Fault diagnosis of rotating electrical machines based on multi-source electrical signal fusion

With the continuous advancement of Industry 4.0 and intelligent manufacturing technologies, the health monitoring and intelligent diagnosis of motor systems have become critical for ensuring the efficient and safe operation of industrial equipment. As typical representatives of rotating motor systems, three-phase asynchronous motors and synchronous generators exhibit significant differences in structure and operating principles, making traditional fault diagnosis methods-based on a single data source or fixed model-difficult to generalize across different motor types. To address this challenge, this paper proposes a hybrid deep learning fault diagnosis method that fuses multi-source electrical signals, and constructs a unified integrated diagnostic framework named RBGLMHAN (ResNet-BiGRU-CNN-LSTM with Multi-head Attention Network). The model integrates multi-modal feature signals such as three-phase current and voltage, combined with multi-level feature extraction and temporal modeling mechanisms, to accurately identify the operating states of various motors under complex working conditions. This study utilizes two representative datasets: the first is a real-world dataset based on high-voltage three-phase asynchronous motors, covering various pole pairs and rotor bar break types; the second is a synchronous generator fault dataset provided by Tominaga et al including typical electrical faults such as inter-turn short circuits, phase-to-phase short circuits, and ground faults. Experimental results demonstrate that the proposed model achieves excellent fault diagnosis performance on both datasets, with maximum accuracies of 99.78% and 98.96%, respectively-significantly outperforming mainstream models and exhibiting strong robustness and cross-platform adaptability. This study not only offers a unified technical path for intelligent monitoring and predictive maintenance of motor systems, but also provides innovative insights and theoretical support for the application of multi-source data fusion and deep learning in heterogeneous industrial equipment.