子网络级模拟故障诊断的神经网络实现

Neural Network Approach to Analog Fault Diagnosis at Sub-network Level

利用容差模拟电路节点电压灵敏度序列守恒定理,得到了模拟电路元件的软、硬故障统一样本。然后利用统一样本集训练BP神经网络,并将神经网络用于子网络级模拟故障诊断。实例验证表明,软、硬故障统一样本集使得用于神经网络训练所需样本数目大大减少,但经过训练的神经网络可以诊断容差模拟电路的全部软、硬故障,而且诊断正确率较高。

In this paper,the unified sample for both soft and hard faults of elements in analog circuits is found according to the invariance of node - voltage sensitivity sequence in analog circuits with tolerance. Then BP neural network is trained by use of the unified sample groups and used to analog fault diagnosis at sub - network - level. The experimental results show that samples are reduced extensively but the neural network can diagnosis both soft and hard faults of tolerance analog circuits and have a high rate of accuracy.