Active mode recognition of dynamic systems

Abstract

The purpose of detecting changes in the operating regime of a system is to identify abrupt changes from one regime to another. This paper proposes a new detection method which can detect regime change instants in sequential time series data. Our approach is based on a sensitivity study of a global model combining, with a multiplicative effect, local models describing the different operating modes without knowing their parameters. More precisely, the operating mode change detection indicator corresponds to the direction of the gradient of the global model of the system (i.e. including all its operating modes) with respect to the variables characterising the system. The proposed procedure is illustrated by the analysis of a dynamic system with several measurable inputs and outputs, which makes it possible to consider auto- and inter-redundancy relationships between inputs and outputs.