Abrupt change detection algorithm: from theory to implementation

Summary: Detection and segmentation of a nonstationary process consist of assuming piecewise stationarity and detecting the instants of change. In some systems, faults produce abrupt changes in signals issued from this system. Early detection of this change reduces the possibility of catastrophic damage of this system. The present work is based on the use of Wavelet Packet (WP) decomposition for abrupt change detection. This aims also to real time implementation of this approach. The first step is to select the only WP of the decomposition tree that is able to highlight changes in the recording. The second step is to find the filters bank related to relevant packet for real time decomposition. The final step is to apply an algorithm of detection like Dynamic Cumulative Sum (DCS) to detect the changes in the packets. The association between wavelet packet, bank filters and DCS algorithm allow to real time detection and segmentation of any frequency change in this signal