Unsupervised Fault Detection of Brushless DC Motors Under Multioperating Conditions via Order Hybrid Interpolated Gaussian Descriptor

Brushless dc motors (BLDCMs) are widely used in the industrial field, and their safe and reliable operation is crucial. Unsupervised fault detection methods only require normal operation data for training, making it highly practical. However, existing methods for fault detection in BLDCMs are susceptible to multioperating conditions, leading to the low fault detection performance. In this case, we propose an unsupervised fault detection method, order hybrid interpolated Gaussian descriptor (OHIGD), to address this problem. First, we construct a preprocessing method for BLDCM current according to stator current operation compensation to mitigate the influence of operating conditions. Second, an order hybrid autoencoder (OHAE) is constructed based on fully connected neural network and convolutional neural network to extract fault features from the processed data. Finally, OHIGD is constructed based on OHAE to enhance its performance, according to the state-of-the-art interpolated Gaussian descriptor. The effectiveness of the proposed method is validated using current data of the BLDCMs under multioperating conditions.