Hi guys!
Ever heard of multi-variate Control Charts?
How do you simultaneously control 2 or more correlated variables/parameters?
What's the most practical (yet effective) approach?
Thanks in advance!!!

>>rey

Hi guys!
Ever heard of multi-variate Control Charts?
How do you simultaneously control 2 or more correlated variables/parameters?
What's the most practical (yet effective) approach?
Thanks in advance!!!

>>rey

From a practical view, I would ask the following questions first:


Do I need to control both? If the two variables are indeed correlated, will control of one of the variables provide adequate control of the other?
Is there a process parameter that is the dominant control of both characteristics? Will control of that process parameter provide sufficient control of the two characteristics?

If you must control both, see the NIST Engineering Statistics Handbook (http://www.itl.nist.gov/div898/handbook/pmc/section3/pmc34.htm) for information on multivariate control charts.

Hi guys!
Ever heard of multi-variate Control Charts?
How do you simultaneously control 2 or more correlated variables/parameters?
What's the most practical (yet effective) approach?
Thanks in advance!!!
>>rey

Ch 15 "Multivariate Charts" in "Advanced Topics in SPC" by Wheeler, SPC Press, gives a detailed discussion.

Some time ago I was interested on the same, but didn't found the "Multivariate charts for dummies" description of such charts (the PCA ones) :(, alltho the interest was in order, I found "Hotelling’s T2" and some references to PCA (Principal Component Analisys) and the Q Chart.

I googled on the NET searching for PCA and Control Charts.

http://www.psa2005.com/downloads/posters/071.doc
www.aueb.gr/pympe/hercma/proceedings2005/H05-FULL-PAPERS-1/H05-STATS-1/hercma-2005-bersimis-panaretos-psarakis.pdf

I wich you more luck, that I have, but this link may give the direction or topics that can take you near to your goal. I you find something, please share with us.:D

I googled on the NET searching for PCA and Control Charts.
...
Take caution with many papers such as these, when it is immediately obvious that the authors consider Control Charts to be probability charts. They are not. Understanding this difference is essential to understanding SPC. Even with simple SPC, popular authors such as Montgomery make this major error.

Thank you all for the responses.

I find this very interesting.


Do I need to control both? If the two variables are indeed correlated, will control of one of the variables provide adequate control of the other?
Is there a process parameter that is the dominant control of both characteristics? Will control of that process parameter provide sufficient control of the two characteristics?



After studying the process more, i think unfortunately i really need to monitor/control both. The process is actually an electrical testing simulation in which using a machine i plot a sine wave of a current. The variable of interests are things such as estimated amplitude, wave noise, etc. (estimated means it is based on some algorithm/formula). This is the case where i have a trade-off between 2 to 8 variables. Improvement in 1 means degrading 1 or more of the others. I am really hoping that im approaching the problem correctly by doing multivariate control charts.

Darius, thanks for the links.These are very useful. I'll keep you posted if i find a very good solution to this.

Artichoke, i USED to be a Montgomery fan. I have a collection of his books. That was before my thinking (as influenced by him, esp on control charts and normal curves, but he is still good btw :)) was challenged by this forum, esp you and Steve P. Now im reading more of Dr. Wheller's works, and maybe after that the original works of Shewhart and Deming. The first thing i realized: The academe gave so much emphasize on assumptions of Normality. :bonk:Something that really gave me a hard time when doing real work in the real world.:frust: :2cents:

After studying the process more, i think unfortunately i really need to monitor/control both. The process is actually an electrical testing simulation in which using a machine i plot a sine wave of a current. The variable of interests are things such as estimated amplitude, wave noise, etc. (estimated means it is based on some algorithm/formula). This is the case where i have a trade-off between 2 to 8 variables. Improvement in 1 means degrading 1 or more of the others. I am really hoping that im approaching the problem correctly by doing multivariate control charts.

Is there an opportunity to perform a series of DOE to better understand the control factors (X), then optimize and control them instead of the output variable (Y)? This would be more direct and easier.