Three-Phase Asynchronous Motor Fault Diagnosis Using Attention Mechanism and Hybrid CNN-MLP By Multi-Sensor Information

Abstract

Single-signal-driven fault diagnosis has been widely applied in motor fault diagnosis, but it cannot meet the diagnostic requirements of complex motor systems. This study proposes a motor fault diagnosis method using attention mechanism (AM) and hybrid CNN-MLP by multi-sensor information. Firstly, Fast Fourier transform and continuous wavelet transform are performed on different signals to obtain the corresponding frequency domain feature information and wavelet time-frequency map images. A hybrid CNN-MLPAM model is used to extract features from spectral feature information and wavelet time-frequency images, respectively, and is trained to obtain preliminary classification results. Finally, a dynamic weight distribution vector is used to obtain the final diagnosis. The proposed method is verified by current, and vibration signals. The results show that the method can dynamically evaluate the sensitivity of different detection signals to different faults. The proposed method is more accurate and stable in fault diagnosis than the traditional method that relies solely on vibration signals. Under the consideration of time cost and diagnostic accuracy, the proposed CNN-MLPAM has higher diagnostic performance compared with CNN-RNNAM and CNN-ELMAM.

Keywords

• Fault diagnosis of motor
• deep learning
• attention mechanism
• multi-sensor information