Font Size: a A A

Analog Circuit Fault Diagnosis Based On Information Fusion And Extreme Learning Machine

Posted on:2012-02-03Degree:MasterType:Thesis
Country:ChinaCandidate:Z M WuFull Text:PDF
GTID:2248330374496200Subject:Electrical engineering
Abstract/Summary:PDF Full Text Request
Analog circuit test and fault diagnosis have been developed from the research ofcomponents solvability since1960s, and have made considerable progress in theory sofar. Unfortunately, due to such factors as tolerance of components, lack of faultinformation, diversification of fault models and nonlinear efforts etc, there is still along way to put the theory into application. Intelligence information processingtechnologies like Neural Networks, Wavelet Analysis, information fusion and so on,provide a new feasible way to solve the problem of analog circuit testing and faultdiagnosis. In addition, infrared diagnosis is supposed to overcome the shortcomingsof traditional contact test, which is an effective supplement of traditional analogcircuits fault diagnosis methods.This paper firstly reviews the research of analog circuit fault diagnosis, thenfocuses on analog circuit fault diagnosis based on intelligent information processingtechnology, and attempts to take a more in-depth study for fault information, featureextraction and fault classification.The rapid development of Analog circuit, especially analog VLSI, has madeconventional contact test constrained in some occasions. In view of insufficientinformation of analog circuit electronic test, this paper researches analog circuit faultlocation by connecting infrared detection, and fault diagnosis by fusing voltageinformation and temperature information. As a result, an analog circuit fault diagnosismethod based on SOFM network and heterogeneous information is proposed. In thismethod, voltage and temperature information are extracted as fault featureinformation, then preprocessed and sent into SOFM neural network classifier as inputsample. By using the competition of output level neurons of SOFM, the winningneuron is attained and the fault recognition for sample data is classified. Simulationresults show that the proposed fusion method has a higher diagnostic accuracy.In order to overcome the difficulties in feature extraction and fault classificationof analog circuit fault diagnosis, the paper focuses on the application of WaveletAnalysis in analog circuit fault diagnosis and Extreme Learning Machine in faultpattern classification. On the basis of using wavelet packet decomposition to extractenergy in each sub-space of output signals as fault feature, Extreme Learning Machineis applied to classify the fault patterns. The diagnosis example results show that the method combining wavelet packet decomposition with Extreme Learning Machine iscorrect in the fault pattern classification.
Keywords/Search Tags:Fault Diagnosis, Analog Circuit, Wavelet Packet Transform, ExtremeLearning Machine, Data Fusion, SOFM network
PDF Full Text Request
Related items