Detecting balling defects using multisource transfer learning in wire arc additive manufacturing

Sainand Jadhav,Duck Bong Kim,Sung-Ho Hong,Seung-Jun Shin

DOI: https://doi.org/10.1093/jcde/qwad067

2023-07-04

Journal of Computational Design and Engineering

Abstract:Wire arc additive manufacturing (WAAM) has gained attention as a feasible process in large-scale metal additive manufacturing due to its high deposition rate, cost efficiency, and material diversity. However, WAAM induces a degree of uncertainty in the process stability and the part quality owing to its non-equilibrium thermal cycles and layer-by-layer stacking mechanism. Anomaly detection is therefore necessary for the quality monitoring of the parts. Most relevant studies have applied machine learning to derive data-driven models that detect defects through feature and pattern learning. However, acquiring sufficient data is time- and/or resource-intensive, which introduces a challenge to applying machine learning-based anomaly detection. This study proposes a multisource transfer learning method that generates anomaly detection models for balling defect detection, thus ensuring quality monitoring in WAAM. The proposed method uses convolutional neural network models to extract sufficient image features from multisource materials, then transfers and fine-tunes the models for anomaly detection in the target material. Stepwise learning is applied to extract image features sequentially from individual source materials, and composite learning is employed to assign the optimal frozen ratio for converging transferred and present features. Experiments were performed using a gas tungsten arc welding-based WAAM process to validate the classification accuracy of the models using low-carbon steel, stainless steel, and Inconel.

Computer Science,Engineering,Materials Science

What problem does this paper attempt to address?

The paper aims to address the issue of defect detection in the Wire Arc Additive Manufacturing (WAAM) process. WAAM, as a large-scale metal additive manufacturing technology, has garnered attention due to its high deposition rate, cost-effectiveness, and material diversity. However, its non-equilibrium thermal cycles and layer-by-layer stacking mechanism lead to uncertainties in process stability and part quality. Therefore, anomaly detection is crucial for quality monitoring of parts.

Most related studies utilize machine learning methods to detect defects through feature and pattern learning. However, acquiring sufficient data is time-consuming and resource-intensive, posing challenges for machine learning-based anomaly detection. To this end, this paper proposes a multi-source transfer learning method to generate an anomaly detection model specifically for defect detection, ensuring quality monitoring during the WAAM process. The method employs Convolutional Neural Networks (CNN) models to extract image features from various materials and transfers and fine-tunes these models for anomaly detection in the target material. Experiments using the WAAM process based on Gas Tungsten Arc Welding (GTAW) validate the classification accuracy of this method on low-carbon steel, stainless steel, and Inconel materials.