Dynamic Neural Network for Detection and Identification Actuator and Sensor Faults

This paper proposes a fault detection and identification method for a particular class of nonlinear flat system using a dynamic higher order sliding mode and a Neural Network. The basic idea consists in calculating the residual using a comparing result between the nominal input and the estimated one using the measured flat output and their successive derivatives. These last ones are estimated by applying a sliding mode differentiator with dynamic gains for taking the measurement noises into account. Then, the computed residual is evaluated to detect the fault. In order to identify in real time, the fault kind (actuator or sensor), a supervised dynamic neural network algorithm is defined. The performances of the proposed method are shown through some simulation tests on the well-known three-tank system.