基于核熵成分分析的模拟电路早期故障诊断方法

针对模拟电路早期故障诊断中存在部分早期故障类别重叠的难点,提出了一种基于核熵成分分析的故障诊断方法。首先应用小波分形分析计算被测电路时域响应信号的小波分形维特征,然后利用核熵成分分析方法进行特征的优选与降维,最后将优选和降维后的特征应用最小二乘支持向量机多类分类器进行区分,其中用于识别重叠故障类别的最小二乘支持向量机的参数由量子粒子群算法优化选择。仿真结果表明,本文提出的核熵成分分析方法能较好地获取故障响应信号的本质特征,并表现出了比其他特征提取方法更好的性能,有助于提高模拟电路早期故障的诊断正确率。

Approach for analog circuit incipient fault diagnosis based on KECA

To solve the overlap of some of the incipient fault classes in the analog circuit incipient fault diagnosis, an approach for analog circuit incipient fault diagnosis based on kernel entropy component analysis (KECA) is presented. The fault response signals are preprocessed by the wavelet-based fractal analysis to obtain the fractal-dimension features, and KECA is employed to extract the optimal features which are used as the inputs to least squares support vector machine (LSSVM) multiclass classifier. Meanwhile, the parameters of the LSSVMs which are used to classify the overlapped incipient fault classes are selected by quantum-behaved particle swarm optimization (QPSO) algorithm. The simulation results show that the proposed approach can acquire the essential features of fault response signals and better performance than other approaches is demonstrated, It is conducive to improve the accuracy of analog circuit incipient fault diagnosis.