动态可调主元分析的多元自相关质量控制方法

The Dynamic Regulated Principal Component Analysis for Multivariate Autocorrelation Process Quality Control Method

为解决在复杂产品制造过程中由多元质量数据自相关引起的大量虚发报警问题,以提高产品制造过程中监测的准确性,提出了一种动态可调的主元分析(DRPCA)方法。通过对主元分析方法引入折息因子,构建了多个参数可调的动态主元分析(PCA)算法。利用实验数据计算各变量的特征值和贡献率,来确定主元个数,降低多元自相关过程的计算复杂度,消除数据间的自相关。根据计算的各变量的每个主元负荷值来确定主导变量,结合多变量控制图和主导变量控制图,对多元自相关制造过程进行了实时监测。研究结果表明:DRPCA方法不仅能够消除数据间的自相关,减少制造过程中的虚发报警,而且可以准确地检测出制造过程中的异常变量。

To solve the fault alarm problems from multivariate quality data autocorrelation in the product manufacturing process, the dynamic regulated principal component analysis （DRPCA） method is proposed. The dynamic PCA algorithm of multiple adjustable parameters is constructed by introducing the discount factors. The principal component number is determined by calculating the cumulative contribution and the characteristic value of variables via experimental data. The autocorrelation among data is eliminated to decrease the computational complexity in autocorrelation process. The main variables are found according to the load values by the variables with the principal component. The multiple variable chart and single value control chart are adopted to real-time monitor of the manufacturing process, which enables to eliminate the autocorrelation among data and reduce the false alarms, and to find out the abnormal variables in manufacturing process accurately.