Research On Testing Diagnosis Theory And Key Technology For Analog Circuits

Analog circuits are important parts of aerospace, defense and modern industry. With theintegration and the increasing scale of electronic devices, modern electronic equipments has higher andhigher requirement on the reliability and testability for analog circuits. The research of analog circuitfault testing diagnosis is of great significance for enriching the modern circuit theory system, andreducing the losses caused by circuit faults and electronic equipments maintenance, and promoting thetechnology development of electronic system fault prediction and health management. In addition toabsorb the advantages of traditional methods, several new methods, such as testability analysis theory,fractional signal processing, support vector data description, are used to diagnose analog circuit faults.This paper deals with four key technologies of test stimulus selection, test node selection, fault featureextraction and intelligent fault diagnosis method. The contributions of this thesis are that:(1) The thesis proposes a test stimulus generation method based on signal waveformdecomposition. Optimal test stimulus selection is a NP hard problem due to infinite options of testsignals and parameters in theory. In order to reduce the complexity of optimal test stimulus selectionand improve the testability of the circuits under test, this paper proposes a new stimulus selectionmethod based on signal waveform decomposition. Two approaches based on Fourier series expansionand wavelet decomposition are introduced to generate random signal. Moreover, the maximum distancewithin/between fault samples is the optimization object to select the optimal stimulus, which isevolved by genetic algorithm. The test results show that the proposed method is effective when isapplied to both of linear circuits and nonlinear circuits and can improve the testability of the circuitunder test.(2) The thesis proposes a node selection method based on fault fuzzy probability. For thetraditional fault dictionary method, the fault fuzzy interval is calculated based on the interval of±0.7V,and the test nodes are selected according to a single standard. However, this traditional method does notensure to obtain the global optimal test node set. In our paper, a test node selection method based onfault fuzzy probability is proposed. The fault fuzzy interval is calculated based on the normaldistribution curve of each class fault. The optimal test node set is selected according to the fusioninformation of normal curve area and fault isolation value. The test results show that the proposedmethod can diagnose most of faults with the least amount of test nodes and reduce the misdiagnosisradio of fuzzy faults.(3) The thesis proposes a fault feature extraction method based on fractional wavelet transform. The fuzzy faults with similar features for analog circuits are difficult to diagnose correctly. In order toimprove the testability of fault, a feature extraction method for analog circuit is introduced in this paper,which is based on fractional wavelet transform theory. Firstly, as fractional wavelet transform is a kindof wavelet transform in fractional Fourier domain, a new discrete algorithm is proposed. Then, a newmethod based on multiple-fractional wavelet transform is proposed to extract fault features of analogcircuits. The diagnosis results show that the extracted fault features in multiple fractional domains canimprove the testability of faults.(4) The thesis designs a VSVDD classifier method based on Vauge set information fusion. Theproposed method is applied to diagnose multiple-class faults for analog circuits. If the descriptionsphere boundary is not compact, fuzzy faults located in crossed spheres would be misdiagnosed. AVSVDD diagnosis method is introduced based on vague set information fusion. Each test sample isassigned a truth membership value and a false membership value according to the sample spacedistribution. Moreover, each description sphere is assigned a weight value based on the size of sphere.Then, the test samples are diagnosed through the decision rule of minimum weight vague distance. Thediagnosis results improve that this method can improves the performance of sphere classification foranalog circuit fault diagnosis.