Abstract: Aiming at the correlation diagnosis problem of multivariate process quality management,an algorithm based on typical correlated component pairs group was proposed.By using theorem of correlation decomposition,the correlation of all quality components was decomposed as a series of correlations of component pairs,and the decomposition results could be represented by a correlations set of typical component pairs group according to the transitivity of component pairs'correlations.Algorithm of optimal typical correlated component pairs group based on maximum correlation spanning tree was proposed,and T2 control charts of typical component pairs were established to form the correlation diagnosis model.Theoretical analysis and practice proved that the proposed algorithm could reduce the scale of diagnosis system and eliminate the redundancy of diagnostic messages effectively.

Key words: multivariate process quality, correlation diagnosis, typical correlated component pairs, maximum correlation spanning tree, T2 control chart

摘要： 针对制造过程中的多元工序质量相关性诊断问题,提出一种基于典型相关分量组的诊断算法。利用相关性分解定理,将质量分量的总体相关关系分解为一系列分量对之间的相关关系;以分量对之间相关关系的传递性为依据,将全部分量对之间的相关关系通过一组典型相关分量对之间的相关关系进行表示;提出了基于最大相关树的最优典型相关分量组求解算法,并针对全部最优典型相关分量对建立T2控制图,构成多元工序质量相关性诊断模型。理论分析和实验表明,基于典型相关分量组的工序质量相关性诊断算法能够有效减小诊断体系的规模,提高诊断效率,并能有效消除诊断信息的冗余性。

关键词: 多元工序质量, 相关性诊断, 典型相关分量组, 最大相关树, T2控制图