Fault Detection and Identification of PMSM Current Sensor Based on Attention Mechanism and Bidirectional Long Short-Term Memory Network

This study proposes a novel fault detection and identification (FDI) method for permanent magnet synchronous motor (PMSM) current sensors. Specifically, the method integrates an attention mechanism (AM) with a bidirectional long short-term memory (BiLSTM) network to effectively capture temporal dependencies in fault signals and emphasize key features, thereby enhancing the accuracy and robustness of FDI. Furthermore, an early stopping mechanism is employed to prevent overfitting and reduce training time. Experimental comparisons with traditional FDI methods highlight the advantages of the proposed method, including its superior robustness and generalization ability under complex operating conditions. Notably, the method can detect and identify up to five common faults (gain, zero-offset, broken-line, stuck, and intermittent faults) with an accuracy rate of over 99%.