基于深度学习的模拟电路软故障诊断  

Soft fault diagnosis of analog circuits based on deep learning

作　　者：常国祥 张京 CHANG Guoxiang;ZHANG Jing(School of Electrical&Control Engineering,Heilongjiang University of Science&Technology,Harbin 150022,China)

机构地区：[1]黑龙江科技大学电气与控制工程学院,黑龙江哈尔滨150022

出　　处：《电气应用》2021年第9期58-66,共9页Electrotechnical Application

摘　　要：为降低模拟电路软故障特征提取与分类的人工成本,提高软故障诊断的通用性,提出一种基于深度学习的软故障特征提取方法。利用通道注意力机制对深度学习中的卷积神经网络进行改进,将时域电压波形数据输入至改进的卷积神经网络中进行卷积池化,实现数据降维和故障特征提取,并利用注意力机制对所得的故障特征进行深度选择,突出通道内关键的故障特征,抑制不重要的特征,最终使用Softmax分类器对故障特征分类。针对四运放双二次高通滤波器进行故障诊断,故障诊断平均准确率为98%。实验结果表明改进的卷积神经网络模型可以实现对模拟电路的故障诊断,并避免了传统故障识别耗费大量人工的故障提取和选择。In order to reduce the labor cost of soft fault feature extraction and classification in analog circuit and improve the versatility of soft fault diagnosis,a soft fault feature extraction method based on deep learning is proposed.Using channel attention mechanism of deep learning convolution neural network is improved,the time domain voltage waveform data are inputted to improved convolution neural network the convolution pooling,data dimensionality,fault feature extraction are realized,and the attention mechanism on the depth of the fault features selection is used,the fault of channel and key features are highlighted,un important features are suppressed,the characteristics of the,Softmax classifier is used to classify the fault features.The average accuracy of fault diagnosis is98%for four-op amplifier double secondary high pass filter.The experimental results show that the improved convolutional neural network model can realize the fault diagnosis of analog circuits and avoid the traditional fault identification which costs a lot of manual fault extraction and selection.

关 键 词：模拟电路 故障诊断 卷积神经网络 通道力机制 特征提取 

分 类 号：TN710[电子电信—电路与系统] TP18[自动化与计算机技术—控制理论与控制工程]