Convolutional Neural Networks Based on Resonance Demodulation of Vibration Signal for Rolling Bearing Fault Diagnosis in Permanent Magnet Synchronous Motors

Abstract

Permanent magnet synchronous motors (PMSMs) are widely used due to their unique advantages. Their transmission system mainly relies on rolling bearings; therefore, monitoring the motor’s working status and fault diagnosis for the rolling bearings are the key focuses. Traditional resonance demodulation methods analyze the vibration signals of bearings to achieve bearing fault diagnosis, but the limiting condition is that the inherent frequency needs to be known. Based on the resonance demodulation method, deep learning methods, such as the convolutional neural network (CNN) model designed in this article, have improved the practicality and effectiveness of diagnosis. A physical explanation of the deep learning model for bearing fault diagnosis is presented in this article, the relationship between resonance demodulation and the 1D CNN is analyzed, and the model is trained and validated. The experimental results show that the CNN model can identify different types of bearing faults. The analysis results of the trained CNN model and the intermediate results indicate that the CNN model is consistent with the resonance demodulation method. The optimized method is verified, proving that the model can achieve the classification and diagnosis of fault bearing data collected under different environments after the optimized training method is adopted.