Can I use the % Tolerance to determine the actual process tolerance?

[DivingDoug]

I have completed and expanded GRR study, 15 parts, 3 operators, 3 runs, 2 measurement systems.

Each measurement system returns acceptable results on it's own, %Contribution levels below 10% and %Tolerance levels in the mid 20% range.

When the data is process through the expanded tool (using Minitab 16) the results are less than stellar. the %Contribution levels are still acceptable but the %Tolerance levels are at 41% and 56% (testing two points during each run)

The %Tolerance level is based on my required Process Tolerance of course. What I need to determine is, if I run both systems as is, what is my Process Tolerance effectively increased to. In other words, if my P/T is .25 to start, would I, in reality be letting go items in a tolerance range of .35? .45? .75?

I tried to work backwards by repeating the analysis with an increasingly larger P/T range but after several iterations, I graphed the results and the curve flattens out so I can never put in a large enough tolerance to get to a zero %Tolerance. So obviously this method is not valid.

What I need is a method to use the results from the expanded GRR to let me know if I run product across both systems, how much will my range grow for the units that pass? I know this won't give me new limits as that is a question of accuracy, a different test altogether. But knowing the range will let me present the information to management to allow an executive decision to be made.

Many thanks to anyone who can assist me on this.

Doug

 

[Miner]

I sounds like you are saying that you have two measurement systems that do not provide equivalent results. I would focus on getting them to provide equivalent results.

Please attach your data and we can take a look at it. Minitab 16 format is fine.

 

[DivingDoug]

You are absolutely correct in that I have two measuring systems that are providing different results. Based on the output, there is a consistent bias between the two systems. Because each system is constructed with several components, each having there own tolerances, correcting one or both systems to get them to provide results that align is a daunting task and certainly not a short term project.

While this condition will be addressed, I need in the short term to calculate what range of measurements the systems will find as potentially acceptable based on my current results.

While I know I can not achieve a %Tolerance of 0%, if I could then my measurement range would be equal to my Process Tolerance, say .25. But if my %Tolerance is 30%, then what will my effective measurement range be? .35?, .45? .75? This is the calculation I need to figure out.

I can post a file with data but will need to generate a clean file as I have been completing a lot of what-if claculations in this file and it is large and messy. I should be able to post it on Monday, the 28th.

Thanks...... Doug

 

[DivingDoug]

I have included a cleaned up minitab v16 file in this post.


There are five worksheets as follows:

• System A - Raw Data - All data as collected during the GRR trials for System A
• System A - Cleaned Data - Special Cause results have been removed
• System C - Raw Data - All data as collected during the GRR trials for System C
• System C - Cleaned Data - Special Cause results have been removed
• System A & C - Cleaned Data - Special Cause results have been removed from both data sets with SpCs data identified in System A also removed from System C and SpCs data identified in System C removed from System A to generate a fully balanced data set.

I processed the basic GRR for the cleaned results for both System A and System C. I also processed the expanded GRR for the combined data sets.

There are also some notes added to the beginning of the Session Log window


My goal is to determine, for each system individually and/or under the assumption I use both systems in parallel, what the maximum potential Tolerance Range will be based on the calculated results of the GRR.

In other words, if my specified Tolerance Range is 0.4, then a theoretical %Tolerance result of zero (0) on the GRR would mean the results the system finds acceptable will always be within that range. (If this is incorrect, please let me know so that I don't hold on to this assumption)

However, since my %Tolerance result will always be above zero, I wish to calculate what the maximum potential Tolerance Range is that the system will find as acceptable based on this increased %Tolerance.

If any additional information is needed I will be happy to comply and thanks for any information in my quest.....

Doug

 

[Bev D]

can you post the data in EXCEL? many people don't have minitab

 

[DivingDoug]

No Problem, I just did minitab as that is what Miner above indicated would be acceptable.

I will take all the help I can get.

Thanks Doug

 

[Miner]

I notice that there appears to be a constant bias of 0.91 up until a certain level then the bias begins to change linearly.

 

[DivingDoug]

Yes, there is indeed a bias between the two systems. We do intend on addressing this however, because of the types of components used in the systems and the requirements for manufacturing, this is not going to be a short term project.

One system is in full time use with the other system is being prepared to go online. Due to this, access to system A is limited and I have to plan my testing to collect the data that will be the most useful to resolve the issue.

My expectations are that system C will be released. This is due to in part because our management are not grasping the implications of the expanded GRR. "We have never done it on any of our other duplicated systems, why would we do it now?" Even our division quality manager, a green belt (not even a black belt as you would expect out of a division level mgr) doesn't accept the importance of the expanded, multi-system GRR.

So my goal is simply to show them in terms they can understand without me having to teach them (which I am not qualified to do anyway) what the impact of using both systems will be when there is a %Tolerance result of greater than 50%.

I want to be able to say, "So the test spec is to have the part within a .25 range but in reality we will be shipping product that coule be within a xyz range"

I know that there are issues that need to be addressed, but I will never get buy in on it from mgnt unless I can explaing the consequences of not working on those issues.


While I have been researching this, I came across a series of your (Miner) blog entries and interestingly found the one that describes the use of the Wheeler's Honest Gauge Study. I have it downloaded and am going to plug my data into it and see if it may provide additional information that the standard anova method does not.

Have you any thoughts, suggestions, warnings about using this tool ?


As an FYI, the systems are measuring the pneumatic air flow of a pump that produces a pulsating flow against a specified load point.

 

[DivingDoug]

Miner, As I begin to use the Wheeler's Honest Gauge Study worksheet that you provided, a couple questions have presented themselves.

1) Can the sheet easily be modified to accept more than the 10 parts? (say up to 15)

2) Can this method be used to perform an MSA on multiple gauges while using multiple operators ?

Thanks.... Doug

 

[Miner]

It can be modified to do what you are asking, but it would not be a quick process. I do not currently have time to make the modifications necessary in the ANOVA calculations.

Regarding warning, the only thing would be if your customer is expecting the traditional AIAG study, otherwise Wheeler's approach is very sound.

Just a thought on the original issue. Can you apply an adjustment factor to remove the bias?