Melt Pool Monitoring in Arc Additive Manufacturing Using Multi-branch Residual Networks

Wire arc additive manufacturing (WAAM) is a promising technology in aerospace, automotive, and marine industries due to its high efficiency and cost-effectiveness. However, the dynamic behavior of the melt pool significantly affects the forming quality and requires real-time and accurate monitoring. In this paper, a novel multi-branch residual network for melt pool monitoring is proposed, which addresses the limitations of traditional methods and single-branch networks. The multi-branch structure effectively extracts multi-scale features, while the residual module enhances feature fusion and training stability. The lightweight architecture ensures real-time processing in industrial scenarios. Experiments on a self-built dataset of 6,000 images of three melt pool states demonstrate the superior performance of the proposed method, achieving 95.1% accuracy, 94.2% precision, and 13.1 milliseconds inference time. Comparative and ablation studies confirm the significant contributions of the multi-branch structure and residual module, enabling robust and efficient monitoring under complex conditions.