Design and Robustness of Some Statistical Quality Control

Design and Robustness of Some Statistical Quality Control Tools Dr. Maria Calzada Loyola University New Orleans

Statistical Quality Control l Statistical quality control (SQC) is the use of modern statistical methods towards the end of assessing and improving a process. l The father of SQC is Walter A. Shewhart. The most widely used tool in quality control is named after him: the Shewhart chart.

Walter A. Shewhart l Developed concepts in quality control based on his work at Bell Laboratories in the 1920’s. l Basic idea: NO TWO THINGS ARE ALIKE. l Two items could be different solely because of “chance causes. ” l But also, two items could be different because of “assignable causes. ”

Chance Causes l CONTROLLED VARIATION: A stable consistent pattern of variation over time. One is able to predict what will happen in the future based on past data. l This type of variation accepts the fact that even when two items are “created equal” there will be differences between the items.

Visualizing Chance Causes At a fixed point in time Target Over time Tim e et g r Ta Think of a manufacturing process producing distinct parts with measurable characteristics. These measurements vary because of materials, machines, operators, etc. These sources make up chance causes of variation.

Assignable Causes l UNCONTROLLED VARIATION: A pattern of variation that changes over time. One is not able to predict what will happen in the future based on past data. l This type of variation tells you that two items were not “created equal. ”

Visualizing Assignable Causes Tim e t e g Tar Here not only the process not “on-target” but its variance is increasing. Possible reasons for this: machines need adjustment, materials are slightly different, workers need training, etc.

Statistical Quality Control l The use of statistical methods to “catch” assignable causes of variation as soon as possible to do something about them. l The most widely used tool in statistical quality control is the Shewhart chart.

Shewhart Chart l We take measurements at different points in time, average them, and plot them together with 3 control limits. l One should also have a Shewhart chart for monitoring the variation. However, for the moment we will disregard this.

An Example From Montgomery, page 237. l The fill volume of soft drinks is an important characteristic. This can be measured by placing a gauge over the crown and comparing the height of the liquid in the neck of the bottle against a coded scale. A reading of zero corresponds to the correct fill height. Samples of size 10 are taken at regular times

Sample Number Average The target is at zero and the variation is three standard deviations from the target. This process is in-control.

This shows the process leaving the control limits at sample 19. However, we can “see” that there may have been a change in the process earlier than that. The problem should be corrected so that the process returns to statistical control.

Average Run Length l The Average Run Length (ARL) is the average number of points that most be plotted before a point indicates an out-ofcontrol condition. l For the Shewhart chart this can be calculated as ARL=1/p where p is the probability that any one point exceeds the control limit.

l We would like the ARL to be long when the process is in-control and short when the process is out-of-control. l For a 3 Shewhart chart and an underlying normally distributed process we have ARL=1/(. 0027)=370. 37 when the process is in-control l For many practical applications this chart has proven very effective and robust. It should be used more.

The EWMA Control Chart l Given individual readings x 1, x 2, x 3, …, the Exponentially Weighted Moving Average Control Chart is defined by. were 0< 1 is a design parameter. An out-ofcontrol signal is given when |Zi|>h, where h is another design parameter.

EWMA Chart for some data

When to use EWMA l Samples sizes) of size 1 (it can be applied to other l We need to quickly recognize a small change in the mean. l For large changes in the mean the Shewhart chart has been shown to be superior.

ARL for EWMA l If L(u) is the ARL given a start at u, then which is an Fredholm Integral Equation of the second kind.

Robustness of EWMA l Borror et al (1999) showed that a properly designed EWMA chart is robust to nonnormality. l Stoumbos et all (2000) is a study of robustness to non-normality and autocorrelation with recommendations.

New tools are being developed l The Synthetic Control chart was introduced in 2000 by Wu and Trevor. Their paper shows that the synthetic control chart detects moderate shifts in the mean faster than the Shewhart chart and the EWMA chart.

The Synthetic Control Chart

Synthetic Robustness l Calzada and Scariano (2001) is a study of the robustness of the Synthetic control chart to non-normality. It is robust to moderate departures from the normal distribution. l There is still much work to be done here.

The Max. EWMA Chart l Introduced in 2001 by Chen et al. l Designed to jointly monitor changes in the mean and variance. l For normal processes this chart shows all the advantages of the EWMA chart.

The ARL for the Max. EWMA chart can be written as Chen, Scariano, and Calzada are working on optimizing the Max. EWMA chart for users.

Robustness of Max. EWMA l Calzada and Scariano have shown that while the Max. EWMA chart is promising for normally distributed processes, it is not robust to heavy tails or skewness. This chart must be used with caution. l There is still much work to be done here.

Conclusion l Industrial applications, like statistical quality control, can be very rewarding areas of research for people with interests in mathematics. l Jobs in this area include academic, industrial, and consulting jobs. l Strong mathematics, statistics and computations background is important.

References l Wheeler and Chambers, Understanding Statistical Process Control l Montgomery, Introduction to Statistical Quality Control, Fourth Edition l Borrow et al, “Robustness of the EWMA chart to non-normality”, Journal of Quality Technology, Vol. 31, 1999. l Wu and Trevor, “A synthetic control chart for detecting small shifts in the process mean”, Journal of Quality Technology, Vol. 32, 2000. l Chen et al, Monitoring process mean and variability with one EWMA chart, Journal of Quality Technology, Vol. 33, 2001 l Calzada and Scariano, “The robustness of the synthetic control chart to nonnormality, Vol. 30, 2001.