Data-Driven Physics-Informed Fusion for Clothing Material Identification in Washing Machines

To meet the demand for refined laundry care in intelligent washing machines and address the low accuracy, poor robustness, and lack of physical interpretability of existing material recognition technologies, a recognition method integrating physical prior knowledge is proposed. Based on a physical experimental platform for drum washing machines, mechanical vibration signals from a three-axis acceleration sensor and motor electromagnetic signals are collected synchronously, a dataset consisting of soft and hard loads is constructed, and time-domain alignment of heterogeneous signals is realized using adaptive pooling technology. Combined with the mechatronic coupling mechanism in the loosening, deviation detection, and weighing stages of washing machines, a Physics-Aware Dual-Stream Multi-Scale Temporal Convolutional Network (PSA-DSMS-TCN) is designed. The network extracts mechanical and electromagnetic features in parallel through a dual-stream structure, expands the receptive field using multi-scale dilated convolution, and introduces an operating condition-gated attention mechanism to achieve dynamic feature fusion. The results of 5-fold cross-validation show that the model achieves an average recognition accuracy of 94.05%, with consistent performance enhancement and substantial practical robustness. The results demonstrate that the PSA-DSMS-TCN effectively improves the precision of material prediction while maintaining lightweight characteristics, providing reliable technical support for the intelligent matching of laundry care parameters.