Gauge Study for Small Lots

[AerospaceAimee]

Hello,

I work for a company with a manufacturing/assembly facility that more often than not runs lots of about 5 pieces or less. We were issued a CAR from our customer, and engineering's corrective action was to design a tool/gauge that they claim can be used for creating the dimension and measuring it. We are now in the process of trying to validate that the tool/gauge, but have a problem in that there is only one part to validate with.

I'm pretty positive we can't do any kind of capability study for the creation function of the tool, but I'm wondering if there's a way to validate the measuring function with only one part to validate with, and no knowledge of if there will be another run in the future. I'm fairly knowledgeable in Gauge R&Rs, but the smallest study I've performed is 5 parts with 2 operators (and am very aware that was sketchy data); is it possible to go smaller? If so, how would you calculate it without minitab or any other study program?

Thank you!

 

[Ron Rompen]

In theory, you can do a GRR study with only one part, but the results would be meaningless. To do any gauge study, you need a selection of parts which represent the anticipated process spread, AND of course you want to have at least one part which exceeds the tolerance.

 

[Proud Liberal]

Since the purpose of the R&R is to determine if the gaging solution is adequate to the task, can you group similar parts into a single study.

 

[Ron Rompen]

Since the purpose of the R&R is to determine if the gaging solution is adequate to the task, can you group similar parts into a single study.

I have done this in the past, with the provisio that the part features are similar, and the GRR is done to the tightest tolerance. It is similar in concept to doing the MSA study on a 'family' of gauges.
I don't know if this is specifically permitted (or excluded) in any documentation, but I have never had it written up as an audit finding, by either 3rd party auditors or customer auditors.

 

[bobdoering]

In order to have a valid gage R&R, you need to have samples that are sufficiently different enough that the gage can tell the samples apart. It is far easier to correct a gage study with too much variation than one with too little - and one part is certainly too little. Can you simulate the feature, as in make stand-in parts where the feature exists - but not entire parts -made with significant variation between the parts?

 

[bobdoering]

In theory, you can do a GRR study with only one part, but the results would be meaningless. To do any gauge study, you need a selection of parts which represent the anticipated process spread, AND of course you want to have at least one part which exceeds the tolerance.

When you have any of the samples that exceed the expected variation of the process of the life of the process (such as the one part that exceeds the tolerance) you do NOT want to include them in the calculation. You need to plug in the historical standard deviation for the PV to calculate a correct GR&R value.