模拟电路故障诊断的神经网络方法

Fault Diagnosis in Analog Circuits Based on Neural Network

给出了模拟电路软故障诊断的神经网络方法,利用蒙特卡洛分析,取其能反映故障信号特征的成分做为电路故障特征,而且在网络训练之前,利用主元分析降低了网络输入维数,再输入给神经网络,不仅优化了网络结构,并提高了辨识故障类别的能力。实验证明了这种方法的可行性与适用性。

The paper presents the method of fault for tolerance analog circuits with the neural network; using Monte Carlo analysis, it selects the fault signal that can reflect the characteristics of circuit components as fault features, before the network training,it uses principal component analysis to reduce the importation of network dimension,and then enter it to the neural networks. This not only predigests the network structure but also improves the identification capabilities of fault types. Experiments show the feasibility and applicability of this method.