Subdomain Adaptation Order Network for Fault Diagnosis of Brushless DC Motors
Chong Luo, Jianyu Wang, Enrico Zio, Qiang Miao
Abstract
Brushless DC motors (BLDCMs) are widely used in the industrial field, and it is important to diagnose their faults for improving operational reliability. However, existing methods for fault diagnosis of BLDCM face challenges in terms of multiple faults, multiple motor types, and cross-operating conditions. Hence, we propose a subdomain adaptation order network (SAON) to address these challenges. Firstly, a tacholess order tracking method is proposed to transform the phase current of BLDCM from time domain to angular domain to eliminate interference from speed variations. Secondly, an order harmonic extraction method is constructed to reduce the size of data and extract order harmonic features, which are then inputted into a fully connected neural network to form an order neural network (ONN). Finally, local maximum mean discrepancy is utilized to improve the generalization ability of ONN, thus completing the SAON method. Extensive data are collected to validate the proposed method, and the comparison results demonstrate that SAON performs best, with highest accuracy of 96.42%, and has faster convergence speed and good adaptability.