Application of Machine Learning for Analysis of Static Eccentricity Fault in IPMSM using Finite Element Method

Abstract

Interior Permanent Magnet Synchronous Motor (IPMSM) is becoming good competitor of other electrical machines in both industrial and traction applications because of their better performance characteristics and high power density. Among all the faults occurring in IPMSM, the airgap eccentricity fault accounts nearly 5 to 10 percent. In motor fault diagnosis, the fault detection at early moment (incipient) is a challenging task. The analysis of static eccentricity fault in IPMSM is carried out in ANSYS Maxwell FEM (finite element modeling) platform. A 0.55 kW IPMSM is considered for the analysis of different static eccentricity fault severities from 10 to 40 percent. Electromagnetic parameters like stator phase current, speed, distribution of flux density over the machine and flux density around radial airgap are examined for normal and faulty conditions. Additional harmonics are added to the FFT spectrum (spatial) of flux density around airgap (radial component) based on severity of fault and asymmetry found in the flux density distribution over IPMSM. The data of stator phase current and airgap flux density (radial component) are obtained from ANSYS Maxwell FEA tool and analyzed using machine learning and in MATLAB. In machine learning Fine k - NN algorithm achieved 96.3 percent accuracy for predicting static eccentricity fault based on stator phase current data at incipient stage.