Improved high resolution estimation approach for rotor fault diagnosis

Abstract

An improved high resolution fault diagnostic approach is presented, which contains two parts. First, a new time-frequency analysis named self-commissioning short time zoom matrix pencil method has been proposed. The analysis has high computing efficiency and maintains high spectral resolution, and it is completely free of human maneuver in dealing with nonstationary and quasi-stationary operating conditions. Vast estimates can be generated from a very finite length of measured data. Second, a feature extraction analysis is presented to settle the common problem of high resolution spectral approaches in quasi-stationary operating condition, where noise, fluctuations of loads and rotor speed contribute to judgement discrepancy or failure in diagnosis results. The feature extraction analysis provides approximately unbiased fault-related frequencies and sideband amplitudes by introducing Monte Carlo method and specific distribution fitting. The entire approach has been validated on interior permanent magnet motors with different degrees of rotor eccentricity in experiments.