GLR control charts for monitoring the covariance matrix of bivariate normal processes
Jiayi Zhow 1 · Gyo-Young Cho 2
12 Department of Statistics, Kyungpook National University
Received 20 August 2018, revised 13 September 2018, accepted 17 September 2018
Abstract
If we want to detect both small shifts and large shifts in means and variances, we use the generalized likelihood ratio (GLR) control chart, in which the range of shift sizes in the parameter does not need to be specified, but can be estimated from the pro- cess data. There have been some works on developing GLR control charts specifically for the problem of monitoring mean and variance. GLR control charts for monitoring the process mean of normal distribution were investigated. Also GLR control charts for monitoring the Bernoulli process were investigated. In the multivariate case, GLR control charts for monitoring the process mean vector of a bivariate normal distribu- tion were investigated. In this paper, we will investigate the GLR control chart for monitoring the covariance matrix of bivariate normal process.
Keywords: Bivariate normal distribution, change point, covariance matrix, GLR control chart, SSATS.
1. Introduction
Statistical process control (SPC) is the important concept of statistical quality control.
In SPC, when we want to detect the quality characteristic of some variables, it is usually necessary to monitor means and variances of the quality characteristic. In this situation, we use control charts. The purpose of control charts is to detect the assignable causes through the mean and variance. A good control chart produces a few false alarms when the production process is in-control and detects the shift as soon as possible when the production process is out-of-control.
The first control chart is named Shewhart control chart. It can be monitoring the large shifts of means and variances. But if the shifts are small or moderate, it has less effective.
So some people developed the cumulative sum (CUSUM) control chart (Page, 1954) and the exponentially weighted moving average (EWMA) control chart (Robert, 1958). But if the shift that occurs is not close the specified size, these control charts are also have less effective. If we want to detect both small shifts and large shifts in means and variances, we
1
Graduate student, Department of Statistics, Kyungpook National University, Daegu, 702-701, Korea.
2