Rotor Broken Bar Fault Diagnosis Method Based on Spectral Attention Mechanism

To solve the problems of concealed early features of rotor broken bar faults in induction motors, difficulties in extracting fault features under variable-speed conditions, as well as the over-reliance of existing methods on specific working conditions, weak anti-interference capability and insufficient real-time performance, we propose a fault diagnosis method based on dual-path convolution with spectral attention mechanism (DCS) and bidirectional gated recurrent unit (DCS-BiGRU). Firstly, the fault current signals are preprocessed through DC component removal, notch filtering and band-pass filtering to enhance the fault-related frequency components. Secondly, DCS is used to extract the instantaneous spectral and local waveform features of the signals, combined with the spectral attention mechanism (SAM) to dynamically increase the weights of key fault features. Finally, BiGRU is employed to capture the long-term dependencies of feature sequences, thereby improving the model’s adaptability to speed fluctuations. Experimental results demonstrates that the DCS-BiGRU model has strong generalization ability and robustness. It can effectively achieve accurate diagnosis of rotor broken bar faults under variable-speed conditions, satisfying the requirements of real-time monitoring in industrial sites.