Gage R&R for sorting gage - Leaktester to inspect for leakrates on parts



We currently use a leaktester to inspect for leakrates on parts we produce. This inspection is done as a 100% sort. We exspect approx 5-25% of the parts we inspect to not pass. Our customer has requested that we reduce our upper limit on our test based on gage R and R results. I understand the reasoning behind his request, but is there any data/information I could use to justify testing at 100% of the upper spec tolerance?

Laura M


I was in a similar situation with leak testing. Our test limit was at 10% less than the spec for the same reason - based on manufactures stated error - and verified with leak test masters. Example - if spec was 3 cc/min, upper test limit was 2.7.

I really don't have any advice to change things. Like Dr. Deming said "no two things are the same - and if they were you couldn't measure them accurately enough to prove it" or something like that.

Interesting thing in our case was we didn't typically have too many that failed in the margin of error. They were either "real good" or "real bad." Borderline parts rarely existed. You may be able to learn something from the parts in between the test spesc and actual spec that can help. Without knowing more about the process, that's the best I can offer.

jane_ackerman - 2010

Have you ever performed an R&R on this leak tester? If so, could you share how this was performed so I can share the info with someone.

Bill Ryan - 2007

Have you ever performed an R&R on this leak tester? If so, could you share how this was performed so I can share the info with someone.
We use air decay exclusively (thus far). I know that we have taken our die castings and sent them out to get plated (once we verified leak rates - sometimes with an outside lab). I have yet to see a "pretty" MSA study on a leak tester. We, typically, take 5 parts, 2 operators, 2 trials. A few examples of the variables we've run across are part temperature, ambient temperature, part cleanliness, orifice calibration, ....

jane_ackerman - 2010

Thanks Bill... this question is actually for my husband's shop. I don't know much about their processes, but they have a QE who is trying to do a R&R and is claiming the test equipment is bad. But from what my husband explains, as you do the leak test, it actually gets better and better the longer it's tested. Thus, construing an actual R&R result.
Another question asked was... is the same formula used to calculate this type of R&R. I have never dealt with this type of issue and but it seems the answer would be no... and I don't know what kind of formula would be used to verify Repeatability and Reproducability of this type of test, when it is expected to see the results improve over time.
They do have known samples (failures and a range of good) that correlate (within less than 10%) with the machine test vs a manual test, and they use these samples to verify the leak tester weekly, but this does not constitute anything to do with an R&R.
Any suggestions?

It Was I Who

SQA Supplier Quality Assurance Manager
Im jumping in into this thread becuase i have some questions about this aswell or maby get some feedback on how i will conduct the test

Im set to do a MSA study on our leaktester, but im confused if im going to do a variable or a attribute MSA.

If i do a attribute i need to know how capable my equipment is so i need to preforme an capability study to se the machine's spread otherwise i dont know if i can trust the figurs that is comming out. Nevertheless
this is what the tester do just says ok or not ok out from the limits it is set for
but if i do a variable MSA on it i need to make some coolers that i know how much they leak,:bonk: that is no problem for 2 of the tests but for the internal test it might be little tricky (The tester preformes 3 test during the cycle, ExternalA, B and internal between A and B)

My instinct says that i should do a Variable one with the leaks just above the leak limit that the program should find so i dont do to big leaks so that the results is not trustable

so i will make 5 leaking products for each side A,B and (Try) internal and add 5 non leak, i will use the non leaking coolers for all test with the leaking A resp. B and Internal so the sum will be 10 for each test run, i will run these 5 times ending up with 50 meassurings X 3.

I will then do this at three different times to reproduce the operator (Cuzz the operator cant effect the test itself) and inbetween these 3 times they will run other product this only to se if the machine itself has any variance

Do you think this is a good way to approuch this ??
Hope to get answer on it

Thanks for a great forum Tommy:agree1:

Bev D

Heretical Statistician
Super Moderator
Do you normally have failures? If so, using them is preferable to 'making' failures unless you absolutely sure that you can 'make' failures that will be exactly like real failures.

Operator can effect the test results if the parts are put into a fixture and technique can change how the parts are placed in the fixture. Other than that there is no need to 'fake' operator variation.

You should test variables data, not attributes data. you will learn more from the variables data, need fewer samples and get better resolution.

There is also no need to test 3 times. twice is sufficient.

You dont' have to span the entire range of variation, there are ways to estimate the error realtive to the specification that will work. The 10 parts * 3 measures * 3 operators 'canned' version of a Guage R&R is no the only (or the best) way to perform a measurement system evaluation.

If you post your data along with the specification limits we can show how to do the analysis...

It Was I Who

SQA Supplier Quality Assurance Manager
Hi Bev
and thanks for a quick replay :)
There are guides in the fixtures that will take care of the centering of the product, but i will make a test to se if the placement has any effect on the test but it shouldent. and the placment is done by robot
and the test is easeir done with manual operation then it is to do it with the whole cell inwolved
I can maby take down the number of runs to 2 but i might think twice on that.
For the failures i will make them for the external test A and B but i will search for the internal (If they arn't to big), why i want to make the failures is that i want to make them just over the topp of the leak limit
oterwise i think that the bigger leaks will not showe me the facts that im searching for (Due to greyzones of leaktesting) and it is the bordelines im most interested in, the big leak is more or less no problem they will be detected by first sekvence testing before they enter the real fine test
(Im glad that this is gas detection and not reference testing with overpreasure)
thanks for the replay

It Was I Who

SQA Supplier Quality Assurance Manager
Hey Bev

I dont know if you know much about the leak testing buissnies?
But as i describe before i just want to have coolers that leak just over the limit, instead of having a big leak that will be no problem to find, The most common leak is the big ones but those i'm not afraid of it is the borderlines im interesting in.
and if i take the big leak im affraid that it will not show me the right results due to the fact that **** happens when the leaks gets smaller and our leak limits are below 20-30mm3/S @ 2-6 bars preassure so it is very small leaks

so i'm afraid that the results will be missguiding or do you have an other view on it, so am i glad if you could share it to me, cuzz i dont whant to be wrong in my test and results
(Exscuse me for my bad english im not trying to be rude or somthing so please don't take it like that it's my spelling of frases) :)
Best regards Tommy

It Was I Who

SQA Supplier Quality Assurance Manager
Hi Bev
We have made the MSA study on the leaktester now and we can see that we are safe towards the customer but we are scraping "Fales Leaks", they are not leaking due to that we have some disturbance in the testing loop, the equipment looks at it blurry so to speak

what the conclusion was is that both machines rate them same and the rate the non leaking leaks the same way aswell

why we dont send anything to the customer that is wrong is that we might have the wrong leak limit (It is set to low) but that is from the recommendations from the supplier
But what i want to do now is to get hold of a calibratet leak (Professinal Calibrated) and put those int to the test equipment at different preassures to verify the leak rate that we are testing and calculate out from there
And out from that i can make the test even more accurate and come up with a solution on witch limits they should be set to with a good margin to not let any leak pass out

Top Bottom