容差模拟电路软故障诊断的神经网络方法

Neural Network Based Soft Fault Diagnosis of Analog Circuits With Tolerances

为了提高诊断容差模拟电路软故障的速度与准确性,提出了一种随机算法、灵敏度分析、免疫遗传算法与神经网络相结合的软故障诊断方法。该法首先利用基于随机算法的灵敏度分析来解决电路发生软故障时测试节点与激励信号频率选择困难的问题,然后对待测电路施加所选的激励并在所选择的测试节点处提取节点电压,这些电压值再经主元分析与归一化处理作为故障特征,输入神经网络。为了解决传统BP算法本身固有的易陷入局部最优等缺点,引入免疫遗传算法来进行优化,形成基于免疫遗传算法的BP神经网络,进行故障分类。本文详述了其诊断原理及诊断步骤,并通过电路诊断实例,验证了所提方法的有效性。

In order to increase the speed and improve the accuracy of soft fault diagnosis in tolerance analog circuits, a new soft fault diagnosis approach, which is based on Randomized algorithms （RAs）, sensitivity analysis, immune genetic algorithms （IGAs） and neural networks, is proposed. First, the proposed RAs based sensitivity analysis method allows for removing the difficulties in the selections of input stimuli frequencies and the most suitable test nodes for faulty circuits. Then, the system uses the selected stimuli to excite the circuit, samples its outputs and preprocesses them by principal component analysis （PCA） and normalization to generate optimal features for training the neural network. In order to overcome the shortcomings that back propagation （BP） algorithms suffer from the problem of getting stuck at local minima, the IGAs are introduced to optimize the BP neural networks （BPNNs） and IGA-BPNNs based fault diagnosis system is formed. The diagnosis principles and steps are described. Finally, the reliability of the method is shown by a practical example.