Dear guys,

I’m quite curious as what MSA is tried to say if the calculation shows 0% for reproducibility. Yes understand from MSA guide line if the AV (appraisal variation) value is appearing in negative form, assuming 0 to precede the calculation.

The total GR&R is calculating base on EV (equipment variation). May I know what the reason is?

Is it mean inconsistent of repeatability definitely we can’t deliver a good “reproducibility”

The sign show, I have to correct the repeatability before count or looking into reproducibility

Correct me if my understanding is wrong.


Periannan :D

Mathematically, reproducibility is 0 or super good when one calculates a result and prior to performing a square root, finds the result is a negative number. Of course, we cannot get the square root of a negative number so the result is zero.

The R & R value is not just the value of Repeatability.

This result has happened to me many times and the first time, I really didn't know what to do.

Hope this helps.

Dear David,

Understood about the math calculation, definately we can't square root negative value. But why the MSA guide line shows assume 0 when such value appear in negative to proceed the calculation.

Why such rule been given, definately there will be a reason behind. Is it Inconsistent of repeatibility will lead to negative value for AV ?

What shall we do to improve when AV show in negative value. first of all, is it an hidden error?

Periannan :confused:

I cannot give you a rational for this but considered yourself blessed if your AV is 0.

How to improve? Simple - work on your repeatability but that is only if your gauge needs improvement.

MSA 3 dictates that the R & R is to be a percentage of the process and not the tolerance range as in the past. Make sure that the parts you use for this test are not consecutive parts (no range between the parts) and spread the range of the tolerance otherwise one will attain a high percentage.

Your AV is 0% for one of two reasons, both relate to the ndc. If you look at the ndc, you probably have a ndc of below 5, and I am guessing it is 1. This means your GRR might not be good depending on your situation.

1) The sample does not reflect the actual process. Samples are consecutive and/or do not represent spread of the process range. Repeat the analysis with a sample that has parts close to the upper, lower, and center of the range.

2) Your process variance is so small your gauge cannot detect the variability. You will need to look for a guage that has better discrimination if you want to do process control.

Page 115 of the MSA 3rd edition standard states "If a negative value is calculated under the square root sign, the appraiser variation (AV) defaults to zero".

It does not state that there is something wrong with the study nor should one try to make improvements to achieve a positive number. Be happy!! You probably have other problems in the plant that are greater than this one.

I am not sure where in the book it talks about analysing the results, but it would be nice if they described it on that same page.

Here is a link with someone who got GRR result similar to what you are asking. I recommend reading this Thread.

http://elsmar.com/Forums/showthread.php?t=19482&highlight=ndc

Depending on your situation, you must interperet your results differently. This is a bit of an extreme example, but I cant think of another:

Say you found a process where a part was machined to 3.000" +/- 0.010on a surface grinder. You do a GRR using a micrometer with 0.001 discrimination. Your results might show that the AV is 0%. You probably can use the gage for inspection with no problem. However, there could be a number of other possiblilities here. Maybe you are wasting money on an over kill process/overkill inspection. Maybe you would like to improve this process/ tighten the tolerance, or do process control. You will need a better gage.

Yea you might be able to move on. It just depends on your the application of your gage and your goals.

ahem: LOOK at the data. remember what the %variation means...

there IS NO OPERATOR TO OPERATOR (reproducibility) variation; the operators are performing the same! (at least within the statistical resolution given the amount of measurement error)

Equipment error is good too. NDC = 6. Not super but good.

Woops, for some reason I did not see the file attached, and was making assumtions. duh duh :D Sorry about that. Guess I have never seen a situation where the process variation was so high. I am used to less process variation and not being able to find capbable gages.

Yup Im wrong the gage is probable ok for process control/improvement in this case.

Thanks for clearing me up!

These results are very interesting to me. Can you tell us more about the process, gage, units, tolerance?

Dear Micsim,

Let me give you a clear picture, the tester is to make sure the brush bar is spinning evenly. Uneven will cause abnormal noise when assemble to machine level. This jig equips with software which is reveal the reading in mg-mm once the spinning rpm is above 1000. The data will capture left and right. The end judgment will fail if either one is fail. Allowance range is from 0 to 17,000 mg-mm.i have tried to plot the bell curve for each sample that tried for 9 times. this to see how far the range will be

hope this will give you some idea to think about

Periannan :)

So, just for giggles - what would happen if you ran this study to evaluate it as a % of Tolerance instead of standard Study Variation for the variable feature that you are measuring - what does the R&R come out to be in that case?

So, just for giggles - what would happen if you ran this study to evaluate it as a % of Tolerance instead of standard Study Variation for the variable feature that you are measuring - what does the R&R come out to be in that case?

It will be even better. the range of actual measurements is less than the tolerance zone (UNLESS there is significant measurement variation at the outer limits of the sepc range...this would be an interaction of the measurement system with the process)

The reason the reproducibility is so small (negative forced to zero?) is that the standard deviation for the differences in operators is very small while the measuremetn system std dev is very large. it's a simple math problem of squaring subtracting then taking a square root. try it by HAND instead of by minitab.

However, one thing you cannot see at all with the traditional method of gage R&R plotting is that operator 2 has extremely good repeatability, on the order of 19 distinct categories! I always recommend plotting the repeated measurements on a scatter diagram with a 45 degree line on it. thsi shows you a LOT more than the standard graphs and it can be done in one graph instead of 4...