An Adaptive Threshold Based on RBF Neural Network for Fault Detection of a Nonlinear System

Abstract

Considering the drawback of the big error when using fixed threshold in fault detection, this paper presented a practical approach to combine RBF-based fault detection observer with an adaptive threshold. Firstly, a fault detection observer based on RBF neural network was applied to generate a residual error signal. Secondly, several key factors in adaptive threshold model were outlined, such as modeling error, random disturbance, input instructions, system status and etc. In order to avoid the above errors, the observer model was modified to combine with an adaptive threshold based on RBF neural network as well. Finally, the suitability of the proposed technique was illustrated through its application to the condition monitoring of an E-cabin temperature control system. It is very effective to adaptively adjust the fault threshold according to a variety of influencing factors.