Search the Elsmar Cove!
**Search ALL of Elsmar.com** with DuckDuckGo including content not in the forum - Search results with No ads.

Gage R&R on alloy

#1
I have a basic question about determining the samples. We manufacture alloys which have a specification, for example Ca 0.100 - 0.120%. We have been requested by our customer to perform a Gage R&R on this specific alloy. Do I pull samples from several lots which range between the 0.100 - 0.120% or do I pull samples from just one specific lot number?
 

Mikey324

Involved In Discussions
#3
You should pull samples that span the expected range of the process. If the measurement spread is too small, you wont have enough "part to part" variation. Your EV or AV will be high.
 
#4
We have debated this several times at my group meetings. My take was to get samples over a range with different lots, which still may not cover the entire spec range, from several lot numbers, while some of my colleagues have argued it should be all samples from a single lot number. One lot number will not span the entire specification.
 

Bev D

Heretical Statistician
Staff member
Super Moderator
#5
there is no reason to debate this. the requirement of the study design is to get parts that span the range of expected variation. The actual process variation may be more - or less than the specification. There is no advantage, excuse or reason of any kind to restrict your samples to one lot - unless of course you are hoping to bias the results in some fashion. There is no requirement to sample from many lots unless it's necessary to meet the actual requirement of spanning the range of expected variation.

Do you have any historical data that would inform you of the range of expected variation?
 
#6
I have historical data for some of the other alloys we manufacture and report Cpk / Ppk values to them quarterly, but on this new alloy they are requesting a Gage R&R to be completed for the PPAP. There is a new quality manager and has included a Gage R&R to be included in the PPAP. They are only requiring 5 samples for the PPAP study, so I guess I will take a couple samples from each lot. We have only produced 1 lot of this so far.
 
#7
I have historical data for some of the other alloys we manufacture and report Cpk / Ppk values to them quarterly, but on this new alloy they are requesting a Gage R&R to be completed for the PPAP. There is a new quality manager and has included a Gage R&R to be included in the PPAP. They are only requiring 5 samples for the PPAP study, so I guess I will take a couple samples from each lot. We have only produced 1 lot of this so far.
And therein lies the problem. You only have one lot to pull from, which doesn't span the expected range. So how can you do a Gage R&R for PPAP?
 
#8
Yes. I have made this comment to my customer and they still require a Gage R&R to be completed. I really don't want to do this as a useless exercise, and want to make it useful to me, but it seems that I may just need to go through the motions for this customer request.
 

Mikey324

Involved In Discussions
#9
And therein lies the problem. You only have one lot to pull from, which doesn't span the expected range. So how can you do a Gage R&R for PPAP?
You are absolutely right. This is going to be a bigger problem in the future. With improvements to process controls, you wont be able to get the range needed to do an R&R for a PPAP, especially for those who have just done the 300 pc production run.

In my experience, we make good parts with great capability results. Then we do an R&R with parts that have very little variation. So what are your options? Purchase a gauge that is overpowered for the job?? I wont buy a $10k gauge for a $2 job.

The best option, in my opinion, is to use similar parts/products that have been running longer. That way you may have the variation needed to do the R&R.
 

Welshwizard

Starting to get Involved
#10
GRR studies will rarely represent the range of variation that you will encounter in production. The studies do a reasonable job of estimating measurement variation and it sounds like in your case this this will be the limitation.
One of the aspects which doesn't help is the way that AIAG type studies try to separate part variation from measurement variation. This is a topic which is too long to be talked about here, however if its important to your company to know how the measurement variation interacts with the part variation i recommend that you report on the Intraclass Correlation Coefficient (ICC).

Don't be put off by the title or by the fact that normal software doesn't compute this directly, the ICC can be computed straight from your normal GRR software by utilising ICC = 1- GRR squared/PV squared ........assuming of course that the measurement process is consistent.

This number is far more useful than trying to split out the PV like the AIAG studies do and it properly describes how measurement variation interacts with product variation.

In summary, if you have to describe how your measurements interact with part measurements and you have a choice choose ICC and give yourself at least a chance of proper representation. If the ICC is not what showing what you want then you will need a better estimate of part variation from somewhere, this is more likely to be true of the AIAG which will tend to overestimate the effects of measurement variation on the parts.

For more information on ICC see the reading room on Dr Donald Wheeler's website www.spcpress.

Hope this helps somewhat
 
Top Bottom