Operating Condition Generalization Network for Fault Diagnosis of Brushless DC Motors
Chong Luo, Jianyu Wang, Enrico Zio, Qiang Miao
Abstract
Existing fault diagnosis methods for brushless dc motors (BLDCMs) encounter problems of limited adaptability in terms of multiple faults, multiple motor types, and cross-operating conditions. Therefore, we propose an operating condition generalization network (OCGN) based on the recently proposed domain-invariant feature exploration (DIFEX) method to address it. First, a tacholess order tracking method is constructed to alleviate the influence of rotating speed variation on the line current. Second, order harmonic features are extracted from the angular line current and inputted into a fully connected neural network to create an order neural network (ONN). Finally, ONN is used as the teacher network in the knowledge distillation framework of DIFEX, and the student network is improved with a Gaussian random projection layer. Based on extensive fault data of BLDCMs, OCGN is compared with other state-of-the-art methods and confirmed to have superior performance.