Fault Diagnosis for a Class of Nonlinear Systems by means of a Polynomial Observer J. L. Mata-Machuca, R. Mart´ ınez-Guerra, D. Hern´ andez-S´ anchez Department of Automatic Control, CINVESTAV,IPN, Mexico D.F., Mexico Phone (+52)555-747-3800 ext.4222 Fax (+52)555-747-3982 E-mail:jmata@ctrl.cinvestav.mx Abstract— This paper deals about the problem of fault diagnosis of a class of nonlinear systems using the theory of state observers. In order to reconstruct the faults of the system, a polynomial observer is proposed, which includes in its structure correction terms of high order. The methodology consists of add to the original system the faults as new states, which increase the order of the observer. This scheme reconstructs simultaneously faults and state variables. In addition, as a comparative study, it has been designed an observer of reduced order. Both techniques are applied to fault diagnosis of a three- tank system. Index Terms— fault diagnosis, observers, nonlinear system. I. INTRODUCTION A fault is a not allowed deviation of at least one charac- teristic property or parameter of the system with respect to its usual, normal or acceptable condition. In the last three decades a considerable number of papers related to the problem of fault diagnosis have been reported [1]–[4]. Among the publications around the diagnosis of faults have different approaches, for example, in [5]–[7] used differential geometric tools; papers [8]–[10] are based on an alternative method using differential algebraic techniques. This work considers the problem of fault diagnosis of a class of nonlinear systems. The outputs are mainly obtained by sensors measurements. The outputs and the number of faults determine if the system is diagnosable or not [11], [12]. The fault diagnosis problem is considered as a problem of fault signals observation. In this sense, the diagnosability of a system is given by the algebraic observability condition of the fault [8]. The main contribution of this article consists of the solution of the fault diagnosis problem by means of a polynomial observer. This work combines the ideas of [13] on the design of observers for nonlinear Lipschitz systems with the method introduced in [11], adding high order correction terms 1 , such that the asymptotic convergence of the observer is guaranteed. In addition, is designed an observer of reduced order based on the free model approach, which converges asymptotically. The proposed schemes to the fault diagnosis and reconstructions of states are applied to a three-tank system. 1 As it is well known, an Extended Luenberger observer can be seen as a first order Taylor series around the estimated state, therefore to improve the estimation performance high order terms are included in the observer structure. II. PROBLEM STATEMENT Let us consider the following nonlinear system: ˙ x(t)= g(x, u, f ) y(t)= h(x, u) (1) where x R n is the state vector, u R l is the known input vector, f =(f 1 ,f 2 , ..., f μ ) R μ is the faults vector (unknown inputs), and y R p is the output vector. With g,h analytic functions. The fault vector f is unknown, which can be interpreted as a state with uncertainties. The faults estimation is obtained by an extended state. Consider the following extended system: ˙ x(t)= g(x, u, f ) ˙ f j (t)= Ω j (x, u, f ) , 1 j μ y(t)= h(x, u) (2) where Ω = (Ω 1 , Ω 2 ,..., Ω μ ) R μ is an unknown bounded function, that is to say, Ω(x, u, f )≤ N< (3) The system (2) can be expressed in the following form, ˙ x(t)= Ax +Ψ(x, u, f ) ˙ f j (t)= Ω j (x, u, f ) , 1 j μ y(t)= Cx (4) where Ψ(x, u, f ) is a nonlinear function that satisfies the Lipschitz condition, i.e., Ψ(x, u, f ) - Ψ(ˆ x, u, f )≤ Lx - ˆ x(5) III. OBSERVABILITY AND DIAGNOSABILITY: DIFFERENTIAL ALGEBRAIC APPROACH Before proposing an observer for the extended system (4), in this section are presented some definitions about observability and diagnosability. III-A. Definitions The observability notion (diagosability) of a system, linear or nonlinear, consists of the possibility of reconstructing the state x (fault f ), having the knowledge of the output y and, the input u, and possibly a finite number of their derivatives, (y (k) , k 0 and u (l) (t), l 0). Next some definitions concerning differential algebra [10], [11], [14] appear. Definition 1: Let L and K be differentials fields. A dif- ferential field extension L/K is given by two differentials fields K, L, such that: 2010 7th International Conference on Electrical Engineering, Computing Science and Automatic Control (CCE 2010) Tuxtla Gutiérrez, Chiapas, México. September 8-10, 2010. IEEE Catalog Number: CFP10827-ART ISBN: 978-1-4244-7314-4 978-1-4244-7314-4/10/$26.00 ©2010 IEEE 158