Real-time fault diagnosis of vehicle control motors for variable operating conditions based on acoustic vibration fusion

For existing offline real-time fault diagnosis systems for vehicle control motors, efficient and accurate fault detection remains challenging due to weak vibration fault signals under varying working conditions, significant noise interference, and high model complexity. This paper proposes a fault diagnosis method based on the acoustic-vibration signal fusion. The diagnostic framework comprises four modules: raw signal acquisition, data preprocessing, multi-sensor signal fusion, and network-based diagnosis. First, sound and vibration signals are collected using an NVH test rig. The signals are then decomposed and reconstructed using Pearson correlation coefficient filtering, with parameters of the Variational Mode Decomposition (VMD) optimized by the Gray Wolf Optimization (GWO) algorithm. Next, the reconstructed signals undergo continuous wavelet transform (CWT) to produce time-frequency representations, and a multimodal feature fusion network is employed to integrate the sound and vibration signals. Finally, fault detection and classification are achieved by training an enhanced MobileNetV2 network. Experimental results show that this method outperforms single-sensor approaches and other comparative techniques in diagnostic accuracy under variable operating conditions. Furthermore, the total processing time for offline inspection is less than 30 seconds, meeting the real-time diagnostic requirements for automotive control motors.