Answers are grounded in this paper and cite their sources.
Accurate recognition of weld states is essential for ensuring structural integrity in cold metal transfer (CMT) welding. However, it remains challenging to develop models that are both robust to industrial disturbances and physically interpretable. In addition, the inherent opacity of current data-driven methods often limits their reliability in complex manufacturing environments. This study proposes ASPIS-Net, a robust physics-informed network for interpretable CMT weld-state monitoring. The proposed network adopts a dual-branch architecture with an adaptive gated fusion mechanism, which dynamically suppresses environmental noise while preserving critical acoustic and electrical features. To improve model transparency, a symbolic regression framework is further introduced to convert the learned representations into explicit physical equations, thereby aligning deep-learning predictions with the underlying welding thermodynamics. The proposed method was validated using an automated CMT welding platform, from which 5,000 synchronised acoustic and electrical signal samples were collected under four typical penetration states. The results show that ASPIS-Net achieves a classification accuracy of 99.36% and retains a Macro-F1 score of 75.59% under the 0 dB composite-noise condition. This work provides a transparent, physics-guided approach for monitoring multi-physics manufacturing processes, establishing a rational basis for assessing structural reliability in critical welded joints.
No takes yet. Share an insight, caveat, or question.
Zhang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: