Gage R&R without tolerance
- Thread starter Jafri
- Start date
Can you provide additional detail about the test? Measurement of count (Poisson) data might require a different approach than usual. If the typical count measurement is > 10, the Poisson distribution closely approximates the normal distribution and you might be able to perform a standard R&R. However, the integer nature of the data could also cause a resolution problem.
In general, you can perform an R&R study and evaluate it as a percent of the study variation, or of the process variation. This would be typical if the measurement device were to be used for process control or any type of statistical study.
In general, you can perform an R&R study and evaluate it as a percent of the study variation, or of the process variation. This would be typical if the measurement device were to be used for process control or any type of statistical study.
In this leak test, water flows through passage ways in the part, and then number of leak points are (carefully) observed.
Leaks found are usually no more than 3, so I guess the Poisson approximation will not work.
But in this case, can I use 0 as tolerance (0 = no leak).
Thanks.
Leaks found are usually no more than 3, so I guess the Poisson approximation will not work.
But in this case, can I use 0 as tolerance (0 = no leak).
Are you referring to what Minitab calls "Total GR&R %SV"? Criteria being: >30% fail, <10% pass.
Thanks.
You may be over thinking this. Unless you are just looking. To the check the box for a Customer, think about what you really want to know about the leak testing process. There are two critical components to the effectiveness of the test: the inspector’s ability to ‘observe’ a leak, and the test method’s effectiveness of initiating a leak without creating a leak that wouldn’t happen naturally in use.
Leak testing is often not destructive but it is a ‘functional’ Test and so we must use different approaches than the standard AIAG approach. Just in general I would gather several pieces that leak and dont’ leak. Then test them TWICE with each of 2-3 inspectors. Then simply look at cross tabs for each inspector. Even a graph at this point is probably overkill. Things to look for: does the test method repeat leaks and non leaks. Does each inspector repeat within themselves and then are the results replicated across the inspectors? How many pieces you use are flexible. Try to get enough pieces and a good proportion of leakers to non leakers that the inspectors won’t remember what they saw. You can help this by testing the second round a couple of days after the first round...
You might find this paper on MSA for Functional Tests helpful. There is a section on water leak testing for cars
Leak testing is often not destructive but it is a ‘functional’ Test and so we must use different approaches than the standard AIAG approach. Just in general I would gather several pieces that leak and dont’ leak. Then test them TWICE with each of 2-3 inspectors. Then simply look at cross tabs for each inspector. Even a graph at this point is probably overkill. Things to look for: does the test method repeat leaks and non leaks. Does each inspector repeat within themselves and then are the results replicated across the inspectors? How many pieces you use are flexible. Try to get enough pieces and a good proportion of leakers to non leakers that the inspectors won’t remember what they saw. You can help this by testing the second round a couple of days after the first round...
You might find this paper on MSA for Functional Tests helpful. There is a section on water leak testing for cars
Last edited by a moderator:
Yes, %SV = % Study Variation. You can also use a historical standard deviation to obtain %PV = % Process Variation.
In this particular scenario, I would agree with Bev's comments. If you wanted to stick with a more formal approach, you could perform an Attribute Agreement Analysis and use Kendall's coefficient for ordinal data.
You may be over thinking this. Unless you are just looking. To the check the box for a Customer, think about what you really want to know about the leak testing process. There are two critical components to the effectiveness of the test: the inspector’s ability to ‘observe’ a leak, and the test method’s effectiveness of initiating a leak without creating a leak that wouldn’t happen naturally in use.
Leak testing is often not destructive but it is a ‘functional’ Test and so we must use different approaches than the standard AIAG approach. Just in general I would gather several pieces that leak and dont’ leak. Then test them TWICE with each of 2-3 inspectors. Then simply look at cross tabs for each inspector. Even a graph at this point is probably overkill. Things to look for: does the test method repeat leaks and non leaks. Does each inspector repeat within themselves and then are the results replicated across the inspectors? How many pieces you use are flexible. Try to get enough pieces and a good proportion of leakers to non leakers that the inspectors won’t remember what they saw. You can help this by testing the second round a couple of days after the first round...
You might find this paper on MSA for Functional Tests helpful. There is a section on water leak testing for cars
What BEV says!
She nailed it.
Forget Poisson, Chebychev, and the fancy math [emoji854].
Looks like attribute testing. Follow what she says, IMHO.
Sent from my iPhone using Tapatalk
Try this link: https://elsmar.com/elsmarqualityforum/attachments/msa-for-functional-tests-pdf.18663/
I corrected the link in the post as well.
