基于径向基函数网络的模拟电路故障诊断设计

模拟电路的固有特点使其故障诊断较数字电路困难.相对于BP网络,RBF神经网络具有最佳逼近性能且收敛快、无局部极小,可引入解决上述困难.根据具体电路,定义故障,选定测试点,确定网络结构,用Pspice获得训练样本,经过训练得到RBF网络.网络的输入为从测试点得到的输入向量,输出为对应的故障.为了验证网络的泛化性能,对每种...

Fault diagnosis design for analog circuit based on Radial Basis Function neural networks

By contrast with digital circuits,fault diagnosis for analog circuit is more difficult due to its intrinsic characteristic.Compared with BP network,Radial Basis Function neural network can be introduced to tackle this problem because of its optimal approximation,quick convergence,and no local minimum.According to the specific circuits,the construction steps for RBF network are as follows: define fault,and choose to designate test point,then establish network structure and gain training sample through Pspice,finally obtain RBF network by training.The input for the network is input vector gaining from test point,and the output is the corresponding fault.To verify Generalization performance of the network,we make Monte carol analysis for healthy component under the circumstance that the uniform distribution is 5% for component tolerances,then get validation sample.The diagnostic accuracy is 86.6%,thus verifying validity of this diagnosis.