We have a customer asking us to do Gage R&R on a particle measurement tool. Our spec is for a delta in count, i.e. pre and post processing, to determine the allowed number of particles the processing itself can add to our product. We designed and performed an ANOVA. The problem we're having is that when we calculate the results using our spec (i.e. delta number), the value comes out extremely high. For example, a product has to have less than 50 to be used. The spec is a delta of no more than 50. If we have product that has an initial count of 25 and a post count of 74, it passes. The delta is 49, which is acceptable, but that 49 takes up almost all of the window, making it seem as if the tool is incapable. We also looked at calling this an attribute, i.e. pass/fail, but we have to apply that against the delta spec instead of the real physical measurement. We found a short study method for attribute in the AIAG manual 2nd edition, however our customer will probably not like this as there are no numbers attached. Does anyone have any good ideas? Thanks.

Hi

Did the parts in your study reflect just process variation or the specification spread? Did you make a test set of parts from just below to just above spec with several in between? Or did you take parts from a typical run within minimal part to part variation. This has a huge impact.

Can you post the data from the study? I'm sure 4 or 5 people here will jump on it and provide more insight?

You may want to invest a few hours searching and reading. I pretty much always find that my question has been answered here somewhere 3 or 4 times already. And I learn other things on the way.

Thanks for considering this problem. I think we've solved it. We can't do a typical attribute study because generating specific intervals of particles is nigh onto impossible. What we did was to find 3 levels of particles, low, medium and high and ran them multiple times for various tools, operators, shifts and programs. Calculating %GR&R is impossible because the control limits are, for all intents and purposes, in another unit of measure. We ran the ANOVA, determined % variance and standard deviation from each factor and are calling it good.

Our engineers spent almost 3 days researching the right way to calculate %GR&R when the control chart is in a different unit and couldn't find anything, so we reported the data and have considered what we have done to be sufficient. It was an interesting exercise to go through.