基于FRFT-KPCA的模拟电路非线性故障特征提取  

Nonlinear fault features extraction for analog circuit based on FRFT-KPCA

作　　者：孙靖杰[1] 赵建军[2] 王汉昌[1] 乔凤兰[1] 

机构地区：[1]海军航空工程学院青岛校区航空军械火控系,山东青岛266041 [2]海军航空工程学院兵器科学与技术系,山东青岛264001

出　　处：《电机与控制学报》2013年第8期100-106,共7页Electric Machines and Control

基　　金：国家自然科学基金(60802088);教育部新世纪优秀人才支持计划(NCET-05-0912)

摘　　要：针对模拟电路受非线性及元件容差影响而导致响应信号在时域和频域都出现耦合,造成故障特征提取困难的问题,结合分数阶傅里叶变换和核主成分分析理论提出一种非线性故障特征提取方法。利用分数阶傅里叶变换对耦合信号进行预处理,采用粒子群优化算法寻找最优分数阶p,实现耦合信号在分数阶域最大程度的分离。采用核主成分分析对非线性特征进行维数压缩,实现故障特征提取。实验结果表明,在时域或频域相互耦合的信号经分数阶傅里叶变换后,在分数阶域上耦合程度明显减弱,核主成分分析能够有效处理信号中的非线性信息,特征提取效果要优于其他线性特征提取方法。经过分数阶傅里叶变换和核主成分分析相结合的方法所提取的故障特征使故障模式具有更好的可分性。Aiming at the difficulty of fault features extraction since response signals of analog circuit cou- pling in both time domain and frequency domain under the effect of nonlinear and tolerance, a nonlinear fault features extraction method based on fractional Fourier transform （FRFT） and kernel principal com- ponent analysis （KPCA） was proposed. FRFT was used as decoupling pretreatment for signals to seek the optimal fractional order p to separate signals greatly in fractional domain by using particle swarm optimization （PSO）. KPCA was applied to compress the dimension of nonlinear features to extract fault features. The experiment results show that after FRFT, signals coupling in time domain of frequency domain decouple obviously in fractional domain. Compared with other linear feature extraction method- s, KPCA obtains better extraction effect since it analyzes nonlinear information of signals. The obtain features extracted by the method based on FRFT-KPCA enhances the divisible characteristic of different modes.

关 键 词：分数阶傅里叶变换 核主成分分析 模拟电路 类内类间散布矩阵 特征提取 

分 类 号：TP306[自动化与计算机技术—计算机系统结构]