Laser-Induced Noise Recognition for Underwater Laser Paint Removal Process Monitoring via Physically Interpretable Deep Learning

This paper investigates acoustic monitoring for underwater laser paint removal from metal surfaces. To achieve reliable cleaning-state recognition under complex underwater noise conditions, a physics-guided deep neural network is developed to classify acoustic emissions into paint-removal dominant (PRD) and substrate-ablation dominant (SAD) states. A physics-guided acoustic frontend extracts informative representations from experimentally collected signals, which are learned by the backbone network for state assessment. The predicted state can serve as feedback for timely laser termination, reducing excessive heat accumulation and substrate damage. Experiments on real sampled data and low-SNR tests validate the effectiveness and robustness of the proposed framework for intelligent underwater sensing and state-aware control.