Fault diagnosis based on Grey-box Neural Network identification model

This paper presents a fault diagnosis (FD) scheme for a class of nonlinear dynamic systems using a novel Grey-Box Neural Network Model (GBNNM). In this GBNNM, a composite structure, including MLP (multi-layer perception) NN (Neural Network) and integer term, is proposed to approximate both nonlinearity and dynamics of object system. Its approximation ability is then proved theoretically. And a self-defined exciting strategy is introduced into NN training to improve NN's generalization ability. Unlike previous NN model based fault diagnosis methods, a quantitative residual, which is obtained from system output and its GBNNM model output, can accurately indicates inconsistency caused by fault, so the improved residual is not essential for our scheme. The proposed FD scheme is applied in a high-fidelity Reaction Wheel (RW) in Satellite Attitude Control System (SACS) in our case study. The results of the case study demonstrate the effectiveness and superiority of our FD scheme.