Research On Fault Diagnosis Technology Of Analog Circuits Based On Support Vector Machine

Analog circuits are widely used in various electronic devices. With the rapid development of modern electronic technology and the increasing demand for reliability of electronic devices, there is more and more attention payed to analog circuit fault diagnosis. Support Vector Machine (SVM) with low computational complexity has the advantage that it is very fit for limited samples decision and it can dig out the hidden classified feature information from limited data samples in great extent. It is significant to research on fault diagnosis technology of analog circuit based on Support Vector Machine in order to improve diagnositic performance.This paper studies on analog circuit fault diagnosis based on SVM, and the main achievements are concluded as follows:(1) A novel method of analog circuit fault diagnosis by combining Error Correcting Output Code (ECOC) and SVM is proposed. In this method, Fuzzy C-Means (FCM) algorithm is employed firstly to cluster the fault samples, and an ECOC matrix is obtained from binary tree structure, then the fault classifiers is designed according to the ECOC matrix by using SVMs. After training and testing fault samples, the final results are obtained by decoding the test results with improved Hamming decoding method. The experiment results show that the proposed method which extends SVM to the multi classification problems of analog circuit fault diagnosis successfully improves the diagnosis speed and accuracy.(2) A novel method of analog circuit fault diagnosis based on Support Vector Machine ensemble is proposed. The method firstly constitutes the training sample set of the next iteration by adding the unclassifiable samples to the training sample set of this iteration, constructs the diversified classifiers. The final result is obtained by choosing classifiers whose accuracy are no less than the average accuracy to ensemble. The experiment results show that the proposed method ensembles SVMs successfully. It also increases the accuracy of single“1-v-a”SVM and is applicable to both linear and nonlinear analog circuits fault diagnosis.(3) Analog circuit test and diagnosis data acquisition system is built based on Advantech PCI-1714Ul, and a typical analog circuit PCB board is developed. Physics experiment on PCB board is completed, and the impact of the change of classifier parameters on the diagnositic performance is analyzed. Compared with the simulation results, it is proved that the two novel analog circuit fault diagnostic methods are effective and robustness.The research of this paper is funded by National Natural Science Foundation of China (60871009), Aeronautical Science Foundation of China (2009ZD52045) and Nanjing University of Aeronautics and Astronautics special research projects of basic research (No.NS2010063).