MSA on CMM Hard Learned Lessons - Look Before you Leap

[font=Times New (W1)][font=Times New (W1)]Look Before you Leap - MSA on CMM - Hard Learned Lessons[/font][/font]
[font=Times New (W1)][/font]

[font=Times New (W1)] I wish I had read and searched the Cove first. [/font]


[font=Times New (W1)]We just did a big 10-3-3 study on our co ordinate measuring machine. We were surprised to see some really terrible results. [/font]

[font=Times New (W1)]So now, after the fact I have re-read the MSA manual (it almost killed me) and have searched a big chunk of this thread on the Cove.[/font]

[font=Times New (W1)]Here is what I learned, followed by 2 questions for the experts.[/font]
[font=Times New (W1)][/font]

[font=Times New (W1)]1)[/font] [font=Times New (W1)]The study must be observed by a knowledgeable person. We ran the study across 3 shifts and could not explain the fliers we saw on the Range chart. It now looks like we had a CMM crash without a re calibration on nights. We’re going to watch and make notes next time.[/font]


[font=Times New (W1)]2)[/font] [font=Times New (W1)]We had our CMM print outs rounding off to the same decimal place as the specification. Next run we will report one extra decimal place. I believe that this makes the CMM look worse than it is. I now see many, many threads about this.[/font]

[font=Times New (W1)]3)[/font] [font=Times New (W1)]We grabbed 10 parts in a row off a current production run. We did not want to make special parts for the study. There was essentially no part to part variation. The parts did not represent the normal operating range of the process (6 sigma). I believe that this inflates the R&R portion. Next run we will get parts that span the normal process range. The average chart did not show more than half of the averages out of control, and Minitab showed only 1 category. I now see many, many threads about this.[/font]


[font=Times New (W1)]It appears from my “after the fact” readings in this forum that these are common and often repeated mistakes.[/font]


[font=Times New (W1)]So, on to my questions. If I am now going to say the study was done incorrectly, I will need good arguments to use.[/font]


[font=Times New (W1)]1)[/font] [font=Times New (W1)]Can someone provide a non mathematical discussion about measuring and reporting 1 extra decimal after the tolerance. We were taught in science class to round off to the tolerance.[/font]


[font=Times New (W1)]1.1) Does measuring and reporting 1 extra decimal after the tolerance apply to control charts and capability studies as well?[/font]


[font=Times New (W1)]2)[/font] [font=Times New (W1)]Can someone provide a non mathematical discussion concerning the impact of using parts that do not span the operating range of the process? Why does it make the R&R look so poor? The best I can do is to say that the error has to go somewhere, so if there is little variation in the parts it will all go to the R&R portions.[/font]


[font=Times New (W1)]Thanks in advance![/font]

I am afraid that I cannot answer but here is a link to the rounding off issue
http://www.variation.com/techlib/vr-1.html

Hope it helps

The result of the R&R is essentially the variation in measurement results compared to the variation in the process. Using parts showing less than the real process variation thus inflates the result.

Hi,

Just want to know more on how do you go GR&R on CMM. Do you perform GR&R separately on linear -x, -y, -z axis and for ID and OD measurement?

b.rgds

This sounds a bit of a chicken and egg situation.

If I need to use parts that span the process then I need to measure them to find them, if my measurement process is not capable how do I know I am truly using parts across the real process spread ?

Chicken and egg - that is indeed the problem. If you don't know your historical process standard deviation, which is almost always the case, the best bet is to collect samples over a "reasonable" time period. The question then becomes what is reasonable? If it's truly a blank slate it's hard to decide. If you do know something about how the process behaves you can guess.

Or you can use the tolerance width rather than the process variation to calculate R&R I believe ?

What are the pro's and con's of this ??

My approach has always been to use parts spanning the entire tolerance range and then some - including some parts slightly outside of tolerance. If the measurement system is expected to detect out of tolerance conditions, we should include them in the study too.

Absolutely acceptable, depending on what you want the gage to do. Using your tolerance range, if the gage passes it would be acceptable to use for accept/reject decisions but not as a tool to improve the process. If you want to use the gage to study the process, it needs to be able to see the actual process variation. So using the process range if the gage passes it can be used for statistical studies, process improvement, etc.

howste said:

My approach has always been to use parts spanning the entire tolerance range and then some - including some parts slightly outside of tolerance. If the measurement system is expected to detect out of tolerance conditions, we should include them in the study too.

Click to expand...

This is acceptable if your intent is to assess the CMM's acceptability to make Pass/Fail inspection decisions using the P/T Ratio (% Tolerance).

If your intent is to assess the CMM's acceptability for process control or capability studies, then you must use %GRR and the parts must reflect the actual process variation, NOT the full tolerance. Using the full tolerance will inflate the process variation (PV) and artificially reduce the %GRR.

Also, note that the study should include removing and reloading the product in permanet fixtures between repeats. If temporary fixturing is used instead of permanent fixturing, it should also include tearing down and rebuilding the temporary fixture.