Data-driven early fault diagnostic methodology of permanent magnet synchronous motor

Abstract

Permanent magnet synchronous motor (PMSM) is one of the common core power components in modern industrial systems. Early fault diagnosis can avoid major accidents and plan maintenance in advance. However, the features of early faults are weak, and the symptoms are not obvious. Meanwhile, the fault signal is often overwhelmed by noise. Accordingly, fault diagnosis for early faults is difficult, and the diagnostic accuracy is generally low. A Bayesian-network-based data-driven early fault diagnostic methodology of PMSM is proposed with vibration and acoustic emission data. The wavelet threshold denoising and minimum entropy deconvolution methods are used to improve the signal-to-noise ratio. The complementary ensemble empirical mode decomposition method is used to extract signal eigenvalues, and Bayesian networks are applied to identify the early, middle, and permanent faults. Experimental data carried out with Tyco ST8N80P100V22E medium PMSM are used to train the fault diagnostic model and validate the proposed fault diagnostic methodology. Result shows that the accuracy for early faults is more than 90% when acoustic emission signal is used, and it is higher than the accuracy with vibration signal. The influence of load on diagnostic accuracy is also investigated, and it indicates that the accuracy with acoustic emission signal is higher than that with vibration signal under different loads.