Gage R&R with only one operator using JMP

[McDesign25]

For example, I measure the width 3 gage blocks, 1.000", 2.000", and 3.000". I measure them 10 times each and repeat it once as the same operator. I would like to know the amount of the tolerance that is being taken up in my measurement? I would calculate the P/T ratio. Do I have to repeat my measurement twice, or since I am the only operator can I do it only once? The JMP manual does not help in this regard. If you could point me to a good book for reference on Gage R&R's that would be great as well I could not find any.

Thanks.

 

[melone]

The classical GR&R requires 3 operators, 10 random samples and 3 measurements per sample per operator. This is required to measure Repeatability (the ability of a single operator to measure the same value on a device) and Reproducibility (the ability of multiple operators to measure the same value on a device). You will need to meet those requirements before JMP can help.

As far as references on GR&R, you should be able to search the net. There are LOTS of good references out there.

 

[McDesign25]

Yes thank you melone.

I am aware of the 3-10-3 idea. Our application involves a CMM machine that is automatically doing the measurement. Only one operator will be running the machine. Could JMP calculate the Repeatability section using only 1 operator? I know this has been encountered in industry before.

 

[melone]

Have 3 different operators write the program. Run the program on 3 different days. Run the program at different times of the year (to account for temperature / humidity /etc.). I have run MANY GR&R's on CMM's, and have always been able to meet the 3-10-3 rule.

 

[melone]

There is another acceptable version of the GR&R that is useful for reduced operators. I am going from memory here, but I believe that the sample size increases to 25, the repeat trial increases to 5, and the number of operators decreases to 2. Once I get back to work, I can dig through some of my material and look for the specifics. Basically, if you decrease the number of operators, you need to increase the samples and number of repeat measurements.

Also, when performing your repeat measurements, it is CRITICAL that you redo your setup. That is to say, you MUST tear down your fixturing and set it all back up. (Moving your fixture to another location on the actual CMM would be another operator....). Ideally, you would tear down your setup in between every measurement, but his often proves too time consuming. Therefore, I have always torn down my setup in between trials, i.e.
1) Measure units 1-10 in ascending order
2) Tear down and build up my fixture (if using fixturing blocks)
3) Calibrate the CMM to my fixture
4) Measure units 1-10 in descending order
5) Repeat 2 & 3
6) Measure units 1-10 in random order.

With a CMM, there usually isn't a dependency on measurement order, but it is a fairly trivial step to remove all doubt.

Hope this helps!

 

[McDesign25]

Thank you Melone, I will try that. I am in the process of re-doing the measurements.

Also, have you heard of a gage verification study? I was reading a new paperbook titled, "all about six sigma" about the authors opinion about how a gage verification study may be a good option versus a gage R&R. The biggest difference is you use a known standard part instead of actual production parts, and you find the total variation by 3 people each measuring the known standard 7 times. You find the average and standard deviation of all 21 measurements, and to find the P/T ratio, all you do is:

P/T = ((5.15*StDev)+ ABS(Known Standard - AVG)) / (USL-LSL)

The author of this book feels that anything from 10 to 30 percent is good, and anything >30 percent needs to be improved. It seems much simplier in some cases. What is you opinion on that. The 5.15 refers to 99% of the variation of a normal distribution. A known standard could be a gage block or a sample part that a lab with greater measurement equipment has agreed upon. I am very interested in this area. Any feedback would be appreciated.