I need some help please.

I have been trying to understand MSA analysis, particularly Gage R&R. I have been going through the MSA Third Edition manual, this forum as well as my statistics book in order to understand ANOVA method.

I have a decent understanding of the math behind the method, the problem I am having is looking at the results and determining if the study that was conducted is acceptable or not. I am currently conducting the study on micrometers that we use for SPC, part verification and PPAPs. When conducting our PPAPs we use the Average and Range method to determine if the process is valid, however, when doing a study on the general measurement system we do not have a specific tolerance. So, I started looking at the ANOVA method to determine if the measurement system was good. However, I have not been able to figure out a way to look at the data and determine this. Is the ANOVA method a good method to use for this? or is the Average and Range method better?

I have attached two sheets, one that has the details of our last PPAP that we conducted and the other the trial general GRR study that we are doing.
What I had believed to be the determining factor on whether the system was appropriate using the ANOVA method was the f factor rejection, the ndc value, and if there was enough part variation. However I am not sure if the f factor rejection is an appropriate metric to use.

I wrote this spreadsheet to conduct the calculations and have checked them but please feel free to plug the data into minitab or elsewhere to double check and verify that everything is calculated right and to help analyse the data.

Any help that anyone can give would be greatly appreciated.

Thanks

I have a decent understanding of the math behind the method, the problem I am having is looking at the results and determining if the study that was conducted is acceptable or not. I am currently conducting the study on micrometers that we use for SPC, part verification and PPAPs. When conducting our PPAPs we use the Average and Range method to determine if the process is valid, however, when doing a study on the general measurement system we do not have a specific tolerance. So, I started looking at the ANOVA method to determine if the measurement system was good. However, I have not been able to figure out a way to look at the data and determine this. Is the ANOVA method a good method to use for this? or is the Average and Range method better?The ANOVA method is slightly better because it provides information about a possible Operator x Part interaction. The Range method cannot provide this information. If no interaction exists, the results between the two methods will be virtually identical. Both methods should provide a %GRR by each element (Repeatability and Reproducibility). This is the criteria for acceptance in addition to ndc.

I have attached two sheets, one that has the details of our last PPAP that we conducted and the other the trial general GRR study that we are doing.
What I had believed to be the determining factor on whether the system was appropriate using the ANOVA method was the f factor rejection, the ndc value, and if there was enough part variation. However I am not sure if the f factor rejection is an appropriate metric to use. The F-ratio simply tells you whether that source of variation is statistically significant, not whether it is of practical significance. The ndc and %GRR tell you the practical significance.

I wrote this spreadsheet to conduct the calculations and have checked them but please feel free to plug the data into minitab or elsewhere to double check and verify that everything is calculated right and to help analyse the data.
I see some differences between your results and Minitab. Some appear to be due to rounding differences. The parts appear to have been specifically selected to cover a wide range of variation. The variation among the 10 parts should accurately reflect the actual variation in the process. If they have a greater variation than the actual process, it will inflate (and invalidate) your %GRR results.

Thanks for the clarafication on that.

I have another question concerning what the proper parts selection should be. We have different size micrometers in this case, 0" - 1" measurement. Should the parts selected be for the entire range of the micrometer? The micrometer is used to check product and SPC for its entire range on our production lines. As seen in the previously attached spreadsheets when the parts are selected over this entire range the Part variation overrides everything else, however, when we select parts from a specific size and tolerance range we get a very low ndc of 1 or 2. An example of this is the PPAP spreadsheet above. A typical tolerance is 2 or 3 thousandths of an inch and our micrometers measure 0.00005 inch increments.

Thanks

I have another question concerning what the proper parts selection should be. We have different size micrometers in this case, 0" - 1" measurement. Should the parts selected be for the entire range of the micrometer? No. A Linearity study may be used to verify the gage over the entire range of measurement.

The micrometer is used to check product and SPC for its entire range on our production lines. As seen in the previously attached spreadsheets when the parts are selected over this entire range the Part variation overrides everything else, however, when we select parts from a specific size and tolerance range we get a very low ndc of 1 or 2. An example of this is the PPAP spreadsheet above. A typical tolerance is 2 or 3 thousandths of an inch and our micrometers measure 0.00005 inch increments.
Since you state that the gages are used for SPC, %GRR and ndc are the appropriate criteria to use to determine gage suitability. The parts should be selected from a specific size of product and should represent the variation of that specific product. If your study still results in an ndc of 1 - 2, your gage is inadequate for SPC. It might still be acceptable for use as an inspection device to the tolerance. The P/T Ratio is the appropriate criteria to determine this.

Since you state that the gages are used for SPC, %GRR and ndc are the appropriate criteria to use to determine gage suitability. The parts should be selected from a specific size of product and should represent the variation of that specific product. If your study still results in an ndc of 1 - 2, your gage is inadequate for SPC. It might still be acceptable for use as an inspection device to the tolerance. The P/T Ratio is the appropriate criteria to determine this.

I would like to make sure I understand this correctly.

Everything I have read about how sensitive a gage needs to be in order to be suitable is 10 times more accurate than your tolerance increment. We use 0.001 tolerance and our gage is 0.00005, which is well over 10 times more accurate. We took 10 random samples in a standard operating run on our line. The spec limits of the bars were 1.1348 to 1.1378. Our SPC data shows the parts were run between 1.13625 and 1.13675. The measurements on the samples bars ranged from 1.13600 to 1.13705. Are our samples not showing enough variation and that is causing the problems? We took these samples off the line and is typical variation from that line. Do we need to get parts with a greater variation than this even though the line does typically holds around one thousandth tolerance?

Thanks

Everything I have read about how sensitive a gage needs to be in order to be suitable is 10 times more accurate than your tolerance increment. We use 0.001 tolerance and our gage is 0.00005, which is well over 10 times more accurate.This is the gage resolution. A gage with inadequate resolution will cause poor R&R results. Unfortunately, a gage with good resolution does not guarantee good R&R results. Your gage resolution is 5% of the tolerance, which means resolution should not be an issue for inspection to tolerance. However, your gage is 10% of your process variation. This should still be acceptable but it is different from the first.
We took 10 random samples in a standard operating run on our line. The spec limits of the bars were 1.1348 to 1.1378. Our SPC data shows the parts were run between 1.13625 and 1.13675. The measurements on the samples bars ranged from 1.13600 to 1.13705. Are our samples not showing enough variation and that is causing the problems? We took these samples off the line and is typical variation from that line. Do we need to get parts with a greater variation than this even though the line does typically holds around one thousandth tolerance?The variation of your samples appear to reflect the variation in your process, so your select of samples is appropriate. Do not select samples with greater variation.

Reviewing your study, you have problems in both Reproducibility and Repeatability. Review the MSA manual. There is a section of the manual that lists many possible causes for each. One area that typically causes MSA problems is variation in form of the parts themselves. When you measure products with such small variations in size, the variation in shape may be a large part of this.

Hi... i am new here.... I am an Engineer in a certain company here in the Philippines... Just want to ask if it is ok to use ANOVA- 10 parts, 3 Appraisers and 2 trials? I recently generated a GRR procedure here in our company... i used that method. Is it ok for just 2 trials instead of 3?

Hi... i am new here.... I am an Engineer in a certain company here in the Philippines... Just want to ask if it is ok to use ANOVA- 10 parts, 3 Appraisers and 2 trials? I recently generated a GRR procedure here in our company... i used that method. Is it ok for just 2 trials instead of 3?

Yes, this is acceptable. There is a limit to how small you can go. For example, 5 parts, 2 apprasiers, 2 trials is really pushing it.

The ANOVA method takes the sample sizes into account in the degrees of freedom (df). As the df gets smaller, the test becomes less sensitive to detecting the effect of Repeatability, Reproducibility and Operator x Part interactions.

Here is a discussion (http://elsmar.com/Forums/showpost.php?p=197901&postcount=5) on the samples required.

Since you state that the gages are used for SPC, %GRR and ndc are the appropriate criteria to use to determine gage suitability.

When looking at the %GRR numbers, what is the breaking point between acceptable and unacceptable. Below is some data from a study we did.

Variation | St Dev | %TV | %Cont
Eq Variance (EV) 3.681E-08 | 1.151E-03 | 0.47% | 37.06%
Interaction (INT) 2.152E-08 | 1.291E-07 | 0.00% | 21.67%
Appraiser (AV) 2.103E-08 | 8.701E-04 | 0.35% | 21.17%
System (GRR) 7.936E-08 | 1.443E-03 | 0.59% | 79.90%
Part (PV) 1.996E-08 | 8.478E-04 | 0.34% | 20.10%
Total Variation 9.933E-08 | 2.465E-01 | 100.00%

Thanks

Thanks Miner! I appreciate your response!

Thanks Miner! I appreciate your response!

No. A Linearity study may be used to verify the gage over the entire range of measurement.


Why not use parts across the entire range? Just put part into the ANOVA model to eliminate part to part variaiblity (this will be highly significant) then you can get a pooled estimate across the range. If you do a linearity study you assume constant variance across the range (assuming you use regression analysis), so why not get a pooled estimate across the entire range?

Hi

For Attribute Gage R&R what is the Sample size
The TS auditors likes to measure 100 parts
but i could not find the this number in the MSA manual

Regards

D.Prabhu

Hi

For Attribute Gage R&R what is the Sample size
The TS auditors likes to measure 100 parts
but i could not find the this number in the MSA manual

Regards

D.Prabhu

Hello Prabhudvp,

Can you clarify why the TS Auditors "required" 100 parts for an attribute gage study? Was there a Customer-specific requirement? If not, you could ask the auditor: "where is the requirement that states that we must use 100 parts?"

I am curious about the reponses from MSA Specialists here at The Cove.

Stijloor.

Why not use parts across the entire range? Just put part into the ANOVA model to eliminate part to part variaiblity (this will be highly significant) then you can get a pooled estimate across the range. If you do a linearity study you assume constant variance across the range (assuming you use regression analysis), so why not get a pooled estimate across the entire range?

You can get away with this if your study aim is soley for this purpose, or to obtain the P/T Ratio. If you want the %GRR, your samples MUST reflect the actual process variation. Otherwise, you invalidate the results of the study.

When looking at the %GRR numbers, what is the breaking point between acceptable and unacceptable. Below is some data from a study we did.

Variation | St Dev | %TV | %Cont
Eq Variance (EV) 3.681E-08 | 1.151E-03 | 0.47% | 37.06%
Interaction (INT) 2.152E-08 | 1.291E-07 | 0.00% | 21.67%
Appraiser (AV) 2.103E-08 | 8.701E-04 | 0.35% | 21.17%
System (GRR) 7.936E-08 | 1.443E-03 | 0.59% | 79.90%
Part (PV) 1.996E-08 | 8.478E-04 | 0.34% | 20.10%
Total Variation 9.933E-08 | 2.465E-01 | 100.00%

Thanks

Typically, < 10% is excellent, 10 - 20% is acceptable, 20 - 30% is okay if the dimension is non-critical and the expense of improving the gage is not warranted, and > 30% is not acceptable. Also, there is more detail in other posts within this forum.

Typically, < 10% is excellent, 10 - 20% is acceptable, 20 - 30% is okay if the dimension is non-critical and the expense of improving the gage is not warranted, and > 30% is not acceptable. Also, there is more detail in other posts within this forum.

Variation | St Dev | %TV | %Cont
Eq Variance (EV) 3.681E-08 | 1.151E-03 | 0.47% | 37.06%
Interaction (INT) 2.152E-08 | 1.291E-07 | 0.00% | 21.67%
Appraiser (AV) 2.103E-08 | 8.701E-04 | 0.35% | 21.17%
System (GRR) 7.936E-08 | 1.443E-03 | 0.59% | 79.90%
Part (PV) 1.996E-08 | 8.478E-04 | 0.34% | 20.10%
Total Variation 9.933E-08 | 2.465E-01 | 100.00%

So looking at this data you say that the GRR should be < 30%. Is the % Total Variation number or the % Contribution number the one to look at? The %TV is 0.59%, extremely low while the % Contribution is 79.90% completely unacceptable.
If neither of these values are the correct one then what do you look at?

Thanks

Variation | St Dev | %TV | %Cont
Eq Variance (EV) 3.681E-08 | 1.151E-03 | 0.47% | 37.06%
Interaction (INT) 2.152E-08 | 1.291E-07 | 0.00% | 21.67%
Appraiser (AV) 2.103E-08 | 8.701E-04 | 0.35% | 21.17%
System (GRR) 7.936E-08 | 1.443E-03 | 0.59% | 79.90%
Part (PV) 1.996E-08 | 8.478E-04 | 0.34% | 20.10%
Total Variation 9.933E-08 | 2.465E-01 | 100.00%

So looking at this data you say that the GRR should be < 30%. Is the % Total Variation number or the % Contribution number the one to look at? The %TV is 0.59%, extremely low while the % Contribution is 79.90% completely unacceptable.
If neither of these values are the correct one then what do you look at?

Thanks

Please attach your raw data. Your %TV numbers do not "look right" to me.

%GRR should be assessed against the %TV or % Study Var (Minitab). The %Cont just provides a quick Pareto analysis of the sources of variation.

Please attach your raw data. Your %TV numbers do not "look right" to me.

%GRR should be assessed against the %TV or % Study Var (Minitab). The %Cont just provides a quick Pareto analysis of the sources of variation.

I did find an error in my calculations which I did fix. The numbers look "more real" now. But I still do not if this would be good or not.
The raw data is in the "Calc1" worksheet. Thanks

Now it looks right! I see that your decimal points were off by 2 in the results.

This gage is unacceptable. Your results are off somewhat from Minitab. I attached the Minitab Analysis for your review.

Your %GRR results of %Total Variation = 86.22% (Minitab's %SV = 89.39%) is much greater than the 30% max acceptable, so this gage is not acceptable for use in process control provided that the variation of the parts accurately reflects the actual process variation. An ndc = 1 is also unacceptable. Since the P/T Ratio is also unacceptable at 56%, you have a problem with the gage.

The % Contribution shows that the sources of measurement variation are large in all categories (Repeatability, Reproducibility and Operator x Part Interaction). Are your parts subject to variation in form (shape)?

Hi guys,

If the process capability is adequate (Cp > 1.33) the lack of precision and inadequate GRR is not a immediate priority. As process improvements are made, the situation is not last. Determine the time line to improve the measurement improvement.

If the process capabiliy is between 1.00 and 1.33, unacceptable GRR variation may make difference in whether the observed process is report as capable.

If process capability is clearly inadequate, Cp <1.0, both measurement and process must be addressed. Measurement system should be in the priority.

Now it looks right! I see that your decimal points were off by 2 in the results.

This gage is unacceptable. Your results are off somewhat from Minitab. I attached the Minitab Analysis for your review.

Your %GRR results of %Total Variation = 86.22% (Minitab's %SV = 89.39%) is much greater than the 30% max acceptable, so this gage is not acceptable for use in process control provided that the variation of the parts accurately reflects the actual process variation. An ndc = 1 is also unacceptable. Since the P/T Ratio is also unacceptable at 56%, you have a problem with the gage.

This leads into my problem with understanding how this all should work. My company, when we have done PPAPs in the past for customers would always take the Average and Range data Total Variation Percent of Tolerance to determine if it passes or fails. In this example this value is 10.36%, which is acceptable. However, the GRR is 78.60% but 8.15% of tolerance. The ANOVA GRR is 86.22%. If you look at this in regard to the tolerance then it is acceptable. But if you just look at the variation percent you have an unacceptable level. Which way are you suppose to look at this?


The % Contribution shows that the sources of measurement variation are large in all categories (Repeatability, Reproducibility and Operator x Part Interaction). Are your parts subject to variation in form (shape)?


The parts that we are doing are turned hot roll bars with a size of 1.1378 +0/-0.003 and are not subject to variation in form.

In response to Yew Jin, the Cp number is 2.69 and Cpk 2.41 for this part.

Thanks for all of the help, it is much appreciated.

This leads into my problem with understanding how this all should work. My company, when we have done PPAPs in the past for customers would always take the Average and Range data Total Variation Percent of Tolerance to determine if it passes or fails. In this example this value is 10.36%, which is acceptable. However, the GRR is 78.60% but 8.15% of tolerance. The ANOVA GRR is 86.22%. If you look at this in regard to the tolerance then it is acceptable. But if you just look at the variation percent you have an unacceptable level. Which way are you suppose to look at this?

The parts that we are doing are turned hot roll bars with a size of 1.1378 +0/-0.003 and are not subject to variation in form.

There is another error in your spreadsheet. When you calculated P/T Ratio (% Tol) you divided s by the tolerance. The correct formula is 6s/Tol. This changes your 10.36% to 62.16%, which is also unacceptable.

It is possible for one indicator to be acceptable and the other to be unacceptable. The one that is important is the one that is relevant to the gage usage. For inspection to tolerance, use P/T Ratio (a.k.a. % Tol). For process control (i.e., SPC) or any type of statistical study (e.g., DOE, t-test, etc.), use %GRR (a.k.a. %TV or %SV).

Regarding your turned parts, you may want to have a few samples measured on a roundness tester for roundness, taper, and hourglass/barrel-shape. You may have variation in form that your gage cannot see. My experience has been that when Repeatability, Reproducibility and the Operator x Part Interaction are all large, variation in forum is usually the reason. And, I have seen a lot of turned shafts with form variation.