Rolling horizon for active fault detection

This paper presents a feasible design of subopti- mal active fault detection system in multiple-model framework. The optimal solution for finite horizon is approximated by means of well known rolling horizon scheme which belongs to the class of limited look-ahead policies. The suboptimal input signal which is chosen from given discrete set is obtained by l- step closed loop optimization. It is shown that such input signal can improve fault detection. I. INTRODUCTION A change detection problem rises in countless appli- cations, from time series analysis to fault detection and isolation (FDI). Primary aim of fault detection is early recognition of a change in behavior of observed system using available measurements. A survey of various methods to the detector design can be found within many of others in (1), (2), (3). More complex methods were proposed with growing availability of faster and cheaper computers (4). The detector design is usually based on a mathematical model of a real process. Such detector passively monitors available measurements and makes decisions. A lot of the approaches consider that the detection consists of two steps. The first step is the generation of residuals which represent inconsistencies between the true, and ex- pected behavior of the observed system and the decisions based on the residuals are made in the second step. A comprehensive review of methods used for residual generator design can be found in (5). The work (6) is aimed to design of both parts of detector and this design is based on statistical approach mainly. In many papers great attention is aimed to robustness problem in the residual generator design (7), (8), (9). On the other side, the problem of system excitation is often either omitted or only the request for appropriate excitation is stated even if change detection methods sensitive to the system excitation are used. Design of a special input signal which can improve system recognition is well known from system identification (10). This idea was consequently applied to change detection prob- lem in multiple-model (MM) framework. In (11) the input signal maximizes the Baram's distance between considered models. More systematical approach was proposed in (12) where the optimal input signal is designed by the appropriate