Problems of Software Detection of Periodic Features in a Time Series for Diagnosis

Abstract

Problems arise when attempts are made to extract automatically, visually obvious periodic features indicative of defects in a vibration time series for diagnosis using computers. Such problems may be interpretational in nature arising either from insufficient knowledge of the mechanism, or the convolution of the source signal with the system structure effect. Some of the periodic structures may turn out to be interferences originating from the cyclic nature of events taking place in the mechanism. Such periodic interferences complicate considerably the selection of suitable parameter values to be used in the period extracting algorithms. There is also the problem of deciding which of the multiple peaks or nulls resulting from impact instabilities in the time series is the true peak or null to be used in estimating the impact period. These problems must be solved if the extracted period is to be used successfully for source identification of defective components. In this paper, we present the ways we have developed to addressing these problem.