and Technology of welding and joining, vol. 14, no. 2, pp. 146–151, 2009.</ref>, and <ref>E. Wei, D. Farson, R. Richardson, and H. Ludewig, “Detection of weld surface porosity by statistical analysis of arc current in gas metal arc welding,” Journal of Manufacturing Processes, vol. 3, no. 1, pp. 50–59, 2001.</ref>.

== Experiment setup ==
The welding system uses TPS 320i C Pulse (figure 15a), produced by Fronius, to supply power, feed wire, and system cooling. For movement control of the welding torch, the system use ABB’s IRB 1520ID (figure 15b), which is a dedicated arc welding robot. The Data Acquisition device is NI's DAQ. Each channel (voltage and current) is sampled at 25kHz while the dominant frequency for voltage and current signal are around 130Hz (for CMT process) and 190Hz (for PMC process).

<div><ul>
<li style="display: inline-block;"> [[File:Fronius_power_supply.jpg |thumb|250px|none|Figure 19a: Fronius' Power Supply]] </li>
<li style="display: inline-block;"> [[File:Welding_robot.jpg |thumb|250px|none|Figure 19b: ABB's welding robot]] </li>
</ul></div>

[[File:Ni_daq.jpg|thumb|Figure 16: NI's DAQ device]]

== Statistical Analysis ==

=== Preprocessing ===
As the captured signal coming from Fronius' power supply, the signals also pick up switching power noise. The noise will interfere with the accuracy of the analysis if not filtered. In this project, we applied a 1000Hz digital Low Pass Butterworth filter. The effect of the filter can be seen in figure 20, while the comparation of using different filters can be seen in the gallery.

<div><ul>
<li style="display: inline-block;"> [[File:Filter_effect.png |thumb|750px|none|Figure 20: Effect of 1000Hz low pass filter]] </li>
<li style="display: inline-block;"> [[File:Freq_response.png |thumb|750px|none|Figure 21: Frequency response of the in-use filter]] </li>
</ul></div>

<gallery mode="packed-hover" caption="Gallery: Comparison of different filtering method" perrow=2>
File:Effect_of_filtering2.png|In use filter vs 300Hz low pass filter
File:Effect_of_filtering3.png|In use filter vs 500Hz low pass filter
File:Effect_of_filtering4.png|In use filter vs 750Hz low pass filter
File:Effect_of_filtering5.png|In use filter vs 2000Hz low pass filter
</gallery>

=== Power Spectral Density ===
==== Prerequisite knowledge ====
Looking at the spectrum is another way to analyse the fundamental information of the signal, in frequency domain <ref name="signal_processing_book">Prandoni, P., & Vetterli, M. (2008). Signal processing for communications. EPFL press.</ref>. Note that the term "Power" in Power Spectral Density does not refer to the convention power definition. In signal processing, energy of a discrete time signal is defined as follow:

[[File:Signal_energy_formular.png|150px]]

The way to interpret this definition is if we consider the signal to be of "Voltage" unit, ''E'' is the total energy (in joules) dissipated over a 1 Ohm resistor <ref name="signal_processing_book"></ref>. The power spectral density of a signal over time interval [-M,M] is defined as follow:

[[File:Psd_formular.png|300px]]

where [[File:Fourier series symbol.png|67px]] is the truncated Fourier transform:

[[File:Fourier_series_formular.png|250px]]

==== Analysis ====
Our goal for this project is to detect the defect welds and predict the layer heights (the contact tip to work piece distance, CTWD, to be more accurate) in real time (refer to [[#Relating Contact Tip to Work-piece Distance to layer height|this section]] to understand the relationship between Contact Tip to Work-piece Distance and Layer height). An analysis of the current power spectral density (PSD) shows that there is an inverse proportional relationship between the CTWD of a weld and the frequency where its '''current signal''' PSD reach the peak, and also the magnitude of the peak. Since the peak of PSD can be fairly sensitive, we only consider the frequency of the peak PSD.

Figure 22 shows the PSD plot of voltage, current, and power signal of the weld using PMC process. In this experiment, we created multiple one-line, 8 seconds welds. These eight-second welds then get segmented into smaller two-second chunk to increase the number of data and decrease the processing time.

<div><ul>
<li style="display: inline-block;"> [[File:Psd pmc good bad all lowfreq.png |thumb|750px|none|Figure 22: Power spectral density of PMC process]] </li>
<li style="display: inline-block;"> [[File:One line weld.jpg |thumb|578px|none|Figure 23: Simple one-line weld experiment]] </li>
<li style="display: inline-block;"> [[File:Cmt_psd_stable_allCTWD_zoomed.png |thumb|750px|none|Figure 24: Simple one-line weld experiment]] </li>

</ul></div>

It seems like the inverse proportional relationship holds true for most cases. Other experiments showed that for a short period at the beginning and ending of the weld, the signal can behave differently relative to the rest of the weld. Currently, to migrate with this problem, data will not be collected and analysed (both offline and online) at these periods. However, the above results seem to only hold true for PMC processes. With CMT process, the current PSD plot indicates no relationship between the CTWD and the frequency at peak (figure 24).

For the next stage of the project, our plan is to develop a linear regression model to predict the CTWD for PMC process. With CMT process, we will have to look into different analysis techniques.

=== Support Vector Machine ===
==== Prerequisite knowledge ====
Refer to [[#Support Vector Machine|Support Vector Machine]] section for more detail.

As mentioned before, we model the dataset as a linear model: [[File:Linear model.png|150px]]. In the analysis, we try 2 different kernel tricks: linear and radial basis function kernel.

Linear kernel: [[File:Linear_kernel_Trick.png|170px]]

Radial Basis Kernel: [[File:Rbf kernel trick.png|240px]]

Let '''x''' be the training dataset in [[File:Traing_dataset_space.png|45px]] domain that contain N data points and for simplicity, each of the training data only has 1 features. The new kernel computed from these dataset will now have a dimension of N. The key takeaway from this is the new kernel's dimension is proportional to the original training dataset and thus, it is difficult to visualised the dataset after using the kernel trick (such as in figure 14).

==== Analysis ====

=== Features extraction from grayscale image ===
==== Prerequisite knowledge ====

==== Analysis ====

== Result discussion ==

== Conclusion ==

== Future work ==

== References ==
<references />

File:Traing dataset space.png

2019-10-07T23:30:51Z

A1713568: A1713568 uploaded a new version of "File:Traing dataset space.png"

File:Traing dataset space.png

2019-10-07T23:28:53Z

A1713568:

File:Rbf kernel trick.png

2019-10-07T23:24:11Z

A1713568:

File:Linear kernel Trick.png

2019-10-07T23:23:47Z

A1713568:

Projects:2019s2-20001 Using Machine Learning to Determine Deposit Height and Defects for Wire + Arc Additive Manufacture (3D printing)

2019-10-07T23:02:04Z

A1713568: /* Kernel tricks */

[[Category:Projects]]
[[Category:Final Year Projects]]
[[Category:2019s2|24501]]
== Abstract ==
Additive manufacturing is an emerging technology that is complementary to subtracting manufacturing. AM is able to create complex model at the cost of production time. An example of AM is Wire + Arc Additive Manufacture. After slicing the manufacturing process of one model to multiple layers, WAAM uses electric arc as a source of heat to feed metal wire and create each layer. However, a slight height inaccuracy in this process will result in an expensive faulty, as error can be multiplied across many layers. Thus, in-process height measurement and faulty detection are critical to the production process. In this thesis, we will look into a machine learning approach to develop an on-the-line algorithm that automatically detects faulty processes and make height measurement by analysing electrical signal (current and voltage) of a Gas Metal Arc Welding machine. This project will introduce the
approaches to tackle the same problem by other researchers, describe the design of our experiment, development of our machine learning model and discuss the final result.

'''Keywords''': Additive manufacturing, Gas metal arc welding, Metal inert gas, Machine learning, Fault detection, Signal analysis
== Introduction ==
3D printing is an emerging technology that has the potential to significantly reduce material usage through the production of near net-shape parts. Many of the systems for 3D printing are based on lasers and powders; however the deposition rate with such systems is very low making the production of large-scale parts difficult. AML Technologies specialises in the use of Wire + Arc Additive Manufacture (WAAM) where deposition is based on arc welding processes and the deposition rates are an order of magnitude greater. When building 3D printed parts, even a relatively small layer height error of only 0.1 mm can produce large build height errors when multiplied across the many layers of a typical build. This can make path planning difficult, so in-process layer height measurement is an essential building block of any production 3D printing system. A variety of techniques can be used for monitoring the layer height including laser scanners, and arc monitoring. It is the latter technique that will be explored in this project due to its robustness, and low cost of implementation – it only requires the measurement of arc current and voltage. Furthermore, it can potentially be used to detect defects by identifying waveform irregularities.
=== Project team ===
==== Project students ====
* Anh Tran
* Nhat Nguyen
==== Supervisors ====
* Dr. Brian Ng
* Dr. Paul Colegrove (AML3D, sponsored company)

[[File:aml3d_logo.jpg|100px]]
=== Objectives ===
The objective of this project is to increase the efficiency of the manufacture process at AML3D. In order to do so, the team will investigate into the possibility of automating and optimising the quality control processes. The two quality control processes that are currently being implemented at AML3D are measuring layer height using laser sensors, and human supervision for detecting defects. These processes add overhead into production time and usage of human resource, which is not desired. To achieve the goal, it is expected that machine learning methods will be used extensively to analyse the electrical signatures of the weld process.

== Background ==
=== Wire Arc Additive Manufacturing ===
Wire and Arc Additive Manufacturing (WAAM) is a type of additive manufacturing that uses electric arc as the heat source and material wire to feed the manufacture process <ref name="WAAM"> S. W. Williams, F. Martina, A. C. Addison, J. Ding, G. Pardal & P. Colegrove (2016) Wire + Arc Additive Manufacturing, Materials Science and Technology, 32:7, 641-647, DOI: 10.1179/1743284715Y.0000000073 </ref>. WAAM has been investigated since the 1990s <ref name="WAAM" />, but only recently that it received more attention from the manufacture world.

Its significant comes from the ability to manufacture complex model with less time and less material. Figure 1a and 1b show real parts that was manufactured by AML3D. Such custom made parts might takes months before be ready to be shipped, but with WAAM, the production time can reduce down to weeks. Currently, the industries that benefit the most from WAAM are maritime and aerospace.

Similar to other additive manufacture methods, WAAM achieves such results by sliding the models into multiple layers, and then build the model layer by layer. The movement control is normally handled by a robotic arm (AML3D uses ABB's Arc Welder robot), and the welding path is generated by a Computer Aid Manufacture (CAM) software.

[[File:welding_part1.JPG|thumb|Figure 1a: A propeller]]
[[File:welding_part2.jpg|thumb|Figure 1b: Another part of the ship]]
[[File:abb_robot.jpg|thumb|Figure 2: ABB's IRB 1520ID Arc Welder]]

Currently at AML3D, the manufacture process for a part can take anywhere from days to weeks. However, the contact tip (where the welding gun deposits the wire) needs replacement every couple of hours. A worn contact tip such as figure 3 will lead to unstable weld and the this effect will propagate and accumulate through multiple layers. Such defects in the manufacture process will no doubt be financially taxing for AML3D. Apart from constant human supervision, another quality control process implemented at AML3D is height measurement of each layer, after the layer is built. This process is implemented to ensure the height of each layer is in the acceptable range (1-2mm). As mentioned before, these quality control processes are costing AML3D in term of time and finance. Our project will focus on replace them with a less time consuming, more automating process.

<div><ul>
<li style="display: inline-block;"> [[File:contact_tip.jpg|thumb|none|400px|Figure 3a: A worn contact tip]] </li>
<li style="display: inline-block;"> [[File:layers.jpeg|thumb|none|530px|Figure 3b: Close up photo of layers]] </li>
</ul></div>

==== Relating Contact Tip to Work-piece Distance to layer height ====
At AML3D, the sensor captures the Contact Tip to Work-piece Distance (CTWD), and from there, we can deduce the layer height. A more detail explanation is of follow:

<div><ul>
<li style="display: inline-block;"> [[File:Example_of_layer_height_measurement.jpg|thumb|none|400px|Figure 4: From CTWD to layer height]] </li>
</ul></div>

The CTWD can be set at the beginning of the weld, however, as the welding process running, the actual CTWD fluctuate around the set CTWD. For example, after finish constructing one layer, the laser sensor capture the model height at 3 different position: A, B and C. Let the 3 model height be m1, m2 and m3. Calculate n_mean and for argument sake, if the set CTWD is 15 mm, the actual CTWD at A is [[File:Ctwd_at_a.png|180px]]. The height of one layer can be calculated as [[File:Layer_height_at_a.png|210px]]. A worked example can be found at table 1.

{| class="wikitable" name="Table 1"
|+Table 1: Worked example
|-
! CTWD
! m1
! m2
! m3
! m_mean
! c1
! c2
! c3
|-
| 15
| 20
| 22
| 25
| 22.3
| 17.3
| 15.3
| 17.7
|}

=== Gas metal arc welding ===
<div><ul>
<li style="display: inline-block;"> [[File:gmaw_welding.png|thumb|500px|none|Figure 5: GMAW welding process <ref name="welding_book"> Norrish, J. (2006). Advanced welding processes. Elsevier. </ref>]] </li>
</ul></div>

Gas metal arc welding (GMAW), or also known as metal inert gas (MIG), is the welding process used at AML3D for WAAM. Due to its high deposition rate and economic benefits, GMAW has became more popular. We will explore two variants of GMAW, which are Pulse Multi Control and Cold Metal Transfer. Both processes are developed by Fronius, an Austrian welding company. Note that it is not required to understand the welding physics to follow this wiki. The next two following sections' purpose is to highlight the high variability, high dynamic nature in welding.

[[File:fronius.png|thumb|Figure 6: Fronius logo]]

==== Pulse Multi Control ====
Pulse Multi Control (PMC) welding is Fronius' modified Pulsed GMAW. The advantage of Pulsed GMAW is the ability to control the metal droplet transfer in welding. Note that from figure 7, the metal droplet is characterised by a downward slope of the current pulse. Figure 8 shows the current signal captured from experiments conducted at AML3D (PMC is used).

<div><ul>
<li style="display: inline-block;"> [[File:Pulsed_gmaw_transfer_mode.png|thumb|500px|none|Figure 7: Relationship between pulse current and droplet transfer<ref name="welding_book" />]] </li>
<li style="display: inline-block;"> [[File:Pmc_current_in_time_domain.png|thumb|650px|none|Figure 8: Current signal captured from experiments conducted at AML3D (PMC process)]] </li>
</ul></div>
==== Cold Metal Transfer ====
Cold Metal Transfer (CMT) is a complex welding process, also developed by Fronius. By detecting a short circuit (mode C and D in figure 9), CMT can make adjustment such as retracting the welding material to cool down, and therefore create a smoother, more stable weld. The complexity of CMT can be seen in the current signal of the process (figure 10). Compare to PMC, CMT's signal has more variation during one signal period.

<div><ul>
<li style="display: inline-block;"> [[File:Transfer_mode.png|thumb|800px|none|Figure 9: Mechanism of metal droplet transfer mode]] </li>
<li style="display: inline-block;"> [[File:Cmt1_current_in_time_domain.png|thumb|735px|none|Figure 10: Current signal captured from experiments conducted at AML3D (CMT process)]] </li>
</ul></div>
=== Machine Learning ===
Machine learning (or statistical learning) refer to a set of tools to give data scientists an insight into the data and make better decisions based on those information <ref name="introduction_ml_book"> James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An introduction to statistical learning (Vol. 112, p. 18). New York: Springer. </ref>. In this project, we will explore the capabilities of classical machine learning methods (i.e anything but Deep Learning) due to the limitation of our available data. By the time of writing this wiki, we have only explored Support Vector Machine.

==== Support Vector Machine ====
The main idea of Support Vector Machine (SVM) is to draw hyperplanes that best separate multiple classes of data. The original SVM judges best decision hyperplane based on the separation margin between data classes (figure 11). However, there are multiple variants of SVM, such as hard/soft margin SVM, Nu-SVM, One class SVM (to solve anomaly detection problem). In our project, we choose to use a soft margin implementation of SVM. The benefit of soft margin SVM (as oppose to hard margin SVM) is that by allowing some misclassifications, we get a wider margin and therefore the solution becomes more "generalise" (i.e work better with a new set of dataset that is independent from the training dataset).

<div><ul>
<li style="display: inline-block;"> [[File:svm_margin.png|thumb|400px|none|Figure 11: Separation margin <ref name="bishop_book"> Bishop, C. M. (2006). Pattern recognition and machine learning. Springer. </ref>]] </li>
</ul></div>

SVM algorithm makes use of [https://en.wikipedia.org/wiki/Lagrange_multiplier Lagrange multiplier]. Let's the linear model of the dataset be:

[[File:Linear_model.png|150px]]

where [[File:Phi_math.png|10px]] denotes the feature space transformation, '''w''' denotes the weights and ''b'' denotes bias constant.

Let [[File:vector_x_n.png|17px]] be the input vector (where each index in the vector denotes a feature), and [[File:target_n.png|15px]] be the target (the class of input vector). It is showed that in order to maximise the (soft) margin, we will need to minimise the following:

[[File:Soft margin minimise equation.png|150px]]

with subject to:

[[File:Soft margin minimise condition.png|150px]]

where N is the total number of training data, C is the regularisation parameter to control the effect of slack variable [[File:slack_variable.png|17px]]. A more visual presentation of slack variables can be found in figure 12.

<div><ul>
<li style="display: inline-block;"> [[File:svm_slack_variable.png|thumb|400px|none|Figure 12: Slack variable <ref name="bishop_book"></ref>]] </li>
</ul></div>

If we were to apply the Lagrangian method to solve the above minimisation problem, we can find the decision hyperplane for the SVM algorithm. However, we shall not discuss the solution here as it is not the main focus of the project.

==== Kernel tricks ====
Both figure 11 and 12 are of linearly separable cases for SVM. For dataset that is not easy to separate with a hyperplane in their original features space (such as figure 13), the data often get projected into a higher dimension space where it is easier to separate (figure 14). Such technique is called the Kernel Tricks. It is important to note that Kernel Tricks does not actually "map" the data from '''X''' dimension to '''Z''' dimension, but rather pairwise compare the similarity of the input space in '''Z''' dimension. However, we shall not dive too deep into this as it is not the main focus of the project.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_nonseparable.png |thumb|350px|none|Figure 13: Nonseparable dataset <ref name="bishop_book"></ref>]] </li>
<li style="display: inline-block;"> [[File:Svm_kernel_trick.png |thumb|650px|none|Figure 14: Simplified example of the Kernel tricks]] </li>
</ul></div>

== Previous studies ==
There are many researches into defect detection in GMAW process. The three most dominant analysis methods are spectroscopic analysis, acoustic analysis, and electrical signal analysis.

=== Spectroscopic analysis ===
In <ref name="spectroscopic_analysis">D. Bebiano and S. Alfaro, “A weld defects detection system based on a spectrometer,” Sensors, vol. 9, no. 4, pp. 2851–2861, 2009.</ref>, the authors captured the intensity of radiation emission of electric arc. It is obvious from figure 15 that good weld's intensity stays stable while defect one's is characterised by a sudden peak. Therefore, a threshold is set to detect the defect welds.

<div><ul>
<li style="display: inline-block;"> [[File:spectroscopic_analysis.png|thumb|400px|none|Figure 15: Spectroscopic analysis <ref name="spectroscopic_analysis"></ref>]] </li>
</ul></div>

Other research into using spectroscopic data to monitor that can be of interest is <ref>D. Naso, B. Turchiano, and P. Pantaleo, “A fuzzy-logic based optical sensor for online weld defect-detection,” IEEE transactions on Industrial Informatics, vol. 1, no. 4, pp. 259–273, 2005.</ref>.

=== Acoustic analysis ===
With <ref name="acoustic_analysis">E. Cayo and S. C. Alfaro, “A non-intrusive GMA welding process quality monitoring system using acoustic sensing,” Sensors, vol. 9, no. 9, pp. 7150–7166, 2009.</ref>, the author discovered that arc ignition is characterised by a distinct sound and therefore it is possible to capture the frequency of ignition through acoustic sensors. A tolerance band was set and both ignition frequency and sound pressure level were used to monitor the quality of the weld.

<div><ul>
<li style="display: inline-block;"> [[File:acoustic_analysis.png|thumb|400px|none|Figure 16: Acoustic analysis <ref name="acoustic_analysis"></ref>]] </li>
</ul></div>

Other research into using acoustic signal to monitor that can be of interest are <ref>M. Fidali, “Detection of welding process instabilities using acoustic signals,” in International Congress on Technical Diagnostic, pp. 191–201, Springer, 2016.</ref>, <ref>L. Zhang, A. C. Basantes-Defaz, D. Ozevin, and E. Indacochea, “Real-time monitoring of welding process using air-coupled ultrasonics and acoustic emission,” The International Journal of Advanced Manufacturing Technology, vol. 101, no. 5-8, pp. 1623–1634, 2019.</ref>.

=== Electrical signal analysis ===
With <ref name="electrical_analysis_1">A. Sumesh, K. Rameshkumar, A. Raja, K. Mohandas, A. Santhakumari, and R. Shyambabu, “Establishing correlation between current and voltage signatures of the arc and weld defects in GMAW process,” Arabian Journal for Science and Engineering, vol. 42, no. 11, pp. 4649–4665, 2017.</ref>, the author plot the probability density distribution of voltage signal and discovered that the stable weld's signal concentrate in one region while unstable signal spread out. While in the paper, the authors did not developed a monitoring method using their findings, it shows that there is a significant differences between stable and unstable weld's electrical signature.

<div><ul>
<li style="display: inline-block;"> [[File:Electrical analysis pdd.png|thumb|400px|none|Figure 17: Voltage PDD <ref name="electrical_analysis_1"></ref>]] </li>
</ul></div>

In <ref name="electrical_analysis_2">S. Simpson, “Signature images for arc welding fault detection,” Science and Technology of Welding and Joining, vol. 12, no. 6, pp. 481–486, 2007.</ref>, the authors shows that by plotting the signature image of voltage and current scatter plot, both arcing and short-circuiting region's signature image can notify a defects welds. A more visual representation of short circuiting and arcing region can be found in [[#Gas metal arc welding|Gas metal arc welding section]].

<div><ul>
<li style="display: inline-block;"> [[File:Electrical voltage signature.png|thumb|400px|none|Figure 18a: Electrical voltage signature of stable weld <ref name="electrical_analysis_2"></ref>]] </li>
<li style="display: inline-block;"> [[File:Electrical signature 2.png|thumb|400px|none|Figure 18b: Electrical voltage signature of unstable weld <ref name="electrical_analysis_2"></ref>]] </li>
</ul></div>

Other research into using electrical signal to monitor that can be of interest are <ref>R. Madigan, “Arc sensing for defects in constant-voltage gas metal arc welding,” Welding Journal, vol. 78, pp. 322S–328S, 1999.</ref>, <ref>Y. Huang, K. Wang, Z. Zhou, X. Zhou, and J. Fang, “Stability evaluation of short-circuiting gas metal arc welding based on ensemble empirical mode decomposition,” Measurement Science and Technology, vol. 28, no. 3, p. 035006, 2017.</ref>,<ref>Z. Zhang, X. Chen, H. Chen, J. Zhong, and S. Chen, “Online welding quality monitoring based on feature extraction of arc voltage signal,” The International Journal of Advanced Manufacturing Technology, vol. 70, no. 9-12, pp. 1661–1671, 2014</ref>,<ref>X. Li and S. Simpson, “Parametric approach to positional fault detection in short arc welding,” Science and Technology of welding and joining, vol. 14, no. 2, pp. 146–151, 2009.</ref>, and <ref>E. Wei, D. Farson, R. Richardson, and H. Ludewig, “Detection of weld surface porosity by statistical analysis of arc current in gas metal arc welding,” Journal of Manufacturing Processes, vol. 3, no. 1, pp. 50–59, 2001.</ref>.

== Experiment setup ==
The welding system uses TPS 320i C Pulse (figure 15a), produced by Fronius, to supply power, feed wire, and system cooling. For movement control of the welding torch, the system use ABB’s IRB 1520ID (figure 15b), which is a dedicated arc welding robot. The Data Acquisition device is NI's DAQ. Each channel (voltage and current) is sampled at 25kHz while the dominant frequency for voltage and current signal are around 130Hz (for CMT process) and 190Hz (for PMC process).

<div><ul>
<li style="display: inline-block;"> [[File:Fronius_power_supply.jpg |thumb|250px|none|Figure 19a: Fronius' Power Supply]] </li>
<li style="display: inline-block;"> [[File:Welding_robot.jpg |thumb|250px|none|Figure 19b: ABB's welding robot]] </li>
</ul></div>

[[File:Ni_daq.jpg|thumb|Figure 16: NI's DAQ device]]

== Statistical Analysis ==

=== Preprocessing ===
As the captured signal coming from Fronius' power supply, the signals also pick up switching power noise. The noise will interfere with the accuracy of the analysis if not filtered. In this project, we applied a 1000Hz digital Low Pass Butterworth filter. The effect of the filter can be seen in figure 20, while the comparation of using different filters can be seen in the gallery.

<div><ul>
<li style="display: inline-block;"> [[File:Filter_effect.png |thumb|750px|none|Figure 20: Effect of 1000Hz low pass filter]] </li>
<li style="display: inline-block;"> [[File:Freq_response.png |thumb|750px|none|Figure 21: Frequency response of the in-use filter]] </li>
</ul></div>

<gallery mode="packed-hover" caption="Gallery: Comparison of different filtering method" perrow=2>
File:Effect_of_filtering2.png|In use filter vs 300Hz low pass filter
File:Effect_of_filtering3.png|In use filter vs 500Hz low pass filter
File:Effect_of_filtering4.png|In use filter vs 750Hz low pass filter
File:Effect_of_filtering5.png|In use filter vs 2000Hz low pass filter
</gallery>

=== Power Spectral Density ===
==== Prerequisite knowledge ====
Looking at the spectrum is another way to analyse the fundamental information of the signal, in frequency domain <ref name="signal_processing_book">Prandoni, P., & Vetterli, M. (2008). Signal processing for communications. EPFL press.</ref>. Note that the term "Power" in Power Spectral Density does not refer to the convention power definition. In signal processing, energy of a discrete time signal is defined as follow:

[[File:Signal_energy_formular.png|150px]]

The way to interpret this definition is if we consider the signal to be of "Voltage" unit, ''E'' is the total energy (in joules) dissipated over a 1 Ohm resistor <ref name="signal_processing_book"></ref>. The power spectral density of a signal over time interval [-M,M] is defined as follow:

[[File:Psd_formular.png|300px]]

where [[File:Fourier series symbol.png|67px]] is the truncated Fourier transform:

[[File:Fourier_series_formular.png|250px]]

==== Analysis ====
Our goal for this project is to detect the defect welds and predict the layer heights (the contact tip to work piece distance, CTWD, to be more accurate) in real time (refer to [[#Relating Contact Tip to Work-piece Distance to layer height|this section]] to understand the relationship between Contact Tip to Work-piece Distance and Layer height). An analysis of the current power spectral density (PSD) shows that there is an inverse proportional relationship between the CTWD of a weld and the frequency where its '''current signal''' PSD reach the peak, and also the magnitude of the peak. Since the peak of PSD can be fairly sensitive, we only consider the frequency of the peak PSD.

Figure 22 shows the PSD plot of voltage, current, and power signal of the weld using PMC process. In this experiment, we created multiple one-line, 8 seconds welds. These eight-second welds then get segmented into smaller two-second chunk to increase the number of data and decrease the processing time.

<div><ul>
<li style="display: inline-block;"> [[File:Psd pmc good bad all lowfreq.png |thumb|750px|none|Figure 22: Power spectral density of PMC process]] </li>
<li style="display: inline-block;"> [[File:One line weld.jpg |thumb|578px|none|Figure 23: Simple one-line weld experiment]] </li>
<li style="display: inline-block;"> [[File:Cmt_psd_stable_allCTWD_zoomed.png |thumb|750px|none|Figure 24: Simple one-line weld experiment]] </li>

</ul></div>

It seems like the inverse proportional relationship holds true for most cases. Other experiments showed that for a short period at the beginning and ending of the weld, the signal can behave differently relative to the rest of the weld. Currently, to migrate with this problem, data will not be collected and analysed (both offline and online) at these periods. However, the above results seem to only hold true for PMC processes. With CMT process, the current PSD plot indicates no relationship between the CTWD and the frequency at peak (figure 24).

For the next stage of the project, our plan is to develop a linear regression model to predict the CTWD for PMC process. With CMT process, we will have to look into different analysis techniques.

=== Support Vector Machine ===
==== Prerequisite knowledge ====

==== Analysis ====

=== Features extraction from grayscale image ===
==== Prerequisite knowledge ====

==== Analysis ====

== Result discussion ==

== Conclusion ==

== Future work ==

== References ==
<references />

Projects:2019s2-20001 Using Machine Learning to Determine Deposit Height and Defects for Wire + Arc Additive Manufacture (3D printing)

2019-10-07T09:57:01Z

A1713568: /* Features extraction from grayscale image */

[[Category:Projects]]
[[Category:Final Year Projects]]
[[Category:2019s2|24501]]
== Abstract ==
Additive manufacturing is an emerging technology that is complementary to subtracting manufacturing. AM is able to create complex model at the cost of production time. An example of AM is Wire + Arc Additive Manufacture. After slicing the manufacturing process of one model to multiple layers, WAAM uses electric arc as a source of heat to feed metal wire and create each layer. However, a slight height inaccuracy in this process will result in an expensive faulty, as error can be multiplied across many layers. Thus, in-process height measurement and faulty detection are critical to the production process. In this thesis, we will look into a machine learning approach to develop an on-the-line algorithm that automatically detects faulty processes and make height measurement by analysing electrical signal (current and voltage) of a Gas Metal Arc Welding machine. This project will introduce the
approaches to tackle the same problem by other researchers, describe the design of our experiment, development of our machine learning model and discuss the final result.

'''Keywords''': Additive manufacturing, Gas metal arc welding, Metal inert gas, Machine learning, Fault detection, Signal analysis
== Introduction ==
3D printing is an emerging technology that has the potential to significantly reduce material usage through the production of near net-shape parts. Many of the systems for 3D printing are based on lasers and powders; however the deposition rate with such systems is very low making the production of large-scale parts difficult. AML Technologies specialises in the use of Wire + Arc Additive Manufacture (WAAM) where deposition is based on arc welding processes and the deposition rates are an order of magnitude greater. When building 3D printed parts, even a relatively small layer height error of only 0.1 mm can produce large build height errors when multiplied across the many layers of a typical build. This can make path planning difficult, so in-process layer height measurement is an essential building block of any production 3D printing system. A variety of techniques can be used for monitoring the layer height including laser scanners, and arc monitoring. It is the latter technique that will be explored in this project due to its robustness, and low cost of implementation – it only requires the measurement of arc current and voltage. Furthermore, it can potentially be used to detect defects by identifying waveform irregularities.
=== Project team ===
==== Project students ====
* Anh Tran
* Nhat Nguyen
==== Supervisors ====
* Dr. Brian Ng
* Dr. Paul Colegrove (AML3D, sponsored company)

[[File:aml3d_logo.jpg|100px]]
=== Objectives ===
The objective of this project is to increase the efficiency of the manufacture process at AML3D. In order to do so, the team will investigate into the possibility of automating and optimising the quality control processes. The two quality control processes that are currently being implemented at AML3D are measuring layer height using laser sensors, and human supervision for detecting defects. These processes add overhead into production time and usage of human resource, which is not desired. To achieve the goal, it is expected that machine learning methods will be used extensively to analyse the electrical signatures of the weld process.

== Background ==
=== Wire Arc Additive Manufacturing ===
Wire and Arc Additive Manufacturing (WAAM) is a type of additive manufacturing that uses electric arc as the heat source and material wire to feed the manufacture process <ref name="WAAM"> S. W. Williams, F. Martina, A. C. Addison, J. Ding, G. Pardal & P. Colegrove (2016) Wire + Arc Additive Manufacturing, Materials Science and Technology, 32:7, 641-647, DOI: 10.1179/1743284715Y.0000000073 </ref>. WAAM has been investigated since the 1990s <ref name="WAAM" />, but only recently that it received more attention from the manufacture world.

Its significant comes from the ability to manufacture complex model with less time and less material. Figure 1a and 1b show real parts that was manufactured by AML3D. Such custom made parts might takes months before be ready to be shipped, but with WAAM, the production time can reduce down to weeks. Currently, the industries that benefit the most from WAAM are maritime and aerospace.

Similar to other additive manufacture methods, WAAM achieves such results by sliding the models into multiple layers, and then build the model layer by layer. The movement control is normally handled by a robotic arm (AML3D uses ABB's Arc Welder robot), and the welding path is generated by a Computer Aid Manufacture (CAM) software.

[[File:welding_part1.JPG|thumb|Figure 1a: A propeller]]
[[File:welding_part2.jpg|thumb|Figure 1b: Another part of the ship]]
[[File:abb_robot.jpg|thumb|Figure 2: ABB's IRB 1520ID Arc Welder]]

Currently at AML3D, the manufacture process for a part can take anywhere from days to weeks. However, the contact tip (where the welding gun deposits the wire) needs replacement every couple of hours. A worn contact tip such as figure 3 will lead to unstable weld and the this effect will propagate and accumulate through multiple layers. Such defects in the manufacture process will no doubt be financially taxing for AML3D. Apart from constant human supervision, another quality control process implemented at AML3D is height measurement of each layer, after the layer is built. This process is implemented to ensure the height of each layer is in the acceptable range (1-2mm). As mentioned before, these quality control processes are costing AML3D in term of time and finance. Our project will focus on replace them with a less time consuming, more automating process.

<div><ul>
<li style="display: inline-block;"> [[File:contact_tip.jpg|thumb|none|400px|Figure 3a: A worn contact tip]] </li>
<li style="display: inline-block;"> [[File:layers.jpeg|thumb|none|530px|Figure 3b: Close up photo of layers]] </li>
</ul></div>

==== Relating Contact Tip to Work-piece Distance to layer height ====
At AML3D, the sensor captures the Contact Tip to Work-piece Distance (CTWD), and from there, we can deduce the layer height. A more detail explanation is of follow:

<div><ul>
<li style="display: inline-block;"> [[File:Example_of_layer_height_measurement.jpg|thumb|none|400px|Figure 4: From CTWD to layer height]] </li>
</ul></div>

The CTWD can be set at the beginning of the weld, however, as the welding process running, the actual CTWD fluctuate around the set CTWD. For example, after finish constructing one layer, the laser sensor capture the model height at 3 different position: A, B and C. Let the 3 model height be m1, m2 and m3. Calculate n_mean and for argument sake, if the set CTWD is 15 mm, the actual CTWD at A is [[File:Ctwd_at_a.png|180px]]. The height of one layer can be calculated as [[File:Layer_height_at_a.png|210px]]. A worked example can be found at table 1.

{| class="wikitable" name="Table 1"
|+Table 1: Worked example
|-
! CTWD
! m1
! m2
! m3
! m_mean
! c1
! c2
! c3
|-
| 15
| 20
| 22
| 25
| 22.3
| 17.3
| 15.3
| 17.7
|}

=== Gas metal arc welding ===
<div><ul>
<li style="display: inline-block;"> [[File:gmaw_welding.png|thumb|500px|none|Figure 5: GMAW welding process <ref name="welding_book"> Norrish, J. (2006). Advanced welding processes. Elsevier. </ref>]] </li>
</ul></div>

Gas metal arc welding (GMAW), or also known as metal inert gas (MIG), is the welding process used at AML3D for WAAM. Due to its high deposition rate and economic benefits, GMAW has became more popular. We will explore two variants of GMAW, which are Pulse Multi Control and Cold Metal Transfer. Both processes are developed by Fronius, an Austrian welding company. Note that it is not required to understand the welding physics to follow this wiki. The next two following sections' purpose is to highlight the high variability, high dynamic nature in welding.

[[File:fronius.png|thumb|Figure 6: Fronius logo]]

==== Pulse Multi Control ====
Pulse Multi Control (PMC) welding is Fronius' modified Pulsed GMAW. The advantage of Pulsed GMAW is the ability to control the metal droplet transfer in welding. Note that from figure 7, the metal droplet is characterised by a downward slope of the current pulse. Figure 8 shows the current signal captured from experiments conducted at AML3D (PMC is used).

<div><ul>
<li style="display: inline-block;"> [[File:Pulsed_gmaw_transfer_mode.png|thumb|500px|none|Figure 7: Relationship between pulse current and droplet transfer<ref name="welding_book" />]] </li>
<li style="display: inline-block;"> [[File:Pmc_current_in_time_domain.png|thumb|650px|none|Figure 8: Current signal captured from experiments conducted at AML3D (PMC process)]] </li>
</ul></div>
==== Cold Metal Transfer ====
Cold Metal Transfer (CMT) is a complex welding process, also developed by Fronius. By detecting a short circuit (mode C and D in figure 9), CMT can make adjustment such as retracting the welding material to cool down, and therefore create a smoother, more stable weld. The complexity of CMT can be seen in the current signal of the process (figure 10). Compare to PMC, CMT's signal has more variation during one signal period.

<div><ul>
<li style="display: inline-block;"> [[File:Transfer_mode.png|thumb|800px|none|Figure 9: Mechanism of metal droplet transfer mode]] </li>
<li style="display: inline-block;"> [[File:Cmt1_current_in_time_domain.png|thumb|735px|none|Figure 10: Current signal captured from experiments conducted at AML3D (CMT process)]] </li>
</ul></div>
=== Machine Learning ===
Machine learning (or statistical learning) refer to a set of tools to give data scientists an insight into the data and make better decisions based on those information <ref name="introduction_ml_book"> James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An introduction to statistical learning (Vol. 112, p. 18). New York: Springer. </ref>. In this project, we will explore the capabilities of classical machine learning methods (i.e anything but Deep Learning) due to the limitation of our available data. By the time of writing this wiki, we have only explored Support Vector Machine.

==== Support Vector Machine ====
The main idea of Support Vector Machine (SVM) is to draw hyperplanes that best separate multiple classes of data. The original SVM judges best decision hyperplane based on the separation margin between data classes (figure 11). However, there are multiple variants of SVM, such as hard/soft margin SVM, Nu-SVM, One class SVM (to solve anomaly detection problem). In our project, we choose to use a soft margin implementation of SVM. The benefit of soft margin SVM (as oppose to hard margin SVM) is that by allowing some misclassifications, we get a wider margin and therefore the solution becomes more "generalise" (i.e work better with a new set of dataset that is independent from the training dataset).

<div><ul>
<li style="display: inline-block;"> [[File:svm_margin.png|thumb|400px|none|Figure 11: Separation margin <ref name="bishop_book"> Bishop, C. M. (2006). Pattern recognition and machine learning. Springer. </ref>]] </li>
</ul></div>

SVM algorithm makes use of [https://en.wikipedia.org/wiki/Lagrange_multiplier Lagrange multiplier]. Let's the linear model of the dataset be:

[[File:Linear_model.png|150px]]

where [[File:Phi_math.png|10px]] denotes the feature space transformation, '''w''' denotes the weights and ''b'' denotes bias constant.

Let [[File:vector_x_n.png|17px]] be the input vector (where each index in the vector denotes a feature), and [[File:target_n.png|15px]] be the target (the class of input vector). It is showed that in order to maximise the (soft) margin, we will need to minimise the following:

[[File:Soft margin minimise equation.png|150px]]

with subject to:

[[File:Soft margin minimise condition.png|150px]]

where N is the total number of training data, C is the regularisation parameter to control the effect of slack variable [[File:slack_variable.png|17px]]. A more visual presentation of slack variables can be found in figure 12.

<div><ul>
<li style="display: inline-block;"> [[File:svm_slack_variable.png|thumb|400px|none|Figure 12: Slack variable <ref name="bishop_book"></ref>]] </li>
</ul></div>

If we were to apply the Lagrangian method to solve the above minimisation problem, we can find the decision hyperplane for the SVM algorithm. However, we shall not discuss the solution here as it is not the main focus of the project.

==== Kernel tricks ====
Both figure 11 and 12 are of linearly separable cases for SVM. For dataset that is not easy to separate with a hyperplane in their original features space (such as figure 13), the data often get projected into a higher dimension space where it is easier to separate (figure 14). Such technique is called the Kernel Tricks. It is important to note that Kernel Tricks does not actually "map" the data from '''X''' dimension to '''Z''' dimension, but rather pairwise compare the similarity of the input space in '''Z''' dimension. However, we shall not dive too deep into this as it is not the main focus of the project.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_nonseparable.png |thumb|350px|none|Figure 13: Nonseparable dataset <ref name="bishop_book"></ref>]] </li>
<li style="display: inline-block;"> [[File:Svm_kernel_trick.png |thumb|650px|none|Figure 14: Kernel tricks]] </li>
</ul></div>

== Previous studies ==
There are many researches into defect detection in GMAW process. The three most dominant analysis methods are spectroscopic analysis, acoustic analysis, and electrical signal analysis.

=== Spectroscopic analysis ===
In <ref name="spectroscopic_analysis">D. Bebiano and S. Alfaro, “A weld defects detection system based on a spectrometer,” Sensors, vol. 9, no. 4, pp. 2851–2861, 2009.</ref>, the authors captured the intensity of radiation emission of electric arc. It is obvious from figure 15 that good weld's intensity stays stable while defect one's is characterised by a sudden peak. Therefore, a threshold is set to detect the defect welds.

<div><ul>
<li style="display: inline-block;"> [[File:spectroscopic_analysis.png|thumb|400px|none|Figure 15: Spectroscopic analysis <ref name="spectroscopic_analysis"></ref>]] </li>
</ul></div>

Other research into using spectroscopic data to monitor that can be of interest is <ref>D. Naso, B. Turchiano, and P. Pantaleo, “A fuzzy-logic based optical sensor for online weld defect-detection,” IEEE transactions on Industrial Informatics, vol. 1, no. 4, pp. 259–273, 2005.</ref>.

=== Acoustic analysis ===
With <ref name="acoustic_analysis">E. Cayo and S. C. Alfaro, “A non-intrusive GMA welding process quality monitoring system using acoustic sensing,” Sensors, vol. 9, no. 9, pp. 7150–7166, 2009.</ref>, the author discovered that arc ignition is characterised by a distinct sound and therefore it is possible to capture the frequency of ignition through acoustic sensors. A tolerance band was set and both ignition frequency and sound pressure level were used to monitor the quality of the weld.

<div><ul>
<li style="display: inline-block;"> [[File:acoustic_analysis.png|thumb|400px|none|Figure 16: Acoustic analysis <ref name="acoustic_analysis"></ref>]] </li>
</ul></div>

Other research into using acoustic signal to monitor that can be of interest are <ref>M. Fidali, “Detection of welding process instabilities using acoustic signals,” in International Congress on Technical Diagnostic, pp. 191–201, Springer, 2016.</ref>, <ref>L. Zhang, A. C. Basantes-Defaz, D. Ozevin, and E. Indacochea, “Real-time monitoring of welding process using air-coupled ultrasonics and acoustic emission,” The International Journal of Advanced Manufacturing Technology, vol. 101, no. 5-8, pp. 1623–1634, 2019.</ref>.

=== Electrical signal analysis ===
With <ref name="electrical_analysis_1">A. Sumesh, K. Rameshkumar, A. Raja, K. Mohandas, A. Santhakumari, and R. Shyambabu, “Establishing correlation between current and voltage signatures of the arc and weld defects in GMAW process,” Arabian Journal for Science and Engineering, vol. 42, no. 11, pp. 4649–4665, 2017.</ref>, the author plot the probability density distribution of voltage signal and discovered that the stable weld's signal concentrate in one region while unstable signal spread out. While in the paper, the authors did not developed a monitoring method using their findings, it shows that there is a significant differences between stable and unstable weld's electrical signature.

<div><ul>
<li style="display: inline-block;"> [[File:Electrical analysis pdd.png|thumb|400px|none|Figure 17: Voltage PDD <ref name="electrical_analysis_1"></ref>]] </li>
</ul></div>

In <ref name="electrical_analysis_2">S. Simpson, “Signature images for arc welding fault detection,” Science and Technology of Welding and Joining, vol. 12, no. 6, pp. 481–486, 2007.</ref>, the authors shows that by plotting the signature image of voltage and current scatter plot, both arcing and short-circuiting region's signature image can notify a defects welds. A more visual representation of short circuiting and arcing region can be found in [[#Gas metal arc welding|Gas metal arc welding section]].

<div><ul>
<li style="display: inline-block;"> [[File:Electrical voltage signature.png|thumb|400px|none|Figure 18a: Electrical voltage signature of stable weld <ref name="electrical_analysis_2"></ref>]] </li>
<li style="display: inline-block;"> [[File:Electrical signature 2.png|thumb|400px|none|Figure 18b: Electrical voltage signature of unstable weld <ref name="electrical_analysis_2"></ref>]] </li>
</ul></div>

Other research into using electrical signal to monitor that can be of interest are <ref>R. Madigan, “Arc sensing for defects in constant-voltage gas metal arc welding,” Welding Journal, vol. 78, pp. 322S–328S, 1999.</ref>, <ref>Y. Huang, K. Wang, Z. Zhou, X. Zhou, and J. Fang, “Stability evaluation of short-circuiting gas metal arc welding based on ensemble empirical mode decomposition,” Measurement Science and Technology, vol. 28, no. 3, p. 035006, 2017.</ref>,<ref>Z. Zhang, X. Chen, H. Chen, J. Zhong, and S. Chen, “Online welding quality monitoring based on feature extraction of arc voltage signal,” The International Journal of Advanced Manufacturing Technology, vol. 70, no. 9-12, pp. 1661–1671, 2014</ref>,<ref>X. Li and S. Simpson, “Parametric approach to positional fault detection in short arc welding,” Science and Technology of welding and joining, vol. 14, no. 2, pp. 146–151, 2009.</ref>, and <ref>E. Wei, D. Farson, R. Richardson, and H. Ludewig, “Detection of weld surface porosity by statistical analysis of arc current in gas metal arc welding,” Journal of Manufacturing Processes, vol. 3, no. 1, pp. 50–59, 2001.</ref>.

== Experiment setup ==
The welding system uses TPS 320i C Pulse (figure 15a), produced by Fronius, to supply power, feed wire, and system cooling. For movement control of the welding torch, the system use ABB’s IRB 1520ID (figure 15b), which is a dedicated arc welding robot. The Data Acquisition device is NI's DAQ. Each channel (voltage and current) is sampled at 25kHz while the dominant frequency for voltage and current signal are around 130Hz (for CMT process) and 190Hz (for PMC process).

<div><ul>
<li style="display: inline-block;"> [[File:Fronius_power_supply.jpg |thumb|250px|none|Figure 19a: Fronius' Power Supply]] </li>
<li style="display: inline-block;"> [[File:Welding_robot.jpg |thumb|250px|none|Figure 19b: ABB's welding robot]] </li>
</ul></div>

[[File:Ni_daq.jpg|thumb|Figure 16: NI's DAQ device]]

== Statistical Analysis ==

=== Preprocessing ===
As the captured signal coming from Fronius' power supply, the signals also pick up switching power noise. The noise will interfere with the accuracy of the analysis if not filtered. In this project, we applied a 1000Hz digital Low Pass Butterworth filter. The effect of the filter can be seen in figure 20, while the comparation of using different filters can be seen in the gallery.

<div><ul>
<li style="display: inline-block;"> [[File:Filter_effect.png |thumb|750px|none|Figure 20: Effect of 1000Hz low pass filter]] </li>
<li style="display: inline-block;"> [[File:Freq_response.png |thumb|750px|none|Figure 21: Frequency response of the in-use filter]] </li>
</ul></div>

<gallery mode="packed-hover" caption="Gallery: Comparison of different filtering method" perrow=2>
File:Effect_of_filtering2.png|In use filter vs 300Hz low pass filter
File:Effect_of_filtering3.png|In use filter vs 500Hz low pass filter
File:Effect_of_filtering4.png|In use filter vs 750Hz low pass filter
File:Effect_of_filtering5.png|In use filter vs 2000Hz low pass filter
</gallery>

=== Power Spectral Density ===
==== Prerequisite knowledge ====
Looking at the spectrum is another way to analyse the fundamental information of the signal, in frequency domain <ref name="signal_processing_book">Prandoni, P., & Vetterli, M. (2008). Signal processing for communications. EPFL press.</ref>. Note that the term "Power" in Power Spectral Density does not refer to the convention power definition. In signal processing, energy of a discrete time signal is defined as follow:

[[File:Signal_energy_formular.png|150px]]

The way to interpret this definition is if we consider the signal to be of "Voltage" unit, ''E'' is the total energy (in joules) dissipated over a 1 Ohm resistor <ref name="signal_processing_book"></ref>. The power spectral density of a signal over time interval [-M,M] is defined as follow:

[[File:Psd_formular.png|300px]]

where [[File:Fourier series symbol.png|67px]] is the truncated Fourier transform:

[[File:Fourier_series_formular.png|250px]]

==== Analysis ====
Our goal for this project is to detect the defect welds and predict the layer heights (the contact tip to work piece distance, CTWD, to be more accurate) in real time (refer to [[#Relating Contact Tip to Work-piece Distance to layer height|this section]] to understand the relationship between Contact Tip to Work-piece Distance and Layer height). An analysis of the current power spectral density (PSD) shows that there is an inverse proportional relationship between the CTWD of a weld and the frequency where its '''current signal''' PSD reach the peak, and also the magnitude of the peak. Since the peak of PSD can be fairly sensitive, we only consider the frequency of the peak PSD.

Figure 22 shows the PSD plot of voltage, current, and power signal of the weld using PMC process. In this experiment, we created multiple one-line, 8 seconds welds. These eight-second welds then get segmented into smaller two-second chunk to increase the number of data and decrease the processing time.

<div><ul>
<li style="display: inline-block;"> [[File:Psd pmc good bad all lowfreq.png |thumb|750px|none|Figure 22: Power spectral density of PMC process]] </li>
<li style="display: inline-block;"> [[File:One line weld.jpg |thumb|578px|none|Figure 23: Simple one-line weld experiment]] </li>
<li style="display: inline-block;"> [[File:Cmt_psd_stable_allCTWD_zoomed.png |thumb|750px|none|Figure 24: Simple one-line weld experiment]] </li>

</ul></div>

It seems like the inverse proportional relationship holds true for most cases. Other experiments showed that for a short period at the beginning and ending of the weld, the signal can behave differently relative to the rest of the weld. Currently, to migrate with this problem, data will not be collected and analysed (both offline and online) at these periods. However, the above results seem to only hold true for PMC processes. With CMT process, the current PSD plot indicates no relationship between the CTWD and the frequency at peak (figure 24).

For the next stage of the project, our plan is to develop a linear regression model to predict the CTWD for PMC process. With CMT process, we will have to look into different analysis techniques.

=== Support Vector Machine ===
==== Prerequisite knowledge ====

==== Analysis ====

=== Features extraction from grayscale image ===
==== Prerequisite knowledge ====

==== Analysis ====

== Result discussion ==

== Conclusion ==

== Future work ==

== References ==
<references />

Projects:2019s2-20001 Using Machine Learning to Determine Deposit Height and Defects for Wire + Arc Additive Manufacture (3D printing)

2019-10-07T09:56:41Z

A1713568: /* Support Vector Machine */

[[Category:Projects]]
[[Category:Final Year Projects]]
[[Category:2019s2|24501]]
== Abstract ==
Additive manufacturing is an emerging technology that is complementary to subtracting manufacturing. AM is able to create complex model at the cost of production time. An example of AM is Wire + Arc Additive Manufacture. After slicing the manufacturing process of one model to multiple layers, WAAM uses electric arc as a source of heat to feed metal wire and create each layer. However, a slight height inaccuracy in this process will result in an expensive faulty, as error can be multiplied across many layers. Thus, in-process height measurement and faulty detection are critical to the production process. In this thesis, we will look into a machine learning approach to develop an on-the-line algorithm that automatically detects faulty processes and make height measurement by analysing electrical signal (current and voltage) of a Gas Metal Arc Welding machine. This project will introduce the
approaches to tackle the same problem by other researchers, describe the design of our experiment, development of our machine learning model and discuss the final result.

'''Keywords''': Additive manufacturing, Gas metal arc welding, Metal inert gas, Machine learning, Fault detection, Signal analysis
== Introduction ==
3D printing is an emerging technology that has the potential to significantly reduce material usage through the production of near net-shape parts. Many of the systems for 3D printing are based on lasers and powders; however the deposition rate with such systems is very low making the production of large-scale parts difficult. AML Technologies specialises in the use of Wire + Arc Additive Manufacture (WAAM) where deposition is based on arc welding processes and the deposition rates are an order of magnitude greater. When building 3D printed parts, even a relatively small layer height error of only 0.1 mm can produce large build height errors when multiplied across the many layers of a typical build. This can make path planning difficult, so in-process layer height measurement is an essential building block of any production 3D printing system. A variety of techniques can be used for monitoring the layer height including laser scanners, and arc monitoring. It is the latter technique that will be explored in this project due to its robustness, and low cost of implementation – it only requires the measurement of arc current and voltage. Furthermore, it can potentially be used to detect defects by identifying waveform irregularities.
=== Project team ===
==== Project students ====
* Anh Tran
* Nhat Nguyen
==== Supervisors ====
* Dr. Brian Ng
* Dr. Paul Colegrove (AML3D, sponsored company)

[[File:aml3d_logo.jpg|100px]]
=== Objectives ===
The objective of this project is to increase the efficiency of the manufacture process at AML3D. In order to do so, the team will investigate into the possibility of automating and optimising the quality control processes. The two quality control processes that are currently being implemented at AML3D are measuring layer height using laser sensors, and human supervision for detecting defects. These processes add overhead into production time and usage of human resource, which is not desired. To achieve the goal, it is expected that machine learning methods will be used extensively to analyse the electrical signatures of the weld process.

== Background ==
=== Wire Arc Additive Manufacturing ===
Wire and Arc Additive Manufacturing (WAAM) is a type of additive manufacturing that uses electric arc as the heat source and material wire to feed the manufacture process <ref name="WAAM"> S. W. Williams, F. Martina, A. C. Addison, J. Ding, G. Pardal & P. Colegrove (2016) Wire + Arc Additive Manufacturing, Materials Science and Technology, 32:7, 641-647, DOI: 10.1179/1743284715Y.0000000073 </ref>. WAAM has been investigated since the 1990s <ref name="WAAM" />, but only recently that it received more attention from the manufacture world.

Its significant comes from the ability to manufacture complex model with less time and less material. Figure 1a and 1b show real parts that was manufactured by AML3D. Such custom made parts might takes months before be ready to be shipped, but with WAAM, the production time can reduce down to weeks. Currently, the industries that benefit the most from WAAM are maritime and aerospace.

Similar to other additive manufacture methods, WAAM achieves such results by sliding the models into multiple layers, and then build the model layer by layer. The movement control is normally handled by a robotic arm (AML3D uses ABB's Arc Welder robot), and the welding path is generated by a Computer Aid Manufacture (CAM) software.

[[File:welding_part1.JPG|thumb|Figure 1a: A propeller]]
[[File:welding_part2.jpg|thumb|Figure 1b: Another part of the ship]]
[[File:abb_robot.jpg|thumb|Figure 2: ABB's IRB 1520ID Arc Welder]]

Currently at AML3D, the manufacture process for a part can take anywhere from days to weeks. However, the contact tip (where the welding gun deposits the wire) needs replacement every couple of hours. A worn contact tip such as figure 3 will lead to unstable weld and the this effect will propagate and accumulate through multiple layers. Such defects in the manufacture process will no doubt be financially taxing for AML3D. Apart from constant human supervision, another quality control process implemented at AML3D is height measurement of each layer, after the layer is built. This process is implemented to ensure the height of each layer is in the acceptable range (1-2mm). As mentioned before, these quality control processes are costing AML3D in term of time and finance. Our project will focus on replace them with a less time consuming, more automating process.

<div><ul>
<li style="display: inline-block;"> [[File:contact_tip.jpg|thumb|none|400px|Figure 3a: A worn contact tip]] </li>
<li style="display: inline-block;"> [[File:layers.jpeg|thumb|none|530px|Figure 3b: Close up photo of layers]] </li>
</ul></div>

==== Relating Contact Tip to Work-piece Distance to layer height ====
At AML3D, the sensor captures the Contact Tip to Work-piece Distance (CTWD), and from there, we can deduce the layer height. A more detail explanation is of follow:

<div><ul>
<li style="display: inline-block;"> [[File:Example_of_layer_height_measurement.jpg|thumb|none|400px|Figure 4: From CTWD to layer height]] </li>
</ul></div>

The CTWD can be set at the beginning of the weld, however, as the welding process running, the actual CTWD fluctuate around the set CTWD. For example, after finish constructing one layer, the laser sensor capture the model height at 3 different position: A, B and C. Let the 3 model height be m1, m2 and m3. Calculate n_mean and for argument sake, if the set CTWD is 15 mm, the actual CTWD at A is [[File:Ctwd_at_a.png|180px]]. The height of one layer can be calculated as [[File:Layer_height_at_a.png|210px]]. A worked example can be found at table 1.

{| class="wikitable" name="Table 1"
|+Table 1: Worked example
|-
! CTWD
! m1
! m2
! m3
! m_mean
! c1
! c2
! c3
|-
| 15
| 20
| 22
| 25
| 22.3
| 17.3
| 15.3
| 17.7
|}

=== Gas metal arc welding ===
<div><ul>
<li style="display: inline-block;"> [[File:gmaw_welding.png|thumb|500px|none|Figure 5: GMAW welding process <ref name="welding_book"> Norrish, J. (2006). Advanced welding processes. Elsevier. </ref>]] </li>
</ul></div>

Gas metal arc welding (GMAW), or also known as metal inert gas (MIG), is the welding process used at AML3D for WAAM. Due to its high deposition rate and economic benefits, GMAW has became more popular. We will explore two variants of GMAW, which are Pulse Multi Control and Cold Metal Transfer. Both processes are developed by Fronius, an Austrian welding company. Note that it is not required to understand the welding physics to follow this wiki. The next two following sections' purpose is to highlight the high variability, high dynamic nature in welding.

[[File:fronius.png|thumb|Figure 6: Fronius logo]]

==== Pulse Multi Control ====
Pulse Multi Control (PMC) welding is Fronius' modified Pulsed GMAW. The advantage of Pulsed GMAW is the ability to control the metal droplet transfer in welding. Note that from figure 7, the metal droplet is characterised by a downward slope of the current pulse. Figure 8 shows the current signal captured from experiments conducted at AML3D (PMC is used).

<div><ul>
<li style="display: inline-block;"> [[File:Pulsed_gmaw_transfer_mode.png|thumb|500px|none|Figure 7: Relationship between pulse current and droplet transfer<ref name="welding_book" />]] </li>
<li style="display: inline-block;"> [[File:Pmc_current_in_time_domain.png|thumb|650px|none|Figure 8: Current signal captured from experiments conducted at AML3D (PMC process)]] </li>
</ul></div>
==== Cold Metal Transfer ====
Cold Metal Transfer (CMT) is a complex welding process, also developed by Fronius. By detecting a short circuit (mode C and D in figure 9), CMT can make adjustment such as retracting the welding material to cool down, and therefore create a smoother, more stable weld. The complexity of CMT can be seen in the current signal of the process (figure 10). Compare to PMC, CMT's signal has more variation during one signal period.

<div><ul>
<li style="display: inline-block;"> [[File:Transfer_mode.png|thumb|800px|none|Figure 9: Mechanism of metal droplet transfer mode]] </li>
<li style="display: inline-block;"> [[File:Cmt1_current_in_time_domain.png|thumb|735px|none|Figure 10: Current signal captured from experiments conducted at AML3D (CMT process)]] </li>
</ul></div>
=== Machine Learning ===
Machine learning (or statistical learning) refer to a set of tools to give data scientists an insight into the data and make better decisions based on those information <ref name="introduction_ml_book"> James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An introduction to statistical learning (Vol. 112, p. 18). New York: Springer. </ref>. In this project, we will explore the capabilities of classical machine learning methods (i.e anything but Deep Learning) due to the limitation of our available data. By the time of writing this wiki, we have only explored Support Vector Machine.

==== Support Vector Machine ====
The main idea of Support Vector Machine (SVM) is to draw hyperplanes that best separate multiple classes of data. The original SVM judges best decision hyperplane based on the separation margin between data classes (figure 11). However, there are multiple variants of SVM, such as hard/soft margin SVM, Nu-SVM, One class SVM (to solve anomaly detection problem). In our project, we choose to use a soft margin implementation of SVM. The benefit of soft margin SVM (as oppose to hard margin SVM) is that by allowing some misclassifications, we get a wider margin and therefore the solution becomes more "generalise" (i.e work better with a new set of dataset that is independent from the training dataset).

<div><ul>
<li style="display: inline-block;"> [[File:svm_margin.png|thumb|400px|none|Figure 11: Separation margin <ref name="bishop_book"> Bishop, C. M. (2006). Pattern recognition and machine learning. Springer. </ref>]] </li>
</ul></div>

SVM algorithm makes use of [https://en.wikipedia.org/wiki/Lagrange_multiplier Lagrange multiplier]. Let's the linear model of the dataset be:

[[File:Linear_model.png|150px]]

where [[File:Phi_math.png|10px]] denotes the feature space transformation, '''w''' denotes the weights and ''b'' denotes bias constant.

Let [[File:vector_x_n.png|17px]] be the input vector (where each index in the vector denotes a feature), and [[File:target_n.png|15px]] be the target (the class of input vector). It is showed that in order to maximise the (soft) margin, we will need to minimise the following:

[[File:Soft margin minimise equation.png|150px]]

with subject to:

[[File:Soft margin minimise condition.png|150px]]

where N is the total number of training data, C is the regularisation parameter to control the effect of slack variable [[File:slack_variable.png|17px]]. A more visual presentation of slack variables can be found in figure 12.

<div><ul>
<li style="display: inline-block;"> [[File:svm_slack_variable.png|thumb|400px|none|Figure 12: Slack variable <ref name="bishop_book"></ref>]] </li>
</ul></div>

If we were to apply the Lagrangian method to solve the above minimisation problem, we can find the decision hyperplane for the SVM algorithm. However, we shall not discuss the solution here as it is not the main focus of the project.

==== Kernel tricks ====
Both figure 11 and 12 are of linearly separable cases for SVM. For dataset that is not easy to separate with a hyperplane in their original features space (such as figure 13), the data often get projected into a higher dimension space where it is easier to separate (figure 14). Such technique is called the Kernel Tricks. It is important to note that Kernel Tricks does not actually "map" the data from '''X''' dimension to '''Z''' dimension, but rather pairwise compare the similarity of the input space in '''Z''' dimension. However, we shall not dive too deep into this as it is not the main focus of the project.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_nonseparable.png |thumb|350px|none|Figure 13: Nonseparable dataset <ref name="bishop_book"></ref>]] </li>
<li style="display: inline-block;"> [[File:Svm_kernel_trick.png |thumb|650px|none|Figure 14: Kernel tricks]] </li>
</ul></div>

== Previous studies ==
There are many researches into defect detection in GMAW process. The three most dominant analysis methods are spectroscopic analysis, acoustic analysis, and electrical signal analysis.

=== Spectroscopic analysis ===
In <ref name="spectroscopic_analysis">D. Bebiano and S. Alfaro, “A weld defects detection system based on a spectrometer,” Sensors, vol. 9, no. 4, pp. 2851–2861, 2009.</ref>, the authors captured the intensity of radiation emission of electric arc. It is obvious from figure 15 that good weld's intensity stays stable while defect one's is characterised by a sudden peak. Therefore, a threshold is set to detect the defect welds.

<div><ul>
<li style="display: inline-block;"> [[File:spectroscopic_analysis.png|thumb|400px|none|Figure 15: Spectroscopic analysis <ref name="spectroscopic_analysis"></ref>]] </li>
</ul></div>

Other research into using spectroscopic data to monitor that can be of interest is <ref>D. Naso, B. Turchiano, and P. Pantaleo, “A fuzzy-logic based optical sensor for online weld defect-detection,” IEEE transactions on Industrial Informatics, vol. 1, no. 4, pp. 259–273, 2005.</ref>.

=== Acoustic analysis ===
With <ref name="acoustic_analysis">E. Cayo and S. C. Alfaro, “A non-intrusive GMA welding process quality monitoring system using acoustic sensing,” Sensors, vol. 9, no. 9, pp. 7150–7166, 2009.</ref>, the author discovered that arc ignition is characterised by a distinct sound and therefore it is possible to capture the frequency of ignition through acoustic sensors. A tolerance band was set and both ignition frequency and sound pressure level were used to monitor the quality of the weld.

<div><ul>
<li style="display: inline-block;"> [[File:acoustic_analysis.png|thumb|400px|none|Figure 16: Acoustic analysis <ref name="acoustic_analysis"></ref>]] </li>
</ul></div>

Other research into using acoustic signal to monitor that can be of interest are <ref>M. Fidali, “Detection of welding process instabilities using acoustic signals,” in International Congress on Technical Diagnostic, pp. 191–201, Springer, 2016.</ref>, <ref>L. Zhang, A. C. Basantes-Defaz, D. Ozevin, and E. Indacochea, “Real-time monitoring of welding process using air-coupled ultrasonics and acoustic emission,” The International Journal of Advanced Manufacturing Technology, vol. 101, no. 5-8, pp. 1623–1634, 2019.</ref>.

=== Electrical signal analysis ===
With <ref name="electrical_analysis_1">A. Sumesh, K. Rameshkumar, A. Raja, K. Mohandas, A. Santhakumari, and R. Shyambabu, “Establishing correlation between current and voltage signatures of the arc and weld defects in GMAW process,” Arabian Journal for Science and Engineering, vol. 42, no. 11, pp. 4649–4665, 2017.</ref>, the author plot the probability density distribution of voltage signal and discovered that the stable weld's signal concentrate in one region while unstable signal spread out. While in the paper, the authors did not developed a monitoring method using their findings, it shows that there is a significant differences between stable and unstable weld's electrical signature.

<div><ul>
<li style="display: inline-block;"> [[File:Electrical analysis pdd.png|thumb|400px|none|Figure 17: Voltage PDD <ref name="electrical_analysis_1"></ref>]] </li>
</ul></div>

In <ref name="electrical_analysis_2">S. Simpson, “Signature images for arc welding fault detection,” Science and Technology of Welding and Joining, vol. 12, no. 6, pp. 481–486, 2007.</ref>, the authors shows that by plotting the signature image of voltage and current scatter plot, both arcing and short-circuiting region's signature image can notify a defects welds. A more visual representation of short circuiting and arcing region can be found in [[#Gas metal arc welding|Gas metal arc welding section]].

<div><ul>
<li style="display: inline-block;"> [[File:Electrical voltage signature.png|thumb|400px|none|Figure 18a: Electrical voltage signature of stable weld <ref name="electrical_analysis_2"></ref>]] </li>
<li style="display: inline-block;"> [[File:Electrical signature 2.png|thumb|400px|none|Figure 18b: Electrical voltage signature of unstable weld <ref name="electrical_analysis_2"></ref>]] </li>
</ul></div>

Other research into using electrical signal to monitor that can be of interest are <ref>R. Madigan, “Arc sensing for defects in constant-voltage gas metal arc welding,” Welding Journal, vol. 78, pp. 322S–328S, 1999.</ref>, <ref>Y. Huang, K. Wang, Z. Zhou, X. Zhou, and J. Fang, “Stability evaluation of short-circuiting gas metal arc welding based on ensemble empirical mode decomposition,” Measurement Science and Technology, vol. 28, no. 3, p. 035006, 2017.</ref>,<ref>Z. Zhang, X. Chen, H. Chen, J. Zhong, and S. Chen, “Online welding quality monitoring based on feature extraction of arc voltage signal,” The International Journal of Advanced Manufacturing Technology, vol. 70, no. 9-12, pp. 1661–1671, 2014</ref>,<ref>X. Li and S. Simpson, “Parametric approach to positional fault detection in short arc welding,” Science and Technology of welding and joining, vol. 14, no. 2, pp. 146–151, 2009.</ref>, and <ref>E. Wei, D. Farson, R. Richardson, and H. Ludewig, “Detection of weld surface porosity by statistical analysis of arc current in gas metal arc welding,” Journal of Manufacturing Processes, vol. 3, no. 1, pp. 50–59, 2001.</ref>.

== Experiment setup ==
The welding system uses TPS 320i C Pulse (figure 15a), produced by Fronius, to supply power, feed wire, and system cooling. For movement control of the welding torch, the system use ABB’s IRB 1520ID (figure 15b), which is a dedicated arc welding robot. The Data Acquisition device is NI's DAQ. Each channel (voltage and current) is sampled at 25kHz while the dominant frequency for voltage and current signal are around 130Hz (for CMT process) and 190Hz (for PMC process).

<div><ul>
<li style="display: inline-block;"> [[File:Fronius_power_supply.jpg |thumb|250px|none|Figure 19a: Fronius' Power Supply]] </li>
<li style="display: inline-block;"> [[File:Welding_robot.jpg |thumb|250px|none|Figure 19b: ABB's welding robot]] </li>
</ul></div>

[[File:Ni_daq.jpg|thumb|Figure 16: NI's DAQ device]]

== Statistical Analysis ==

=== Preprocessing ===
As the captured signal coming from Fronius' power supply, the signals also pick up switching power noise. The noise will interfere with the accuracy of the analysis if not filtered. In this project, we applied a 1000Hz digital Low Pass Butterworth filter. The effect of the filter can be seen in figure 20, while the comparation of using different filters can be seen in the gallery.

<div><ul>
<li style="display: inline-block;"> [[File:Filter_effect.png |thumb|750px|none|Figure 20: Effect of 1000Hz low pass filter]] </li>
<li style="display: inline-block;"> [[File:Freq_response.png |thumb|750px|none|Figure 21: Frequency response of the in-use filter]] </li>
</ul></div>

<gallery mode="packed-hover" caption="Gallery: Comparison of different filtering method" perrow=2>
File:Effect_of_filtering2.png|In use filter vs 300Hz low pass filter
File:Effect_of_filtering3.png|In use filter vs 500Hz low pass filter
File:Effect_of_filtering4.png|In use filter vs 750Hz low pass filter
File:Effect_of_filtering5.png|In use filter vs 2000Hz low pass filter
</gallery>

=== Power Spectral Density ===
==== Prerequisite knowledge ====
Looking at the spectrum is another way to analyse the fundamental information of the signal, in frequency domain <ref name="signal_processing_book">Prandoni, P., & Vetterli, M. (2008). Signal processing for communications. EPFL press.</ref>. Note that the term "Power" in Power Spectral Density does not refer to the convention power definition. In signal processing, energy of a discrete time signal is defined as follow:

[[File:Signal_energy_formular.png|150px]]

The way to interpret this definition is if we consider the signal to be of "Voltage" unit, ''E'' is the total energy (in joules) dissipated over a 1 Ohm resistor <ref name="signal_processing_book"></ref>. The power spectral density of a signal over time interval [-M,M] is defined as follow:

[[File:Psd_formular.png|300px]]

where [[File:Fourier series symbol.png|67px]] is the truncated Fourier transform:

[[File:Fourier_series_formular.png|250px]]

==== Analysis ====
Our goal for this project is to detect the defect welds and predict the layer heights (the contact tip to work piece distance, CTWD, to be more accurate) in real time (refer to [[#Relating Contact Tip to Work-piece Distance to layer height|this section]] to understand the relationship between Contact Tip to Work-piece Distance and Layer height). An analysis of the current power spectral density (PSD) shows that there is an inverse proportional relationship between the CTWD of a weld and the frequency where its '''current signal''' PSD reach the peak, and also the magnitude of the peak. Since the peak of PSD can be fairly sensitive, we only consider the frequency of the peak PSD.

Figure 22 shows the PSD plot of voltage, current, and power signal of the weld using PMC process. In this experiment, we created multiple one-line, 8 seconds welds. These eight-second welds then get segmented into smaller two-second chunk to increase the number of data and decrease the processing time.

<div><ul>
<li style="display: inline-block;"> [[File:Psd pmc good bad all lowfreq.png |thumb|750px|none|Figure 22: Power spectral density of PMC process]] </li>
<li style="display: inline-block;"> [[File:One line weld.jpg |thumb|578px|none|Figure 23: Simple one-line weld experiment]] </li>
<li style="display: inline-block;"> [[File:Cmt_psd_stable_allCTWD_zoomed.png |thumb|750px|none|Figure 24: Simple one-line weld experiment]] </li>

</ul></div>

It seems like the inverse proportional relationship holds true for most cases. Other experiments showed that for a short period at the beginning and ending of the weld, the signal can behave differently relative to the rest of the weld. Currently, to migrate with this problem, data will not be collected and analysed (both offline and online) at these periods. However, the above results seem to only hold true for PMC processes. With CMT process, the current PSD plot indicates no relationship between the CTWD and the frequency at peak (figure 24).

For the next stage of the project, our plan is to develop a linear regression model to predict the CTWD for PMC process. With CMT process, we will have to look into different analysis techniques.

=== Support Vector Machine ===
==== Prerequisite knowledge ====

==== Analysis ====

=== Features extraction from grayscale image ===

== Result discussion ==

== Conclusion ==

== Future work ==

== References ==
<references />

Projects:2019s2-20001 Using Machine Learning to Determine Deposit Height and Defects for Wire + Arc Additive Manufacture (3D printing)

2019-10-07T09:55:13Z

A1713568: /* Analysis */

[[Category:Projects]]
[[Category:Final Year Projects]]
[[Category:2019s2|24501]]
== Abstract ==
Additive manufacturing is an emerging technology that is complementary to subtracting manufacturing. AM is able to create complex model at the cost of production time. An example of AM is Wire + Arc Additive Manufacture. After slicing the manufacturing process of one model to multiple layers, WAAM uses electric arc as a source of heat to feed metal wire and create each layer. However, a slight height inaccuracy in this process will result in an expensive faulty, as error can be multiplied across many layers. Thus, in-process height measurement and faulty detection are critical to the production process. In this thesis, we will look into a machine learning approach to develop an on-the-line algorithm that automatically detects faulty processes and make height measurement by analysing electrical signal (current and voltage) of a Gas Metal Arc Welding machine. This project will introduce the
approaches to tackle the same problem by other researchers, describe the design of our experiment, development of our machine learning model and discuss the final result.

'''Keywords''': Additive manufacturing, Gas metal arc welding, Metal inert gas, Machine learning, Fault detection, Signal analysis
== Introduction ==
3D printing is an emerging technology that has the potential to significantly reduce material usage through the production of near net-shape parts. Many of the systems for 3D printing are based on lasers and powders; however the deposition rate with such systems is very low making the production of large-scale parts difficult. AML Technologies specialises in the use of Wire + Arc Additive Manufacture (WAAM) where deposition is based on arc welding processes and the deposition rates are an order of magnitude greater. When building 3D printed parts, even a relatively small layer height error of only 0.1 mm can produce large build height errors when multiplied across the many layers of a typical build. This can make path planning difficult, so in-process layer height measurement is an essential building block of any production 3D printing system. A variety of techniques can be used for monitoring the layer height including laser scanners, and arc monitoring. It is the latter technique that will be explored in this project due to its robustness, and low cost of implementation – it only requires the measurement of arc current and voltage. Furthermore, it can potentially be used to detect defects by identifying waveform irregularities.
=== Project team ===
==== Project students ====
* Anh Tran
* Nhat Nguyen
==== Supervisors ====
* Dr. Brian Ng
* Dr. Paul Colegrove (AML3D, sponsored company)

[[File:aml3d_logo.jpg|100px]]
=== Objectives ===
The objective of this project is to increase the efficiency of the manufacture process at AML3D. In order to do so, the team will investigate into the possibility of automating and optimising the quality control processes. The two quality control processes that are currently being implemented at AML3D are measuring layer height using laser sensors, and human supervision for detecting defects. These processes add overhead into production time and usage of human resource, which is not desired. To achieve the goal, it is expected that machine learning methods will be used extensively to analyse the electrical signatures of the weld process.

== Background ==
=== Wire Arc Additive Manufacturing ===
Wire and Arc Additive Manufacturing (WAAM) is a type of additive manufacturing that uses electric arc as the heat source and material wire to feed the manufacture process <ref name="WAAM"> S. W. Williams, F. Martina, A. C. Addison, J. Ding, G. Pardal & P. Colegrove (2016) Wire + Arc Additive Manufacturing, Materials Science and Technology, 32:7, 641-647, DOI: 10.1179/1743284715Y.0000000073 </ref>. WAAM has been investigated since the 1990s <ref name="WAAM" />, but only recently that it received more attention from the manufacture world.

Its significant comes from the ability to manufacture complex model with less time and less material. Figure 1a and 1b show real parts that was manufactured by AML3D. Such custom made parts might takes months before be ready to be shipped, but with WAAM, the production time can reduce down to weeks. Currently, the industries that benefit the most from WAAM are maritime and aerospace.

Similar to other additive manufacture methods, WAAM achieves such results by sliding the models into multiple layers, and then build the model layer by layer. The movement control is normally handled by a robotic arm (AML3D uses ABB's Arc Welder robot), and the welding path is generated by a Computer Aid Manufacture (CAM) software.

[[File:welding_part1.JPG|thumb|Figure 1a: A propeller]]
[[File:welding_part2.jpg|thumb|Figure 1b: Another part of the ship]]
[[File:abb_robot.jpg|thumb|Figure 2: ABB's IRB 1520ID Arc Welder]]

Currently at AML3D, the manufacture process for a part can take anywhere from days to weeks. However, the contact tip (where the welding gun deposits the wire) needs replacement every couple of hours. A worn contact tip such as figure 3 will lead to unstable weld and the this effect will propagate and accumulate through multiple layers. Such defects in the manufacture process will no doubt be financially taxing for AML3D. Apart from constant human supervision, another quality control process implemented at AML3D is height measurement of each layer, after the layer is built. This process is implemented to ensure the height of each layer is in the acceptable range (1-2mm). As mentioned before, these quality control processes are costing AML3D in term of time and finance. Our project will focus on replace them with a less time consuming, more automating process.

<div><ul>
<li style="display: inline-block;"> [[File:contact_tip.jpg|thumb|none|400px|Figure 3a: A worn contact tip]] </li>
<li style="display: inline-block;"> [[File:layers.jpeg|thumb|none|530px|Figure 3b: Close up photo of layers]] </li>
</ul></div>

==== Relating Contact Tip to Work-piece Distance to layer height ====
At AML3D, the sensor captures the Contact Tip to Work-piece Distance (CTWD), and from there, we can deduce the layer height. A more detail explanation is of follow:

<div><ul>
<li style="display: inline-block;"> [[File:Example_of_layer_height_measurement.jpg|thumb|none|400px|Figure 4: From CTWD to layer height]] </li>
</ul></div>

The CTWD can be set at the beginning of the weld, however, as the welding process running, the actual CTWD fluctuate around the set CTWD. For example, after finish constructing one layer, the laser sensor capture the model height at 3 different position: A, B and C. Let the 3 model height be m1, m2 and m3. Calculate n_mean and for argument sake, if the set CTWD is 15 mm, the actual CTWD at A is [[File:Ctwd_at_a.png|180px]]. The height of one layer can be calculated as [[File:Layer_height_at_a.png|210px]]. A worked example can be found at table 1.

{| class="wikitable" name="Table 1"
|+Table 1: Worked example
|-
! CTWD
! m1
! m2
! m3
! m_mean
! c1
! c2
! c3
|-
| 15
| 20
| 22
| 25
| 22.3
| 17.3
| 15.3
| 17.7
|}

=== Gas metal arc welding ===
<div><ul>
<li style="display: inline-block;"> [[File:gmaw_welding.png|thumb|500px|none|Figure 5: GMAW welding process <ref name="welding_book"> Norrish, J. (2006). Advanced welding processes. Elsevier. </ref>]] </li>
</ul></div>

Gas metal arc welding (GMAW), or also known as metal inert gas (MIG), is the welding process used at AML3D for WAAM. Due to its high deposition rate and economic benefits, GMAW has became more popular. We will explore two variants of GMAW, which are Pulse Multi Control and Cold Metal Transfer. Both processes are developed by Fronius, an Austrian welding company. Note that it is not required to understand the welding physics to follow this wiki. The next two following sections' purpose is to highlight the high variability, high dynamic nature in welding.

[[File:fronius.png|thumb|Figure 6: Fronius logo]]

==== Pulse Multi Control ====
Pulse Multi Control (PMC) welding is Fronius' modified Pulsed GMAW. The advantage of Pulsed GMAW is the ability to control the metal droplet transfer in welding. Note that from figure 7, the metal droplet is characterised by a downward slope of the current pulse. Figure 8 shows the current signal captured from experiments conducted at AML3D (PMC is used).

<div><ul>
<li style="display: inline-block;"> [[File:Pulsed_gmaw_transfer_mode.png|thumb|500px|none|Figure 7: Relationship between pulse current and droplet transfer<ref name="welding_book" />]] </li>
<li style="display: inline-block;"> [[File:Pmc_current_in_time_domain.png|thumb|650px|none|Figure 8: Current signal captured from experiments conducted at AML3D (PMC process)]] </li>
</ul></div>
==== Cold Metal Transfer ====
Cold Metal Transfer (CMT) is a complex welding process, also developed by Fronius. By detecting a short circuit (mode C and D in figure 9), CMT can make adjustment such as retracting the welding material to cool down, and therefore create a smoother, more stable weld. The complexity of CMT can be seen in the current signal of the process (figure 10). Compare to PMC, CMT's signal has more variation during one signal period.

<div><ul>
<li style="display: inline-block;"> [[File:Transfer_mode.png|thumb|800px|none|Figure 9: Mechanism of metal droplet transfer mode]] </li>
<li style="display: inline-block;"> [[File:Cmt1_current_in_time_domain.png|thumb|735px|none|Figure 10: Current signal captured from experiments conducted at AML3D (CMT process)]] </li>
</ul></div>
=== Machine Learning ===
Machine learning (or statistical learning) refer to a set of tools to give data scientists an insight into the data and make better decisions based on those information <ref name="introduction_ml_book"> James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An introduction to statistical learning (Vol. 112, p. 18). New York: Springer. </ref>. In this project, we will explore the capabilities of classical machine learning methods (i.e anything but Deep Learning) due to the limitation of our available data. By the time of writing this wiki, we have only explored Support Vector Machine.

==== Support Vector Machine ====
The main idea of Support Vector Machine (SVM) is to draw hyperplanes that best separate multiple classes of data. The original SVM judges best decision hyperplane based on the separation margin between data classes (figure 11). However, there are multiple variants of SVM, such as hard/soft margin SVM, Nu-SVM, One class SVM (to solve anomaly detection problem). In our project, we choose to use a soft margin implementation of SVM. The benefit of soft margin SVM (as oppose to hard margin SVM) is that by allowing some misclassifications, we get a wider margin and therefore the solution becomes more "generalise" (i.e work better with a new set of dataset that is independent from the training dataset).

<div><ul>
<li style="display: inline-block;"> [[File:svm_margin.png|thumb|400px|none|Figure 11: Separation margin <ref name="bishop_book"> Bishop, C. M. (2006). Pattern recognition and machine learning. Springer. </ref>]] </li>
</ul></div>

SVM algorithm makes use of [https://en.wikipedia.org/wiki/Lagrange_multiplier Lagrange multiplier]. Let's the linear model of the dataset be:

[[File:Linear_model.png|150px]]

where [[File:Phi_math.png|10px]] denotes the feature space transformation, '''w''' denotes the weights and ''b'' denotes bias constant.

Let [[File:vector_x_n.png|17px]] be the input vector (where each index in the vector denotes a feature), and [[File:target_n.png|15px]] be the target (the class of input vector). It is showed that in order to maximise the (soft) margin, we will need to minimise the following:

[[File:Soft margin minimise equation.png|150px]]

with subject to:

[[File:Soft margin minimise condition.png|150px]]

where N is the total number of training data, C is the regularisation parameter to control the effect of slack variable [[File:slack_variable.png|17px]]. A more visual presentation of slack variables can be found in figure 12.

<div><ul>
<li style="display: inline-block;"> [[File:svm_slack_variable.png|thumb|400px|none|Figure 12: Slack variable <ref name="bishop_book"></ref>]] </li>
</ul></div>

If we were to apply the Lagrangian method to solve the above minimisation problem, we can find the decision hyperplane for the SVM algorithm. However, we shall not discuss the solution here as it is not the main focus of the project.

==== Kernel tricks ====
Both figure 11 and 12 are of linearly separable cases for SVM. For dataset that is not easy to separate with a hyperplane in their original features space (such as figure 13), the data often get projected into a higher dimension space where it is easier to separate (figure 14). Such technique is called the Kernel Tricks. It is important to note that Kernel Tricks does not actually "map" the data from '''X''' dimension to '''Z''' dimension, but rather pairwise compare the similarity of the input space in '''Z''' dimension. However, we shall not dive too deep into this as it is not the main focus of the project.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_nonseparable.png |thumb|350px|none|Figure 13: Nonseparable dataset <ref name="bishop_book"></ref>]] </li>
<li style="display: inline-block;"> [[File:Svm_kernel_trick.png |thumb|650px|none|Figure 14: Kernel tricks]] </li>
</ul></div>

== Previous studies ==
There are many researches into defect detection in GMAW process. The three most dominant analysis methods are spectroscopic analysis, acoustic analysis, and electrical signal analysis.

=== Spectroscopic analysis ===
In <ref name="spectroscopic_analysis">D. Bebiano and S. Alfaro, “A weld defects detection system based on a spectrometer,” Sensors, vol. 9, no. 4, pp. 2851–2861, 2009.</ref>, the authors captured the intensity of radiation emission of electric arc. It is obvious from figure 15 that good weld's intensity stays stable while defect one's is characterised by a sudden peak. Therefore, a threshold is set to detect the defect welds.

<div><ul>
<li style="display: inline-block;"> [[File:spectroscopic_analysis.png|thumb|400px|none|Figure 15: Spectroscopic analysis <ref name="spectroscopic_analysis"></ref>]] </li>
</ul></div>

Other research into using spectroscopic data to monitor that can be of interest is <ref>D. Naso, B. Turchiano, and P. Pantaleo, “A fuzzy-logic based optical sensor for online weld defect-detection,” IEEE transactions on Industrial Informatics, vol. 1, no. 4, pp. 259–273, 2005.</ref>.

=== Acoustic analysis ===
With <ref name="acoustic_analysis">E. Cayo and S. C. Alfaro, “A non-intrusive GMA welding process quality monitoring system using acoustic sensing,” Sensors, vol. 9, no. 9, pp. 7150–7166, 2009.</ref>, the author discovered that arc ignition is characterised by a distinct sound and therefore it is possible to capture the frequency of ignition through acoustic sensors. A tolerance band was set and both ignition frequency and sound pressure level were used to monitor the quality of the weld.

<div><ul>
<li style="display: inline-block;"> [[File:acoustic_analysis.png|thumb|400px|none|Figure 16: Acoustic analysis <ref name="acoustic_analysis"></ref>]] </li>
</ul></div>

Other research into using acoustic signal to monitor that can be of interest are <ref>M. Fidali, “Detection of welding process instabilities using acoustic signals,” in International Congress on Technical Diagnostic, pp. 191–201, Springer, 2016.</ref>, <ref>L. Zhang, A. C. Basantes-Defaz, D. Ozevin, and E. Indacochea, “Real-time monitoring of welding process using air-coupled ultrasonics and acoustic emission,” The International Journal of Advanced Manufacturing Technology, vol. 101, no. 5-8, pp. 1623–1634, 2019.</ref>.

=== Electrical signal analysis ===
With <ref name="electrical_analysis_1">A. Sumesh, K. Rameshkumar, A. Raja, K. Mohandas, A. Santhakumari, and R. Shyambabu, “Establishing correlation between current and voltage signatures of the arc and weld defects in GMAW process,” Arabian Journal for Science and Engineering, vol. 42, no. 11, pp. 4649–4665, 2017.</ref>, the author plot the probability density distribution of voltage signal and discovered that the stable weld's signal concentrate in one region while unstable signal spread out. While in the paper, the authors did not developed a monitoring method using their findings, it shows that there is a significant differences between stable and unstable weld's electrical signature.

<div><ul>
<li style="display: inline-block;"> [[File:Electrical analysis pdd.png|thumb|400px|none|Figure 17: Voltage PDD <ref name="electrical_analysis_1"></ref>]] </li>
</ul></div>

In <ref name="electrical_analysis_2">S. Simpson, “Signature images for arc welding fault detection,” Science and Technology of Welding and Joining, vol. 12, no. 6, pp. 481–486, 2007.</ref>, the authors shows that by plotting the signature image of voltage and current scatter plot, both arcing and short-circuiting region's signature image can notify a defects welds. A more visual representation of short circuiting and arcing region can be found in [[#Gas metal arc welding|Gas metal arc welding section]].

<div><ul>
<li style="display: inline-block;"> [[File:Electrical voltage signature.png|thumb|400px|none|Figure 18a: Electrical voltage signature of stable weld <ref name="electrical_analysis_2"></ref>]] </li>
<li style="display: inline-block;"> [[File:Electrical signature 2.png|thumb|400px|none|Figure 18b: Electrical voltage signature of unstable weld <ref name="electrical_analysis_2"></ref>]] </li>
</ul></div>

Other research into using electrical signal to monitor that can be of interest are <ref>R. Madigan, “Arc sensing for defects in constant-voltage gas metal arc welding,” Welding Journal, vol. 78, pp. 322S–328S, 1999.</ref>, <ref>Y. Huang, K. Wang, Z. Zhou, X. Zhou, and J. Fang, “Stability evaluation of short-circuiting gas metal arc welding based on ensemble empirical mode decomposition,” Measurement Science and Technology, vol. 28, no. 3, p. 035006, 2017.</ref>,<ref>Z. Zhang, X. Chen, H. Chen, J. Zhong, and S. Chen, “Online welding quality monitoring based on feature extraction of arc voltage signal,” The International Journal of Advanced Manufacturing Technology, vol. 70, no. 9-12, pp. 1661–1671, 2014</ref>,<ref>X. Li and S. Simpson, “Parametric approach to positional fault detection in short arc welding,” Science and Technology of welding and joining, vol. 14, no. 2, pp. 146–151, 2009.</ref>, and <ref>E. Wei, D. Farson, R. Richardson, and H. Ludewig, “Detection of weld surface porosity by statistical analysis of arc current in gas metal arc welding,” Journal of Manufacturing Processes, vol. 3, no. 1, pp. 50–59, 2001.</ref>.

== Experiment setup ==
The welding system uses TPS 320i C Pulse (figure 15a), produced by Fronius, to supply power, feed wire, and system cooling. For movement control of the welding torch, the system use ABB’s IRB 1520ID (figure 15b), which is a dedicated arc welding robot. The Data Acquisition device is NI's DAQ. Each channel (voltage and current) is sampled at 25kHz while the dominant frequency for voltage and current signal are around 130Hz (for CMT process) and 190Hz (for PMC process).

<div><ul>
<li style="display: inline-block;"> [[File:Fronius_power_supply.jpg |thumb|250px|none|Figure 19a: Fronius' Power Supply]] </li>
<li style="display: inline-block;"> [[File:Welding_robot.jpg |thumb|250px|none|Figure 19b: ABB's welding robot]] </li>
</ul></div>

[[File:Ni_daq.jpg|thumb|Figure 16: NI's DAQ device]]

== Statistical Analysis ==

=== Preprocessing ===
As the captured signal coming from Fronius' power supply, the signals also pick up switching power noise. The noise will interfere with the accuracy of the analysis if not filtered. In this project, we applied a 1000Hz digital Low Pass Butterworth filter. The effect of the filter can be seen in figure 20, while the comparation of using different filters can be seen in the gallery.

<div><ul>
<li style="display: inline-block;"> [[File:Filter_effect.png |thumb|750px|none|Figure 20: Effect of 1000Hz low pass filter]] </li>
<li style="display: inline-block;"> [[File:Freq_response.png |thumb|750px|none|Figure 21: Frequency response of the in-use filter]] </li>
</ul></div>

<gallery mode="packed-hover" caption="Gallery: Comparison of different filtering method" perrow=2>
File:Effect_of_filtering2.png|In use filter vs 300Hz low pass filter
File:Effect_of_filtering3.png|In use filter vs 500Hz low pass filter
File:Effect_of_filtering4.png|In use filter vs 750Hz low pass filter
File:Effect_of_filtering5.png|In use filter vs 2000Hz low pass filter
</gallery>

=== Power Spectral Density ===
==== Prerequisite knowledge ====
Looking at the spectrum is another way to analyse the fundamental information of the signal, in frequency domain <ref name="signal_processing_book">Prandoni, P., & Vetterli, M. (2008). Signal processing for communications. EPFL press.</ref>. Note that the term "Power" in Power Spectral Density does not refer to the convention power definition. In signal processing, energy of a discrete time signal is defined as follow:

[[File:Signal_energy_formular.png|150px]]

The way to interpret this definition is if we consider the signal to be of "Voltage" unit, ''E'' is the total energy (in joules) dissipated over a 1 Ohm resistor <ref name="signal_processing_book"></ref>. The power spectral density of a signal over time interval [-M,M] is defined as follow:

[[File:Psd_formular.png|300px]]

where [[File:Fourier series symbol.png|67px]] is the truncated Fourier transform:

[[File:Fourier_series_formular.png|250px]]

==== Analysis ====
Our goal for this project is to detect the defect welds and predict the layer heights (the contact tip to work piece distance, CTWD, to be more accurate) in real time (refer to [[#Relating Contact Tip to Work-piece Distance to layer height|this section]] to understand the relationship between Contact Tip to Work-piece Distance and Layer height). An analysis of the current power spectral density (PSD) shows that there is an inverse proportional relationship between the CTWD of a weld and the frequency where its '''current signal''' PSD reach the peak, and also the magnitude of the peak. Since the peak of PSD can be fairly sensitive, we only consider the frequency of the peak PSD.

Figure 22 shows the PSD plot of voltage, current, and power signal of the weld using PMC process. In this experiment, we created multiple one-line, 8 seconds welds. These eight-second welds then get segmented into smaller two-second chunk to increase the number of data and decrease the processing time.

<div><ul>
<li style="display: inline-block;"> [[File:Psd pmc good bad all lowfreq.png |thumb|750px|none|Figure 22: Power spectral density of PMC process]] </li>
<li style="display: inline-block;"> [[File:One line weld.jpg |thumb|578px|none|Figure 23: Simple one-line weld experiment]] </li>
<li style="display: inline-block;"> [[File:Cmt_psd_stable_allCTWD_zoomed.png |thumb|750px|none|Figure 24: Simple one-line weld experiment]] </li>

</ul></div>

It seems like the inverse proportional relationship holds true for most cases. Other experiments showed that for a short period at the beginning and ending of the weld, the signal can behave differently relative to the rest of the weld. Currently, to migrate with this problem, data will not be collected and analysed (both offline and online) at these periods. However, the above results seem to only hold true for PMC processes. With CMT process, the current PSD plot indicates no relationship between the CTWD and the frequency at peak (figure 24).

For the next stage of the project, our plan is to develop a linear regression model to predict the CTWD for PMC process. With CMT process, we will have to look into different analysis techniques.

=== Support Vector Machine ===

=== Features extraction from grayscale image ===

== Result discussion ==

== Conclusion ==

== Future work ==

== References ==
<references />

Projects:2019s2-20001 Using Machine Learning to Determine Deposit Height and Defects for Wire + Arc Additive Manufacture (3D printing)

2019-10-07T09:51:54Z

A1713568: /* Wire Arc Additive Manufacturing */

[[Category:Projects]]
[[Category:Final Year Projects]]
[[Category:2019s2|24501]]
== Abstract ==
Additive manufacturing is an emerging technology that is complementary to subtracting manufacturing. AM is able to create complex model at the cost of production time. An example of AM is Wire + Arc Additive Manufacture. After slicing the manufacturing process of one model to multiple layers, WAAM uses electric arc as a source of heat to feed metal wire and create each layer. However, a slight height inaccuracy in this process will result in an expensive faulty, as error can be multiplied across many layers. Thus, in-process height measurement and faulty detection are critical to the production process. In this thesis, we will look into a machine learning approach to develop an on-the-line algorithm that automatically detects faulty processes and make height measurement by analysing electrical signal (current and voltage) of a Gas Metal Arc Welding machine. This project will introduce the
approaches to tackle the same problem by other researchers, describe the design of our experiment, development of our machine learning model and discuss the final result.

'''Keywords''': Additive manufacturing, Gas metal arc welding, Metal inert gas, Machine learning, Fault detection, Signal analysis
== Introduction ==
3D printing is an emerging technology that has the potential to significantly reduce material usage through the production of near net-shape parts. Many of the systems for 3D printing are based on lasers and powders; however the deposition rate with such systems is very low making the production of large-scale parts difficult. AML Technologies specialises in the use of Wire + Arc Additive Manufacture (WAAM) where deposition is based on arc welding processes and the deposition rates are an order of magnitude greater. When building 3D printed parts, even a relatively small layer height error of only 0.1 mm can produce large build height errors when multiplied across the many layers of a typical build. This can make path planning difficult, so in-process layer height measurement is an essential building block of any production 3D printing system. A variety of techniques can be used for monitoring the layer height including laser scanners, and arc monitoring. It is the latter technique that will be explored in this project due to its robustness, and low cost of implementation – it only requires the measurement of arc current and voltage. Furthermore, it can potentially be used to detect defects by identifying waveform irregularities.
=== Project team ===
==== Project students ====
* Anh Tran
* Nhat Nguyen
==== Supervisors ====
* Dr. Brian Ng
* Dr. Paul Colegrove (AML3D, sponsored company)

[[File:aml3d_logo.jpg|100px]]
=== Objectives ===
The objective of this project is to increase the efficiency of the manufacture process at AML3D. In order to do so, the team will investigate into the possibility of automating and optimising the quality control processes. The two quality control processes that are currently being implemented at AML3D are measuring layer height using laser sensors, and human supervision for detecting defects. These processes add overhead into production time and usage of human resource, which is not desired. To achieve the goal, it is expected that machine learning methods will be used extensively to analyse the electrical signatures of the weld process.

== Background ==
=== Wire Arc Additive Manufacturing ===
Wire and Arc Additive Manufacturing (WAAM) is a type of additive manufacturing that uses electric arc as the heat source and material wire to feed the manufacture process <ref name="WAAM"> S. W. Williams, F. Martina, A. C. Addison, J. Ding, G. Pardal & P. Colegrove (2016) Wire + Arc Additive Manufacturing, Materials Science and Technology, 32:7, 641-647, DOI: 10.1179/1743284715Y.0000000073 </ref>. WAAM has been investigated since the 1990s <ref name="WAAM" />, but only recently that it received more attention from the manufacture world.

Its significant comes from the ability to manufacture complex model with less time and less material. Figure 1a and 1b show real parts that was manufactured by AML3D. Such custom made parts might takes months before be ready to be shipped, but with WAAM, the production time can reduce down to weeks. Currently, the industries that benefit the most from WAAM are maritime and aerospace.

Similar to other additive manufacture methods, WAAM achieves such results by sliding the models into multiple layers, and then build the model layer by layer. The movement control is normally handled by a robotic arm (AML3D uses ABB's Arc Welder robot), and the welding path is generated by a Computer Aid Manufacture (CAM) software.

[[File:welding_part1.JPG|thumb|Figure 1a: A propeller]]
[[File:welding_part2.jpg|thumb|Figure 1b: Another part of the ship]]
[[File:abb_robot.jpg|thumb|Figure 2: ABB's IRB 1520ID Arc Welder]]

Currently at AML3D, the manufacture process for a part can take anywhere from days to weeks. However, the contact tip (where the welding gun deposits the wire) needs replacement every couple of hours. A worn contact tip such as figure 3 will lead to unstable weld and the this effect will propagate and accumulate through multiple layers. Such defects in the manufacture process will no doubt be financially taxing for AML3D. Apart from constant human supervision, another quality control process implemented at AML3D is height measurement of each layer, after the layer is built. This process is implemented to ensure the height of each layer is in the acceptable range (1-2mm). As mentioned before, these quality control processes are costing AML3D in term of time and finance. Our project will focus on replace them with a less time consuming, more automating process.

<div><ul>
<li style="display: inline-block;"> [[File:contact_tip.jpg|thumb|none|400px|Figure 3a: A worn contact tip]] </li>
<li style="display: inline-block;"> [[File:layers.jpeg|thumb|none|530px|Figure 3b: Close up photo of layers]] </li>
</ul></div>

==== Relating Contact Tip to Work-piece Distance to layer height ====
At AML3D, the sensor captures the Contact Tip to Work-piece Distance (CTWD), and from there, we can deduce the layer height. A more detail explanation is of follow:

<div><ul>
<li style="display: inline-block;"> [[File:Example_of_layer_height_measurement.jpg|thumb|none|400px|Figure 4: From CTWD to layer height]] </li>
</ul></div>

The CTWD can be set at the beginning of the weld, however, as the welding process running, the actual CTWD fluctuate around the set CTWD. For example, after finish constructing one layer, the laser sensor capture the model height at 3 different position: A, B and C. Let the 3 model height be m1, m2 and m3. Calculate n_mean and for argument sake, if the set CTWD is 15 mm, the actual CTWD at A is [[File:Ctwd_at_a.png|180px]]. The height of one layer can be calculated as [[File:Layer_height_at_a.png|210px]]. A worked example can be found at table 1.

{| class="wikitable" name="Table 1"
|+Table 1: Worked example
|-
! CTWD
! m1
! m2
! m3
! m_mean
! c1
! c2
! c3
|-
| 15
| 20
| 22
| 25
| 22.3
| 17.3
| 15.3
| 17.7
|}

=== Gas metal arc welding ===
<div><ul>
<li style="display: inline-block;"> [[File:gmaw_welding.png|thumb|500px|none|Figure 5: GMAW welding process <ref name="welding_book"> Norrish, J. (2006). Advanced welding processes. Elsevier. </ref>]] </li>
</ul></div>

Gas metal arc welding (GMAW), or also known as metal inert gas (MIG), is the welding process used at AML3D for WAAM. Due to its high deposition rate and economic benefits, GMAW has became more popular. We will explore two variants of GMAW, which are Pulse Multi Control and Cold Metal Transfer. Both processes are developed by Fronius, an Austrian welding company. Note that it is not required to understand the welding physics to follow this wiki. The next two following sections' purpose is to highlight the high variability, high dynamic nature in welding.

[[File:fronius.png|thumb|Figure 6: Fronius logo]]

==== Pulse Multi Control ====
Pulse Multi Control (PMC) welding is Fronius' modified Pulsed GMAW. The advantage of Pulsed GMAW is the ability to control the metal droplet transfer in welding. Note that from figure 7, the metal droplet is characterised by a downward slope of the current pulse. Figure 8 shows the current signal captured from experiments conducted at AML3D (PMC is used).

<div><ul>
<li style="display: inline-block;"> [[File:Pulsed_gmaw_transfer_mode.png|thumb|500px|none|Figure 7: Relationship between pulse current and droplet transfer<ref name="welding_book" />]] </li>
<li style="display: inline-block;"> [[File:Pmc_current_in_time_domain.png|thumb|650px|none|Figure 8: Current signal captured from experiments conducted at AML3D (PMC process)]] </li>
</ul></div>
==== Cold Metal Transfer ====
Cold Metal Transfer (CMT) is a complex welding process, also developed by Fronius. By detecting a short circuit (mode C and D in figure 9), CMT can make adjustment such as retracting the welding material to cool down, and therefore create a smoother, more stable weld. The complexity of CMT can be seen in the current signal of the process (figure 10). Compare to PMC, CMT's signal has more variation during one signal period.

<div><ul>
<li style="display: inline-block;"> [[File:Transfer_mode.png|thumb|800px|none|Figure 9: Mechanism of metal droplet transfer mode]] </li>
<li style="display: inline-block;"> [[File:Cmt1_current_in_time_domain.png|thumb|735px|none|Figure 10: Current signal captured from experiments conducted at AML3D (CMT process)]] </li>
</ul></div>
=== Machine Learning ===
Machine learning (or statistical learning) refer to a set of tools to give data scientists an insight into the data and make better decisions based on those information <ref name="introduction_ml_book"> James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An introduction to statistical learning (Vol. 112, p. 18). New York: Springer. </ref>. In this project, we will explore the capabilities of classical machine learning methods (i.e anything but Deep Learning) due to the limitation of our available data. By the time of writing this wiki, we have only explored Support Vector Machine.

==== Support Vector Machine ====
The main idea of Support Vector Machine (SVM) is to draw hyperplanes that best separate multiple classes of data. The original SVM judges best decision hyperplane based on the separation margin between data classes (figure 11). However, there are multiple variants of SVM, such as hard/soft margin SVM, Nu-SVM, One class SVM (to solve anomaly detection problem). In our project, we choose to use a soft margin implementation of SVM. The benefit of soft margin SVM (as oppose to hard margin SVM) is that by allowing some misclassifications, we get a wider margin and therefore the solution becomes more "generalise" (i.e work better with a new set of dataset that is independent from the training dataset).

<div><ul>
<li style="display: inline-block;"> [[File:svm_margin.png|thumb|400px|none|Figure 11: Separation margin <ref name="bishop_book"> Bishop, C. M. (2006). Pattern recognition and machine learning. Springer. </ref>]] </li>
</ul></div>

SVM algorithm makes use of [https://en.wikipedia.org/wiki/Lagrange_multiplier Lagrange multiplier]. Let's the linear model of the dataset be:

[[File:Linear_model.png|150px]]

where [[File:Phi_math.png|10px]] denotes the feature space transformation, '''w''' denotes the weights and ''b'' denotes bias constant.

Let [[File:vector_x_n.png|17px]] be the input vector (where each index in the vector denotes a feature), and [[File:target_n.png|15px]] be the target (the class of input vector). It is showed that in order to maximise the (soft) margin, we will need to minimise the following:

[[File:Soft margin minimise equation.png|150px]]

with subject to:

[[File:Soft margin minimise condition.png|150px]]

where N is the total number of training data, C is the regularisation parameter to control the effect of slack variable [[File:slack_variable.png|17px]]. A more visual presentation of slack variables can be found in figure 12.

<div><ul>
<li style="display: inline-block;"> [[File:svm_slack_variable.png|thumb|400px|none|Figure 12: Slack variable <ref name="bishop_book"></ref>]] </li>
</ul></div>

If we were to apply the Lagrangian method to solve the above minimisation problem, we can find the decision hyperplane for the SVM algorithm. However, we shall not discuss the solution here as it is not the main focus of the project.

==== Kernel tricks ====
Both figure 11 and 12 are of linearly separable cases for SVM. For dataset that is not easy to separate with a hyperplane in their original features space (such as figure 13), the data often get projected into a higher dimension space where it is easier to separate (figure 14). Such technique is called the Kernel Tricks. It is important to note that Kernel Tricks does not actually "map" the data from '''X''' dimension to '''Z''' dimension, but rather pairwise compare the similarity of the input space in '''Z''' dimension. However, we shall not dive too deep into this as it is not the main focus of the project.

<div><ul>
<li style="display: inline-block;"> [[File:Svm_nonseparable.png |thumb|350px|none|Figure 13: Nonseparable dataset <ref name="bishop_book"></ref>]] </li>
<li style="display: inline-block;"> [[File:Svm_kernel_trick.png |thumb|650px|none|Figure 14: Kernel tricks]] </li>
</ul></div>

== Previous studies ==
There are many researches into defect detection in GMAW process. The three most dominant analysis methods are spectroscopic analysis, acoustic analysis, and electrical signal analysis.

=== Spectroscopic analysis ===
In <ref name="spectroscopic_analysis">D. Bebiano and S. Alfaro, “A weld defects detection system based on a spectrometer,” Sensors, vol. 9, no. 4, pp. 2851–2861, 2009.</ref>, the authors captured the intensity of radiation emission of electric arc. It is obvious from figure 15 that good weld's intensity stays stable while defect one's is characterised by a sudden peak. Therefore, a threshold is set to detect the defect welds.

<div><ul>
<li style="display: inline-block;"> [[File:spectroscopic_analysis.png|thumb|400px|none|Figure 15: Spectroscopic analysis <ref name="spectroscopic_analysis"></ref>]] </li>
</ul></div>

Other research into using spectroscopic data to monitor that can be of interest is <ref>D. Naso, B. Turchiano, and P. Pantaleo, “A fuzzy-logic based optical sensor for online weld defect-detection,” IEEE transactions on Industrial Informatics, vol. 1, no. 4, pp. 259–273, 2005.</ref>.

=== Acoustic analysis ===
With <ref name="acoustic_analysis">E. Cayo and S. C. Alfaro, “A non-intrusive GMA welding process quality monitoring system using acoustic sensing,” Sensors, vol. 9, no. 9, pp. 7150–7166, 2009.</ref>, the author discovered that arc ignition is characterised by a distinct sound and therefore it is possible to capture the frequency of ignition through acoustic sensors. A tolerance band was set and both ignition frequency and sound pressure level were used to monitor the quality of the weld.

<div><ul>
<li style="display: inline-block;"> [[File:acoustic_analysis.png|thumb|400px|none|Figure 16: Acoustic analysis <ref name="acoustic_analysis"></ref>]] </li>
</ul></div>

Other research into using acoustic signal to monitor that can be of interest are <ref>M. Fidali, “Detection of welding process instabilities using acoustic signals,” in International Congress on Technical Diagnostic, pp. 191–201, Springer, 2016.</ref>, <ref>L. Zhang, A. C. Basantes-Defaz, D. Ozevin, and E. Indacochea, “Real-time monitoring of welding process using air-coupled ultrasonics and acoustic emission,” The International Journal of Advanced Manufacturing Technology, vol. 101, no. 5-8, pp. 1623–1634, 2019.</ref>.

=== Electrical signal analysis ===
With <ref name="electrical_analysis_1">A. Sumesh, K. Rameshkumar, A. Raja, K. Mohandas, A. Santhakumari, and R. Shyambabu, “Establishing correlation between current and voltage signatures of the arc and weld defects in GMAW process,” Arabian Journal for Science and Engineering, vol. 42, no. 11, pp. 4649–4665, 2017.</ref>, the author plot the probability density distribution of voltage signal and discovered that the stable weld's signal concentrate in one region while unstable signal spread out. While in the paper, the authors did not developed a monitoring method using their findings, it shows that there is a significant differences between stable and unstable weld's electrical signature.

<div><ul>
<li style="display: inline-block;"> [[File:Electrical analysis pdd.png|thumb|400px|none|Figure 17: Voltage PDD <ref name="electrical_analysis_1"></ref>]] </li>
</ul></div>

In <ref name="electrical_analysis_2">S. Simpson, “Signature images for arc welding fault detection,” Science and Technology of Welding and Joining, vol. 12, no. 6, pp. 481–486, 2007.</ref>, the authors shows that by plotting the signature image of voltage and current scatter plot, both arcing and short-circuiting region's signature image can notify a defects welds. A more visual representation of short circuiting and arcing region can be found in [[#Gas metal arc welding|Gas metal arc welding section]].

<div><ul>
<li style="display: inline-block;"> [[File:Electrical voltage signature.png|thumb|400px|none|Figure 18a: Electrical voltage signature of stable weld <ref name="electrical_analysis_2"></ref>]] </li>
<li style="display: inline-block;"> [[File:Electrical signature 2.png|thumb|400px|none|Figure 18b: Electrical voltage signature of unstable weld <ref name="electrical_analysis_2"></ref>]] </li>
</ul></div>

Other research into using electrical signal to monitor that can be of interest are <ref>R. Madigan, “Arc sensing for defects in constant-voltage gas metal arc welding,” Welding Journal, vol. 78, pp. 322S–328S, 1999.</ref>, <ref>Y. Huang, K. Wang, Z. Zhou, X. Zhou, and J. Fang, “Stability evaluation of short-circuiting gas metal arc welding based on ensemble empirical mode decomposition,” Measurement Science and Technology, vol. 28, no. 3, p. 035006, 2017.</ref>,<ref>Z. Zhang, X. Chen, H. Chen, J. Zhong, and S. Chen, “Online welding quality monitoring based on feature extraction of arc voltage signal,” The International Journal of Advanced Manufacturing Technology, vol. 70, no. 9-12, pp. 1661–1671, 2014</ref>,<ref>X. Li and S. Simpson, “Parametric approach to positional fault detection in short arc welding,” Science and Technology of welding and joining, vol. 14, no. 2, pp. 146–151, 2009.</ref>, and <ref>E. Wei, D. Farson, R. Richardson, and H. Ludewig, “Detection of weld surface porosity by statistical analysis of arc current in gas metal arc welding,” Journal of Manufacturing Processes, vol. 3, no. 1, pp. 50–59, 2001.</ref>.

== Experiment setup ==
The welding system uses TPS 320i C Pulse (figure 15a), produced by Fronius, to supply power, feed wire, and system cooling. For movement control of the welding torch, the system use ABB’s IRB 1520ID (figure 15b), which is a dedicated arc welding robot. The Data Acquisition device is NI's DAQ. Each channel (voltage and current) is sampled at 25kHz while the dominant frequency for voltage and current signal are around 130Hz (for CMT process) and 190Hz (for PMC process).

<div><ul>
<li style="display: inline-block;"> [[File:Fronius_power_supply.jpg |thumb|250px|none|Figure 19a: Fronius' Power Supply]] </li>
<li style="display: inline-block;"> [[File:Welding_robot.jpg |thumb|250px|none|Figure 19b: ABB's welding robot]] </li>
</ul></div>

[[File:Ni_daq.jpg|thumb|Figure 16: NI's DAQ device]]

== Statistical Analysis ==

=== Preprocessing ===
As the captured signal coming from Fronius' power supply, the signals also pick up switching power noise. The noise will interfere with the accuracy of the analysis if not filtered. In this project, we applied a 1000Hz digital Low Pass Butterworth filter. The effect of the filter can be seen in figure 20, while the comparation of using different filters can be seen in the gallery.

<div><ul>
<li style="display: inline-block;"> [[File:Filter_effect.png |thumb|750px|none|Figure 20: Effect of 1000Hz low pass filter]] </li>
<li style="display: inline-block;"> [[File:Freq_response.png |thumb|750px|none|Figure 21: Frequency response of the in-use filter]] </li>
</ul></div>

<gallery mode="packed-hover" caption="Gallery: Comparison of different filtering method" perrow=2>
File:Effect_of_filtering2.png|In use filter vs 300Hz low pass filter
File:Effect_of_filtering3.png|In use filter vs 500Hz low pass filter
File:Effect_of_filtering4.png|In use filter vs 750Hz low pass filter
File:Effect_of_filtering5.png|In use filter vs 2000Hz low pass filter
</gallery>

=== Power Spectral Density ===
==== Prerequisite knowledge ====
Looking at the spectrum is another way to analyse the fundamental information of the signal, in frequency domain <ref name="signal_processing_book">Prandoni, P., & Vetterli, M. (2008). Signal processing for communications. EPFL press.</ref>. Note that the term "Power" in Power Spectral Density does not refer to the convention power definition. In signal processing, energy of a discrete time signal is defined as follow:

[[File:Signal_energy_formular.png|150px]]

The way to interpret this definition is if we consider the signal to be of "Voltage" unit, ''E'' is the total energy (in joules) dissipated over a 1 Ohm resistor <ref name="signal_processing_book"></ref>. The power spectral density of a signal over time interval [-M,M] is defined as follow:

[[File:Psd_formular.png|300px]]

where [[File:Fourier series symbol.png|67px]] is the truncated Fourier transform:

[[File:Fourier_series_formular.png|250px]]

==== Analysis ====
Our goal for this project is to detect the defect welds and predict the layer heights (the contact tip to work piece distance, CTWD, to be more accurate) in real time (refer to [[#Objectives|Objective]]). An analysis of the current power spectral density (PSD) shows that there is an inverse proportional relationship between the CTWD of a weld and the frequency where its '''current signal''' PSD reach the peak, and also the magnitude of the peak. Since the peak of PSD can be fairly sensitive, we only consider the frequency of the peak PSD.

Figure 22 shows the PSD plot of voltage, current, and power signal of the weld using PMC process. In this experiment, we created multiple one-line, 8 seconds welds. These eight-second welds then get segmented into smaller two-second chunk to increase the number of data and decrease the processing time.

<div><ul>
<li style="display: inline-block;"> [[File:Psd pmc good bad all lowfreq.png |thumb|750px|none|Figure 22: Power spectral density of PMC process]] </li>
<li style="display: inline-block;"> [[File:One line weld.jpg |thumb|578px|none|Figure 23: Simple one-line weld experiment]] </li>
<li style="display: inline-block;"> [[File:Cmt_psd_stable_allCTWD_zoomed.png |thumb|750px|none|Figure 24: Simple one-line weld experiment]] </li>

</ul></div>

It seems like the inverse proportional relationship holds true for most cases. Other experiments showed that for a short period at the beginning and ending of the weld, the signal can behave differently relative to the rest of the weld. Currently, to migrate with this problem, data will not be collected and analysed (both offline and online) at these periods. However, the above results seem to only hold true for PMC processes. With CMT process, the current PSD plot indicates no relationship between the CTWD and the frequency at peak (figure 24).

For the next stage of the project, our plan is to develop a linear regression model to predict the CTWD for PMC process. With CMT process, we will have to look into different analysis techniques.

=== Support Vector Machine ===

=== Features extraction from grayscale image ===

== Result discussion ==

== Conclusion ==

== Future work ==

== References ==
<references />

File:Ctwd at a.png

2019-10-07T09:51:40Z

A1713568: A1713568 uploaded a new version of "File:Ctwd at a.png"

File:Layer height at a.png

2019-10-07T09:48:48Z

A1713568: