Fault Diagnosis In Analog Circuit Based On Support Vector Machine With Composite Kernel Function

With the rapid development of modern analog integrated circuit, the complexity and intensity of modern analog integrated circuit increases constantly, which results in increasingly strict requirement with the reliability of analog circuit, the analog circuit fault diagnosis research field is taken more and more seriously. This article uses the wavelet theory and kernel principal component analysis method to extract the fault feature. And select the appropriate compound kernel function, build support vector machine model of a composite kernel function optimized by the particle swarm and chaos particle swarm algorithm, analyzing the circuit fault feature by classification model, the main work is as follows:First. The paper uses the Monte Carlo analysis to gain fault feature data, extracting the failure data according to the principle of wavelet. Using nuclear principal feature extraction method to extract the fault feature, make the fault samples reduce the dimension.Second. This paper chooses libsvm as troubleshooting tools. This paper analyzes the typical circuit fault diagnosis by constructing a model of analog circuit fault diagnosis based on libsvm, building the single-core kernel function, dual-core nuclear function and three-nuclear nuclear function, It is concluded that compound kernel function is most suitable for the circuit through the comparisons of classification performanceAt last. This paper puts forward two kinds of the composite kernel function of support vector machine (SVM) parameters method of optimization. By using particle swarm optimization (PSO) algorithm and chaos particle swarm optimization algorithm for optimizing the parameters of composite kernel function(Punishment parameters, nuclear parameters and composite kernel function parameters) of support vector machine. Putting the optimized parameters in support vector machine (SVM), and building the analog circuit fault diagnosis model of the composite kernel function libsvm based on particle swarm optimization algorithm and chaos particle swarm algorithm. Having the analysis of two typical circuit fault diagnosis, classification results prove that composite kernel function of support vector machine is superior to the single core support vector machine, and particle swarm optimization algorithm and chaotic particle swarm algorithm to optimize the classification of the composite kernel function of support vector machine model is better than wavelet support vector machine (SVM) method and the PSO support vector machine method In two examples of circuit analysis.