Analog Circuits Fault Diagnosis On Neural Network

With the rapid development of electronic technology, the electronic equipmentreliability requirements are highly required, at the same time, we need to solve theproblem of circuit fault automatic diagnosis system. Since the non-linear, and the diverse,complex of fault phenomenon, so that the analog circuit fault diagnosis is beset withdifficulties, the actual diagnosis correct rate is not high, it has not been fully maturediagnosis system. The neural network has the characteristics of nonlinear approximation,it provides a better idea for the fault diagnosis of analog circuit with large-scale andtolerances, which is a hotspot research of the field.To solve the problem of analog circuit fault diagnosis with the Back PropagationNeural Network, we design three kinds of improved diagnostic methods, to insure thenetwork convergence and decrease the time, at the same time improve the accuracy inanalog circuit fault diagnosis.Firstly, for the local minimum problem of BP algorithms achieving the analogcircuit fault diagnosis, we modify it on the basis of additional momentum factor andvariable learning rate method, to improve the problem. We simulate some experimentswith this method. And contrast to the standard BP algorithm and the additionalmomentum algorithm and variable learning rate method.Secondly, reference the fuzzy neural network and the integrated neural networkalgorithm, we design a kind of fuse multiple T-S fuzzy neural networks of analog circuitfault diagnosis algorithm, the training samples are assigned to multiple sub networks. Weadapt the variable learning rate method to train the parameters during the process ofparameters iterative. It is achieve the best between the network convergence speed andaccuracy. We contrast the simulations of circuit fault diagnosis between this method andthe network fusion methods of fix learning rate.Finally, we design a improved multi-population co-evolutionary genetic algorithmsto achieve the fault diagnosis of analog circuit. Using a tent map which addition chaoticdisturbance initializes uniformly distributed populations, each population use different control parameters to control the searching range of the algorithm, and we introduce theimmigration operator to realize the transverse connection between populations, wecontrast the simulations of circuit fault diagnosis between this method and the standardgenetic algorithm and multi-population genetic algorithm.