Anomaly detection for equipment condition via cross-correlation approximate entropy

Abstract

In this paper, a new method named cross-correlation approximate entropy is proposed based on the correlation analysis and the approximate entropy theory. It can detect anomaly of running state in a quantitative manner without any priori knowledge. The method takes a section of signal with fixed-length of running state of equipment as a window. By sliding the window through the state signal, the paper calculates the cross-correlation function of the first window and latter ones, and then figure out their approximate entropy values. This paper sets the approximate entropy value of cross-correlation function of the first and second windows as the standard value. If there is an anomaly, the approximate entropy value of cross-correlation function of windows will be far larger than the standard value. Finally, a case is studied to test the validity and stability of this method by using the normal vibration signals of normal and faulty rolling bearing.