摘要
针对具有非高斯、非线性及多工况特性的批次过程,提出一种基于特征量最近邻统计指标的过程监视方法.首先,将批次过程正常工况原始数据投影到其特征空间,提取主元T和平方预测误差SPE,并进行特征量k最近邻距离平方和的求解.然后,采用核密度估计法获得概率密度分布函数,确定统计监视控制限.特征空间的主元T和SPE特征量能全面代表原始数据的有用信息.采用特征量k最近邻建立监视模型将会节省存储空间,提高建模样本数量与变量之比以及检测异常工况的速度.另外,利用局部近邻数据建模可以解决过程具有的非线性和多工况问题,而应用核密度估计法可以解决过程数据具有的非高斯分布问题.最后,在半导体生产过程的成功应用表明了所提方法的有效性.
A process monitoring method based on feature space k nearest neighbor is proposed for batch process with the characteristics such as non-Gaussian, nonlinear and multi-mode. Firstly, the normal mode data are projected into the feature space, the principal components T and squared prediction error （SPE） are extracted, and the sum of k nearest neighbor squared distance is computed. Then the kernel density estimation is used to set the statistical threshold of the normal mode. Feature statistic can represent useful information of raw data comprehensively. Modeling using feature statistic can economize storage space, improve the ratio between modeling sample numbers and variables, and increase speed of fault detection. Furthermore, modeling using local neighbor data can solve nonlinear and multi-modes questions of batch process. The use of kernel density estimation method can solve non-Gaussian characteristics of modeling data. The effectiveness of the proposed method is illustrated by applying it to the monitoring of a semiconductor process.
出处
《自动化学报》
EI
CSCD
北大核心
2014年第1期135-142,共8页
Acta Automatica Sinica
基金
国家自然科学基金(60774070
61034006
61174119)资助~~
关键词
特征空间k最近邻
过程监视
批次过程
半导体生产过程
Feature space k nearest neighbor （FS-kNN）, process monitoring, batch process, semiconductor process
