Closed-loop performance diagnosis using prediction error identification

Abstract

This paper presents a methodology to detect the origin of closed-loop performance degradation of model-based control systems. The approach exploits the statistical hypothesis testing framework. The decision rule consists of examining if an identified model of the true system lies in a set containing all models that fulfill the closed-loop performance requirements. This allows us to determine whether performance degradation arises from changes in system dynamics or from variations in disturbance characteristics. The probability of making an erroneous decision is estimated a posteriori using the known distribution of the identified model with respect to the unknown true system.