Diagnosis and prognosis of in-service electric machine in the absence of historic data related to faults and faults progression

Abstract

Extensive work has been presented in the literature related to fault diagnosis and prognosis of machines and related components. Prime focus of the proposed techniques is on either on assembly line checkout of machines or newly installed machines as a large number of methods are based on supervised learning. In this paper, fault diagnosis algorithm of in-service DC starter motor is presented. The proposed approach encompasses on the development of predefined fault progression curves. Features to develop these curves are extracted from machine current in time frequency domain. According to the proposed method, a number of curves are developed each of different order and slope. As the machine fault progresses, the fault features are projected on these curves and the % fault severity is identified. The results are presented and conclusions are made.