基于BP网络的线性电路故障诊断

LINEAR CIRCUIT FAULT DIAGNOSIS BASED ON BP NETWORKS

本文研究基于神经网络的有容差线性电路故障诊断问题，提出了把反向传播（ＢＰ）神经网络训练成故障诊断器的具体方法．仿真实验结果表明，该故障诊断器不仅在诊断精度方面超过了所有传统算法，而且使测后诊断速度提高了至少一个数量级．此外，它对神经网络内部硬件故障具有一定的容错能力．

Based on the neural networks, the diagnosis problem of linear circuit with element parameter tolerances is addressed in this paper. Specific strategies are proposed for training the back-propagation (BP) neural network to be a fault diagnoser. The experimental results demonstrate that the diagnoser performs better than any classical approaches in terms of accuracy,and provides at least an order-ofmagnitude improvement in the speed of post-fault diagnosis. In addition, it can tolerate its internal hardware damage to a certain extent.