Fault Diagnosis For Analog Circuit Based On Multiple Data-dependent Kernels

With the rapid development of semiconductor technology, the analog circuits used in modern electronic systems are more complex, the function is much stronger. The fault probability is increasing, so different types of circuit fault diagnosis method are proposed by researchers to ensure that the analog circuit has good availability and reliability. In these methods, the intelligent fault diagnosis for analog circuits based on support vector machine(SVM) method has a solid theoretical foundation and promotion of features which can solve the problem of learning of small samples and become a hot research topic in the field of circuit fault diagnosis. The nonlinear method of support vector machine(SVM) using a single kernel function input data mapped to high-dimensional space linearly separable, But research suggests that performance differences due to the usage of different kernel functions, the classification results differ greatly in various data sets. And the choice of kernel function and the structure is still not perfect theoretical basis. In addition, the inherent characteristics of analog circuit itself make the fault samples data of irregular and uneven distribution of high dimensional feature space, using a single kernel function is often difficult to obtain good diagnosis effect.To solve the above problems, this paper proposes an optimization strategy based on multiple data-dependent kernel matrix, to improve the standard support vector machine which relies too much on training model and parameters, and the multiple data-depended kernel method is applied to solve the problem of analog circuit fault diagnosis. Firstly, this paper studies the structure principle of multiple kernel function, using kernel alignment to optimize the weights of basic kernels. Then the kernel function optimization method in the empirical feature space, optimize the multiple kernel matrix structure data-dependent kernel. This paper proposed fault diagnosis based on multiple data-dependent kernel for analog circuit, multiple kernel will be convex combination several basic kernels structure more nuclear matrix, and optimize the multiple kernel matrix by the method of data-dependent kernel, kernel matrix by using the optimized alternative standard single kernel function of support vector machines(SVM), in order to improve the diagnostic accuracy of fault diagnosis method. Finally, the paper using Hspice software for several typical circuits, the simulation experiment in the circuit into common fault types, using Haar wavelet transform and PCA method for feature extraction, fault samples were obtained, based on many data to verify the thesis put forward the multiple data-dependent kernel circuit fault diagnosis method is effective.Simulation experiment results show that, compare with the standard support vector machine, the single multi-kernel learning method and data-dependent kernel method, the analog fault diagnosis method based on multiple data-dependent kernel can make better fault diagnosis accuracy and shorten diagnosis model establishment time and improve the diagnostic efficiency.