Abstract
Rotating machine failures are a major cause of downtime in a wide variety of industrial processes and are a burden on maintenance personnel and facilities. Some of these failures occur suddenly and are seemingly unpredictable. However, the overwhelming majority develop slowly over time and produce characteristic warning signs. A system capable of detecting and diagnosing these incipient faults before they become critical would significantly reduce downtime and serve to facilitate maintenance and repair of these machines. The ability to accurately distinguish between different types of incipient faults would be a critical aspect of such a system. In this research, a model-based method for diagnosing motor faults is examined and tested using two squirrel-cage AC induction motors with staged fault conditions. The proposed method involves the multi-resolution signal analysis of the current residuals. These residuals are generated by comparing the measured motor current with the current predicted by a recurrent neural network. The frequency content of the distortion of the residuals is used to identify the type of fault present. Although "steady-state" conditions are examined exclusively in this research, the nonstationarities of the current signals are sufficient to warrant the use of multi-resolution analysis. The fault diagnosis system is tested using data taken from an 800 hp motor and a 3 hp motor. The method is successful in identifying residual distortion in the frequency range expected for broken-bar faults. Because the magnitude of the distortion grows with increasing fault severity, the method is also useful for evaluating fault severity for broken-bar faults. However, the current distortions caused by rotor eccentricities and damaged bearings are too small to be identified in a statistically significant manner using this approach. Nevertheless, this research demonstrates the feasibility of a general method by which the characteristic frequencies produced by a particular type of fault can be identified in the output of a system.
McFatter, Justin Robert (2002). Discrimination among mechanical fault types in induction motors using electrical measurements. Master's thesis, Texas A&M University. Available electronically from
https : / /hdl .handle .net /1969 .1 /ETD -TAMU -2002 -THESIS -M3832.