Fault Detection For A Class Of Nonlinear Sampling Systems Based On Deterministic Learning

Most engineering systems are essentially p resented in the form of nonlinear sampling data ones with higher performance of automation in contrast to general ones in engineering practice.However,the increasing scale of modern industrial systems has brought increasing complexity of framework and po ssibility of fault in this class of systems.To enhance the system stability,a lot of work on the fault detection technology for this class of systems has been done by researchers worldwide with some progress.Despite all these,confounding of nonlinear s ampling system and unknown internal sampling information exist,and the fault detection remains difficult.Based on the deterministic learning theory,this paper mainly explores the fault detection of a class of nonlinear sampling data systems.The main re searches of this paper are as follows:Firstly,for a class of nonlinear sampling data systems,the Euler approximate modeling error,modeling error term and fault dynamic term after system discretization are set as unknown sampling dynamic information.Ba sed on the deterministic learning theory,the local neurons of the system can satisfy the local continuous excitation condition and reach the convergence of neural weights.In this way,the unknown sampling dynamics can be approximated and modeled by the R BF neural network,and then the fault database can be established and updated step by step.When processing high-dimensional data,only the neurons in the neighborhood of the system trajectory are activated,which will reduce the computational load.Second ly,a set of monitors is constructed on the basis of the fault database to obtain the unknown sampling dynamic information in the tested system.The system dynamics of the current monitored system are analyzed through the fault database,and then the resid ual specification is generated to realize the system fault detection.Finally,the detection ability of the scheme is characterized by strict analysis.The results indicate that the quality of mismatch function has an important influence on the effectivene ss of the system fault diagnosis.For analysis of the impact of mismatch function on residual,generalizations of mismatch interval and duty cycle are proposed in this paper.The fault detection method provided in this paper can be introduced into the sh ort circuit fault detection of power system.Similarly,the dynamic model of the power system in the case of short-circuit fault is built first.The dynamic information of short-circuit fault is learned by means of constant RBF neural network approximation,and the fault database is established.Then the fault detection of power system in different types of short circuit is accomplished according to fault database.Finally,the feasibility and effectiveness of the detection method are further verified by a practical nonlinear fault system.