The Reaserch On Fault Diagnosis Of Analog Circuits With Tolerance Based On Wavelet Analysis And Artificial Neural Networks

With the increasingly complexity of structure and scale of circuits network andthe wide application of large-scale integrated circuits,it is an urgent question in thearea of practical engineering project that how to identify the fault and the faultelements correctly in the large-scale integrated circuits with tolerance by usingeffective diagnosis technique,and it is also the development tendency of analogcircuit fault diagnosis theory.Analog circuit fault diagnosis,as a synthesis intercrossed subject mixed withnew theory, technology and method from other subjects and fields,has been formed aseries of diagnosis theory and method over years.But as a result oftolerance,non-linear and the diversity and complexity of faults,the methods ofanalog circuits fault diagnosis are limited,therefore,novel methodologies of analogcircuit fault diagnosis are required to be presented.Recent years,wavelet analysis andartificial neural network theory have developed rapidly,and have been widelyapplied in many research areas.It is a new and effective approach for fault diagnosisof analog circuits to combine wavelet analysis theory with artificial neural networktheory,and it is also a hot research field currently.The novel approaches presented by this dissertation are based on waveletanalysis and artificial neural network theory,as the good time-frequency localizationproperty of wavelets,the fault signals of analog circuits are preprocessed by waveletanalysis and the feature vectors are extracted;the artificial neural network is used toclassify the faults because of its advanced property of pattern recognition,theproposed methods are verified by the experimental examples. The main novel worksin this dissertation are as follows:Firstly,the research on artificial neural network approach of analog circuitsfault diagnosis has been presented,the general steps of this method have beenanalyzed and stated;the advantages and disadvantages of the BP neural networkmethod of analog circuits fault diagnosis have also been described,an improved BPneural network model has been presented and it has been applied in fault diagnosisof analog circuits.Secondly,the wavelet analysis method of analog circuit fault diagnosis has beenproposed.The sampled fault signal from the analog circuit under test is preprocessed by wavelet analysis,which possesses good time-frequency characteristic.The featureparameters extracted from the sampled signals are used as input sample of artificialneural network,and the neural network is trained for implementing analog circuitfault diagnosis and identifying all kinds of faults.Thirdly,the approach based on S-transform and wavelet neural network for faultdiagnosis of analog circuit with tolerance has been presented.S-transform is used foranalyzing and processing fault signals from analog circuit under test and extractingfeature vecters,as a result of its properties that the width of time window is variedwith frequency and good time-frequency localization;a kind of neural network whichuses wavelet function as activation function of BP neural network is fabricated,and itis applied as a classifier for fault diagnosis of analog circuit;It is researched that thepropose method is effective for fault diagnosis of analog circuit with tolerance,evenunder the circumstance of high level noises,overlap in fault feature vectors andsufficient accessible nodes.