摘要
为克服FD-kNN算法的计算量和存储量特别大,PC-kNN主元仅仅能体现过程中线性信息的不足,提出一种基于KPC-kNN的故障诊断方法.在KPCA提取非线性信息后,在核主元空间里应用kNN算法,计算k个最近样本的距离平方和作为统计指标,使用核密度估计方法计算训练空间的控制限.半导体工业实例的实验结果验证了所提方法的有效性.
Aiming at the problem that large computational and storage capacity of FD-kNN and inefficient in reflection nonlinear information of PC-kNN,a fault detection method based on KPC-kNN(Kernel Principal Component-k Nearest Neighbor) is proposed.kNN algorithm is applied in KPCA feature space after the nonlinear information has been extracted by KPCA.Then the sum of k nearest neighbor squared distances of KPC is computed as stastical indicators and kernel density estimation is used to set the statistical threshold of normal mode.The experiment results of semiconductor industry show the good performance of the proposed KPC-kNN method in fault detection.
出处
《沈阳化工大学学报》
CAS
2014年第2期170-174,共5页
Journal of Shenyang University of Chemical Technology
基金
国家自然科学基金资助项目(60774070
61034006
61174119)
辽宁省教育厅科学研究项目(L2013155)