Faulty Diagnoses of PMSM in Flywheel Energy Storage Based on Phase Current Signal and Convolutional Neural Network

Flywheel energy storage system, as a high-efficiency physical energy storage method, has superior performance in the field of regenerative braking for urban rail vehicles. As an energy conversion device with wide speed range, high efficiency and high power density, the permanent magnet synchronous motor (PMSM) is more suitable for application in flywheel energy storage system. However, a disadvantage of PMSM is that the permanent magnets on the motor rotor may generate demagnetization failure under severe operating conditions, which will further lead to other safety hazards. Therefore, the diagnosis of PMSM demagnetization faults is crucial for the safe operation of flywheel energy storage systems. Traditional fault diagnosis methods mainly rely on manual extraction of signal features and combine with machine learning for fault classification, while the drawback of this method is that it relies too much on expert knowledge and diagnostic experience. Therefore, an automatic diagnosis method based on deep learning is proposed in this paper. The method first converts the three-phase current data of PMSM into an image signal using a Gaussian mixture model (GMM) and trains a convolutional neural network (CNN) to achieve demagnetization fault detection. To verify the effectiveness of the proposed model, three PMSMs current data with different health conditions are used for training the model in this paper, and the trained model is also compared with a one-dimensional convolutional neural network (1D CNN). The experimental results show that the proposed method has better fault diagnosis accuracy than the 1D CNN.