Copyright © IFAC System Identification, Kitakyushu, Fukuoka, Japan, 1997 SYSTEM IDENTIFICATION AND RESIDUAL GENERATION FOR ROBUST FAULT DIAGNOSIS D. Sauter , H. Garnier , C. Christophe , M. Mensler Centre de Recherche en Automatique de Nancy CNRS URA 821 Universite Henri Poincare, Nancy 1 BP 239, F-54506 Vandreuvre Cedex, France, sauter@cran. u-nancy.fr Phone : (33) 038391 2037 Abstract : This paper deals with the connection problem of system identification and residual generation for robust Fault Detection and Isolation (FDI). For robustness enhancement against disturbances, a procedure is presented when the de coupling of the fault effect from the system perturbations is not possible. The design is based on the optimization of a performance index expressed in the frequency domain. It is also shown that the estimation of the residual generator parameters must be connected to the design of the residual. A classical frequency domain approach via spectral analysis is used to identify the system for designing the robust fault detection filter. Keywords: Fault detection, Frequency domain identification, Linear systems, Robustness, Spectral analysis. 1. INTRODUCTION The fault detection and isolation problem has received a great deal of attention in the last two decades and a wide variety of model-based ap- proaches has been proposed (Isermann, 1984; Bas- seville , 1988; Frank , 1990). In particular methods which make use of the analytical redundancy via state estimation including techniques based on state observer design (Frank, 1993) or parity rela- tions (Patton and Chen, 1991) have been widely theoretically developed and successfully applied to industry (Isermann and Balle, 1996). Different solutions have been proposed in the liter- ature for robustness achievement when uncertain- ties belong to the structured type (Frank, 1994; Patton and Chen, 1991) and one of the most ac- tive research area in the design of robust residual generator uses frequency domain techniques (Ding et al ., 1993; Qiu and Gertler, 1993). 1123 In this paper , linear systems subject to unknown input disturbances are considered and an ap- proach is developed which considers both system identification and residual generation for robust fault diagnosis. It states the problem of the con- nection between FDI and system identification. In order to identify the system, the classical spectral analysis method is firstly used to estimate a non- parametric model. A parametric model is then ob- tained from the frequency responses of the system, which are used in connection with the information included in the disturbance spectrum estimates to design the robust fault detection filter. The residual generator is composed of the follow- ing two stages. The residuals are first designed so that they are closed to zero in the fault free case and deviate from zero otherwise. The frequency characteristics of transfer matrices related to dis- turbances and fault unknown inputs are then used to ensure the robustness performances. A method leading to the in the frequency domain of Copyright © IFAC System Identification, Kitakyushu, Fukuoka, Japan, 1997 SYSTEM IDENTIFICATION AND RESIDUAL GENERATION FOR ROBUST FAULT DIAGNOSIS D. Sauter , H. Garnier , C. Christophe , M. Mensler Centre de Recherche en Automatique de Nancy CNRS URA 821 Universite Henri Poincare, Nancy 1 BP 239, F-54506 Vandreuvre Cedex, France, sauter@cran. u-nancy.fr Phone : (33) 038391 2037 Abstract : This paper deals with the connection problem of system identification and residual generation for robust Fault Detection and Isolation (FDI). For robustness enhancement against disturbances, a procedure is presented when the de coupling of the fault effect from the system perturbations is not possible. The design is based on the optimization of a performance index expressed in the frequency domain. It is also shown that the estimation of the residual generator parameters must be connected to the design of the residual. A classical frequency domain approach via spectral analysis is used to identify the system for designing the robust fault detection filter. Keywords: Fault detection, Frequency domain identification, Linear systems, Robustness, Spectral analysis. 1. INTRODUCTION The fault detection and isolation problem has received a great deal of attention in the last two decades and a wide variety of model-based ap- proaches has been proposed (Isermann, 1984; Bas- seville , 1988; Frank , 1990). In particular methods which make use of the analytical redundancy via state estimation including techniques based on state observer design (Frank, 1993) or parity rela- tions (Patton and Chen, 1991) have been widely theoretically developed and successfully applied to industry (Isermann and Balle, 1996). Different solutions have been proposed in the liter- ature for robustness achievement when uncertain- ties belong to the structured type (Frank, 1994; Patton and Chen, 1991) and one of the most ac- tive research area in the design of robust residual generator uses frequency domain techniques (Ding et al ., 1993; Qiu and Gertler, 1993). 1123 In this paper , linear systems subject to unknown input disturbances are considered and an ap- proach is developed which considers both system identification and residual generation for robust fault diagnosis. It states the problem of the con- nection between FDI and system identification. In order to identify the system, the classical spectral analysis method is firstly used to estimate a non- parametric model. A parametric model is then ob- tained from the frequency responses of the system, which are used in connection with the information included in the disturbance spectrum estimates to design the robust fault detection filter. The residual generator is composed of the follow- ing two stages. The residuals are first designed so that they are closed to zero in the fault free case and deviate from zero otherwise. The frequency characteristics of transfer matrices related to dis- turbances and fault unknown inputs are then used to ensure the robustness performances. A method leading to the in the frequency domain of