Search the Elsmar Cove!
**Search ALL of Elsmar.com** with DuckDuckGo including content not in the forum - Search results with No ads.
  • Google has changed ad sizes for some reason. I am investigating and will get the sizes reduced to what they have been until now. I apologise for the inconvenience.

Do I need part variation while doing Destructive Variable Gage R&R MSA study

#1
Hello guys,
Can anybody give some insights regarding part to part variation during variable destructive MSA study
Do I really need to take parts covering my entire specification range or I can just randomly select parts from a single lot
 

Jim Wynne

Super Moderator
#2
The idea is to select parts that represent the process range, which might not be the same as the tolerance range. When you have to destroy parts to do the measurement, It should be enough to select parts at intervals during a typical production run. If you're selecting parts post-production, remember that "random" means that every part in the lot has an equal chance of being chosen.
 
#3
Thank you for the reply, Jim.
So it's better to consider taking the parts during production run instead of varying the parameters of the machine to get that tolerance range.
 
#4
"Better" isn't a word I would choose. To me, the right phrasing is it will reduce your chance of getting a meaningless result.

First, remember that a Gage R&R is not a description of a process (such as a capability study or an SPC chart). It is a designed experiment to validate the performance of the gage. And as such, you absolutely can set the parameters of this experiment to better test the gage. You just have to be careful that you understand the effects of these choices on the results.

Here is a simple example to illustrate what I mean:

You wish to test a plug gage that's checking 16 mm +/- 0.5 mm. And your normal process is good, you can easily hit the tolerance. So you run a bunch of parts at 16 mm exactly. Or even 16 +/- 0.01. You get a bunch of operators and you test the gage. Unsurprisingly, everyone passes every part, everyone is in total agreement with no variability. What does this tell you about the effectiveness of the gage? It is clearly wrong to just use parts from your "good" process.

So you redesign the experiment. Now you introduce a few from your process, a few that are 8 mm in diameter, and a few that are 24 mm in diameter (well out of spec). And you test again. Unsurprisingly, everyone passes, measures undersized, measures oversized the same. This is "better" but if you had given a little more thought to the testing of the gage, you maybe wouldn't have picked parts so far out of tolerance. You still have no idea how effective it is at checking your parts against the tolerance.

Try 3, you pick (or make) all your samples near the limits, some just in, some just out. THAT tests your gage - and whether or not everyone uses it the same and gets the same results.

Your goal is to feed it parts that will either prove or disprove the veracity of the gage in a fair way. In other words, TEST the darn thing. You're not trying to get an SQ or a customer off your back, you're trying to convince yourself that this gage works. If we switch back to an analog gage instead of the simplified example, then yes, you want parts throughout the range.

A big part of this is linearity. Should you design your gage R&R experiment and you feed it parts in the lowest 25% of your tolerance and get a good R&R, you don't really "know" it works the SAME in the upper 25%. You can somewhat infer this. You are definitely in a better position to comment than if you had not tested the gage. But you COULD be wrong. The likelihood of this is driven by the particulars of the gage system itself. A micrometer - you'd be pretty safe inferring it worked towards the large end, if the small end passed because there's a linear, direct relationship to thread pitch. But consider the fuel gage on your car.... they definitely seem non-linear. Mine seems to hover near F and E for a while, but quickly moves from F to E. Which means my pickup float is probably on a rocker and not a linear flow. Determining performance near F only, I would improperly infer it's behavior near 1/2.

Sorry this is long. These questions are hard to give you the absolute best answer on because we don't know the particulars. I'd rather give you guidance on how to approach the problem, then you figure it out.
 

Jim Wynne

Super Moderator
#5
"Better" isn't a word I would choose. To me, the right phrasing is it will reduce your chance of getting a meaningless result.

First, remember that a Gage R&R is not a description of a process (such as a capability study or an SPC chart). It is a designed experiment to validate the performance of the gage. And as such, you absolutely can set the parameters of this experiment to better test the gage. You just have to be careful that you understand the effects of these choices on the results.

Here is a simple example to illustrate what I mean:

You wish to test a plug gage that's checking 16 mm +/- 0.5 mm. And your normal process is good, you can easily hit the tolerance. So you run a bunch of parts at 16 mm exactly. Or even 16 +/- 0.01. You get a bunch of operators and you test the gage. Unsurprisingly, everyone passes every part, everyone is in total agreement with no variability. What does this tell you about the effectiveness of the gage? It is clearly wrong to just use parts from your "good" process.

So you redesign the experiment. Now you introduce a few from your process, a few that are 8 mm in diameter, and a few that are 24 mm in diameter (well out of spec). And you test again. Unsurprisingly, everyone passes, measures undersized, measures oversized the same. This is "better" but if you had given a little more thought to the testing of the gage, you maybe wouldn't have picked parts so far out of tolerance. You still have no idea how effective it is at checking your parts against the tolerance.

Try 3, you pick (or make) all your samples near the limits, some just in, some just out. THAT tests your gage - and whether or not everyone uses it the same and gets the same results.

Your goal is to feed it parts that will either prove or disprove the veracity of the gage in a fair way. In other words, TEST the darn thing. You're not trying to get an SQ or a customer off your back, you're trying to convince yourself that this gage works. If we switch back to an analog gage instead of the simplified example, then yes, you want parts throughout the range.

A big part of this is linearity. Should you design your gage R&R experiment and you feed it parts in the lowest 25% of your tolerance and get a good R&R, you don't really "know" it works the SAME in the upper 25%. You can somewhat infer this. You are definitely in a better position to comment than if you had not tested the gage. But you COULD be wrong. The likelihood of this is driven by the particulars of the gage system itself. A micrometer - you'd be pretty safe inferring it worked towards the large end, if the small end passed because there's a linear, direct relationship to thread pitch. But consider the fuel gage on your car.... they definitely seem non-linear. Mine seems to hover near F and E for a while, but quickly moves from F to E. Which means my pickup float is probably on a rocker and not a linear flow. Determining performance near F only, I would improperly infer it's behavior near 1/2.

Sorry this is long. These questions are hard to give you the absolute best answer on because we don't know the particulars. I'd rather give you guidance on how to approach the problem, then you figure it out.
The OP asked a question about a variables gage study, not attributes, which is a different kettle of fish in which good and bad parts must be used. A standard variables GR&R should include parts that represent the range of the process, not the tolerance. There is no need to adjust and readjust the process to get parts that include the whole tolerance range. In this sense, it's not an experiment. It's a test to determine whether the measurement system (part, measurement device, operators) is appropriate to the task and to gain some understanding of risk.
 

Jim Wynne

Super Moderator
#6
Thank you for the reply, Jim.
So it's better to consider taking the parts during production run instead of varying the parameters of the machine to get that tolerance range.
See my reply to ncwalker. If you're doing a variables gage study, adjusting the machine/process can be counterproductive.
 
#8
I agree with Jim and Miner to an extent. There absolutely IS a difference between what the results of a % Study Variation test and a % Tolerance test tell you.

I don't agree that it should use the range of the process only. My reasoning is the goal is testing the gage, not the process. (I could be wrong). My real world experience where NOT using the entire range causing a problem is with mass flow leak testers, which do not have linearity - their accuracy changes based on the leak rate measured. If I am leak testing parts and my current process is a subset of the tolerance, and I use only these parts, at a future date my process may be in a different location in my tolerance. I need to know my gage works there, too. I have seen where it doesn't.

I do concede that once we got understanding of how they performed, we were in a better place to predict how a leak tester worked without having to test the full range. And range-subset test (which reduces the monkeying around with the samples, and therefore the cost) is OK. Looking forward to the discussion on the difference between "test" and "experiment ..." :)

Editted to add: Where I do agree with testing the output "as is" is if you're in some sort of DMAIC study where you are looking for a Red X. You need to know that your gage can effectively spot the Green Y. To do this, you test the process as is and look at % Study Variation.
 

Miner

Forum Moderator
Staff member
Admin
#9
I don't agree that it should use the range of the process only. My reasoning is the goal is testing the gage, not the process. (I could be wrong). My real world experience where NOT using the entire range causing a problem is with mass flow leak testers, which do not have linearity - their accuracy changes based on the leak rate measured. If I am leak testing parts and my current process is a subset of the tolerance, and I use only these parts, at a future date my process may be in a different location in my tolerance. I need to know my gage works there, too. I have seen where it doesn't.

I do concede that once we got understanding of how they performed, we were in a better place to predict how a leak tester worked without having to test the full range. And range-subset test (which reduces the monkeying around with the samples, and therefore the cost) is OK. Looking forward to the discussion on the difference between "test" and "experiment ..." :)
This aspect of the measurement system should be evaluated through a linearity study.

Editted to add: Where I do agree with testing the output "as is" is if you're in some sort of DMAIC study where you are looking for a Red X. You need to know that your gage can effectively spot the Green Y. To do this, you test the process as is and look at % Study Variation.
This is correct, but also includes where the gage is used for SPC and capability studies.
 

Jim Wynne

Super Moderator
#10
Let's also remember that the OP is asking about a gage study that requires destruction of the parts. We can't tell from this distance what the parts are worth, but in any event the destruction should be kept to a minimum, which means that fiddling with the process to get parts that reflect the tolerance range might be irresponsible.

I don't agree that it should use the range of the process only. My reasoning is the goal is testing the gage, not the process.
The goal is to test the measurement system, not just the gage. Gage "testing" is done by calibration. Testing the measurement system necessarily involves factors involving the process that might come into play, such as availability of machine time, and the nature of the manufacturing method. Stamping, casting, extruding and molding all might make it difficult or impossible (at least in practical sense) to create the variation necessary to include the entire tolerance range.
 
Top Bottom