基于粒子群优化的核主元分析的故障检测方法

Fault Detection of Kernel Principal Component Analysis Based on Particle Swarm Optimization

目的提出使用粒子群优化(PSO)方法进行核参数优化,获得混合核KPCA的故障检测方法.方法引入多项式核函数和高斯径向基核函数的混合核方法,使用PSO对各参数同时进行优化,得到最优的混合核函数,再与PCA相结合,得到基于PSO优化的KPCA.结果根据混合非线性主元特征计算出的T2和SPE统计量,实现故障检测.并且其故障检测率高于径向基KPCA,时间成本低于多项式KPCA.结论通过田纳西-伊斯曼(TE)测试过程以及电主轴系统的应用实例说明了KPCA方法的可行性与实用性.

Particle Swarm Optimization（PSO） is used to optimize the kernel parameter, and a fault detection approach based on hybrid-kernel KPCA is proposed. Definition of mixed kernel function is introduced by combing RBF kernel and polynomial kernel with PCA. Firstly the kernel parame- ter of the two kernel functions and weight coefficient are simultaneously optimized by PSO, to ob- tain the optimal mixed kernel function. Second the optimal mixed kernel function combines with Principal Component Analysis （PCA）, to obtain hybrid-kernel KPCA is optimized by PSO. The fault can be detected on-line by monitoring T2 and squared prediction error（SPE）. Tennessee East- man （TE）process and motorized spindle working process are applied to validate the practicability and feasibility of the improved method.