| With the rapid development of computer technology,the proportion of circuit system is increasing.Most faults in digital-analog hybrid circuits come from analog circuits.Analog circuits are characterized by continuity,nonlinearity and tolerance of component parameters,which makes the diagnosis process very complicated.Generally,faults are considered as permanent faults,but there are actually more special instantaneous faults and intermittent faults.The performance of instantaneous faults and intermittent faults is very similar,so there is no strict difference between them.Here,they can be collectively referred to as intermittent faults.The bad working conditions will lead to the frequent occurrence of intermittent faults in analog circuits,which are difficult to reproduce,test and diagnose.The diagnosis of analog circuit fault and intermittent fault is also the focus of many scholars.This paper analyzes and studies the key techniques of fault feature extraction and mode classification in analog circuit fault diagnosis:For the problem of fault feature extraction of analog circuit,two methods are proposed.One is used by the time-domain analysis statistics(such as sample range,mean,standard deviation,skewness,kurtosis and entropy,etc.)as the feature vectors of analog circuit faults.Another method is to propose a time-frequency analysis method based on wavelet packet transformation to extract fault features.The original signal of the circuit is decomposed by multilayer wavelet packet,and the energy entropy of wavelet packet is obtained after the decomposition coefficient is reconstructed.Finally,the current mainstream principal component analysis(PCA)method is used for feature selection,and the fault feature vectors of analog circuits are obtained after further dimensionality reduction.For the problem of fault identification and classification of analog circuits,the construction method of Deep Belief Network(DBN)model is proposed.As a model in deep learning,DBN has good capability of feature recognition and classification,as well as powerful ability of processing high-dimensional nonlinear data and self-learning,which can effectively and accurately locate faults.After building the analog circuit model and collecting the original data set,the time-domain features are extracted,which is then used as the input of DBN network.The experimental results show that this method has good recognition effect and high recognition rate.In order to further improve the analog circuit fault recognition rate and reduce the iteration times of network model,the wavelet packet energy entropy is extracted from the original data,and then the PCA method is used to reduce the dimension to obtain the feature vector,which is input into the DBN network model as the feature vector of analog circuits.After training and learning the DBN network model,the fault diagnosis is completed.Finally,the proposed method is compared with the common fault diagnosis methods,and the experimental results show that the proposed method has a higher fault identification rate in analog circuit fault diagnosis compared with other methods.The time,frequency,probability and fault intensity of intermittent fault in analog circuit system are stochastic,so it is difficult to model and detect.The circuit model of analog circuit intermittent fault is built for data collection,and the method of combining wavelet packet energy entropy and DBN is applied in more special analog circuit intermittent fault diagnosis.This method trains an effective DBN model through the sample set,and realizes the diagnosis of intermittent fault.Compared with the BP back propagation algorithm,the experimental results show that the proposed DBN method is still excellent in the intermittent fault diagnosis of analog circuits. |
