CAJ | 학술논문

针对电子装备模拟电路故障诊断过分依赖专业技术人员和诊断专家的不足，提出利用信息融合技术，综合极限学习机（ELM，Extreme Learning Machine）、支持向量机（SVM，Support Vector Machine）以及BP（Back Propagation）神经网络等智能故障诊断模型，对模拟电路软故障进行诊断的故障诊断方法。通过对不同模型分别输入不同频率的电压信号，得到每个模型的诊断结果；采用DS (Dempster-Shafe)证据理论对每个模型诊断结果的可信度进行评估，确立每个模型诊断结果的组合置信度。通过不同模型诊断结果的决策层融合，最终获得诊断结果。以某型装备滤波电路的故障诊断为例，多模型融合诊断结果的准确率比单一方法模型的诊断准确率有了明显的提高，证明了该方法的有效性。
침대전자장비모의전로고장진단과분의뢰전업기술인원화진단전가적불족，제출이용신식융합기술，종합겁한학습궤（ELM，Extreme Learning Machine）、지지향량궤（SVM，Support Vector Machine）이급BP（Back Propagation）신경망락등지능고장진단모형，대모의전로연고장진행진단적고장진단방법。통과대불동모형분별수입불동빈솔적전압신호，득도매개모형적진단결과；채용DS (Dempster-Shafe)증거이론대매개모형진단결과적가신도진행평고，학립매개모형진단결과적조합치신도。통과불동모형진단결과적결책층융합，최종획득진단결과。이모형장비려파전로적고장진단위례，다모형융합진단결과적준학솔비단일방법모형적진단준학솔유료명현적제고，증명료해방법적유효성。
In view of the over dependence of experts and the shortcomings in traditional expert systems for circuit fault diagnosis,a new method of fault diagnosis using information fusion technology is put forward. First of all,three models including Support Vector Machine (SVM),Back Propagation Neural Networks(BP NN)and Extreme Learning Machine(ELM)are chosen to diagnose the soft faults of a analog circuit from different aspects. Choose three kinds of signals of different frequencies as the input for every model. Then Dempster-Shafer (DS)evidential theory is used to evaluate the credit of diagnose results of the above three models. The credits are combined to calculate the weights with DS evidential theory. The final diagnose result is obtained by the weighted diagnosis methods. At last,a fault diagnosis simulation experiment on a filter circuit shows the effectiveness of the proposed method.