基于Volterra频域核和神经网络的非线性模拟电路故障诊断法

本文提出基于非线性电路频域核的神经网络诊断方法。通过范德蒙特法得到各种工作状态下电路响应的各阶频域核,提取故障特征并适当预处理后与相应的电路工作模式一起构成输入/输出样本集,利用训练样本和测试样本分别对神经网络训练和测试,借助神经网络实现故障诊断。文中给出了各阶频域核的递推算式和求解方法,并通过一个基于递归神经网络的非线性电路诊断实例加以验证。

Fault diagnosis method in nonlinear analog circuit based on Volterra frequency-domain kernel and neural network

The neural network diagnosis method based on frequency-domain kernel of nonlinear circuit is presented. Each order frequency-domain kernel can be obtained using Vandermonde method from the circuit response under all states; circuit fault features are extracted and preprocessed, which are combined with circuit work modes and regarded as input/output samples of the neural network. Training sample and test sample are used to train and test the neural network respectively; and circuit fault diagnosis is realized using the neural network. The recursion formula and its solution method of each order frequency-domain kernel are given; and a nonlinear circuit fault diagnosis example based on recurrent neural network is used to verify this method.