High-Level Feature Extraction based on Correlogram for State Monitoring of Rotating Machinery with Vibration Signals

Abstract

Vibration analysis is one of the most popular methods for state monitoring of rotating machines, and feature extraction is of much importance in the design of the monitoring system. In this paper, a new high-level feature extraction method based on correlograms for vibration signal analysis is proposed, and it includes two phases. Firstly, in the learning process, a codebook is created from training data using the k-means algorithm. Next, in the testing process, for a given data stream collected from a monitoring rotating machine, the correlogram in each cycle is obtained by comparing every data point with all codewords in the codebook at first; the entropy is then computed to form final high-level features to measure the state of the machine. A change decision can be made finally based on features extracted from null hypothesis testing. Based on an experimental setup used in our previous work, the proposed method is evaluated with application to the speed change monitoring of a rotating machine. Experimental results demonstrate the excellent performance and the priority of the method compared with ten typical features.