基于时域信号特征和卷积神经网络的模拟电路故障诊断算法  

Fault diagnosis algorithm of analog circuits based on time domain signal features and convolutional neural network

作　　者：赵师兵 张志明[1] ZHAO Shibing;ZHANG Zhiming(College of Electronics and Information Engineering,Tongji University,Shanghai 200092,China)

机构地区：[1]同济大学电子与信息工程学院,上海200092

出　　处：《计算机应用》2022年第S02期320-326,共7页journal of Computer Applications

基　　金：上海市教育委员会科研创新计划项目(202101070007E00098);教育部产学合作协同育人项目(201902097012,201902016059);同济大学双一流引导专项立项项目(4250145304/001)。

摘　　要：模拟电路是现代电子技术的基础,及时识别与定位电路故障是保证系统正常工作的重要环节。针对此类工程问题,提出一种基于时域信号特征和卷积神经网络(CNN)的模拟电路故障判断算法。首先采集对象电路的激励输入和输出响应信号,经过处理后成为1*N或2*N的时域信号序列输入CNN中,端到端实现从原始输入时域信号到故障识别期望输出的映射。实验仿真和实测结果表明,与经过信号预处理的频谱图+CNN和小波包变换+反向传播(BP)神经网络算法相比,该算法对结构性电路故障的识别正确率明显提高,在参数变化型电路故障的识别效果上总正确率相差不到1个百分点,但对于电路正常的判断正确率由93%提高到97%,避免出现某一具体故障正确率很低的情况,总体性能优于对比算法。该算法能够准确快速地识别和定位模拟电路中的结构性故障和参数变化型故障。In order to effectively identify and locate the faults of analog circuits,a circuit fault diagnosis algorithm based on time domain signal features and Convolutional Neural Network(CNN)was proposed.Firstly,the real-time signals of excitation input and output response of the object circuit were acquired with portable device directly.Secondly,the timedomain signal was reshaped as a 1*N or 2*N sequence and fed into a convolutional neural network subsequently,which was trained to achieve the end-to-end output mapping of the recognition and location of the expected circuit fault.Finally,the typical two-port network circuit models with calculated and measured data were used to verify the combined diagnosis algorithm in different test environments.The experimental simulation and actual measurement results show that,compared with the spectrogram+CNN algorithm and wavelet packet transform+Back Propagation(BP)neural network algorithm,the proposed algorithm can improve the identification accuracy of structural circuit faults while keeping almost the same correct rate of parameter variation fault identification.And the diagnosis accuracy of fine situation is increased from 93%to 97%,avoiding the uneven identification result with low accuracy for a certain type fault.This algorithm can accurately and quickly identify and locate structural faults and parameter variable faults in analog circuits.

关 键 词：双端口网络单元 模拟电路故障诊断 时域信号特征 卷积神经网络 识别和定位 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构] TP183[自动化与计算机技术—计算机科学与技术]