基于一维卷积神经网络的模拟电路故障诊断

Fault Diagnosis of Analog Circuit Based on One-Dimension Convolutional Neural Network

模拟电路是复杂电子设备的重要组成部分,其故障可能导致整个设备停机,造成巨大的财产损失,甚至人员伤亡。传统的模拟电路故障诊断方法主要依赖于复杂的信号处理技术和专家经验,只适用于特定场景。因此提出一种基于一维卷积神经网络的模拟电路故障诊断方法,可以直接从原始时间序列信号中提取故障特征,不依赖于信号处理技术和专家经验。为了减少模型参数,避免出现模型过拟合,采用全局平均池化层取代传统卷积神经网络的全连接层。实验结果表明,相比传统方法,所提出的方法能够有效提取深度故障特征,具有更高的诊断准确率和更稳定的分类性能。

Analog circuits are an important component of complex electrical systems,whose failure may cause a sudden breakdown of the entire equipment,bringing about enormous financial losses or even personnel casualties.Traditional fault diagnosis of analog circuits relies heavily on complex signal processing and expert knowledge,which is ad-hoc.An analog circuit fault diagnosis method based on 1 D CNN(One-Dimension Convolutional Neural Network)is proposed,which extracts fault features directly from raw time-series signals without signal processing and expert knowledge.In order to reduce model parameters and the risk of overfitting,global average pooling is used in our model instead of the fully-connected layers in traditional CNN.The experimental results show that the proposed method can be utilized effectively to extract deep fault features,and has higher diagnostic accuracy and more reliable performance.