A Case Study for a Turbogenerator Accident Using Multiscale Association

Abstract

This paper presents a novel method of multiscale association for analyzing a turbogenerator accident having strange behaviors and serious consequence. Wave index (WI) and credibility of sensor fault are proposed based on multiscale analysis of the recorded data, and then the associational degree of WI is used to detect sensor fault. In addition, mechanism models are built to verify that detection. Furthermore, maximum likelihood method and neural network are applied to estimate the confidence interval of the fault sensor and the true signal. The estimation has been used to clearly explain the cause of this accident.