Operating Range & Measurement System Discrimination:How to select parts for R&R test?

When setting up to do Gage R&R Avg. and range method the MSA manual states "the sample parts must be taken from the process and represent its entire operating range". Does this mean the part tolerance or the the variation shown in your process? This can be quite different and greatly affect your results in regards to data catagories and discrimination.
At our plant we do small production runs that have good CPK's (1.33 to 2.0). In the past the process was altered to produce parts that spread across the tolerance zone. These parts were used to do the R&R studies. They even went so far as to take scrap parts and include them in the samples. I feel this skews your results and makes you think that your gage discriminates good, when actually it may be unacceptable for analyzing your process. I would think you would want to pull your sample parts from your process without influencing it. You would use linearity studies to (see page 18 of MSA) to insure gage works adequately across tolerance range.

[Dawn]

To "represent the gages entire operating range, in my opinion means to use measurements during the study which cover the entire range of the gage. Example: A 0-1" micrometer's operating range is 0-1". Measurements need to be taken at different places in between. If you are only taking measurements from .100 - .200 you are not using the full range of the gage. Any comments?

[Marc]

It says to take parts which represent the entire operating range? Yes - this is typical for calibrating any measurement device. However, if you look at page 45 under Conducting The Study (average and range method), line item 1, it says: "...that represents the actual or expected range of process variation..." From this we know they are not interested in the range of the gage, but rather they are interested in the range of the process variation. It is, in my experience, correct to take samples to represent being just outside of the tolerance at each end as well.

I see no reason in statistics why the parts taken at the tails (out of tolerance) would skew the results. But - I'm not the expert, either.

Comments from others?

NOTE: If we look at the previous method (range method), it says that method is not convertable to R&R, so I'm assuming you are talking about the average and range method.

[Marc]

You might want to take a read through: www.qs9000.com/ubb/Forum4/HTML/000005.html
and www.qs9000.com/ubb/Forum4/HTML/000079.html
and www.qs9000.com/ubb/Forum4/HTML/000165.html
and www.qs9000.com/ubb/Forum4/HTML/000143.html

In an GR&R (AVG & Range method) you are evaluating the gages in its regards to a specific process, so the sample parts should
be represent the entire range of this process, not the gage's full operating range.
This is evaluated in the calibration and linearity tests.
My concern is if you manipulate a process with a high Cpk value to get parts across the whole tolerance zone. When you evaluate your studies for the capability to discriminate, it may look good (high number of avg. outside control limits) which would say it discriminates well for product acceptance. But it may not have adequate discrimination for process analysis (because your control limits should stay the same but your part-to-part variation would be smaller). This may also have the same type of affect on your % varition because it widens part variation in your sample. It would be virtually the same as comparing to the tolerance instead of process varition for your percentages.
If you have a process with .010" tolerance
range and a process varition of .005" I would want to take my sample from the normal process (which should represent the .005" range). That way if my gage passes evaluation on the designated statistical properties, I know it would be acceptable for product acceptance and product analysis because it passed against the smaller of the two ranges.
If I have concerns about it operating differently at the .010" area, my linearity studies should answer this. Anyone have any feedback?

Originally posted by BHarder:
"the sample parts must be taken from the process and represent its entire operating range":

Let's assume the R&R is for Micrometers:
1. Select your samples that have been produced from a standard process run.
2. If your process can produce parts that would utilize micrometers of different sizes (1", 2", 3"), ensure you perform an R&R with each type of micrometer.
3. Prior to each of these R&R studies, ensure the operators verify the micrometer (ensuring that the entire range is checked).

Hopefully this is helpful

[This message has been edited by Marc Smith (edited 26 April 2000).]