Research On Fault Diagnosis Method Of Analog Circuit Based On Empirical Wavelet Transform And Residual Network

Analog circuit plays an essential role in the electronic system.If the fault diagnosis of analog circuits can be effectively implemented,it will help to improve the reliability and stability of the electronic circuit and provide guidance for the operation of the electronic device.However,with the increased integration and complexity of electronic circuits,as well as being vulnerable to circuit noise,component tolerance,fault diagnosis of analog circuit has become more intricate and difficult.In order to effectively suppress the influence of circuit noise and component tolerances on analog circuit fault diagnosis and improve the accuracy of fault diagnosis,this thesis researches and implements the analog circuit fault diagnosis method based on empirical wavelet transform(EWT)and residual network(ResNet)on the basis of the project “XXX equipment fault diagnosis system development”,and the main research contents of this thesis include:(1)Research and design of a fault diagnosis scheme for analog circuits.In this thesis,three standard filter circuits in ISCAS'97 are selected as the simulation validation circuits,the analysis of the circuit under test and the selection of excitation signal,test point and fault diagnosis are completed,the response signal at the output of the circuit under test is collected to build the fault data set,EWT is used for signal preprocessing and ResNet model is used for fault feature extraction and fault classification through the investigation and analysis of the fault diagnosis method of the analog circuit.(2)Research and improvement of EWT-based signal preprocessing method.In this thesis,for the traditional EWT based on local maximum method,combined with the selected stimulus,the stimulus-based improved EWT is proposed,which can still achieve accurate segmentation of Fourier spectrum under the influence of noise.For the traditional scale-space based EWT,this thesis proposes another improved EWT based on the power spectral density theory which can effectively suppress the influence of noise and the isolated forest algorithm with linear time and space complexity,which can not only achieve the accurate segmentation of spectrum under the influence of noise,but also the higher validity of the obtained components.(3)Research and implementation of ResNet model-based fault diagnosis for analog circuits.A 1-dimensional ResNet model is studied and implemented to extract richer abstract fault features from the signal and to solve the problem of gradient degradation caused by deeper network depth.In order to obtain more comprehensive high-level features and fully learn the fault information of each component,a multi-branch ResNet model is studied and implemented in this thesis,which is more complex in structure but has stronger robustness and generalization capability.(4)The design of a fault diagnosis software for analog circuits.Based on the Qt development framework,the design and implementation of the analog circuit fault diagnosis software is completed,and based on this software,an analogue circuit fault diagnosis platform is built to collect the actual data of the signal conditioning circuit in the equipment under test to verify the effectiveness of the method proposed in this thesis in practical application.Three simulated circuits and one actual circuit are used to evaluate and validate the proposed method.The test results show that the proposed method has higher diagnostic accuracy compared to the comparison method.