模拟电路故障诊断中的特征信息提取

Feature Information Extraction in Fault Diagnosis of Analog Circuits

在运用BP神经网络进行模拟电路故障诊断过程中,代表故障特征的网络输入至关重要。分析了常见特征信息提取和故障诊断方法,提出一种基于多测试点、多特征信息原始样本集的新方法。运用这种方法构造原始故障特征集,然后作为BP神经网络的输入对网络进行训练,仿真结果表明,通过该方法构造的样本集训练出来的网络对模拟电路故障诊断的正确率优于传统方法,证明了该方法在模拟电路故障诊断中的可行性,为模拟电路的故障诊断提供了一种新方法。

In the use of BP neural network to diagnose fault in analog circuits,the network input that represents fault signature is very important.The common characteristics of information structure and fault diagnosis method are introduced,and a new method based on multi-test point multi-feature information of the original sample set is proposed.The original fault signature set is constructed as the input of BP neural network to train the network.Simulation results show that the network trained with sample set by this method offers better accuracy than those by traditional methods in fault diagnosis of analog circuits.This novel method for fault diagnosis of analog circuits proves feasible.