Noise Tester GR&R (Gage R&R) study on one sided specification

[Yew Jin]

Hi all,

We are doing the GR&R on our noise tester. The upper specification limit of the noise is 30. The nominal of the noise in current process is 28. When I use the minitab on GR&R ANOVA Crossed method (study variation) it is very hard to get the GR&R below 30%. The repeatability variable is very high although the range among the 3 trial runs from a particular operator and a particular sample is only 1 point something.

Is any method that I can use to study the tester GR&R in this case?

Thanks and regards
YJ

 

[Ron Rompen]

Re: GR&R study on one sided specification

ALthough I haven't had much success in using Minitab for GR&R, I would think that you could include a lower limit (of zero) for your study.

If feasible, this should also be set as a boundary, to show that the value can never go lower.

Failing this (if zero is not acceptable) I would use 'basic logic'. Take the average value of your readings, and 'force' the limits to ± 14. You are determining the effectiveness of your equipment in measuring to this range, not the process capability.

 

[Dave Strouse]

Re: GR&R study on one sided specification

Yew Jin -

Getting a study var GR&R greater than 30% has nothing to do with one or two sided specs. That might affect the P/T ratio if you enter the tolerance of 30 into options.

If you are getting high repeatability it means that the repeatability (pure error) is high with respect to the total variation. I'm afraid that the data you stated on the range is somewhat irrelevant unless evaluated against the total variation. For example,if the variability is 1 but the total variability is only 3, you will get 33% GR&R.


FROM MINITAB HELP (bold is mine)
Study Var
Standard deviation component multiplied by 6. You can change the multiple from 6 to some other number in the Options subdialog box. The default is 6 because 6 is the number of standard deviations needed to capture 99.73% of your process measurements. The last entry in the Study Var column is 6*Total. This number, usually referred to as the study variation, estimates the width of the interval you need to capture 99.73% of your process measurements. Use the multiplier 5.15 to capture 99% of your process measurements.

%Study Var
Percent of the study variation for each component. This is calculated by dividing the standard deviation for each component by the total standard deviation. The percentages in this column do not add to 100.




So investigate the reason for the high ratio. One real big thing to check is the way you did the study.

How many parts/operators/trials did you use?

Do the parts span the normal range of variation in your process?
This one is critical. The criteria of <30% GR&R is presupposed on this. As shown above it is a ratio. If you cherry pick parts to all be near the same value (a common mistake when beginning GR&R) than the total standard deviation will be artificially low for the intended use of the measurement system, The %GR&R will react exactly as you describe.

Conversely , if you span too wide a range e.g. you want to use a thickness gage on three different products that are 1mm, 10mm and 100mm thick. If you include parts from all products in the study, you will likely get a good GR&R, but the gage may not be acceptable for any of the products evaluated separately.

Finally, if you have never previously evaluated the gaging system, it just might be unsuitable for this purpose. It would help if you can post the data .

 

[Yew Jin]

Re: GR&R study on one sided specification

I have attached the raw data for 3 operators with 3 trial runs on each for the 10 samples used.

Please help to advice how to analysis the data in the GR&R with the upper spec is 30. (one sided spec)

 

[Dave Strouse]

Re: GR&R study on one sided specification

Yew JIn:
I'm going to repeat two things from my earlier post.

You can't be helped much until you answer this that was asked before.

Do the parts span the normal range of variation in your process?
This one is critical.

Please help to advice how to analysis the data in the GR&R with the upper spec is 30. (one sided spec)

Getting a study var GR&R greater than 30% has nothing to do with one or two sided specs. That might affect the P/T ratio if you enter the tolerance of 30 into options.

I'm attaching a report from MINITAB. Open and note that the NDC is 1.
Look at the graphs. The rbar is about 1, while the total spread of the part means is also about one. These conditions say BAD GAGE.

Conditional conclusions:

1) IF the parts used are not representative of the range to be tested, than redo the study on parts that span the range to be tested. Perhaps it will be much better if the range is much wider.

2) IF the parts used do span the normal process output, than save the company some time and money. Get a dart board and assign various values in the range 26 to 30 or so. Throw a dart at the board and record whatever it lands on. This measurement system is nothing more than a random number generator. It cannot discern one part from another. Seriously, under this condition, you must improve the gage system even to use it as an effective GO/NO GO tester.

Thanks for sharing. I do believe this is the second worst GR&R result I've ever seen! I remember one that had a %SV of 99 plus. BTW, it took a major redesign of the tester to improve that result.

 

[Yew Jin]

Re: GR&R study on one sided specification

Thanks. But our process capability just can go down to 27 for the mean from the upper spec.

The data that we collected for the GR&R is span in the normal range.

Do you use the study variation to calculate the GR&R? Or we can use the spec tolerance as 30-0 = 30 to calculate the GR&R?

 

[Dave Strouse]

Re: GR&R study on one sided specification

Yew Jin:

Sorry, I did not get the file attached previously.

Apparently you don't believe me when I said it will make no difference on the study var ratio wheter you input a tolerance or not. Enter the tolerance (30) yourself in the options box for GR&R crossed in MINITAB. You'll get a P/T ratio then of about 11%, but none of the other results change.

Using the very small data set given in a capability analysis shows a Cpk of 1.7 or so. If this is true of a more rational cabability study and Cpk is perhaps 1.5 to 2 or better, I would have to retract my earlier statement and say that maybe you could use this system as a go/no-go tester. However, from the data you will simply not be able to disposition parts that are much above 28 as the mesasurement error may be misleading you at about that point.

You will not be able to use it for any kind of analytic /DOE / process improvement type studies.

Hope all that makes sense. I would encourage you to improve this system if it is at all possible and the noise is an important customer issue.

 

[antoine.dias]

Re: GR&R study on one sided specification

I have put your data into the AIAG sheet.

( of course ) same result here.

The combination of "high % EV and low ndc" indicates that something is wrong about the equipment.

Best regards,

Antoine

 

[Yew Jin]

Re: Noise Tester GR&R study on one sided specification

Antoine, the AV formula is missing in the spreadsheet.

However, thanks all. The GR&R is the final step that we will evaluate. We are doing the accuracy, linearity, stability....evaluation to qualify the tester first. Well, sometime it is very difficult to maintain a "old technology" tester......

 

[antoine.dias]

Re: Noise Tester GR&R study on one sided specification

The formula is not missing.

If the result of it is negative, it automatically turns to "0".

added later :

as it also says in the AIAG manual page 115 :
quote
If a negative value is calculated under the square root sign, the appraiser
variation (AV) defaults to zero.
unquote

Best regards,

Antoine