Design of Strong Detection Filters by Eigenvector Assignment: Theory and Application to Real Time Automotive Failure Diagnosis

Considerable attention has recently been devoted by a number of researchers to the design of Failure Detection Filters using the tools of modern control theory. The original results obtained by Beard [1] and Jones [2] in the early 70's have been revised and expanded by Meserole [3], Chow and Willsky [4], White [5], Massoumnia [6], and more recently by Mi [71The common thread in the various methods proposed is the structure of the detection filter, which is that of a state estimator. Typically, in a detection filter the failure event is modeled as an additive term in the state equation for the error residuals, and the feedback gains are then selected to force a given failure residual to lie in a known, unique direction in output space. Recently, the application of these ideas has proven successful in the case of electronically controlled internal combustion engines [7],[8]49]. Parallel to the development of a theory for the design of detection filters, the results of Moore 112] and others (more recently Kwon and Youn [11]) have focused on eigenstructure gnment for application to the design of feedback controlers. This problem is, in a sense, dual to the detection filter design problem. The present note discusses the improved robustness of a detection filter design based on eigenvector assignment techniques. The main improvement with respect to previous techniques, in particular those developed by Min [7] and White [5], is that the possibility of having different failures produce residuals with the same direction in output space is greatly reduced. In Section 2 we will define the system under consideration. In Section 3 Strong Detectability will be introduced, and in Section 4 a theory wil be developed to support the notion of a strong detection operator. The method is illustrated by way of an application in section 5. The application is based on the design of detection filter which can discriminate between certain plant and component failures.