Setting Specification Limits

Hello Group,

I would like to get some inputs for setting up the specification limits, we have done some developmental work and got 90 data for every test (3 tests) performed related to the performance of the product.
I need some input/ guidance on setting the spec limits, from the sigma value can I set 3 sigma on both the sides will it be a good option to begin the production with that spec limits ?

Variable N N* Mean SE Mean StDev Min Q1 Median Q3
(units) 90 0 26.41 0.39 3.75 11.00 24.20 26.65 28.62

Variable Maximum
(units) 34.000

Thanks,
Rk

Re: Setting up Specification limits

Sorry, but you have a major conceptual error here.

Specification limits come from the customer, from the engineering specifications for the product being made. The specification limits are completely independent from what your actual production numbers are.

Now, a lot of folks use "statistical process control" / control charts to see what the data are telling them. The control limits on a control chart are set at the baseline average, plus or minus three standard deviations.

Re: Setting up Specification limits

Thanks Steve for your repsone .

Its a bit tricky in my situation , the customer wants us to decide the spec limits and like I mentioned it was just a developmental study with 90 data points , and the product is from coating a web so there is no dimensional check or stuff like that...the tests are related to web tackiness , shear , peel etc.

Yes you're right

"Now, a lot of folks use "statistical process control" / control charts to see what the data are telling them. The control limits on a control chart are set at the baseline average, plus or minus three standard deviations. "

Thats what I exactly meant...to start with will that be ok to have 3SD limit up ( positive) and down ( negative ) was my question

__________________

Re: Setting up Specification limits

Yes, having the 3SD up and down for UCL and LCL is correct.

Now, there is a need for discussion of the calcuation of the standard deviation. Most authors using some form of a range calculation (the easiest being 2.66 times the average moving range) rather than the statistical standard deviation.

You should also plot your data. Although the overall average and standard deviation may be as you calculated, you want to be sure there is not a trend in the middle of your data that needs to be accounted for in predicting the future results.

Re: Setting up Specification limits

The Upper and Lower limits can be defined using 2.66 x MR bar or +/-3 X sigma.

The trending of the chart in terms of Mixture, clusters, oscillation and trend can be understood by run charts but still what will be the spec limit for the product.

Because Spec and UCL/ LCL or two different things and in my case the spec is not defined....

Re: Setting up Specification limits

I recommend specifications be +/- 4 standard deviations of a capability study, which will permit +/- 3 standard deviation process to be capable, as well as allow for variation not captured in a short term study (see total variance equation). Of course, that applies to normally distributed variation.

Re: Setting Up Specification Limits

While this all sounds interesting - it's a bit like walking into a resturant and ordering a meal - the waiter says "what would you like" and the customer says "I don't care (or know) you decide" and if the customer doesn't like the meal the waiter pays.

Specifications as mentioned earlier are obtained from the customer through design characteristics.

Control Limits come from the variation in the process. (Data pulled from the process)

Being capable is good, however if you can't show stability everything else is in question.

I would suggest that 90 data points would not allow you to show valid capability or stability - not enough information to capture all the variation over time.

You did mention that this was only a developmental study - so your not really leaving yourself open to liability - if this were under different circumstances (RFQ) you would most likely question the reason they don't know what they want.

Tom

Re: Setting Up Specification Limits

TShepard has it exactly correct. Your customer is asking to slide a very steep slope to a burning pit.

My recommendation would be to have a very frank discussion with them to discuss what their functional, performance and assembly needs are. THEY should provide you with the specifications. Or in a very collaborative relationship, they would provide you with their product and assembly needs are and you would develop your final specifications that ensure their needs are met.

On the other hand, if you are preparing process specifications for characteristics not specified by the customer - +/- 4 SD works pretty darn well.

bobdoering said:

On the other hand, if you are preparing process specifications for characteristics not specified by the customer - +/- 4 SD works pretty darn well.

Click to expand...

Definitely true, as the supplier you have it pretty well made if the customer accepts those limits - ASSUMING that the data were indeed stable for the duration of your study.