Fault Diagnosis Of Electronic Circuit Based On Deep Feature Learning

Fault diagnosis of electronic circuit is always the research hotspot in circuit testing field.Due to the increasing integration and scale of electronic circuit,and with its selfproblem that such as nonlinear and tolerance,fault feature of electronic circuit has emerged a more complex nonlinear situation.Therefore,it puts a higher requirement to feature extraction technology and diagnostic techniques.In this paper,a new fault diagnosis method of electronic circuit based on deep learning is proposed.In fact,the essence of fault feature extraction is to find a nonlinear mapping representation of fault response signal,and the deep learning is exactly a technology that extract feature by nonlinear mapping representation of data layer by layer.So based on their similar characteristics,we research the application of deep learning in the deep fault feature extraction of circuit.The two main research work in this paper are shown as follows:(1)We propose a novel fault feature extraction method of electronic circuit based on SAE-SOFTMAX.This method combines Stack-AE with SOFTMAX layer to construct deep learning framework.By training this framework with unsupervised pre-training and supervised global fine-tuning,we improve the feature extraction performance of SAE and use it to extract deep fault feature of electronic circuit.(2)We give two SAE-based fault diagnosis models for electronic circuit,including diagnosis model based on SAE-SOFTAMX and diagnosis model based on SAE-SVM.The former model merge SAE and SOFTMAX classify layer together,which leads to a quick fault feature extraction and fault diagnosis;and the later model merge SAE and SVM which has a good robustness together to implement fault diagnosis.Finally,the simulation experiment results of two circuit shows that our proposed feature extract technology has an outstanding performance than wavelet analysis and PCA-based method,reflecting in the high evaluation value of feature.And,two our proposed fault diagnosis model based on SAE both get no misclassification which shows their great diagnosis performance.We found that the model based on SAE-SOFTMAX is superior to several traditional BPNN diagnosis models by comparing their diagnosis performance and effcet.