How do you do a gage R&R when the sample is destroyed or modified during the test? I have several tests such as a rupture test that we want to define the errors but cannot repeat the test with the same samples. Suggestions?

Up until two days ago, I would have said no way. However, I just finished reviewing the PPAPall.xls file - found in the PDF Zone elsewhere in this site. On one of the pages, near the 'regular R&R' page, it has a fill in the blanks form with calculations. While I am NOT any type of math wiz, I do have some objection to this philosophically. However, I believe it is blessed by Delphi, so have at it.

The spreadsheet Batman mentioned appears to be statistically sound. Destructive testing could also be R&R'd as attribute data, but the attribute portion of the same spreadsheet appears to be somewhat 'lacking.'

Regards,
Don

Yea, I guess I should expound upon my last post. Sorry about that.

…the calculations are 'lacking,' so could one get in trouble using it? Trouble with auditors, trouble using the data to make decisions?

Using logic, if the portion of the spreadsheet for attribute R&R is somewhat ‘lacking’ (IMHO), then an argument could be made that other portions of the same spreadsheet may be lacking as well, thus you must be able to justify the use of the calculations to your assessors (assuming the question is even asked) and yourself. My particular problem with the attribute page of PPAPall.xls was that is did not consider marginal samples. The samples either had to all pass or all fail for the gage test to be considered acceptable (100% same results per sample. Plueeeeeeze. Does differentiate ring a bell?). BUT, I did not design nor have input to its design. Therefore, an argument could be made that the authors had a legitimate reason for the calculations, but I cannot explain these reasons to my assessors, nor myself.

If Delphi accepts the destructive portion (or any portion, for that matter) and the calculations of PPAPall.xls, that should satisfy most assessors. But, I would not be willing to gamble my registration on it. I would want to be able to explain how if the subject came up. How I came to the conclusion that the R&R was acceptable, etc. and PPAPall.xls offers no derivations on the calculations involved. Perhaps they may be available from the source, perhaps not. Do not get me wrong, I am neither accepting nor rejecting PPAPall.xls nor its authors. They obviously put a lot of good work into it. And, I applaud Marc for providing it to the public. These are just my thoughts on this matter.

CAN you legitimately perform a Gage R&R in destructive testing?

Having reviewed the destructive portion of PPAPall.xls, these are my particular issues, other than those outlined above. Under the assumption that the test was tolerance based, you could use that if you wished. Under the assumption that the test was attribute based, you may not be able to. Therefore, I humbly propose an alternative. Use process capability.

Process capability tests are available for both variable data and attribute data. Process capability does not require use of the same sample. AND process capability is something that should be acceptable as a measure of R&R, designed correctly. You perform capability on the gage, not the process. However, if the assessors do not want to accept process capability numbers, translate process capability numbers into R&R numbers. I have the method laying around here somewhere, if anyone is interested.

In short, my answer to Batman’s question is yes, assuming you are willing to work outside the normal paradigm.

Regards,
Don

Sorry for the delay, been having trouble logging into the site.
So Don, to answer Orlando's question, CAN you legitimately perform a Gage R&R in destructive testing? Even though there is a form in that PPAPall spreadsheet? As you pointed out, the calculations are 'lacking,' so could one get in trouble using it? Trouble with auditors, trouble using the data to make decisions?

I'd like to see the interpolation from capability to R&R.

Will send by e-mail this weekend. Format is Word 6.0.

Enjoy and critical analysis welcome.

Regards,
Don

Thanks for all the input. I would like to see the method to translate process capability into R&R.

I have been using control charts and cpk for the approvals of destructive methods at this time. The charts must be in control and cpk greater than 1.3 to accept the gage and part. This makes the assumption a bad gage R&R will not allow a cpk of 1.3 or greater. Of course if the part varies widely you are condemning the gage as well.

I cannot download the PPAPall.xls file mentioned. Is their a trick? By the way I cannot get to the listing through the Cayman system. For some reason I have to go straigt to it via a bookmark to "index of /pdf_files".

I cannot download the PPAPall.xls file mentioned. Is their a trick? No trick - it's in how you set up your browser. By the way I cannot get to the listing through the Cayman system. For some reason I have to go straigt to it via a bookmark to "index of /pdf_files".I have no idea why, if you input the correct address, you cannot get a listing in the pdf_files directory. Works fine for me.

I was able to download and open the zip file. Thanks.

I'm a fan of MSA from way back, but isn't a MSA study outside of the scope of the existing MSA manual if the measurement is not repeatable. Refer to MSA manual. By definition if we have a destructive test this infers that we cannot repeat.

From a registrar's perspective, it is impossible for them to give a NC if no MSA has been conducted on any IMTE that conducts a destructive type test because it is outside of the scope of the MSA manual.

Just my thoughts...

Yes, Jada, I agree. I just can't get past the non-repeatability of the thing. Actually, you don't even get reproducability, since the operators can't measure the same parts again.
???

Batman,

R&R on a destructive test is always intersting. When you initially think about it you might think that the majority of the error would come from the equipment(repeatability) but we found that the majority of the error was actually from the the operator(reproducibilty) because of the variation that was involved in the operator setting up the part (in this case a hardness test on a foam pad). So although the R&R study was not valid the raw data we obtained and the subsequent Exploratory Data Analysis (EDA) was very worthwhile.

Has anyone else found similar results?

Jada

Don,

Where is this conversion of Process Capability to GR&R?

Marc was kind enough to post it here as part of a process capability paper I collected. You can find it here:

Elsmar Cove pdf files directory (http://Elsmar.com/pdf_files/) and look for CPK.pdf

You can also find it at my site below in the SQC section.

Regards,
Don

Don,

Thanks for the information. Come to find out I had already downloaded the information back in March. I just had not used it, yet. I do have some destructive type measurement systems to work on in a fluid/batch type environment. I also enjoyed your site, and have placed it in my favorites along with Caymen Systems.

Doug

P.S. I was not able to view the information from your links, but was able to download just fime.

hello,

we have recently improved an existing method to assess repeatability in destructive gauge R&R. It is based on a two time procedure
where repeatability is first counfouded with sites, and then with parts.
Solving a simple equation, it is then possible to estimate repeatabilty and then %R&R and %P/T.

A paper is going to be published in Quality Engineering before the end of the year. I can tell you more if you want.

Regards,

Francois

Bergeret, Franco

i need the Atomative Industry Action Group reference manual about the mesurement system evaluation and the book of evaluating the measurement process by wheeler.
if anybody can help me to get those book.

The AIAG book is available from the AIAG (http://www.aiag.org) and the Wheeler book is probably available from the American Society for Quality (http://www.asq.org)

Yea, I guess I should expound upon my last post. Sorry about that.

Would you please to send the interpolation from process capability numbers into R&R numbers to me by email?

Thanks and best regards,
Chi

I do not have this information readily available on the weekend, but you can derive it fairly easily.

Capability indices are based on the StdDev of the observed process variation.

The observed process variation is composed of the actual process variation and the measurement variation. This means that you may determine the actual process variation by removing the effect of the measurement variation.

The equation is:
(StdDev Actual)^2 = (StdDev Observed)^2 - (StdDev Measurement System)^2

Note: (StdDev)^2 is the StdDev squared (or the Variance).

You may now use StDDev Actual to calculate Cp actual (or another index).

You may use these calculations to develop a series of graphs that plot capability index observed to capability index actual along a line of a given measurement error. You can add additional lines of different measurement errors on the same graph.

I have a question regarding destructive attribute agreement analysis. I tried using the search but maybe I missed it...regardless I could use some help.

How do you set up your sample selection for a destructive attribute Gage R&R? Minitab requires two trials per appraiser, so how do you set that up given the criteria?

Any pointers or direction would be much appreciated!

The MSA shows the steps required to calculate the destructive GR&R Studies.
I found a way to get 11-18% GRR using a good lot/bad lot approach. Our company is accepting GR&R with 20% or less. Let's say that I'm getting values ranging from 50-100 lbs for the pull test (spec is 30 lbs min.). This variation is too high and we will likely have GRR's ranging from 90-100%. If I combine bad lots in the study with values ranging from 8-15 lbs, the GRR will decrease up to 12% in some examples I have.
We are pull testing silicon tubes, PVC syringes and also performing seal strength for pouches and packaging trays. All of the GRR studies are acceptable since they are giving 11% thru 18%.
If I'm doing anything wrong by reducing the variation this way, please let me know.

can you post your data and your math?

I'm attaching two files. Dataset1.pdf has the data info for the pull test. The measurement column shows the odd rows with readings within specs and the even rows with reading below specs. The GRR results shows 13.5% GRR. Column C6 shows all readings within specs and a 100% GRR. The file named as chart1.pdf shows the charts for both Measurement and C6 columns. We deleted the "even" row values from C6 and replaced with values coming from bad lots (i.e.: 138.8 is a good lot, 22.9 is a bad lot, since the spec is 80 lbs min. That means if we combine good and bad lots we will highly reduce the variation for a GRR study.
Any Comments are welcome.

I am confused...what constitutes your matched pairs?
you should be doign repeat measurements on the same part (or for destruct, portions of a part or sequentially produced parts if they can't be 'cut up')

what exactly makes the first and second reading per operator?

We are using one lot per reading. Parts are similar to the other ones.
Readings 1, 11, 21, 31 are coming from a "good lot"- Odd lot
Readings 2, 12, 22, 32 are coming from a "bad lot" - Even lot
and so on.
As you can see on the attached data, the variation on the "odd lots" is significantly higher than the "even lots", so the total variation is reduced.
You can compare that variation within trials and appraisers.

I forgot to mention that the way you setup the "bad" parts will determine the total variation. Let's suppose that the pull test requirement is 30 lbs minimum. You can test about 60 samples (2 operators, 3 trials). Since each trial will be 10 sample parts, then you will need to have the same quantity of good and bad parts for every trial (5 each).
The bad parts preparation will simulate an incorrect process or material. Let's assume that our bad parts will have readings from 5-10 lbs so we will ensure that all of them are bad parts of the same lot. We can start the first trial using a good part in the first sample, then we will use a bad part for the 2nd sample, a good part for the 3rd sample, and so on, until completion of the first trial. The second and 3rd trial will have the same criteria as the first trial. The second operator will repeat the same steps until completion of all samples. You will note that the resulting GR&R variation will be less than 20% so the GR%R may be acceptable. You can use the same approach with almost every destructive test.