Sample Range for a valid GRR using ANOVA in Minitab

[vinny]

Hello Covers,
I'm in a discussion with one of my customers from an automotive OEM. They require us to conduct GRR using the ANOVA method in Minitab. We're in a debate over the range of samples used for the study relative to the specification. For example, let's say that the characteristic we are studying has a tolerance of 0.0 +/- 3.0 mm. In order to judge the gage and create as many distict catagories as possible, I want to have my GRR samples to at least span the entire range of the spec limit, and preferably go slightly beyond those spec limits. The customer is saying he would never allow samples for the GRR to be out of spec to the part tolerance. I'm saying it doesn't really matter since we are evaluating the measurement system, not my samples relative to the specification. Validating the measurement system does not have a bias toward one end of the spec or another seems important so we should try to at a minimum include samples from both extremes of the spec (so at least one sample of +3.0 mm and at least one sample of -3.0 mm. I'd actually prefer my samples to range from -4.0mm to +4.0 mm).
I'd appreciate any support or gentle correction to my perspective the group can offer.

[Stijloor]

Vinny,

You will find all the ammunition to win the argument with your customer in the fantastic blogs from our Friend Miner.

Intro to Measurement System Analysis.

Look at sections 5a and 5b.

Very much worth the read and study!

Good Luck!

[Barbara B]

In Reply to Parent Post by vinny

The customer is saying he would never allow samples for the GRR to be out of spec to the part tolerance. I'm saying it doesn't really matter since we are evaluating the measurement system, not my samples relative to the specification.

Of course you're right: The samples for a Gage R&R should cover the whole range (process and/or tolerance +/- something). Maybe the customer has a knowledge gap here and you can solve the issue by providing some informations.

As Stijloor already mentioned Miner's blog on MSA is very helpful, see section "Part Selection" in this entry: Intro to Measurement System Analysis (MSA) of Continuous Data – Part 5b: R&R. Additionally Minitab's White Paper on Gage R&R could provide further information, e.g. comments to Xbar chart on p.8.

Or you can try the "common sense approach" with your customer:
The MSA is conducted to verify that the measurement system is able to

1. distinguish between good and bad parts AND
2. monitor process performance accurately (enough).

For both goals the measurement system has to provide reliable results within the whole range of possible values. Since a part outside of the tolerance should be detected (reliably) by its value of the measurement system (regardless how likely it is that such an out of spec value occurs), those parts should be used within a Gage R&R. You can't prove that a measurement system works well on the whole range (tolerance+/-some more or process) if you only have parts for a smaller part of the whole range.

But if your customer doesn't want to have parts outside of the tolerance in the Gage R&R (in contrast to MSA4 requirements), just do what he suggests. The customer should get what he wants - he's paying the bill.

To avoid small ndc numbers you can use the Minitab (R16) Assistant menu for MSA ANOVA (Assistant > Measurement System Analysis). The ndc isn't part of the output Or does the customer have specific requirements regarding the Minitab menu (ndc is part of MSA within the quality tools menu) and/or the numbers he wants to see?

The best way to get your customer happy (in case he remains ignorant to MSA requirements) and to evaluate the measurement system according to MSA guidelines would be to select enough parts of the "customer accepts"-range and additional parts of the "customer doesn't accept"-range (e.g. out of tolerance). After taking the readings you run two analyses, one for the "customer accepts"-part to get the numbers for a satisfied customer and one analysis for internal use to evaluate measurement uncertainty properly according to MSA4.

Barbara