基于PCA-LDA算法的模拟电路复杂故障在线诊断研究

Research on Diagnosing Complex Analog Circuit Failure on Line Based on An Algorithm of Coupling PCA and LDA

摘要:

模拟电路在出现多个元器件同时故障情形时，由于容差多样、耦合关系复杂等因素，难以对其进行准确在线故障诊断。对模拟电路的在线故障诊断过程进行了定量数学描述，提出了幅频特征值获取方法，将PCA方法和LDA方法相结合，构建属性协方差矩阵、类间散度矩阵、类内散度矩阵，对模拟电路的复杂故障样本数据进行降维与分类，采用BP神经网络对样本数据集与故障模式集进行准确匹配。实验结果表明，论文提出的方法对数据分类降维有效、诊断结果正确，样本数据维度由31降到3，故障分类准确率达到100%，较LDA、PCA、KPCA和KPCA-LDA等其它四种方法，本文方法的指标更优。

Abstract:

It was difficult to diagnose the complex analog circuit on line on the situation which some elements failed because of its various tolerance error and multiplex coupling relationship. Firstly, the quantitative mathematics process of diagnosing the complex analog circuit was given in this paper, and the method of picking voltage-frequency characteristic value was proposed, too. Secondly, the coupling method of principal component analysis and linear discriminant analysis was applied, thus the attribute covariance matrix, the between class scatter matrix and the within class scatter matrix were found. In this way the failure sample dimensions of the complex analog circuit were depressed. Finally, the failure data were precisely matched with the failure modes by applying the back propagation neural network. It was proved by the experiment result that the method of this paper was effective in the domain of depressing samples, classifying data and diagnosing failure. The accurate rate of classifying failure is a hundred percent,and the data sample dimension was reduced from 31 to 3. Compared with other four methods, such as discriminant analysis, principal component analysis and kernelized principal component analysis, the method of this paper was superior to these in respect of classifying data, depressing samples, etc.