基于最大异类距离和正则极端学习机的模拟电路在线故障诊断

Online Fault Diagnosis of Analog Circuit Based on Maximum Interclass Distance and Regularized Extreme Learning Machine

针对模拟电路在线故障诊断问题,提出了一种将最大异类距离和正则极端学习机结合起来的新方法。首先,利用最大异类距离对初始特征样本进行特征提取,压缩样本规模,获取维数更低、可分性更好的特征样本集;然后,将提取的特征样本送入正则极端学习机进行训练,再利用训练好的正则极端学习机对待测电路实时状态进行诊断,并不断根据诊断结果和新旧样本之间的相似度更新训练样本集;最后,将所提方法用于模拟电路在线诊断中。结果表明:所提方法能够有效实现模拟电路单、双故障的在线诊断,比ELM效果好。

According to the problem of analog circuit online fault diagnosis,a new method based on maximum interclass distance（MID） and regularized extreme learning machine（RELM） was proposed.Firstly,dimension of feature sample set was reduced using MID to get the new feature sample set with lower dimension and better separation.Then,the new sample set was sent for training RELM,sequently,the trained RELM can be used for real-time diagnosis of analog circuit,and on this process,feature sample set was renewed according to diagnosis result and similar degree between new test samples and old training samples.In the end,the proposed method was used in analog circuit diagnosis online.Simulation results of analog circuit online diagnosis show that the proposed method can diagnose both single and double faults effectively,which is better than ELM.