Operating Range - Definition for Gage R&R Study - Micrometers measuring thickness

[chalapathi]

R&R Studies for Product Families

I am confused. Do you mean mixing parts with different nominals in one GRR study? If so, you are artificially increasing the Part Variation. This is a sure way of getting good (low) GRR numbers!

This is a practical issue and needs discussion between different stakeholders - customer, supplier and auditing agency. Consider an example, in a steel company, they are checking sheet thickness and there are more than 50 variety of products (different thickness and chemestry) are checked. If they have to do R&R studies for each and every product. Practically it is not feasible and also does not add any value.

The practical solution to this problem is taking product families (low, medium and high thickness groups) and do three studies. In this case, we are grouping similar products (the part variation will not increase drastically).

The MSA manual is not giving any guidelines on how to practically apply R&R studies. They have just described the statistical technique and we need to make guidelines, which are acceptable to all the stakeholders. The manual, many places suggests that we need to take customer into confidence and use these methods.

In my experience, I have seen that despite such grouping many Measurement methods have shown abnormal R&R values and we have improved the measurement system. These studies have served the purpose.

I invite stakeholders to respond to this important aspect.

 

[Atul Khandekar]

In order to reduce the number of GRR studies to a practicable level, I have no problem with your suggestion of grouping the products. And I may have indeed misunderstood the point of your earlier post (hence the confusion). You can certainly group products together, say, based on their thicknesses, provided that the tolerances and process variations are not significantly different and then do a GRR study on any one representative part from each group. My only contention is that you cannot mix samples with different nominal values in one GRR study. That is if you have 5 part numbers in the 'Low thickness' family, you cannot take two parts from each to make the 10 samples for a GRR study.

 

[chalapathi]

In order to reduce the number of GRR studies to a practicable level, I have no problem with your suggestion of grouping the products. And I may have indeed misunderstood the point of your earlier post (hence the confusion). You can certainly group products together, say, based on their thicknesses, provided that the tolerances and process variations are not significantly different and then do a GRR study on any one representative part from each group. My only contention is that you cannot mix samples with different nominal values in one GRR study. That is if you have 5 part numbers in the 'Low thickness' family, you cannot take two parts from each to make the 10 samples for a GRR study.

We can group different nominal values also. In R&R studies the focus is on Measurement Capability and it is perfectly OK. To take care of your concern, we can see the ANOVA table (P value) This will tell us clearly the sources of variation which are significant.
Nominal value is important for process control and in R&R we are not interested about the process control or capability. We are interested in finding the reliability of the Measurement system.
In six sigma, I have mentored few hundred projects and my experience is that we can group few products and conduct the R&R study. It is practical and reliable. In many cases, we could actually improve the R&R and conformed by repeating the study second time.

 

[Jim Wynne]

We can group different nominal values also. In R&R studies the focus is on Measurement Capability and it is perfectly OK. To take care of your concern, we can see the ANOVA table (P value) This will tell us clearly the sources of variation which are significant.
Nominal value is important for process control and in R&R we are not interested about the process control or capability. We are interested in finding the reliability of the Measurement system.
In six sigma, I have mentored few hundred projects and my experience is that we can group few products and conduct the R&R study. It is practical and reliable. In many cases, we could actually improve the R&R and conformed by repeating the study second time.

You are correct insofar as being free to construct a study any way you want to do it, so long as everyone agrees on the method. But just because everyone agrees on the method doesn't mean that the results will be accurate or useful in a statistical sense. If you're not interested in process control, why bother doing GR&R at all? The methods that have been developed for GR&R (and other statistical studies) make certain assumptions about the conduct of the studies, and results are reliable only when those assumptions are true. If you disregard the standard assumptions and do whatever you want to do for the sake of convenience, you shouldn't be claiming to be doing GR&R in the traditional sense, because you're not.

 

[Miner]

We can group different nominal values also. In R&R studies the focus is on Measurement Capability and it is perfectly OK. To take care of your concern, we can see the ANOVA table (P value) This will tell us clearly the sources of variation which are significant.
Nominal value is important for process control and in R&R we are not interested about the process control or capability. We are interested in finding the reliability of the Measurement system.
In six sigma, I have mentored few hundred projects and my experience is that we can group few products and conduct the R&R study. It is practical and reliable. In many cases, we could actually improve the R&R and conformed by repeating the study second time.

This could be done, but not using the standardized GRR format. You would have to add the nominal dimension as an additional factor and analyze using ANOVA. If there is no interaction between the nominal level and other factors, you then could validate the assumption that you can group the nominals under a single gage. If an interaction is present, you cannot make this assumption.

I would only recommend this for a statistically experienced individual.

 

[Tim Folkerts]

2. Group the products into 2/3 families and do it for the families. In case of steel sheets - Low thickness and high thickness.

I am confused. Do you mean mixing parts with different nominals in one GRR study? If so, you are artificially increasing the Part Variation. This is a sure way of getting good (low) GRR numbers!

I think that the intent was to do GR&R on a few representative types of parts. For example, if you make sheet metal every 1/32 of an inch from 1/16 - 1/2, that would be 30 different products. You could do GR&R on all 30, but if the same gage is good for 1/16 (=2/32) and 1/8 (= 4/32) , it is presumably good for 3/32 (in the middle of the other two).

My interpretation of chalapathi's suggestion is to pick one thickness near 1/32, one near 1/2, and perhaps one more near 1/4. If the GR&R is good for each of these separately, then it would be fairly safe to believe the same gage is also good for the other thicknesses in between. (Of course, if it is really critical, then do more parts - possibly all parts if the situation warrents.)

Tim F