Lightweight Fault Diagnosis Method for Railway Point Machines Based on Acoustic Signals

As a critical electromechanical component of railway signaling systems, railway point machines exhibit a relatively high failure rate among all system components. Acoustic signals offer distinct advantages for fault diagnosis, including noncontact measurement, ease of acquisition, and direct representation of the railway point machine's operational processes. Current diagnostic approaches typically extract time-domain, frequency-domain, or time-frequency-domain features for classifier-based diagnosis, yet struggle to achieve an optimal balance between model parameter efficiency and diagnostic accuracy. To address these limitations, this study proposes a lightweight diagnostic framework based on acoustic signal visualization. The methodology comprises three key innovations: (1) transformation of acoustic signals into Mel spectrograms using short-time Fourier transform (STFT) and Mel-frequency filter banks, (2) integration of grouped convolutions into MobileNetV2 architecture to reduce parameter count, and (3) incorporation of an enhanced squeeze-and-excitation (SE) attention mechanism to improve recognition accuracy. Experimental validation was conducted using 844 acoustic samples across seven operational conditions. Results demonstrate a fault diagnosis accuracy rate of 97.66 % with only 1.14 million parameters. Compared to existing methods, the proposed approach achieves superior parameter efficiency while maintaining high diagnostic accuracy, offering significant advantages for railway point machine fault diagnosis applications.