Remote detection of abnormal behavior in mechanical systems

Document Type

Conference Proceeding

Publication Date

1-1-2019

Abstract

© 2019, Society for Experimental Mechanics, Inc. Machinery can undergo many different types of operational faults during use. Mitigating these faults through scheduled maintenance programs or upon failure is costly, inefficient, and can introduce safety hazards. An alternative approach to identifying abnormal behavior in mechanical systems is through condition based monitoring, which is applied by taking measurements of machinery and using spectral analysis to diagnose faults. The signals produced by machines contain characteristic frequencies describing the operating state. The application of local sensors to monitor the health of mechanical systems is often expensive or not feasible; it is of interest to find methods of remote sensing to allow for detection of abnormalities in mechanical systems without the need for sensors at the local point of operation. This work intends to utilize the method of remote detection to study the diminishment of fault feature observability as a function of sensor location relative to operating equipment. In this study, the procedure used for fault detection consists of the following: (1) collect signals for rotating machinery operating in a healthy, faulty, and off state (2) identify features that are unique to a faulty operating state (3) use those features to identify faulty behavior. This procedure is performed at three locations. For each location, the observable feature for a healthy, faulty and motor off state is calculated and compared using the statistical mean, standard deviation, p-value and Bhattacharyya distance. The relationship between fault observability and distance is presented as a function of Bhattacharyya distance and classified based on a one-way ANOVA test. Machine fault diagnosis becomes more difficult as the measurement location increases. At greater distances, there is no statistical significance between healthy and faulty machine operation. Using a predictable frequency response, machine faults can be correctly identified up to a certain threshold determined by environment noise.

Publication Title

Conference Proceedings of the Society for Experimental Mechanics Series

Share

COinS