Fault Diagnosis Based On HBF Neural Network Observer

With the development of the control system,the scale has gradually become larger and more complicated,and the probability of failure of the system during operation is increasing.Therefore,the security problem during the operation of the system has become a hot issue in the current research.In order to improve the safety of the system operation,it is necessary to detect the fault in time when the fault occurs and analyze and make decisions on the fault characteristics to prevent further losses.In this paper,based on the traditional observer fault diagnosis method,combined with HBF neural network and HBF neural network learning algorithm is studied and improved,which verifies that the improved algorithm optimization HBF neural network has the advantages of fast convergence speed and high accuracy in nonlinear system tracking.Finally,a fault diagnosis method based on HBF neural network observer is proposed and its stability is proved.The main research work is as follows:Firstly,the theory and knowledge related to fault diagnosis and neural network are neural network are analyzed,and HBF neural network is introduced.Since there is no comprehensive analysis on the performance of HBF neural network in the existing literature,this paper verifies that HBF neural network has the advantages of fast convergence speed and small error in nonlinear tracking through the analysis of local error and global error.Secondly,two learning algorithms for HBF neural networks are introduced.First,a new PSO algorithm is proposed to improve the inertia weight of the traditional PSO algorithm,and it was proposed to improve the accuracy of the trained network in nonlinear tracking.Secondly,a modified gradient descent method based on traditional gradient descent method is proposed,which makes the convergence speed faster in the initial operation of the algorithm.Simulation results show that the tow methods can further improve the approximation ability of neural network to nonlinear system,greatly reduce the approximation error and improve the training speed.In addition,the growth and purning algorithm is analyzed,which can be used to train HBF neural network dynamically to get a compact structure.The mathematical expression of neuron contribution in the algorithm is derived under different probability density distributions,and the gaussian mixture probability model is used to make the algoriyhm adapt to various probability density distributions of input samples,which reduces the computation and improves the practicability of the algorithm.Finally,HBF neural network and the proposed algorithm optimized HBF neural network are designed into an observer for fault diagnosis.Simulation shows that this method can overcome the system disturbance in time,detect the fault quickly and accurately and keep to with the changes of the system.