改进LDA在模拟电路故障预测中的应用  

An improved LDA algorithm and its application to fault prognostic for anglog circuits

作　　者：金川[1] 李小珉[1] 田沿平 

机构地区：[1]海军工程大学电子工程学院,湖北武汉430033

出　　处：《电子设计工程》2015年第8期103-105,共3页Electronic Design Engineering

基　　金：总装预研基金(9140A27020113JB11393)

摘　　要：针对模拟电路早期故障识别难度大的问题,提出一种改进线性辨别分析法和隐马尔科夫相结合的故障预测方法。首先设置元件的参数,提取幅频特征;然后采用改进的线性辨别分析(LDA)对电压特征进行提取消除特征的冗余性和高维性;最后将提取的特征用于训练和测试HMM,以实现模拟电路的状态监测。通过实验验证了其具有良好的模拟电路早期故障监测能力。Diagnosis of incipient faults for analog circuits is important yet difficult .By combing LDA to reduce feature dimension HMM to identify fault category to evaluate its health performance. Firstly, the corresponding frequency features are extracted from analog circuit with its components change gradually and the voltage feature vectors are extracted from analog circuits. Secondly, due to redundancy and high domain of original features, the superior features are obtained by using the improved LDA . Finally, the processed feature vectors are used to form the observation observation sequences, which are sent to HMM to accomplich the state monitoring . Our method owns the best ability to monitor analog circuit state.

关 键 词：线性辨别分析 特征提取 故障诊断 状态监测 隐马尔科夫模型 

分 类 号：TN710[电子电信—电路与系统]