Πολυδιάστατα διαγράμματα ελέγχου διεργασιών

Σοφικίτου, Ελισάβετ Μ.

Multivariate control charts

Master Thesis

Author

Σοφικίτου, Ελισάβετ Μ.

Date

2013-02-28

Abstract

Nowadays, the concept of Statistical Quality Control is strongly associated with the process control charts. In fact, these charts are the main statistical tool that is used to identify and improve the quality of a single product and the quality of the entire production process as well. A control chart is used for the early detection of "abnormal" behavior of a process. Typically, this is achieved by launching separate-univariate control charts, each one monitoring one of the characteristics to be investigated. However, it is quite common the quality of a product to be determined by the value of more than one quality characteristics. The possible dependence which is usually present among these characteristics makes it imperative to use multivariate control charts, which allow the simultaneous control of all dependent characteristics of interest. If such a diagram indicates that the process is out of control, one may also be interested to detect which characteristic or characteristics that caused the out of control shift of the process. The present dissertation initially provides an introduction to Statistical Quality Control and its history, a description of a typical control chart and its categories. Next, the main multivariate control charts, both parametric and non-parametric, are presented analytically. These charts are a straightforward generalization of the univariate ones that are presented in brief as well. In addition, we illustrate several methods that are capable of identifying the characteristic or characteristics that have caused an out-of-control signal. Finally, we present a detailed comparison of the several univariate and/or multivariate control charts, which have recently appeared in the literature.