Analog Circuit Fault Diagnosis Based On Extreme Learning Machine

With the rapid development of semiconductor technology, the structure ofelectronic equipment which has been applied widely become more complex and thescale become huger. And analog circuits are more likely to get wrong in electronicsystem, so in order to improve the reliability of electronic equipment, higher andnewly demands on circuit test are established. Also the test of analog circuit is limitedconditions for the integrated-circuit industry with the large scale of circuits and thedecrease of number of accessible testing nodes.So the research on analog circuit testis significant in the theory and practice.Extreme learning machine (ELM) has quick learning speed and goodgeneralization performance, which randomly chooses the input weights andanalytically determines the output weights. The dissertation combined the method ofELM and theory of analog circuit fault diagnosis and also conducted the research andrealization for the method of analog circuit fault intelligent diagnosis based on ELM.This paper bringhts out principal component analysis(PCA) and extreme learningmachine applying to analog circuit fault diagnosis. First, fault features are extractedfrom voltage amplitude of efficient points in frequency response curve directly infrequency domain and fault feature extracti on from circuit response are realized byPCA. Secondly, Through the ELM classifier to classify the fault, we can get goodclassification results.Based on it, to improve its generalization capability and diagnostic accuracy, theELM-RBF method is proposed.The radial basis function neural network whichExtreme Learning Machine algorithm is applied in, combines the advantages of both,and decrease the number of hidden layer, optimize network structure.Taking into account that the impact of the ELM performance of major network isconnection weights and threshold values, Combining with swarm intelligence appliedinto Analog circuit fault diagnosis is researched.Using global optimization ability ofintelligent algorithm, differential evolution algorithm and group search algorithm, forexample, to optimize the connection weights of the ELM and the threshold, we can getoptimal network and good effect.As integrated learning can significantly improve the generalization ability of thelearning system, the Extreme Learning Machine integration method is proposed which selective Extreme Learning Machine integration is done by combining baggingthought with k-means clustering effectively.It has high classification accuracy.The simulation results of examples given in this dissertation show that the faultdiagnosis methods proposed above have good diagnosis effect and feasibility inanalyzing the fault response of analog circuits and can locate the faults in analogcircuits correctly.