Robust parametric fault detection and isolation for nonlinear systems

Although most practical failure situations require nonlinear modeling, quite often the nonlinearity associated with the fault function is partially known based on past experience of plant technicians, with the uncertainty arising due to unknown parameters. Motivated by such practical considerations, this paper presents a robust fault diagnosis scheme for parametric faults in nonlinear dynamical systems. A detection and approximation observer is used for online health monitoring. Once a fault is detected, a bank of nonlinear adaptive observers are activated for the purpose of fault isolation. A key contribution is the derivation of robust isolability conditions for the class of nonlinear systems under consideration. The effectiveness of the proposed methodology is illustrated by simulations on a benchmark example of a three-tank system