Appropriate Statistical Methods to set up an Alert/Warning Limit

[duouser]

hi, i am currently doing background research on appropriate statistical methods to set up an alert/warning limit to prompt if a process is near the spec limit, could anyone point me in the direction of suitable methods and tools. thanks

 

[Bev D]

Welcome!

you will no doubt get many responses asking:

• why do you want warning limits? what action will you take if you violate a warning limit
• what kind of process are you dealing with?
• How stable and capable is your process?
• How much do you understand about the inputs? can you reliably use them to alter the output in a predictable manner?

Answers to these questions will help us give you the best answers that will actually be useful for your situation.

In general, we wouldn't use alert or warning limits on specifications UNLESS the process is predictable (the relationship of the inputs to the outputs is known and predictable) AND we have the ability to adjust the inputs to reliably achieve the desired output. If we don't have these two conditions we are most likely going to make things worse - this is known as tampering.

But to directly answer your stated question: the tool that fits this is PRECONTROL. Like any tool there is an appropriate time to use it and inappropariate times...I have attached a briefing on Precontrol that I use within my organization.

 

[duouser]

Thanks for your help, much appreciated.

 

[Steve Prevette]

hi, i am currently doing background research on appropriate statistical methods to set up an alert/warning limit to prompt if a process is near the spec limit, could anyone point me in the direction of suitable methods and tools. thanks

Traditionally, Statistical Process Control is used to set up such alarms. Now, it will give an alarm if the process data are shifting. If the process routinely makes things outside the spec limit, you won't get an alarm as this is a routine event!

The idea is to get it so the statistical warning limits are within the specification limits (by improving the process to reduce variation) and then you will be able to detect if the process is starting to drift towards a specification limit.

I have some work on SPC posted at http://www.efcog.org/wg/esh_es/Statistical_Process_Control/index.htm

 

[duouser]

Thank you Steve, your info has helped.

 

[duouser]

could you please tell me which spc chart would be most appropriate for setting an alert limit using historical sample data. thanks

 

[Barbara B]

Alerting or warning limits could be added to all kind of control charts. They give the area which covers approximately 95% of the usual process data (+/-2*S for individual numeric data following a normal distribution), whereas the control limits cover 99.73% of the usual process data (+/-3*S for normally distributed individual numeric data).

In many formulas a "3" or k-factor appears (with default k=3). If you alter this to "2" (k=2) you'll get the warning limits. Formulas could be found e.g. in Engineering Statistics Handbook, Ch. 6.3.1.

 

[Bev D]

could you please tell me which spc chart would be most appropriate for setting an alert limit using historical sample data. thanks

What is the process? How large is your sample size? How often do you sample?

We really need more information to answer this question

And if you could post your data it would help...

 

[Steve Prevette]

could you please tell me which spc chart would be most appropriate for setting an alert limit using historical sample data. thanks

Depends on the type of data. Generally, you can't go wrong with X-individuals, moving range.

 

[duouser]

thanks for your all your help, i have figured out what i need to do.