Combat Low Part to Part Variation in GRR

[brauenbuehler]

Hi All,

I am new to Gauge R&R and am conducting a study on a relatively powerful gauge that has the capability to measure out to 6 decimal places.

Our GRR analysis takes place with 10 parts x 3 inspectors x 3 repetitions. Our acceptance criteria call for 5 distinct categories and a Total Gauge R&R %Study Var of <30%.

The part is a small plastic piece that has been injection molded and the dimensions we are looking for are controlled by the tool steel and very consistent. As such, our GRR says all variation in the study is attributed to the gauge/measurement process (fair since the parts are so consistent), but blows our acceptance criteria out of the water because it is a percentage (often times exceeding a %Study Var of 60%).

My leading questions are: are these acceptance criteria the best to use in this instance? Our tolerance window is only out to 3 decimal places where there is no detectable difference in parts. Do we need to introduce variation to the parts to meet our acceptance criteria or are there any other potential solutions?

Any help is appreciated.

 

[Miner]

How will you be using this gauge? For inspection or for SPC? ndc and %SV should only be used if the gauge will be used for SPC. If it will be used for inspection, use %Tolerance and ignore ndc/%SV.

 

[brauenbuehler]

Thanks for the quick response!

From my understanding, this gauge will be used only for inspection.

What difference does it make between SPC and Inspection? Is the gauge not evaluated the same way for a final product/feature regardless of how it was controlled to up to that point?

Also, Is %Tolerance the only acceptance criteria that is recommended on a gauge used for inspection, not SPC?

 

[Miner]

The study is run identically regardless of the application. However, the metric used to assess the suitability of the device is dependent on the application. Since you intend to use the device to inspect, you want to minimize the probability of rejecting good product or accepting bad product. % Tolerance is the metric that will help you do this. If the device were intended to control the process (i.e., SPC) you would want to ensure that the device was able to "see" part-to-part variation. To accomplish this, the device needs to have a small %SV (or the redundant ndc). NOTE: I recommend using the historical process variation in place of the study variation. It is almost impossible for 10 parts to truly represent the actual process variation.

 

[brauenbuehler]

Ah this makes much more sense now. Wish I would've found this forum sooner! Thanks again!

 

[Mikey324]

I've noticed that with todays ever improving process performance, its becoming more and more difficult to have an R&R turn out any different than you described above. Seems the point of the game is to reduce part to part variation. So you do that. Your parts are great. Little variation. Everyone is happy, until you need to do an R&R. The circle seems to continue from there. You really need a large quantity of data to try and get the true variation. Even then, there may not be enough.

We had an auditor at one time suggest we buy a gauge that was way over spec'd for what we would ever possibly need. Just to try and get some variation identifiable. What's the point in that? You have a spec of +/-0.5mm. Parts are varied by about +/-.002mm. Cpk is great. Why would you ever need to know if the variation is actually .002075mm??

 

[brauenbuehler]

^^This is exactly the problem that we are running into. For Operational and Performance Qualification, we are thrilled that our parts are this consistent. For our the SV metric, it's turned into somewhat of a nightmare.

 

[Mikey324]

The only way we've found to remedy it is as Miner said, using process historical data. Even that seems like it will be problematic in the future.

 

[brauenbuehler]

Sounds like my instance will be fine using the %Tolerance to accept but it's a good note down the line.