Study On Fault Diagnosis For Nonlinear Circuit Based On Volterra Series And Neural Networks

With the advancement in electronic technology, nonlinear analog are being used aggressively in industry. Nevertheless, fault diagnosis continues to be the bottleneck in producing and using these circuits. For the circuits with nonlinear component, there are no generally proper math model and common simulation program for nonlinear circuits. So the unified method is lacking for computing fault character of nonlinear circuits. The Volterra kernels of nonlinear analog circuits are the inherent characteristic of the system. System identification and nonlinear system analysis based on Volterra series are widely used in resent years. Neuron network has strong patter identification ability and can approximate a nonlinear system accurately. Therefore, this dissertation studied the structure of nonlinear analog circuits fault diagnosis system and the methods of how to extract the fault signatures from the respose of the circuit based on Volterra series and neural networks. Author's main work concentrates on three aspects as follows:1. Faults diagnosis system of analog circuits based on neural networks.Researched the structure of nonlinear analog circuits fault diagnosis system based on neural network. Include the methods of extract fault signatures, generate the train samples, design the neural network structure and the learn arithmetic.2. How to extract fault signatures from frequency response of nonlinear analog circuitsâ‘ Nonlinear analog circuits fault diagnose based on frequency response. This dissertation researched the character of the frequency response of nonlinear analog circuits with Volterra series, and presented a method of how to extract the fault signatures from frequency response of the nonlinear analog circuit. Because the key of fault diagnosis based on frequency response is how to design the stimulus, this dissertation presented two methods to solve this problem. One is designing the frequency of stimulus with least multisine to generate most fault signature. The other is using optimization that used the Volterra kernels as the models of the fault circuits and fault free circuits and search the optimum stimulus based on genetic algorithm. To develop the precision models of nonlinear analog circuits, this dissertation researched how to determine the highest significant order of nonlinear analog circuit and how to measure all the Volterra kernels.â‘¡How to use the Volterra frequency kernels as the fault signatures in diagnosis nonlinear analog circuit. The Volterra kernels of nonlinear analog circuits which are independent of the input are the inherent characteristic of the circuits. So this dissertation presented that using the Volterra frequency kernels as the fault signatures to diagnosis. This dissertation presented a Volterra frequency kernels measurement method that used the frequency response of nonlinear analog circuits directly.3. How to extract fault signatures from transient response of nonlinear analog circuitâ‘ This dissertation presented that use wavelet analysis to extract the transient response signature of nonlinear circuits and compress the signature data, and fed it into neural networks to execute fault identification. This method can largely reduces the input number and training time, simplifies the construction and improves the ability of fault identification for neural networks. This dissertation presented a selecting the best wavelet function method based on the between-category total scatter of signature.â‘¡Studied how to optimum the transient testing stimulus. This dissertation present that use Elman network to develop the models of the fault circuits and fault free circuit and search the optimum stimulus based on genetic algorithm.The simulation results of examples given in this dissertation show that the fault diagnosis methods proposed above have good diagnosis effect and feasibility in analyzing the fault response of nonlinear analog circuits and can locate the fault correctly.