Analog Circuit Fault Diagnosis Based On Wavelet Packet Entropy And Support Vector Machine

With the rapid development of semiconductor technology and computertechnology, the size of modern electronic equipment is increasing, the systemstructure has become increasingly complex, and the use of digital/analog hybridcircuits becomes more and more common. Mixed-signal electronic systems are oftenprone to failure in the analog part. Due to the continuity of its response, thenonlinear characteristics and tolerance of device parameters, the diagnostic processof analog circuits is very complicated. The research on methods of analog circuitfault diagnosis has been the focus of the scholars at home and abroad, while thecircuit fault characteristic extraction and pattern classification are two key problems.For the problem of analog circuit fault characteristic extraction, a extractionmethod based on wavelet packet entropy measure was proposed in this paper. Twokinds of wavelet packet entropy measure(energy entropy and characteristic entropy)were defined to characterize the feature information of the circuit under differentkinds of fault modes. Multi-layer wavelet packet decomposition was performed tothe circuit output response signal, and decomposition coefficients werereconstructed and computed its energy entropy or characteristic entropy values.After normalized, the feature vector of the circuit were obtained.For the problem of analog circuit fault pattern classification, The constructionmethod of the Support Vector Machine(SVM) multi-classification recognizer wereresearched in this paper. On the basis of two-class support vector machine, byselecting the kernel function and introducing the “one-against-one”multi-classification algorithm, the multi-classification recognizer used for circuitfailure mode identification was built. For the structural optimization problem ofSVM, the optimization method of the kernel function parameter and the penaltyfactor were also discussed.Combined the characteristic extraction method based on wavelet packet energyentropy with Support Vector Machine theory, the building method of analog circuitfault diagnosis system based on wavelet packet energy entropy and Support VectorMachine was presented in this paper. The concrete steps to realize diagnosis were also detailed. The results of the simulation examples show that high diagnosticaccuracy could be obtained by using this method. For the problem that thecharacteristic vector dimension size affect the training speed and classificationaccuracy, a diagnostic method based on Least Squares Support Vector Machine(LS-SVM) was further proposed in this paper. And it combined with the extractionmethod of the fault characteristics based on wavelet packet characteristic entropy,together to achieve analog circuit fault diagnosis. The results of diagnosis exampleshowed that the diagnosis method based on wavelet packet characteristic entropy andLS-SVM could ensure a high classification accuracy while effectively reduce thecomplexity of the calculation process at the same time.