Yes I have went thru the other threads and they all link or say look in another thread which does not have any help at all. I only have minitab 13 which I own and I have not upgraded to 14.2 which does the unilaterail GR&R.
I tested for normaility CPK is fine, I got ahold of some sheets that another engineer said it passes, but am not sure how to read it. What exactly makes it passing. My customer looked at another excel sheet I had for 3rd ed and since it says fails he thinks it fails. Does this study pass or fail and WHY?

Thanks

Oh and first post!!!!

I do not need minitab help, unless someone has the $395 upgrade laying around not being used.

I need to be able to explain WHY and HOW the GR&R is acceptable, my customer is QS/TS certified and from the rejected first submission I must be able to pass on in laymans terms how to read the reports I have.

Wow first post and it gets moved to the WRONG section???????

That was not the topic title, I do NOT need minitab help, it was in the GR&R section and that is where this thread belongs. The moderators need to leave stuff alone sometimes. @@!#@!###

First, you may want to ponder the saying "You can catch more flies with honey than you can with vinegar." Your tone in your very first post will put a lot of people off that would normally help.

Now, to your question. The gage is not acceptable for use as an inspection device (P/T Ratio) or as a process control device (% GRR). 30% is considered the maximum acceptable limit for both measures with 10 - 15% the ideal maximum.

You gage does have adequate resolution, but the repeatability is the major source of measurement variation. This is the variation of a single operator repeating their measurements.

Since these are form measurements to a very high resolution, I assume that you are using a CMM. Is it manually controlled or DCC? The source of the high repeatability variation maybe the within-part variation in form. Using more points to define the form and defining the measurement location of each point may help.

Honey and flies, after reading the other responses to the unilateral questions, I could have made a post that would have gotten links to other threads that provide nothing I seek. Now Miner your response is what I am looking for. Yet it makes me wonder if the normaility of the measurements has no bearing on the overall GR&R. And Yes it is on a CMM, DCC, with a SP600M scanning head, and the roundness is from scans. I am tweaking the settings to improve the repeatability, slowing the scan speed and increasing the point count.

One of the other threads on unilat interpetation said to test for normilaity, is there a standard PPK,CPK that should be held for gages?

Our other option is to use a talyrond, which comes at a inspection time increase.

Also Miner please note I did not ask anything about minitab, can I please get a CAR issued to the moderator that moved this thread?:notme:

Miner thanks for the respones as well......... and yes I am a noob with GR&R's

Yet it makes me wonder if the normaility of the measurements has no bearing on the overall GR&R. The sample size of the measurements is really too small to define any kind of distribution. The normality of the distribution of the parts may impact your results because %GRR is a ratio of the measurement variation to the part variation. The part variation should reflect the variation of the parent distribution. If the parent distribution is nonnormal, it may be more difficult to obtain samples that accurately reflect that distribution.

And Yes it is on a CMM, DCC, with a SP600M scanning head, and the roundness is from scans. I am tweaking the settings to improve the repeatability, slowing the scan speed and increasing the point count. Good. On a DCC, the repeatability of fixturing the product may have an influence, especially of the program starts from a reference point on the fixture versus a reference point on the part itself. To minimize the effect of within part form variation, you want to scan the same location each time. For example, if you scan a circle to check roundness of the inner diameter of a hole 1" deep, you should always measure at the same depth each time.

One of the other threads on unilat interpetation said to test for normilaity, is there a standard PPK,CPK that should be held for gages?No. A Gage R&R is not the same as a capability study. The standard is P/T Ratio and/or %GRR should be less than 10% for an ideal gage. 10 - 30% may be okay for less critical applications, but > 30% is usually not acceptable. P/T Ratio is used if the gage is used for Pass/Fail decisions. %GRR is used if the gage is used for SPC or to make statistical decisions such as in a DOE or hypothesis test.

Our other option is to use a talyrond, which comes at a inspection time increase.
Try a study using this. If it does not improve, you may be looking at a fixturing issue or repeatablility of the starting location of the scan.

Yes, I have did the talyrond, yesterday. The tol. is .0002, and the customer wants < 20%. It is beyound the cmm to pass the GR&R, it just does not have the resolution, and can not repete within the .000012 needed to pass.

We have used 3 diffrent .xls GR&R sheets, and minitab, and today I ordered the versuion 14 of minitab so we can do a true unilat GR&R study< I will also have to get a copy of the MSA to figure out how to read the results.

Attached is the study off the talyrond, the sheet is 3rd edition and says fail???? In minitab the same exact numbers give me a 10.52%, so my lack of understanding in what exactly is wrong with the study has a job on hold.

?????????????????????????????????

Thanks again miner....

:topic: I do not need minitab help, unless someone has the $395 upgrade laying around not being used.

Wow first post and it gets moved to the WRONG section???????

That was not the topic title, I do NOT need minitab help, it was in the GR&R section and that is where this thread belongs. The moderators need to leave stuff alone sometimes. @@!#@!###

Miner - Thanks for helping!

justncredible - As an FYI, I re-title many threads and I moved this thread. Sometimes threads can be put in different places and each place would be applicable. Were that every thread could be 'button holed' into one specific forum category. In my opinion you should be happy that so many of us care enough to look at each thread, try to put in titles that (usually) best describe the question or discussion topic (goodness knows I've seen threads with titles as simple as "Help!" which makes it difficult when people are scanning thread titles while trying to find information), and if it might better fit in a different forum move it. All in all I think we do a pretty good job at keeping things organized, all things considered.

I originally retitled this thread only adding "Minitab 13" because you are using Minitab and you are using version 13. The most recent version might be different. I put it in the Minitab forum because you're using Minitab regardless of what function, so the assumption on my end was it would be best if people familiar with Minitab functions and outputs could best help. And I felt it made more sense to put it in the Minitab forum because many people do not use Minitab at all for gage R&R. I made a judgment call.

Don't get too excited. There was a redirect in the Gage R&R related forum. A lot of people here help each other out and most questions do get answered by people who just want to help someone out. As to attitude, people here are judged by their words and actions so I really don't have a comment. Ultimately, what people do to help others is the key. :2cents:

Yes, I have did the talyrond, yesterday. The tol. is .0002, and the customer wants < 20%. It is beyound the cmm to pass the GR&R, it just does not have the resolution, and can not repete within the .000012 needed to pass.

We have used 3 diffrent .xls GR&R sheets, and minitab, and today I ordered the versuion 14 of minitab so we can do a true unilat GR&R study< I will also have to get a copy of the MSA to figure out how to read the results.

Attached is the study off the talyrond, the sheet is 3rd edition and says fail???? In minitab the same exact numbers give me a 10.52%, so my lack of understanding in what exactly is wrong with the study has a job on hold.

?????????????????????????????????

Thanks again miner....

Attached is an analysis of your data using Minitab 14 with a unilateral tolerance of 0.0002 max.

In summary, your CMM has a P/T Ratio of 16.16%, which means that it may be acceptable for Pass/Fail decisions. You customer required < 20%, and it does meet that requirement, so the CMM may be used for Roundness inspection to tolerance.

The %GRR is 58.10%, and the ndc = 1 (5 minimum), which means that the CMM is not acceptable for process control (i.e., SPC).

The Repeatability and Reproducibility are almost identical, so if further improvement is required, I would look closely at the part fixturing and at scanning the same part location every time.

Thanks, Miner!

I have ordered the upgrade for minitab. The last dataset I posted was from the talyrond. I have the quality toolbox 1st and 2nd eds, and the certified quality engineers books sitting right here and they just do not give enough information to understand what is being reported in a GR&R. I will look into getting the MSA 3rd ed. It is rare in this job shop that we have to preform a GR&R and even rarer that the tol is .0002 on the GR&R. Is all good information on our inspection practices we now have a better grasp of what we need to be able to inspect a part with confidence.


Marc, geez, I only look in areas that I might learn or help others, GR&R experts would not look under minitab. Also I am not excited by anything other than it is friday and I am almost done today and will be intoxicated very soon. Others that seek help in GR&R's could benefit from the exchange between miner and me. Since it is your forums, I do not care if you delete the information, I have no vested intrest or ties to the topic. Think of my profession and understand I am use to yelling when a mistake is made.

We don't delete threads here very often and I'm surprised you'd think I would delete the information. I have absolutely no idea why you would even suggest that.

My post to you was essentially to say please 'Chill Out'. You're obviously new here and as Miner hinted, you came on rather strong. You asked very basic questions which, I would bet, are discussed in existing discusson threads, but you wanted information specific to your situation NOW.

I doubt that this thread is buried so deep that no one will find it especially considering the forum Search function. If you go to one forum expecting all discussion threads related to that one topic to be there you will miss quite a lot because many threads can be classified in several ways, including yours. Personally, I Search for information knowing that discussion threads can often be classified as belonging in several forums, and by only looking in one forum I might miss information that has already been discussed.

There is no doubt that others that seek help in GR&R's could benefit from the exchange between Miner and you. Those who know how to Search will find it.

I do thank Miner for taking the time to give you answers specific to your situation very quickly and in a courteous manner.

I am regressing a response variable on 11 regressors;three of which are categorical.(My data points are 50)

MINITAB's best subset gives me a model with good P and R squared values but the residual plots don't look good.

I figured i could take care of this with the boxcox transformation but i don't know what "group size" means in MINITAB'S boxcox transformation. so i'm stuck.

Second, i want to know how to interpret P values obtained from the RUNS TEST.

thanx.

I am regressing a response variable on 11 regressors;three of which are categorical.(My data points are 50)

MINITAB's best subset gives me a model with good P and R squared values but the residual plots don't look good.

I figured i could take care of this with the boxcox transformation but i don't know what "group size" means in MINITAB'S boxcox transformation. so i'm stuck.

Second, i want to know how to interpret P values obtained from the RUNS TEST.

thanx.

Please attach example Minitab reports, and we will try to help.

Typically, a p-value < 0.05 means that you will reject the null hypothesis for whatever test that you are running. A p-value > 0.05 means that you accept the null hypothesis.

Thanx for replying to my thread.

When i regressed the response variable on all predictors i got this output from MINITAB......

The regression equation is
di = 89.8 + 1.17 actual - 1.07 max - 0.0267 sales + 0.0188 assets + 0.0084 age
+ 0.00020 prmarg + 0.0149 roce + 0.0345 current - 0.278 listing
+ 0.41 audit - 0.357 Cong + 1.16 Manufacture


36 cases used, 14 cases contain missing values


Predictor -------P-Value
Constant 0.000
actual 0.000
max 0.000
sales 0.384
assets 0.339
age 0.531
prmarg 0.872
roce 0.750
current 0.153
listing 0.669
audit 0.729
Cong 0.676
Manufacture 0.043


S = 1.26391 R-Sq = 97.8% R-Sq(adj) = 96.7%


Analysis of Variance

Source DF SS MS F P
Regression 12 1650.82 137.57 86.12 0.000
Residual Error 23 36.74 1.60
Total 35 1687.56


/////////////////////////////////////////////////////////////
Question #1.

The response variable, disclosure index(di), is directly calculated from the ratio of "actual" to "max".............
ie di= actual/ max for each company.
hence, their very significant p-values....
I want to know if it makes sense to exclude ONE or BOTH of these predictors ( ie "actual" and "max") in my regression.


Question #2.
Because MINITAB's residual plots for this regression showed that the residuals are not normal, the residual to fitted value plot showed unequal variance, the residual vrs order of the data did not look good and most of the p-values in the regressonal model are not significant........
I decided to transform some variables but i don't wont to use trial and error; however, i don't really understand how to use boxcox transformation in MINITAB 14.

//////////////////////////////////////////////////////

Can you attach a file with the raw data? The plotted shape of the residuals provide clues to the problem. Many times it is caused by imposing the incorrect distribution on the data. This may be corrected by using transforms or nonlinear regressions.

Actual, Max and Manufacture are the factors that should be left in the regression. All other factors with p-values > 0.05 should be removed.

Thanx.
Please find attached the data. I uploaded an Excel file because this thread does not support .MTB file downloads.

When i regressed the response variable on all predictors i got this output from MINITAB......

The regression equation is
di = 89.8 + 1.17 actual - 1.07 max - 0.0267 sales + 0.0188 assets + 0.0084 age
+ 0.00020 prmarg + 0.0149 roce + 0.0345 current - 0.278 listing
+ 0.41 audit - 0.357 Cong + 1.16 Manufacture


36 cases used, 14 cases contain missing values


Predictor -------P-Value
Constant 0.000
actual 0.000
max 0.000
sales 0.384
assets 0.339
age 0.531
prmarg 0.872
roce 0.750
current 0.153
listing 0.669
audit 0.729
Cong 0.676
Manufacture 0.043


S = 1.26391 R-Sq = 97.8% R-Sq(adj) = 96.7%


Analysis of Variance

Source DF SS MS F P
Regression 12 1650.82 137.57 86.12 0.000
Residual Error 23 36.74 1.60
Total 35 1687.56


/////////////////////////////////////////////////////////////
Question #1.

The response variable, disclosure index(di), is directly calculated from the ratio of "actual" to "max".............
ie di= actual/ max for each company.
hence, their very significant p-values....
I want to know if it makes sense to exclude ONE or BOTH of these predictors ( ie "actual" and "max") in my regression.


Question #2.
Because MINITAB's residual plots for this regression showed that the residuals are not normal, the residual to fitted value plot showed unequal variance, the residual vrs order of the data did not look good and most of the p-values in the regressonal model are not significant........
I decided to transform some variables but i don't wont to use trial and error; however, i don't really understand how to use boxcox transformation in MINITAB 14.

//////////////////////////////////////////////////////

I re-read your post, and pick up on something that I missed previously.

You stated that di = actual/max.

I think that you may want to regress your predictors against actual, then repeat against max, leaving the di and actual/max out of the equation.

You already have a model for di, so as you stated actual and max will always correlate strongly.

I am a bit confused here.
I have, therefore, uploaded a full description of the variables I am working with and the regressional model i am building.

Kindly take a look at the attached pdf file and let me know what you think of the problem.
thank you