GR&R impact on pass/fail tolerance 1

Seems like there's two separate problems:

> The variance on the product seems high for the tolerance imposed
> FIX THE FLOW RATE MEASUREMENT

What exactly is supposed to happen if the "over check" fails?

TTFN (ta ta for now)
I can do absolutely anything. I'm an expert! faq731-376 forum1529 Entire Forum list

If the over check fails the part is scrapped.

Product Engineering improved the master flow rig with a better flow meter, GR&R is under 10%, but the meters cost over $10,000 not to mention the cost of changing the software to interface with the new meter. Manufacturing has 4 rather more complex machines (they do several other things in addition to the flow check). They also can't meet production demand so there is no time to modify them.

The question is how much to reduce the production pass/fail criteria to ensure with a reasonable level of confidence that they are not passing non-conforming production with a 20% GR&R.

----------------------------------------

The Help for this program was created in Windows Help format, which depends on a feature that isn't included in this version of Windows.

What's the source of the variation?

Sorry, but your company seems to have things backwards. Your test instrumentation, per any standard you might look at, would demand that the instrument's accuracy be BETTER than the spec, not worse.

As such, you would have to require the process to be at least 20% better than the spec, since your uncertainty from your instrument is so poor. The uncertainties are neither additive nor multiplicative; they require RSS to combine.

TTFN (ta ta for now)
I can do absolutely anything. I'm an expert! faq731-376 forum1529 Entire Forum list

Do you even know what a GR&R is? The accuracy of the machines is better than the spec but not enough. So the machine is using up 20% of the product spec tolerance which is why we want manufacturing to reduce their pass/fail tolerance.

----------------------------------------

The Help for this program was created in Windows Help format, which depends on a feature that isn't included in this version of Windows.

Yeah, sorry, everyone uses their own terminology.

Is the GR&R a standard deviation or a total variation, i.e., 2 or 3 standard deviations?

TTFN (ta ta for now)
I can do absolutely anything. I'm an expert! faq731-376 forum1529 Entire Forum list

If the GR&R is 2 standard deviations, then the original calculation is correct:



TTFN (ta ta for now)
I can do absolutely anything. I'm an expert! faq731-376 forum1529 Entire Forum list

Here is a sample MSA GR&R, not from the machines in question.


----------------------------------------

The Help for this program was created in Windows Help format, which depends on a feature that isn't included in this version of Windows.

Are you using the AIAG method or compound variance method? This spreadsheet, specifically the "Results" tab, can be used to backsolve for the reduction in PV that would be allowed, given whatever EV you have.

Below are the baseline case and the solver case to meet a tighter acceptance criteria. Using compound variance, the net variance would need to be tightened to 86.67% of 5%, so 4.33%




TTFN (ta ta for now)
I can do absolutely anything. I'm an expert! faq731-376 forum1529 Entire Forum list

The spreadsheet has both methods, but I only used the compound variance. You can set the goal on the AIAG total variance to be 1 and then have the solver tweak the product variance to achieve that.

TTFN (ta ta for now)
I can do absolutely anything. I'm an expert! faq731-376 forum1529 Entire Forum list

Thanks for that spreadsheet. I'm not quite following what you did on the Results tab. It looks like you entered 0.5 in cell E8 & E18 and 0.866 in cell E23 to drive the output in E28 to 1. Is that correct? Where did the E8 value of 0.5 come form? That is showing GR&R of 25% in cell L18. In my case the production equipment has a GR&R of 20% so should the input to E8 & E18 be 0.447 which gives 20% in L18 and 0.8946 in E23 resulting in TV of 1.00006?

Then the process limits become 0.8946 * 5% = 4.473%



----------------------------------------

The Help for this program was created in Windows Help format, which depends on a feature that isn't included in this version of Windows.

Sorry about the upload confusion. By now it should be pretty evident that I'm noob on the subject; nevertheless, it is a math problem and therefore interesting.

So here, I've got %EV at 20% of TV, by making EV=0.5, which is what I think your OP stated, where I basically put PV at unity and assumed there was no AV (operator error).



So that puts the TV larger than the allowable spread. The second sheet basically uses Solver to set the TV=1 by adjusting PV. I'm unclear whether this is what you intended, though



TTFN (ta ta for now)
I can do absolutely anything. I'm an expert! faq731-376 forum1529 Entire Forum list

OK, I see what you did in the first screen, but why would you not just use the reciprocal of the TV=1.118? That gives 0.994, which is pretty much want I came up with. Your second screen the GR&R is now 25%. I must be missing something.

----------------------------------------

The Help for this program was created in Windows Help format, which depends on a feature that isn't included in this version of Windows.

The GRR gets larger because the the TV has to decrease to fit the spec. Since that comes at the expense of throwing away otherwise good product, the GRR has to get larger.

To recap, your TV is an RSS of the 3 major components of variability, one of which (AV) is assumed to be zero, since it's an instrument and not an actual gage. The EV, based on the original statement of GRR=20%, is essentially 1/2 as large as the entire PV. If EV were zero, then all of the product would be acceptable at its specified limits. Luckily, you're RSS'ing everything, but, as it stands, the EV is potentially hurting you by potentially passing bad product as good, and failing good product. Assuming that you want to minimize the acceptance of bad product means that you'll now have to throw away good product, so that can only happen if you reduce the acceptance limits from the original 5% to ~4.3%. Since the EV is unchanged, the GRR winds up being larger, since its relative magnitude is now 0.5/0.866 = ~58% as big as the PV.

Hopefully, that makes sense; I think it does, but caveat, this is new material for me.

TTFN (ta ta for now)
I can do absolutely anything. I'm an expert! faq731-376 forum1529 Entire Forum list

I believe that is correct, it makes sense to me. Thanks for the help. MFG is not going to like +/-4.33%, it's a little more than the +/-4.20% I was allowing but not the +/-4.8% they have been arguing for (+/-[1-.2^2]*5%).

Totally agree we need to improve MFG process capability AND gauge repeatably.

One saving grace is that in most applications this product gets used with multiple units in parallel. Typically 5 to 10 of them are fitted into the customers metering block and the flows combined. Assuming sufficient randomization the average combined flow should be more uniform than the individual units.

----------------------------------------

The Help for this program was created in Windows Help format, which depends on a feature that isn't included in this version of Windows.

Thanks for posing the problem; learned something new today.

TTFN (ta ta for now)
I can do absolutely anything. I'm an expert! faq731-376 forum1529 Entire Forum list