小波神经网络在模拟电路故障诊断中的应用

Applying Wavelet Transform and Neural Networks to Fault Diagnosis in Analog Circuits

研究模拟电路优化问题,电路系统存在非线性和漂移会引起系统故障,针对BP神经网络在模拟电路故障诊断上存在的收敛速度慢、易陷入局部最小等不足,为解决上述问题,提出基于小波神经网络的模拟电路故障诊断方法。采用正弦信号仿真模拟电路,应用小波变换对模拟电路响应的采样信号进行故障特征提取,建立故障字典,利用神经网络对各种状态下的特征向量进行分类决策,实现模拟电路的故障诊断。故障诊断仿真表明,保证较高故障诊断正确率RBF网络的训练次数得到了极大地缩小,极大地提高了模拟电路故障诊断的效率,为设计提供了依据。

Aiming at faults diagnosis of analog circuit,a method of fault diagnosis in analog circuits was proposed which Combines the time-frequency location and multiple-scale analyzation of wavelet transform（WT）with the nonlinear mapping and generalizing of neural networks.Sinusoidal input to the analog circuit was simulated and its output was sampled in time domain to collect training data for neural networks.The collected data was preprocessed by WT to generate fault features and build a fault dictionary.Feature vectors under certain states could be classified using neural networks,and fault diagnosis in analog circuits was realized.Simulation results on benchmark circuits show that the training epochs of RBF ANN are less than that of BP ANN dramatically.The shortcomings of BPNN,such as low convergence speed and prone to fall into the local minimum points are overcame.And the diagnosis efficiency is enhanced.