Intelligent model-free diagnosis for multiple faults in a nonlinear dynamic system

In terms of fault diagnosis, there are two general approaches: model-based and model-free. This paper presents the fault diagnosis techniques for a nonlinear dynamic system with multiple faults using the model-free approach. A new concept for fault detection by means of a real-time tracker was employed to predict the system outputs from which the residuals could be quickly generated. To classify faults and determine the degree of each fault, soft computing techniques: fuzzy logic and neural network were used. This study consists of three parts: diagnosis of single faults before the system reaches its steady state, diagnosis of simultaneous multiple faults and diagnosis of sequential multiple faults. A three-tank nonlinear dynamic system was chosen to demonstrate the presented techniques. The result showed promise in using the model-free approach for the diagnosis of multiple faults.