基于投影变换的多元统计过程控制与诊断模型

针对多元统计过程控制图中变异源的识别问题，给出了基于投影变换的多元过程控制和诊断模型，将原始样本数据通过投影变换，转换为相互独立的数据，然后对各独立变量分别构建控制图，从而达到控制多元过程的目的．恰当地构造变换矩阵，使转换后的样本数据的协方差矩阵为单位矩阵．由于转换后的各变量相互独立，可通过多个控制图的联合报警概率的计算，建立联合概率控制图，对过程进行整体控制；利用转换矩阵，可发现导致异常的原因．最后通过对手机摄像头聚焦度测量结果的分析将该方法进行了验证．结果表明，所提的多元过程控制方法与产控帆图的结果基本相同，并可以指示出过程变异的原因．

Multivariate Statistical Process Control and Fault Diagnosis Based on Projection Transformation

In view of the identification of fault source in multivariate statical process control(MSPC)charts,a model of multivariate process controlling and diagnosis based on projection transformation is proposed.In this model the correlated variables are transformed to independent variables which can be controlled with different SPC charts.A properly constructed transform matrix can make the covariance of the transformed sampling data to be an identity matrix.At the same time,a joint probability of these independent variables can be calculated and the process can be controlled with a joint probability control chart and it is useful for process diagnosis with the projection matrix.At last,a process of camera focus measurement in mo-bile phone assembly is studied.It is verified that the proposed method gets almost the same results as the T2 control chart and the fault source can be indicated by the independent control charts.