Study On Actuator Fault Diagnosis Of Nonlinear Systems

The requirement of modern industrial control process to security and reliability is increasing, while the complex control system fails inevitably. How to detect and diagnose faults of the system quickly is not only the guarantee of system stability but also the basic requirement of system operating safely and economically. Actuator faults of nonlinear systems are studied as the object of this paper. The methods of nonlinear system actuator fault diagnosis are mainly researched though wavelet neural network, which is optimized by genetic algorithm and particle swarm algorithm.The intelligent fault diagnosis methods have more obvious advantages than the traditional methods. Intelligent fault diagnosis method based on wavelet neural network is a very effective method. The residual error of system is analyzed and processed after the system observer was designed by wavelet neural network, and the fault diagnosis of nonlinear system will be further implemented. For some flaws of wavelet neural network, such as converging slowly and ling in the minimum easily, genetic algorithm and particle swarm algorithm are adopted to improve wavelet neural network in this paper.For actuator faults of a class of nonlinear time-delay systems with unmeasured states, a method using genetic algorithm to optimize wavelet neural network is put forward. Based on wavelet neural network constructing the state observer, the unmeasured states is estimated by using outputs of the state observer and nonlinear approximation ability of wavelet neural network. When occurring actuator faults, the fault model can be obtained by the fault estimator. Thus the fault diagnosis of the nonlinear time delay system can be realized. This method improves the shortcomings of wavelet neural network, such as slow convergence speed and local optimization, and it also creates conditions for practical applications.Considering other actuator faults of nonlinear systems which exist modeling errors, disturbances and uncertain items, a new method using particle swarm algorithm to optimize wavelet neural network is proposed. Aiming at settling unpredictable actuator faults of the control system, the system observer based on wavelet neural network is designed, which parameters are adjusted by systematic residual error. The control system can maintain stable before and after occurring faults. The feasibility of this method solving the problem of nonlinear fault diagnosis is verified by simulation examples, and the on-line actuator fault diagnosis is realized.