Projects:2021s1-13012 Deep Learning Approach for Automatic Defect Detection during Wire + Arc Additive Manufacture (3D printing)

Introduction

3D metal printing through AML3D's ARCEMY AND WAAM (Wire Arc Additive Manufacturing) Process uses programmed robots and wire-arc welding to generate 3D metal objects and designs. These objects are free form parts utilising wire as feedstock to the wire-arc process. There are economic, environmental, and time benefits from AML3D's ARCEMY and WAAM process. Throughout this process there is potential for equipment and deposition problems which can cause defects. Examples of these types of defects are porosity, lack of fusion, humping, and holes. With this in mind, the project will focus on layer by layer Automatic defect detection during the WAAM process. The primary objective for this project is to use Deep-Learning approaches for Automatic Defect detection during the WAAM process. There are serval ways for this to be achieved such as image processing and recognition, computer vision, machine learning, deep learning, or any combination of these. The detection process for layer-by-layer detection will be (1) A Welding/3D metal printing layer is produced. (2) A photo is taken of the layer and inputted into an algorithm. (3) If there is a defect, then signal a warning and wait until it has been resolved. Otherwise, if there are no defects, then repeat the cycle after the next layer has been completed. To assist a deep learning approach, a large sample of defects will allow the algorithm to learn and grow. This will result with increasing success for identifying defects. An extensive literature review will need to be initially conducted not only in 3D metal printing defect detection, but in all other aspects like image processing, computer visions, and machine learning. Once this is complete, then some experimental analysis techniques can be implemented and tested.

Project team

Project students

• Yuhan Zhou
• Liam Dorward

Supervisors

• Dr Brian Ng

Wire Arc Additive Manufacturing (WAAM)

WAAM technology can deposit a variety of metals such as titanium, nickel alloy, aluminium, and steel [1][2]. Some initial benefits are, 40-60% reduction in fabrication time, 15-20% reduction in post-machining time, and minimising the number of raw materials that are used [1][2].

It is important the weld is being produced at a high quality in each subsequent layer to maximise the product mechanical strength. Types of defects which can occur are deformation and residual stress, porosity, and crack and delamination [2]. Each type of defect influences build integrity uniquely. There are current methods already designed to help improve build quality.

Welding Defects

Through WAAM, the weld is constantly subjected to thermal expansion and shrinkage throughout the building process [2]. This can cause longitudinal and transverse shrinkage, bending, angular, and rotational deformations to occur. Residual stress is tested when all external loading forces have been removed. This residual stress will determine its mechanical properties and fatigue performance. If the stress is high these parts can crack. Cracks can also arise from high strain in the melt pool.

Low mechanical strength and low fatigue properties are potential outcomes with porosity defects [2]. Porosity occurs from surface contamination which may contain hydrocarbons, grease, and moisture. It contaminates the molten pool being deposited. Additionally, a lack of fusion can occur if the metals are not properly clean. This can create gaps within each layer.

Delamination is an example of the layers appearing as separate and not representing one whole component [2]. This type of defect is formed from incomplete melting or insufficient remelting of the underlying solid between the layers. This defect is much more visible and is unlikely to be fixed later.

Motivations

There is significant motivation for this project. If the defect is possible to fix at the time it is discovered, it will improve the mechanical integrity of the build [2]. A layer of the weld could simply be cut off and re-completed if it does not meet manufacturing standards. If the defect cannot be removed, then the bare minimum of product is wasted. Identifying defects after a completed build will require more restrictive post-process operations or the whole design is discarded [2]. This costs additional time, money, materials, and resources to fix or restart the build.

Method

High level Design

The high-level design for this solution begins with WAAM completing a welding layer. Once this is complete a photo of the layer is captured and inputted for image processing. The image processing will apply automatic rotating and cropping, and the result is passed onto the CNN. The CNN is trained prior from determined testing and validation datasets and is saved ready for testing. The saved model receives the image input and analyses it to make a prediction of whether there is a defect or not. If the weld is clear of defects the process repeats from the start, otherwise if it identifies a defect, it will pause production and alerts the user. The software will wait for the user to update whether the defect has been removed or not. If the defect has been removed return to the start, otherwise keep waiting until the defect has been removed. This high-level design can be seen in Figure 1.

Figure 1: High level design for automatic defect detection system

References

[1] AML3D. "Our Technology." AML3D. [Online]. Available: https://aml3d.com/technology/ [Accessed: May 22, 2021]

[2] B Wu, Z Pan, D Ding, D Cuiuri, H Li, J Xu, J Norrish, “A review of the wire arc additive manufacturing of metals: properties, defects and quality improvement.”, Science Direct, Journal of Manufacturing Processes 35, p. 127-139, 2018, [Online serial]. Available: https://www.sciencedirect.com/science/article/pii/S1526612518310739?casa_token=pSYJF2BzVTYAAAAA:kh3GpK3D3cC3PCPCLc4vy3PATqOBzhsjPDIWUmgVWyQfuuqg-vJ9Bd7xzDfBQGe4UbTTmKUb3Ig. [Accessed: Apr. 15, 2021]