Deep learning assisted insights on porosity evolution of WAAM aluminium alloys

This work investigates the porosity distribution of Wire Arc Directed Energy Deposition (WA-DED) Al-12Si alloys using image processing and three deep learning techniques. The UNet model outperformed the two models based on accuracy and Jaccard index score and exhibited strong similarity with manually segmented ground truth images. The porosity tends to increase along the build owing to oxidation anomalies and improper shielding environment. The top region exhibited the highest porosity, whereas the middle region exhibited stable porosity variations owing to interlayer reheating. The work offers an automated deep learning segmentation approach that is fast and accurate, and scalable for XCT analysis of WA-DED.