Gage R&R for Automated Measurement Equipment Systems

Hello,

I searched in blogs ı couldnt find exact answer for my question.

We will buy a measurement robot which will decide part is ok/nok . I want to test it before acceptence.How can i implement GRR for this machine.This machine will be in the end of assembly line and no operator will effect robot system.I search MSA third edition was written by AIAG but I coulnt find explanation about this.

Re: Gage R&R for Automatical Measurement Systems

It seems very difficult to me
I have no idea , above all if it an unique machine.
I would be very curious in understanding the solution.

Re: Gage R&R for Automatical Measurement Systems

inneruniverse said:

................... We will buy a measurement robot which will decide part is ok/nok . I want to test it before acceptence.How can i implement GRR for this machine. ......................

Click to expand...

Welcome to the Cove.

I am sure you can direct these questions to your potential suppliers even at the selection stage. I believe the whole package should include all that you mentioned. If they used terms like IQ/OQ/PQ, ask them to explain if you are not sure what it means and what is the scope.

Meanwhile, this is a search here for links to our threads on Gage R&R for Automated Machines.

Re: Gage R&R for Automated Measurement Systems

Hi,

A GR&R is not run on an automated measurement system. Without operators, who handle the gage, there is no reproducibility term to measure. You need to assess the consistency of your measurement system, its precision, and bias. You are looking to see if the test-retest error is consistent. The basic model is:

X = P + E

Where X is product measurements, P is product values and the predictability of the production process, and E is measurement error and has three constituents; consistency, precision, and bias of your measurement system.

1. Consistency is the predictability of your measurement system
2. Precision is the variation of a predictable system
3. Bias is the mean of the errors produced by a predictable system

Your objective is to be able present objective evidence that the variance(E) is small compared to the variance(P). I would recommend the following to achieve that goal:

1. Set up a XmR chart(s) to measure the consistency of your equipment. I recommend that it be posted next to the equipment as part of the production control board for that equipment.

2. Select several parts, if possible, that will be measured repeatability on a schedule; every day, twice a week, once a week, once a month, etc. If you have multiple operators for this equipment they all follow the same measurement schedule and use the same part(s). You would need to schedule the operators for different times of the day; OP A 07:00, OP B 10:00, OP C 13:00, etc. You have a XmR chart for each operator and each part.

3. It would be best if the part(s) are standards. Standards have a known value that can be used for the bias and linearity study on your equipment. If you can not purchase standards then, if it is feasible, take some parts off the production floor, label them as reference standards, send them to a good calibration lab that can measure them and you have custom standards. You have to decide if your custom standards will be stable over time.

4. Perform a bias and linearity study of your equipment.

5. If you have operator(s) that place the part to be measured into a fixture then you have to determine how much error the operator placing the part into a fixture will introduce into your measurement. This is where the standard helps. Its value has not changed.

I strongly recommend that you purchase Dr. Wheeler's book:

Wheeler, D. J. (2006). Emp III Evaluating the Measurement Process & Using Imperfect Data. SPC Press.
ISBN-13: 978-0945320678


It is the best book I know of for, "how to", evaluating a measurement system.

Here is an article that might help: https://www.qualitydigest.com/inside/quality-insider-column/better-way-do-rr-studies.html

Good luck and here is a quote I like:

"Until a measurement process has been 'debugged' to the extent that it has attained a state of statistical control it cannot be regarded, in any logical sense, as measuring anything at all" -- Churchill Eisenhart

Keith Jones

Re: Gage R&R for Automated Measurement Systems

Keith,

Overall, this is a good post, and I encourage you to continue posting. However, I do take exception to the following comment:

keithljelp said:

A GR&R is not run on an automated measurement system. Without operators, who handle the gage, there is no reproducibility term to measure.

Click to expand...

While this MAY be true in some instances, it is not always true. For example, there may be multiple automated test systems, or a single test system with multiple test stations. I have encountered both types many times at multiple companies.

In this situation, tester to tester, or station to station, variability constitutes Reproducibility and must be included in the study.

Wheeler's article is still relevant. Instead of evaluating operator to operator differences, you evaluate tester to tester, or station to station differences. The method is still the same.

BTW, I am a big fan of Donald Wheeler's various works. It is a shame that AIAG has chosen not to continuously improve their approach. However, we must remember that many who visit this forum are forced to use the AIAG approach by their customers. And, while the AIAG approach is fundamentally flawed, it does work. It may result in rejecting an adequate gauge, but it does identify the various sources of measurement variation for further improvement.

Re: Gage R&R for Automated Measurement Systems

Miner,

I agree with you.

I wrote my reply based on a literal reading of Inneruniverse's post about "...a measurement robot..." one machine. When more than one test system is present the "ball game" statistically speaking changes considerably.

I have heard some people say that if more than one test system is present then the testers can be treated as multiple operators. I disagree with that in that each tester is, in-effect, a different gage as well. I personally would track each tester with a XmR chart to track their consistency and run a t-Test for the Significance of the Difference between the Means of Two Independent Samples to see if the two testers are measuring the same. We have not even touched on the auto-correlation of readings from an automated test system.

After fifteen years in the automotive electronics world as a quality, component, and test engineer I know well the pressure the big three put on second and third tier suppliers and the vagaries of the AIAG MSA Manual. I also found that the lack of specificity of the AIAG manual allowed me to make reasonable, statistically sound recommendations for the qualification of our automated measurement systems and have the recommended proposal accepted by the customer project manager.

Later this year we are going to set up a programmable CMM with ruby stylus and 3D laser scanner head in the CNC area. The recommendations I gave in my post is basicly the same proposal I wrote for our plant.

I think that a detailed discussion on qualification of automated measurements systems would make an excellent conversation over a steak dinner and a few cups of Jamaican Blue Mountain coffee.

Keith Jones

Re: Gage R&R for Automated Measurement Systems

keithljelp said:

I wrote my reply based on a literal reading of Inneruniverse's post about "...a measurement robot..." one machine. When more than one test system is present the "ball game" statistically speaking changes considerably.

Click to expand...

Not a problem. I have often been guilty of the same and have been called to task by others.

keithljelp said:

I have heard some people say that if more than one test system is present then the testers can be treated as multiple operators. I disagree with that in that each tester is, in-effect, a different gage as well.

Click to expand...

Yes, they are different gauges, but remember a measurement system consists of a gauge plus an operator. Change the operator and you have a different measurement system. This is splitting hairs. To me, gauge to gauge reproducibility is the same as operator to operator reproducibility.

keithljelp said:

I personally would track each tester with a XmR chart to track their consistency and run a t-Test for the Significance of the Difference between the Means of Two Independent Samples to see if the two testers are measuring the same. We have not even touched on the auto-correlation of readings from an automated test system.

Click to expand...

These are legitimate approaches. Including the testers as reproducibility variation is legitimate, and evaluating the testers using a Bland-Altman Plot or ISO-Plot are also legitimate approaches. The purpose is to quantify the effect of the different testers. There are many legitimate approaches to achieve the same end result.

keithljelp said:

I think that a detailed discussion on qualification of automated measurements systems would make an excellent conversation over a steak dinner and a few cups of Jamaican Blue Mountain coffee.

Click to expand...

I agree with the discussion, but I drink tea or beer, preferably an assertive beer such as an India pale ale or stout.

Re: Gage R&R for Automated Measurement Systems

Miner said:

...but remember a measurement system consists of a gauge plus an operator.

Click to expand...

To be a little more accurate - the gage, operator and the part within an environment.

Also, if you pick the operator skill levels to be representative of skill levels of people using the gage (inexperienced, experienced and expert) it should be adequate as valid support the gage as applicable - which is the true point of the exercise.

Hello again,

Thank you all, Let me explain that measurement robot.
We have assembly line.In the end of it we will set this robot (camera control system) to check if operators assembled components like sockets properly or not.Do you still make a GRR study? By the way management want us to calculate cmk for this camera control system, How can i implement it?

Essentially you have an attribute (go/no go) gage r. Best thing to do is got borderline good parts and borderline bad parts - about 50 total. Run them through 3 times, then run them again either the next day or shift (to capture stability -since your likely do not have operator variation if they are pulled from nests.) Use the attribute forms and calcs.