Torque Gauge R&R - Non-Replicable GRR Case Study

K

ksanders

We have an operation that torque's an assembly together. The machine is equipped with a nutrunner torque wrench which monitors and collect the torque readings. We are being asked by our customer to provide R&R results from this process. In the past, we have just supplied the calibration cert, and that has been sufficient, but times are a-chang'in. With torque being a non-replicating feature, we are strugling with which approach is "acceptable" for evaluation.

Based on other reading from this site, the most consistent suggestion has been Nested ANOVA. I consider myself to be an above average statistician, but must confess, I don't have the slightest ideas how to preform this study. Can anyone offer some help? I have minitab, which has this feature. I just need help in organizing the study and collecting the correct data.

Has anyone found what they would consider to be the "acceptable" method for this type of study? Maybe this is one that needs to be addressed with AIAG and singled out for a clarification from the standard body.

Any comments are appreciated.:frust:
 
A

Atul Khandekar

ksanders,

Welcome to the Cove!

Here is a white paper (pdf format) from AIAG you may find useful:

***DEAD LINK REMOVED***
 
Last edited by a moderator:
K

ksanders

Thanks for the info. I've conducted some taguchi studies in the past, and, at a glance, this looks kind of like the same thing. I'll take some time to review and give it a go. :D
 
K

ksanders

Results for Comments

Attached find the results of the nested ANOVA, which shows a Total Gage R&R of 5.48% of Tolerance. My question is with the XChart by Operator. The white paper, as well as the Wheeler book, state that at least 50% of the average values should be out of control. In my case, they were all in control. In addition, on the summary results was the statement that there were zero distinct catagories.

I have to admit that the data used to analyze this process was taken consectively from the process (90 consec. readings), and plugged into the ANOVA. The white paper recommended taking consecutive part samples from various lots to simulate repetative readings on the same part, however, based on our statistical data, very little variation exists from lot to lot (CPK=20.0), so my assumption was that waiting on different lots of bolts / nuts was only going to be a waist of time.

Can the results seen on both the summary and the XChart for Operators be attributed to the means for collecting the samples, or to the fact that the process is extremely capable? Additionally, would you consider this GRR acceptable if it was accompanied with the capability data as supporting evidence that the process is (1) capable enough that there is no need for a GRR in the first place, or (2) further proof that gage capability is not in question for this process?
 

Attachments

  • nested gage r&r results.doc
    118.5 KB · Views: 660
C

Charles

ksanders

Hi - you'll see from my post that we face similar issues.

I have chosen not to get embroiled with ANOVA, as it seems that a simple X-bar / R chart, summary stats and capability study (e.g. Cpk) do enough to answer the question whether the wrench is up to the job. I am open to other ideas tho !!

Charles
 

LUV-d-4UM

Quite Involved in Discussions
Re: Results for Comments

I am in the same situation as you are. So we just do the X bar and R chart plus Cpk for the test method that follows the ASTM method.
 
Top Bottom