Research On Analog Circuit Fault Diagnosis Based On Support Vector Machine

Analog circuit fault diagnosis plays a key role in circuit design,equipment production and instrument maintenance.It is one of the most important topics that academic researchers and engineers are challenging in the field of circuit theory technology research.With the rapid development of electronic technology,the complexity and intensity of analog circuits continue to increase,and more stringent requirements are imposed on the reliability of analog circuit operation.Academia and engineering circles have carried out a lot of research on the theory and method of analog circuit fault diagnosis,and have achieved certain stage results.Although the theory and method of analog circuit fault diagnosis have made some progress,there are still many problems that need to be studied and solved.First,the analog circuit fault feature optimization extraction lacks effective optimization criteria.Second,the reliability of the analog circuit fault set denoising is not high.Third,the multi-class combination strategy selection is too blind.The fourth is the lack of an effective solution to the imbalance of analog circuit fault sets.In this paper,combined with wavelet analysis and support vector machine,the following four parts are mainly worked on the above problems: The first part studies the analog circuit fault extraction method based on wavelet packet transform and the wavelet packet basis function optimization method.In this paper,the wavelet basis function optimization criterion based on energy entropy and inter-class separability measure is proposed,which provides a new standard for the selection of optimal wavelet basis function and improves the extraction quality of analog circuit fault samples.This lays the foundation for the diagnosis of later analog circuit faults.In the second part,the support vector machine,fast density peak clustering(FDPC)algorithm,Box-Cox transformation of analog circuit fault sample noise recognition algorithm is studied.Compared with the noise recognition method based on density recognition only,the algorithm solves the sensitivity problem of noise discriminant threshold selection in analog circuit fault diagnosis.As a result,the anti-interference of the support vector machine training model is improved and the influence of noise on the accuracy of analog circuit fault diagnosis is reduced.In the third part,the support vector machine multi-classification combination strategy for analog circuit fault diagnosis is studied,and an improved hierarchical support vector machine algorithm based on minimum spanning tree is proposed.The algorithm solves the problem that the classification strategy design of one-to-one,one-to-many,directed acyclic graph multi-classification strategy is too blind and inefficient.The topology of the classification decision tree is designed reasonably according to the size of the inter-class separability measure.The classification strategy for analog circuit fault diagnosis tends to be reasonable and efficient,which further improves the diagnostic rate of analog circuits.The fourth part studies the solution to the problem that the number of analog circuit fault samples is unbalanced and the diagnosis rate of the sample is insufficient.The w-smote algorithm and w-k-smote algorithm are proposed for two different imbalance situation.A new method for measuring the boundary degree of a small number of samples improves the ability of the classification model to quantify the boundary degree of fault samples.Finally,the diagnostic rate of the analog circuit with fewer sample faults has been effectively improved,and the overall diagnostic rate of the circuit has been improved accordingly.