Can a Gage R&R be used to Determine Suitable Guardband Limits?

I am in the process of creating a work instruction for deriving "Guardband Limits" for ATE.

We want to utilize work performed from doing the Gage R & R studies.

It is intended that a Gage R & R will be performed on all tests done by the ATE as part of the software V&V (actually prior performing V&V).

Can a statistically based Guardband be determine solely from a Gage R & R? Or is Linearity & Bias study required also?

Any help is appreciated!

Yes, statistically based guardbanding can be done using gage R&R data.

Use the gage R&R data to construct a gage performance curve as discussed in the MSA manual. Establish the guardbands at the two points where the probability of acceptance (Pa) becomes equal to 1.

The linearity and bias studies are typically done prior to the gage R&R, and any issues resolved. The results of these studies are not required for the statistically based guardbanding.

Having said all that, guardbanding is economically undesireable if you have a high fallout rate. The more you reduce your R&R, the smaller the guardbands become, and the higher your yield.

I will look into this further. Do you have any examples of this?

Can you not simply take the Standard deviation for the Gage Study and apply 3 sigma to either specification as the guard band. ensure that variance from the gauge(system) is accounted for.

If necessary apply 95% confidence to this StdDev

What are your thoughts?

See my attachment in this post.

The gage performance curve is under the tab by the same name. You may also try using the PME (Probable Measurement Error) under the GRR Report tab as the width of the guardband.

The approach you suggested would result in inflated guardbands that would hurt yields.

I aggree to just using the Total Gage R & R stdDev as the basis for determinig the limits is pefect. I also think that this does not give the confidence you need to support the choosen limits.

What about determining the 95% confidence interval using the Total Gage R & R stdDev and for the DF using the total source measurement from the gage rr ie parts*operators*trails? Then using the large value of the CI range as the actual measurement StdDev then multiple this 3 to obtain guard band limits.

This would still give a somewhat inflatted limits but I would statistical justified limits.

What do you think?
I would look at the calculation and statistic's behind the the Gage Performance Cure?
Also how can I get a copy of MSA?

ajhsys said:

Also how can I get a copy of MSA?

Click to expand...

Hello ajhsys,

Look at this post + attachment: What is NDC (Number of Distinct Categories)? MSA 3rd Edition.

Stijloor.