Handling knowledge of measurement uncertainty in practice - Part verification

[Morten Lunde]

MSA studies results in a %GRR value plus a bias-estimate etc. This can be translated into: The measurement process uses a percentage of the total tolerance you are verifying. The amount used by the measurement process is often denoted "U" when normal spread has been calculated .
Therefore you should only approve the part you are measuring if the result falls inside the specification limits minus "U" on each side of the interval (if bias is also considered, you could add this to the GRR or correct your measurement results systematically....).

This leads to questions like:

If the product has been produced for years before the MSA-study was done, we have actually approved parts measuring: tolerance+ (2x"U"), so should we then increase our tolerance limit with 2 x "U" (a "U" at LCL and UCL), and proceed approving parts inside "old" tolerance limits (new tol.-2x"U")?

Or should we only approve parts within: tol.-(2 x "U") and then exspect higher rejection rates?

If it is a new product and say the %GRR = 20 for a tolerance interval 10-20, could we then approve parts that measures 12-18 (having subtracted 20% of 10 from each side of the interval, not considering bias etc.)?

Have you any inputs on this?
How do you handle the measurement uncertainty in practice with regards to verification of parts?

It is mainly the "political" issues regarding introduction of MSA-studies and there results in a product verification process and in running production, that concerns me.

 

[Miner]

First, be careful with your terminalogy. GRR is a part of measurement uncertainty, but is not all of it. Your question appears to deal more with the GRR and the performance characteristics of the gage.

If your part characteristic is very important, such that no customer risk is acceptable, you may use "guardbanding". This is your tolerance - 2 x "U" scenario. This reduces customer risk to ~ zero, and places 100% of the risk on the manufacturer.

Your tolerance + 2 x "U" scenario does the opposite. 100% of the risk is on the customer and 0% on the manufacturer.

Your current situation places 50% of the risk on each.

The most common practice is to use the existing tolerance and ensure an acceptable %GRR. In a quality-critical application, guardbanding is a somewhat less-common practice that is usually dependent on having a very good capability index.

The MSA 3rd edition manual shows how to construct a gage performance curve that will show the probability of accepting a product throughout the tolerance range. This will help clarify your situation.

 

[Morten Lunde]

Hi Miner

My focus is the entire measurement process and the resulting U as defined in MSA 3rd ed. chapt. 1 sect. F. In my experience, the biggest contribution to U comes from the measure process (handling, levelling, measurepoint etc.) not the gage itself.

Also, as a rule I want to measure only characteristics that are important for costumers down the “production stream” including the end costumer.

Interesting input about sharing the risk. I have not realized this perspective before. Problem arises ofcourse, when a costumer checks the part characteristics eg. at incoming control. In a 50/50 scenario he will find parts outside specification, that manufacturer has judged inside spec. Result: Products falling between two chairs that nobody wants to take resposibility for...

I would prefer guardbanding as a manufacturer, cause if I don’t, it feels as if I am shipping products which I know have defects, unless ofcourse a 50/50 scenario has been specifically agreed.

The “gage performance curve” example referred to, where in the manual is this?

 

[Miner]

Hi Miner

My focus is the entire measurement process and the resulting U as defined in MSA 3rd ed. chapt. 1 sect. F. In my experience, the biggest contribution to U comes from the measure process (handling, levelling, measurepoint etc.) not the gage itself.

Also, as a rule I want to measure only characteristics that are important for costumers down the “production stream” including the end costumer.

Interesting input about sharing the risk. I have not realized this perspective before. Problem arises ofcourse, when a costumer checks the part characteristics eg. at incoming control. In a 50/50 scenario he will find parts outside specification, that manufacturer has judged inside spec. Result: Products falling between two chairs that nobody wants to take resposibility for...

I would prefer guardbanding as a manufacturer, cause if I don’t, it feels as if I am shipping products which I know have defects, unless ofcourse a 50/50 scenario has been specifically agreed.

The “gage performance curve” example referred to, where in the manual is this?

I am away from my manual, so maybe someone else can jump in and assist. I think it might be in one of the appendices, but I will not guarantee it.

In simple terms, you take each tolerance limit and draw one line either side of the each limit at +/- 3s (where s is the measurement error s determined in MSA study).

Everything less than the LSL - 3s and above the USL + 3s has a probability of acceptance of 0%.

Everything between the LSL + 3s and USL - 3s has a probability of acceptance of 100%.

At the LSL and the USL, the probability of acceptance is 50%.

Between LSL - 3s to LSL + 3s (and USL - 3s to USL + 3s), the transition of probability from 0% to 100% (and from 100% back to 0%) follows the probability distribution from a standard z-table.