This paper proposes a dual-modal monitoring system combining visible and infrared imaging to enhance overlap defect detection in wire arc additive manufacturing (WAAM) based on cold metal transfer (CMT) welding for multi-pass builds. Traditional single-modal approaches, primarily relying on melt pool imagery, are often hindered by arc light and spatter interference, which can compromise detection accuracy. In this work, overlap defect refers to insufficient overlap between adjacent tracks, and the dataset is created by inducing overlap defects through inter-track spacing in multi-pass deposition. The proposed dual-modal strategy mitigates these challenges and significantly improves detection precision. A dual-input convolutional neural network model named Multimodal Mutual Fusion Network (MMFNet) was designed, fusing visible and infrared data at the feature level to achieve a prediction accuracy of 98.34%. Comparative experiments with single-modal models demonstrate the superiority of the proposed approach, with single-modal accuracies of only 95.76% (infrared) and 92.85% (visible light). The proposed system provides a robust solution for monitoring of overlap defects in WAAM in the studied multi-pass setting, highlighting the potential of dual-modal systems for improving quality control in additive manufacturing processes.
Wang et al. (Fri,) studied this question.