一种使用深度学习网络进行模拟电路故障特征自动提取的方案  

A Scheme to Automatically Extract Fault Characteristics of Analog Circuits by Deep Learning

作　　者：王月海[1] 王瀚晨 WANG Yuehai;WANG Hanchen(Col.of Information Engineering,North China Univ.of Tech.,100144,Beijing,China)

机构地区：[1]北方工业大学信息学院,北京100144

出　　处：《北方工业大学学报》2020年第5期66-70,共5页Journal of North China University of Technology

基　　金：国家自然科学基金面上项目“基于概率光流计算的仿生仿振/光流辅助INS组合导航方法研究”(61573019).

摘　　要：为提高故障特征提取的通用性,降低故障诊断特征提取的人工成本,本文提出了一种将采样数据直接输入到训练好的卷积神经网络自动提取故障特征的方法.选用四运放高通滤波电路作为实验电路,使用仿真电压响应信号作为输入,经过多次卷积、池化得到诊断特征,然后使用SVM分类器进行故障诊断.仿真结果表明故障诊断准确率可以达到98.3333%,因此利用深度学习可以自动提取故障数据中的特征,从而减少故障特征提取阶段的人工成本.In order to improve the generality of fault feature extraction and reduce the labor cost of fault diagnosis feature extraction,this paper proposes a method which can automatically extract fault features by directly inputting the sampled data into the trained convolutional neural network.We chose the Quadamplifier high pass filter circuit as the experimental circuit.And then,we used the voltage response data as the input of the convolutional neural network.After multiple convolution and pooling,we obtained the fault characteristics.Finally,we used the SVM classifier for fault diagnosis.The simulation results show that the accuracy of fault diagnosis can reach 9.3333%.As a conclusion,deep learning scheme can be used to automatically extract the features of fault data,which can reduce the labor cost of fault feature extraction.

关 键 词：深度学习 卷积神经网络 支持向量机 模拟电路 故障诊断 

分 类 号：TN710[电子电信—电路与系统]