可诊断非线性电路直流故障的神经网络方法

测后诊断速度和诊断精度是模拟电路故障诊断性能的主要衡量指标。文中将神经网络的自学习和分类技术应用于非线性电路直流故障诊断，把反向传播（ＢＰ）网络训练成一部能诊断软、硬单故障的故障字典。考虑元件参数容差对诊断的影响，提出了优选训练样本的具体方法。此外，重新定义了ＢＰ网络的输出误差函数，使网络在训练时有较大的自由度。ＢＰ网络高度并行的信息处理能力决定了这种新型故障字典的诊断速度非常快。仿真实验结果表明，神经网络方法的综合性能要优于传统的故障字典法。

Neural network approach for DC fault diagnosis in nonlinear circuits

For analog circuit diagnosis, the post fault diagnostic speed and the diagnostic accuracy are the main indexes of its performance. In this paper, the self learning ability and the classification function of the artificial neural network have been used to nonlinear DC fault diagnosis, training a back propagation(BP) network as a fault dictionary that can diagnose single soft and hard faults. The specific method for optimally selecting the training patterns has been presented with the element parameter tolerances taken into account. Furthermore, the output error function of BP networks has been redefined to make the network have relatively high freedom degree while training. Therefore the network with relatively small net size can fulfill the same diagnostic task. The obtained novel fault dictionary can diagnose fault fast owing to the high parallel information processing capabilities of the BP networks. Simulation results demonstrate that the neural network approach is superior to the traditional fault dictionary approach.