The Research Of Fault Detection And Diagnosis Based On Neural Network Observer

Nowadays, the modern control systems are becoming larger and more complicated, so the possibility of fault arising in system is completely increasing. In order to improve the security and reliability of the system, detect the breakdowns in real time, analysis the cause and the features of it, so as to prevent the trouble occurred. This research based on neural network observers for fault diagnosis. The main content studied in this dissertation is as follow:Firstly, the dissertation introduces the theory of fault detection and method, combined with years of the fault diagnosis technique development and the present situation, proposed the key work is researched in this paper. Furthermore, introduces the improved algorithm of BP network which proposed in this paper. The simulation proves that this network have a good ability of approximation to any nonlinear functions.Secondly, based on the BP algorithm of the neural-network observer is designed, and the weights updating mechanism is used in the nonlinear system for guarantee the stability of the state observer. Moreover, a fault diagnosis scheme is proposed for a linear discrete system based on BP network observer approach, and the observer can estimate the state variables in order to determine whether failure. The effectiveness and feasibility were verified with the simulation examples.To carry out sensor fault diagnosis further, a method for the nonlinear system based on neural-network observer is established, which the method of detection is similar with observer based on the model. But the former diagnose faults by the ability of approximation with neural networks. The simulation demonstrates the proposed methodology is effective.