Nonlinear Observer Fault Detection for a Multivariable Process Using a Learning Methodology

Abstract

A fault diagnosis method for nonlinear systems is developed in this paper using a designed nonlinear state observer. In the observer system a neural network is utilized to estimate the possible fault on-line. It is proved that when the nonlinear observer output converges to the system states, the on-line estimator will converge to the time varying faults. In this way, not only that the occurring fault can be detected, the size and waveform of the fault can be estimated to achieve fault identification, which is very useful when the fault tolerant control will be further developed. The developed fault diagnosis method is applied to a continuous stirred tank reactor (CSTR) process with some simulated faults. Simulation results demonstrate the effectiveness of the fault diagnosis method.