Actuator fault diagnosis with neural network-integral sliding mode based unknown input observers

This paper proposes an integral sliding mode (ISM) based unknown input observer (UIO) which is able to perform fault diagnosis (FD) in condition of lack of knowledge of the plant model. In particular, a two-layer neural network (NN) is employed to estimate online the drift term of the system dynamics needed to compute the so-called integral sliding manifold. The weights of such a NN are updated online using adaptation laws directly derived from theoretical analysis, carried out in this paper. Finally, the proposal has been assessed in simulation relying on a benchmark model of a DC motor.