Tolerance vs. Part to Part - Is it ndc? When using % of tolerance everything is fine

[Corey]

Hello, I have been browsing these forums for too long. The more I research, the more I get confused.
Woudl someone please check out the attachment and tell me why when using %tol everythign is fine and dandy? However, part to part is horrid?

Is it due to ndc? I am on the road and do not have any reference material that will help me determine the ndc.

 

[Miner]

Re: Tol vs. Part to Part is it ndc?

This spreadsheet evaluates the guage for two very distinct purposes.

The tolerance method is calculating the P/T Ratio. This compares the gage variation to the total tolerance, and assesses the guages suitability for part inspection to tolerance. Your gauge is very suitable for part inspection.

The "Using Part to Part variation" method calculates %GRR. This compares the gauge variation to the part variation, and assesses the guages suitability for process control (e.g., SPC), or for use in statistical studies (e.g., DOE, t-tests, etc.). ndc is a related measure to %GRR. It does not relate to P/T Ratio at all. If the variation of the parts accurately represents the variation of the process, then this guage is not suitable for process control. If you need this guage for process control, you will need to reduce the repeatabilty (within operator) variation.

Most guages are used for inspection or for process control, but not usually for both. Determine what this guage is used for, select the appropriate method (P/T Ratio for inspection; %GRR for process control), then make your decision based on the result of that method. The results of the two methods will not typically match each other, so don't let that worry you.

 

[cuadra - 2010]

Re: Tol vs. Part to Part is it ndc?

Hi Corey:

Your P/T ratio looks great.

Your part to part variation is small hence that is why your second acceptance criteria is giving an artificially high number. A low ndc is not the cause, it is the result of having low part to part variation.

This is easy stuff:
1- The Gage R&R give you an estimate of the standard deviation of your measurement system error, lets call it STD.
2- If you multiply the STD by 6, you get the total error that your measurement system.
3- If you divide the standard deviation of your part to part variation over the STD of your measurement system, this ratio will give you a sense of how many times bigger is the part to part variation with respect the measurement system error (STD).
4- If this ratio is for instance one, the system will not be able to tell one part from the other.
5- In your case, this is not bad, it just means that the parts are so close together that the system will put all of them inside the same measurement "bucket". In other words, the system will have only one distinct category.

Note: Your spreadsheet has some conceptual errors.

Cuadra

 

[Corey]

Re: Tolerance vs. Part to Part - Is it ndc? When using % of tolerance everything is f

Thanks Guys. It made sense till I read the both responses..lol

I will need the gage to control a KPc of a 2 cavity mold.

So if the gage is not acceptable for process control, how can it also say that the p to p variation is having an impact on the %Part to Part. How does it identify /sense the p to p variation?

 

[Miner]

Re: Tolerance vs. Part to Part - Is it ndc? When using % of tolerance everything is f

Thanks Guys. It made sense till I read the both responses..lol

I will need the gage to control a KPc of a 2 cavity mold.

So if the gage is not acceptable for process control, how can it also say that the p to p variation is having an impact on the %Part to Part. How does it identify /sense the p to p variation?

Since you need to control a KPC, the Tolerance method or P/T Ratio does not apply. You must use the %GRR or Part Variation method. This is where your guage has a problem.

cuadra's post is partially correct. I agree with items 1 - 4, but item 5 is not accurate. ndc does matter greatly, and this will be a major issue for process control. If you attempt to use this guage for SPC, your control limits will be based on measurement error, not part-to-part variation. Your control limits will be inflated, and your averages will hug the central line.

The first thing that you should do is verify that the variation of your sample parts match the variation of your process. If they do not reflect the actual variation, your %GRR is not valid. If they do reflect the process variation, you probably have a highly capable process, which is good, but an inadequate guage, which is bad.

All of the variation is in Repeatability or Within-Operator variation. Try evaluating the within-part variation (variation in form) of the characteristic that you are measuring.

 

[cuadra - 2010]

Re: Tolerance vs. Part to Part - Is it ndc? When using % of tolerance everything is f

Thank you for the Clarification on 5:

I did not mean to imply that ndc is not important, in his case, he just probably selected parts that are too close together. The sentence shall read " this is not necessarily bad".

In fact ndc, is conceptually equivalent to the "Signal to Noise" ratio of your measurement system. The equation to compute is ndc = 1.4142 * ( variation of your sample parts / variation of your measurement system).

As Miner already mentioned, the variation of the numerator in this equation needs to reflect the actual variation of your process. In other words, a perfect measurement system could resolve any production part passing thru it, but in real life, the "Noise/Error" will diminish the ability of the measurement system to resolve "who is different/equal from Who" with just a single number, hence the system is bound to discriminate using a "Range/Bucket/Categority"

cuadra
Note: This is beyond the scope of this discussion but under the "right conditions": A measurement system may have a low ndc but still, be able to protect the customer by discriminating a "good" from "bad" part. Of course, it is ideal to have both properties available.