基于灵敏度预分类的 BP 神经网络故障诊断方法

摘要

故障特征信息的获取和处理对电路故障的可靠分类和准确诊断有很大的影响。在电路故障诊断时，对于不同的故障模式，存在信息混叠的现象，需要解决特征信息的有效提取和故障的可靠分类等问题。为此，本文提出了一种结合灵敏度特性分析的 BP 神经网络故障诊断方法。基本思想是通过灵敏度的计算，对电路故障样本作预分类，再根据电路灵敏度的计算结果分别提取相应特征信息，以此构造故障样本特征集，然后作为 BP 神经网络的输入对网络训练，并进行故障诊断。对滤波器的仿真结果表明，该方法能分类不同的元件故障，且对模拟电路故障诊断的平均正确率优于传统方法。%Fault characteristic information acquisition and processing has a great influence on circuit fault classification and accurate reliable diagnosis .In the circuit fault diagnosis ,for different failure modes ,there is a phenomenon of fault characteristic information aliasing ,the problems of feature information effective extrac-tion and fault reliable classification need to be solved .Therefore ,presents the BP neural networks fault diag -nosis method based on the characteristic of sensitivity .The basic idea is through the circuit characterization and sensitivity calculations ,pre classification of circuit fault samples ,according to the sensitivity of the results to extract feature information accordingly ,in order to construct fault samples feature set ,and then as BP neu-ral network input the network training and diagnosis .On the filter simulation results show that ,this method can classify different component failure ,the average accuracy of fault diagnosis of analog circuits is superior to traditional methods .