P/T ratio vs GRR - which is more meaningful and why?

I am an undergraduate currently studying Quality engineering. My question is as follows:

Performance to Tolerance (P/T) ratio uses σ_R&R / (USL - LSL) as an estimate of a gauge capability while GRR method uses σ_R&R / σ_total where σ_total is given by sqrt[(σ_R&R )^2 + (σ_product) ^2].

Which estimate/method (P/T ratio or GRR) gives a more meaningful estimate of the performance of a gauge? As an extension, which is more stringent? Any responses appreciated.

 

Howdy, and welcome to QFO!

Depends: What "meaning" are you looking for?
If you are trying to convince a customer product engineer that your measurement system is adequate for their part to work, P/T ratio.
If you're trying to answer most auditors, or a customer SQE that knows nothing of the part...GRR.
If you're trying to answer correctly on an undergrad test...whatever the professor says is right.

Depends:
If the tolerance is less than

P/T ratio is more stringent.

If tolerance is greater, then GRR is more stringent.
In every case I've ever seen, GRR is more stringent.

FWIW, out in the field I've never heard of P/T ratio...both methods are called "GRR", and only those who have an academic interest or are trying to show off for someone will ask which method was used.

 

Hello! thanks for the welcome, what a great community here. Insightful response, really appreciate it. Thanks again!

Actually, I was thinking about it. Would you say that the GRR method as per what I defined could be more meaningful also because measurement variation due to gauge could be a very small fraction of the tolerance because of high tolerance (wide specification limits) but a large fraction of total observed variation of the process?

I would think that the gauge performance isn't good enough if measurement variation accounts for a large part of observed process variation, but I wouldn't know if I had used the P/T ratio with a high tolerance. Thoughts on this? Is this a possibility in reality?

Sorry, I'm speaking very theoretically and from a lot of inexperience!

 

Agree with Miner...read his blog...lots of gold in there.

I am an engineer responsible for good parts shipping on time in order to make money. I use what you are calling P/T ration exclusively.
I've had scientists insist that what you call GRR is the only allowable way and tried to alter the test method to improve repeatability.
They get one warning that "We aren't here to comply with a math exercise, we're here to ship good parts and {P/T ratio} accomplishes that with no issues. You are wasting time and labor expense."
When they insist again (they always do, it's like an infection) they get transferred to a different area to be someone else's problem.
That's the real life aspect from my experience.

Performing well in GRR using std.dev. is a great bonus...performing well in GRR using USL-LSL is good enough to stop trying to fix it.

Again, Miner covers a lot more in his blog...take a good peruse through there.