medical_eng

Involved In Discussions
#1
Hi Covers :bigwave:

Long time Cove reader but the first time to post. This is a very helpful forum, so I thought I would post my question. I've been through other posts and I didn't come across my question yet, so here it goes.

I've been through all of Wheeler's material on GR&R's, fantastic stuff. If all statistics material could have such clear explanations it might not have such a bad name! :)

My question is with the "Tolerance Analysis" of the Excel worksheet. I see the output for a good gauge per his example and agree with it, but I'm not sure of how a bad gauge would look like.

So here's my question: What would the numbers look like if "the gauge (is) NOT good enough to inspect this part to this tolerance"?

My aplogoies if the answer is staring me in the face and I am not recognizing it!

Many thanks in advance.

Cheers,
medical_eng
 

Ninja

Looking for Reality
Trusted
#2
Take one of your GRR's (or on of the published demos) and do the following:

1. Take ALL trial #1 results and increase them by 25%
2. Take ALL trial #3 results and decrease them by 25%
3. run your GRR with that data.

EV should become a significant % of the spec range, and AV might even become irrational (Sqrt of a negative number since Xdiff is much lower than EV)

Is that what you are asking?
 

medical_eng

Involved In Discussions
#3
Hmmm, I did what you said using Wheeler's default data that came with the spreadsheet.

Of course, the % Total Variation now has a %RR at 81.54 and has been dowgraded to a 4th class monitor. No suprise there, these results I understand.

My purpose is to check parts to spec. and assess "pass" or "fail". Later on I will get fancy with SPC, etc. One step at a time.

So, looking at the Tolerance Analysis section "Is the gauge good enough to inspect the parts", the 96% manufacturing specs go from 152 to 218. This doesn't seem that bad, but no objective comparison criteria seems to exist. % Watershed Tolerance is now 234%, but what does this mean?

Still puzzled....
 

Ninja

Looking for Reality
Trusted
#4
Your first question was easy...your second is not (I don't have Wheeler's previous works and may or may not understand them if I did:confused:.)
Per Wheeler's article, he covers your question completely in Wheeler [2004, 2006] referenced. I have not read these...

Anyone esle?

Toward your initial question of what it would look like if an inappropriate gage was being used...I think you have a good example of that now. Sorry I couldn't help further.
 

medical_eng

Involved In Discussions
#7
Hi Miner,

Yes, I did check out your blog (I had read them all before - very well written!).

Perhaps I am alone here but my question still remains. I am very clear on the GR&R output for ***Process Control*** and what makes a good gage vs a bad one.

I am specifically referring to the area on Wheeler's GR&R for Tolerance Analysis (for Product Control Decisions). What would the graph and numbers look like for a "bad" gage and by what objective criteria would you come to that conclusion?

I apologize if the answer is obvious to every one else. I don't mean to waste anyone's time.

Sincerely,
Medical_Eng
 

Miner

Forum Moderator
Staff member
Admin
#8
In the file attached to my blog, increasing the gauge variation, or reducing the tolerance will change the watershed tolerances, the guard band values and increase the % Watershed tolerance value.

Note: You may have to increase the number of decimal places shown using format cells to actually see the changed values.
 

Miner

Forum Moderator
Staff member
Admin
#10
You would have to ask Donald Wheeler. In his article "A Honest Gauge Study", he used this as a better alternative to the %Tolerance metric, but never provided acceptable guidelines. In a later paragraph, he decried the AIAG criteria for % Tolerance, but never said whether the criteria were good for the %Watershed Tolerance.

The answer may be in his book EMP III, but I do not have a copy.
 

Top Bottom