An Analog Circuit Diagnosis Method Based On Independent Component Analysis And Support Vector Machine

In highly developed process of industrialization, in order to meet the requirements of safety and environment, the circuit testing requirements are also increasing. With the growing competition between nations, the circuit testing technology is also developing. The most concerned problem in the industrial development is how to extract the characteristic value and classify the data in circuit testing effectively. So currently, with the Independent component analysis (ICA) and support vector machine (SVM), the fault diagnosis testing has solved the above problem effectively. Its advantage is that we don’t need to have accurate experience or to build complex mathematical models, we just need to use the online data or off-line data to dig the data relations, and finally obtain the equipment’s running status and complete the fault diagnosis of the system simulation and testing.Independent component analysis is one of the data driven method of fault diagnosis and detection, it is a process control method based on multivariate higher-order statistics, with more effective extraction of characteristic value; support vector machine method can better solve the problems of high dimension, small sample and nonlinear and so on, which shows a significant application value in the circuit testing and analog fault diagnosis, and other field.In this paper, the main research contents are as follows:1. In view of the data, with non-Gaussian characteristic in the process of analog fault diagnosis, we have further studied the adaptive kernel function of independent component analysis. We have made the detailed analysis and description in independent component analysis of the adaptive kernel function with our theory and practice, and also compared with the accuracy of experiment and the time for validation based on the traditional Infomax of ICA and PCA, calculated that the adaptive independent components analysis method is effective and feasible.2. Single gaussian kernel function is a local function with strong learning ability but poor generalization. While multilayer perception set function is a global function with strong generalization but weak learning ability. Considering the system’s different demand in different situations, this article will combine the two kinds of kernel function advantages and take the analog circuit test signal characteristics in account, to build adaptive kernel function for considering. For the obtained processing data with using the adaptive independent component analysis method, will be classified with adaptive kernel function of support vector machine (SVM) method. Which can better solve the problems of high dimension, small sample and nonlinear and so on. At the end of the paper, we used the Gaussian kernel function and multilayer perception kernel function to deal with analog circuits fault, and compared the training time and testing accuracy between samples.3. We have made detailed description of the diagnosis steps of the adaptive non-Gaussian independent component analysis and support vector machine method. And use the data obtained from the tests with using the Salley-key band pass filter, differential amplifier circuit and mobile phone to prove that the method is feasible.