Attribute Gage R&R study - pain or not?

[greenimi]

Since I am new in the quality engineering world I am asking many questions and sometimes I get conflicting answers........ but I keep questioning.
One of the interesting facts: I was told that if you put 3 Quality Engineers in the same room and give them the same problem to solve, you will get 5 different answers.

Anyway, I asked one of our company statisticians about Attribute Gage R&R and here is the answer --see attachment--

Not that I am not trusting, but I am not fully trusting, so I would like experienced members of this forum to comment:
What do you think about Attribute Gage R&R in general?
Is there any “adjustments” or comments you could make pertaining to the statements posted below?
Is Attribute Gage R&R study such a pain?

I guide my learning approach by “'Nullius in verba'” – roughly, 'Take nobody's word for it”, hence I am questioning everything.

Thank you very much for your comments

"“I would rather have questions that can't be answered than answers that can't be questioned” Richard P. Feynman"

 

[Bev D]

The two main sources of misinformation with QEs is that there is no real formal training in Quality Assurance or Quality Engineering in Engineering or business schools. The other source is that there is a substantial difference between theoretical & so called ‘applied’ statistics and industrial engineering statistics. (Check out my resource Essential References for Practical Quality Engineering; this list has been compiled over 40+ years as a QE; all methods have been extensively researched and more importantly proven effective over thousands of studies). You might also find Miner’s Blogs to be very useful…

Your statistician isn’t wrong concerning attribute MSAs. Preparation for these types of studies is more involved than for a standard continuous data MSA, but I have found that most attribute systems, particularly those involving human visual inspection, to have many more problems than continuous gauging systems.

The three key concerns for an attribute study are:
1. Challenge the margins - just good and just bad. Clearly good and bad units are typically easy to detect, while the margins is where peopel and equipment struggles.
2. AFTER fixing the system’s ability to detect correctly at the margins, test the system at production rate, under production conditions (lighting, temperature, etc.) with a data set that has the current defect rate.
3. Because an effective study requires the above it cannot be done before production begins in earnest unless you have real defects created naturally during development and OQ where the allowable extremes of the input tolerances are tested…

You can also review other MSA resources in the Resources section

A final note: Just because something is difficult doesn’t mean it is useless. In fact the opposite is probably true…

 

[greenimi]

Bev D and all,

Thank you very much for you answer and for the references and the link to other members blogs. I appreciate your help. For sure those links and your research will contribute positively to my education in this area.
I am currently involved in the plan for my first attribute gage R&R for a “simple” functional gage (to qualify/ check location control of some holes in a feature control frame -GD&T- modified at MMC; hence fixed gage size at virtual condition).

My follow up question is (see attachment):
Do you consider the guidelines below as good rule of thumb? I found them during my own research to improve my learning in this field of expertise. Now, I guess I have to “measure” them in order to have good “references”, right?
If yes, what happens if I don’t get lucky enough to get the suitable parts? Is it mandatory to get “bad” parts or marginal bad parts?
Is the study is invalid if those bad parts are missing. I have hard time finding those and I cannot deploy the functional gage in the production without a proper MSA.
Is this the chicken and the egg evolutionary perspective?
How can I solve this conundrum?

 

[Ron Rompen]

To answer at least PART of your question, not having all the suitable 'bad' and 'marginal bad' parts may not invalidate your study (unless you have customer specifics to conform to). If your process is very unlikely to produce parts which are marginal, AND IF the risk to your customer and the end user is low, then the study MAY be acceptable.
However, I would prefer to validate my gauge across the entire range of the tolerance, including marginal good/bad parts. I may not have the 35% (each side) that are at or near the upper or lower limit, but I will endeavour to have SOME. If this isn't done, then you have no definitive evidence that your gauge will detect them. Yes, it adds to the pain, but in the end, it provides useful information.

 

[Bev D]

The “rules of thumb” aren’t bad as any increase in marginal good bad parts help with the actual validity of your study.

Yes you should determine ‘truth’ prior to the study. That is a critical element of any attribute study like calibration is for a continuous data study.

Yes it can be a chicken and egg situation. That is the really difficult part of the study. If you have to wait for production to make marginal parts to complete the study then you do.