Research On Fault Diagnostic Methods For Analog Circuits Based On Time-frequency Analysis And Complex Field Analysis

Facing to the high speed development of circuits on integration and scale,study on advanced theory and method of analog circuit fault diagnosis is becoming more and more urgent in order to satisfy the development of electronic technology on security,reliability and testability.The main focus study of analog fault diagnosis is how to extract distinguishable fault features in order to avoid the undistinguishable problem between faults because of mixing.Time-frequency analysis method is the joint distribution of time and frequency information.It is able to clearly represent the signal frequency and amplitude characteristics at any time.Complex field analysis means to transform signal into complex filed and then using the real part and imaginary part of the signal to discrible its distribution on complex plane.Thus both methods can provide more fault details for fault diagnosis system.This dissertation mainly focus on study of new fault diagnosis method of analog circuits using time frequency analysis and complex field analysis.In addition,testability verification experiment is also studied.Primary contents and results in this disseration as follows:(1)Fault diagnosis for analog circuits by using ensemble empirical mode decomposition and extreme learning machine(EEMD-ELM)method is propsed.According to research on Hilbert-Huang transform and empirical mode decomposition,EEMD method is introduced to analog circuit fault diagnosis.A new analog fault diagnosis method is proposed for single faults and multi-faults with the combination of relative entropy,kurtosis and ELM.First,the output voltage signals of fault-free circuit and various fautly circuits are extracted,respectively.Then these signals are decomposed into a set of intrinsic mode functions(IMF),through calculating kurtosis of every IMF and relative entropies between fault-free state IMFs and faulty states IMFs,fault feature vectors are constructed.As input samples these fault feature vectors are fed to ELM to train and test fault diagnosis.Experiment results show that the method can reduce the computation cost of diagnosis system and increase the performance of fault diagnosis.(2)Analog circuit fault feature optimization research based on local mean decomposition(LMD)is researched and a new feature optimization method based on clustering method is established.First,the principle and theory of LMD and its decomposition process are discussed.And use LMD method to decompose circuit response signal into a set of PF signal and then build fault feature vectors with energy entropy,standard deviation and kurtosis of each PF component.However,the fault feature vector dimension will become bigger with the increase of the number of PF.Therefore,this disseration proposes a new feature optimization strategy for dimension reduction with cluster center and cluster radius of each fault mode.New feature vector is built by the method and then it is used to train a LVQ neural network to validate the method.Experiments prove that this method can effectively reduce dimension and time cost.Meanwhile,the performance of analog fault diagnosis is acceptable.(3)A new fault modeling and fault diagnosis method on complex field based on the least square circle fitting is proposed.In traditional fault dictionary method fault characteristic values storaged are discrete.However,because of the continuity of input and output signals,actual fault characteristic values are infinite.This leads to the fact that fault coverage is low in the traditional fault dictionary.In order to solve this problem,according to the theory of the slope fault model and fault complex domain model,fault feature functions are obtained by using least-square circle fitting method.On the basis of the fault feature functions,a corresponding fault diagnosis method is put forward.Both theoretical and practical circuit experiments verified the method can realize the fault diagnosis very well.(4)Failure sample selection based on ant colony algorithm optimization is studied.Random test sample selection method generally casues the omission of some propagation failures with smaller failure rate in traditional testability demonstration experiment method.This phenomenon could lead to serious trouble.This dissertation proposes a new failure sample selection method to solve this problem.First,using directed graph and ant colony algorithm to search an optimal path transmission failure,then to establish a subsequent failure propagation set(SFPS)for each failure module(device).Based on SFPS,an optimized test sample set can be obtained.Experimental results show this method can improve the fault coverage,increase the diagnostic ability and reduce the risk of using.