THRESHOLD ASSESSMENT OF THE STATE OF A TECHNICAL OBJECT BASED ON SEGMENTATION AND IDENTIFICATION OF THE CONTROLLED PARAMETER MODEL

Abstract

To fix the jumps in the average value, a detection method based on the segmentation of the signal under study based on the formation of cumulative sums using the Page-Hinckley criterion is proposed. The use of the Page–Hinckley likelihood criterion makes it possible to detect abrupt changes in the average value of the controlled object parameter in real time under noisy conditions. When using the method, it is assumed that the signal is described by a time series of values of the signal under study. From this series, it is possible to single out separate successive sections, which can be considered as some signal models limited in time. The method is based on the use of criterion statistics, on the basis of which two or three models estimated from different parts of the signal are compared, which makes it possible to detect abrupt changes in the model parameters. The method assumes that a piecewise constant signal with additive noise is considered. At arbitrary moments of time, there are jumps in the average value of this signal. Jumps in the average value of the signal can be different in sign (fixed on different sides of the time axis) and significantly exceed the original value in absolute value. The average value of the signal is a constant value close to zero. But a situation is possible when a repeated jump will be made from a level different from the average value close to zero, both in the direction of increasing and decreasing the average value of the signal and changing the signal polarity (the sign of the signal values). A criterion has been chosen that allows minimizing the delay time in detecting a jump in the average value of the recorded signal with a minimum of false alarms. In this case, segmentation of the signal under study is used based on the formation of cumulative sums using the Page-Hinckley criterion. The use of the Page–Hinckley likelihood criterion makes it possible to detect abrupt changes in the average value of the controlled object parameter in real time under noisy conditions.