Research On Analog Circuit Fault Diagnosis Based On Support Vector Machine

With the rapid progress in modern electronic and computer technology, the structure of electronic equipment become more complex and the scale become huger. Higher and newly demands on circuit test are put in order to improving the reliability of electronic equipment. Analog circuits are more likely to get wrong in electronic system, also the test of analog circuit is limited conditions for the IC industry. Within the background of the new development in modern micro-electronic technology and information technology, the research on analog circuit test is significant in the theory and practice of IC industry.Recently terrible lack of typical fault samples and finding problem of diagnosis knowledge are the dominating difficulties that analog circuit fault intelligent diagnosis system faces. Support Vector Machines (SVM) is a machine learning algorithm based on statistical learning theory. This algorithm accomplishes the structural risk minimization (SRM) principle. The most predominance of SVM is proper for limited samples decision. The nature of the algorithm is acquiring connotative class information to great extent from limited samples. The dissertation combined the method of SVM and theory of analog circuit fault diagnosis and also conducted the research and realization for the method of analog circuit fault intelligent diagnosis based on SVM.According to that standard support vector machine is deduced from two class classification problem, which can't be applied directly to resolving multi-class problem just like analog circuit fault diagnosis, mufti-class arithmetic is put forward which adopts decision directed acyclic graph, and mufti-fault classifier is set up. The selection of kernel function and the adjustment of kernel parameters for SVM are also studied. The key problem is studied and the foundational realizing steps are brought out for support vector machine applying to analog circuit fault diagnosis. Fault features are extracted from voltage amplitude of efficient points in frequency response curve directly in frequency domain and fault feature extraction from circuit response are realized by wavelet packet transform. Two analog circuit fault diagnosis approaches are presented, one is based on efficient points sampling and SVM, and another is based on wavelet packet transform and SVM.Using for reference from the method of fractal diagnosis and classification in mechanical equipment, fractal dimension can reflect the state of analog circuit. Based on fractal theory fractal diagnosis method combined SVM is developed. In practical application the binary fractal compute method by gridding dimension is used. Through calculating the binary gridding dimensions of signal based on various sampling interval as fault feature vectors, mufti-fault SVM classifiers are established and trained. At last they are used for analog circuit fault diagnosis.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. Based on fault samples from Volterra frequency kernels, mufti-class SVM are established, trained and used for analog circuit fault diagnosis.There are some defects with SVM which decrease the stability and the generalization ability of SVM. Ensemble learning can significantly improve the generalization ability of learning system. The technology of SVM ensemble is studied in this dissertation. The affectivity of two different disturbance mechanisms on augmenting the diversities among member classifiers, disturbing feature subspace and disturbing model parameters is analyzed. Through analyzing randomness and ergodicity of Logistic maps in parameter selection, two ensemble SVM algorithms are proposed based on the double disturbance mechanism combined Logistic maps, one is RAB-SVM algorithm and another is 2D-RBaggingSVM. Two ensemble SVM algorithms are used for analog circuit fault diagnosis.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 analog circuits and can locate the faults in analog circuits correctly.