模拟电路故障诊断的多小波神经网络算法

提出了基于多小波变换和能量归一化预处理的模拟电路故障诊断多小波神经网络算法.这种方法能够有效提取故障信号特征,从而减少小波网络训练时输入层和隐层节点的个数,减小网络的规模,降低计算的复杂度,也加快了训练速度.最后在Matlab和模拟电路仿真软件IsSpice4下对算法的收敛性能进行了仿真比较,结果表明基于多小波变换的算法能够对模拟电路的故障进行有效诊断和定位,而且收敛速度比小波包变换更快一些.

Fault Diagnosis Algorithm of Analog Circuit Based on Multiwavelet Neural Network

A new multiwavelet neural network method for fault diagnosis of analogue circuit based on multiwavelet transform and energy normalization preprocessing is presented. By the advantage of multiwavelet, the fault signal feature can be extracted effectively. Thus, the total number of wavelet neural network nodes and computation complexity can be reduced availably, and the learning and convergence speed can be also improved. Finally, The simulation experiments using Matlab and IsSpice4 software show this algorithm is effective for fault diagnosis of analogue circuit, and multiwavelet transform algorithm is better than wavelet packet transform algorithm.