Quantifying risk in choosing the number of parts, operators and replicates in a GR&R

[Cleversn]

Hello,

This is my first post so I hope this isn't breaking any rules. I did search for this to see if it was already posted and couldn't find it.

I am seeking advice on how to interpret the risk in determining the number of parts, operators and replicates in a GR&R.

For example, how can I communicate the risk in running different variations in the number of parts, operators and replicates. I'm used to running 10 part, 3 operators 3 times and am being asked why we can't reduce that because "it will take too long".

Is there some way to put it in terms of how confident we can be if we measure capability and how that is impacted by changing the number of parts, operators or replicates?

Thank you!

 

[Bev D]

Welcome Cleversn

you might find this page of a similar posting helpful...

I will say that the 10X3X3 approach is statistically and mathematically invalid. A better approach is detailed by Dr. Wheeler and my paper. The links can be found at the above page. If you read these two papers it will give you some things to really think about and come back with more questions...

A simple answer for you is that a measurement error study is intended to determine how much measurement error exists in relationship to the product variation and the specification range. measurement error and product variation are both quantified by their standard deviations. 10 parts is simply to small to create a useful estimate of either standard deviation. This is not a simple topic. It is complex and requires deep understanding and critical thought.

As for the "it takes too long" comment, well I guess that's true if you run a poorly designed study that can't really provide any insight especially if you just treat like a pass/fail 'check the box' thing to do. Speaking from years of experience I can't tell you how much more time, money and poor customer experience have resulted from not performing a simple well designed study that only takes a few hours to execute and analyze.

 

[Cleversn]

Hi Bev!

Thank you for the response! I think I had come across your slides in my search before posting. I bet you have forgotten more about this subject than I will ever know.

I do have some follow up questions because there is a lot to unpack here. Luckily I'm able to conclude that since the 10x3x3 format is invalid reducing that to fewer parts/people/trials will be even less valid. So this partially address my original reason for posting.

1. My understanding of the "Honest GRR" is that it takes the percentages adding up to 100% into account (which makes sense) plus the math is different. I think I overlooked how this relates back to a 10x3x3 style. Does it change these variables or is it completely different?

2. I looked through all the links on this site (elsmar.com/elsmarqualityforum/attach-list/?filename=msa) and couldn't find a template for the "Honest GRR" format. I don't want to oversimplify it but I think that will help me visualize it. Do you know if one exists?

3. Am I correct in saying that completing the 10x3x3 is better than doing nothing? I'd definitely like to learn the "Honest GRR" but might need to put a pin in that so I can make some time to learn it enough to apply it.

Thank you!

 

[Miner]

My MSA blogs got kind of buried during the transfer, but here is a link to my blog on Wheeler's method. There is an Excel file attached.

 

[Bev D]

I do have some follow up questions because there is a lot to unpack here. Luckily I'm able to conclude that since the 10x3x3 format is invalid reducing that to fewer parts/people/trials will be even less valid. So this partially address my original reason for posting.

1. My understanding of the "Honest GRR" is that it takes the percentages adding up to 100% into account (which makes sense) plus the math is different. I think I overlooked how this relates back to a 10x3x3 style. Does it change these variables or is it completely different?

The Honest Gauge R&R paper points out the mathematical flaw in taking a ratio of two standard deviations - they will never add up to 100% and the way the math works in the popular method it grossly overstates the measurement error's contribution to the overall variation. This flaw is unrelated to the number of parts, repeated measures of each part or operators. The sample size numbers create a separate and distinct problem from the ratio of standard deviations.

2. I looked through all the links on this site (elsmar.com/elsmarqualityforum/attach-list/?filename=msa) and couldn't find a template for the "Honest GRR" format. I don't want to oversimplify it but I think that will help me visualize it. Do you know if one exists?

in addition to Miner's link - have just posted "MSA Tools" in the resources section, it will do the math and createe the plots for you. You must use only 2 repeated measures per part.

3. Am I correct in saying that completing the 10x3x3 is better than doing nothing? I'd definitely like to learn the "Honest GRR" but might need to put a pin in that so I can make some time to learn it enough to apply it.

[/QUOTE]
Don't confuse the sample size of 10X3X3 with the honest gauge R&R approach.
10X3X3 sample size can work with the honest gauge R&R math - although plotting it on a youden plot is impossible. The best way to learn is to just do it. get 30 parts measure them twice and enter the data into the spreadsheet. compare that with the results of measuring 10 of those 30 parts 3 times. it's really easy.