An Improved Rail Damage Detection Method Based on Multi-Sensor Data Adaptive Weight Fusion and Mel-spectrogram Roll-off Point Feature

High-speed rail plays a pivotal role in modern transport, and the safety of rails is crucial to preventing accidents and ensuring uninterrupted operation. Methods based on singlesensor acoustic emission signal are commonly used in rail damage detection, but the problems of incomplete information and poor stability seriously affect the accuracy and reliability. Therefore, this paper proposes an adaptive weight multi-sensor data fusion algorithm based on the signal consistency variance and geometric position of the sensors. This innovative approach enhances the confidence level of the acoustic emission signals used for detection. Further, mel-spectrogram roll-off point feature is extracted from the fused signals as the detection metric, and combined with statistical threshold methods for damage detection. Finally, the experiment based on the vehicle-mounted rail damage detection platform proves the effectiveness and superiority of the proposed method.