Characterizing Accuracy

[kite005]

When someone asks you guys about accuracy of your CMM how you characterize that? I tried taking a standard (gage block) and doing an MSA study on it and just reported those results. Does that sound legit to you guys?

[MrComment]

Depends on what you measure?

If you measure gage blocks for a living you might have a case.

Otherwise you need to do a gage study on what you intend to measure.

[kmcmm]

The accuracy should be defined on your last calibration report.

[kite005]

ok I mean when someone asks you generally "Whats the accuracy of this machine?". The cal report has x,y,z, linear accuracy and some uncertainty numbers. You would give them those numbers? Maybe I should try to understand uncertainty better.

[Hilton Roberts]

When someone asks you guys about accuracy of your CMM how you characterize that? I tried taking a standard (gage block) and doing an MSA study on it and just reported those results. Does that sound legit to you guys?

That is probably not enough. The manufacturer of your machine has a pretty good handle on the overall accuracy of your machine and that should probably be your starting point.

How did you do your study on a gage block? Are you comparing the measured size to the stated/calibrated size?

You generally use all the attributes of your cmm when you measure parts and I would think you need some kind of part that represents the kind of features you will find on the parts you produce and measure that part over time and analyze those results.

I am not sure you are getting enough information.

If you are actually using your machine the way you would use it to measure a gage block, then maybe what you have is a good start. If the parts you produce must be inspected with multiple tools/probe angles, then you probably need more.

H

[Hilton Roberts]

ok I mean when someone asks you generally "Whats the accuracy of this machine?". The cal report has x,y,z, linear accuracy and some uncertainty numbers. You would give them those numbers? Maybe I should try to understand uncertainty better.

I am trying to figure out how to apply an uncertainty factor to my measurements. I am having a time ( ) doing it. The folks who can ( and do ) develop the uncertainty statements do not measure parts. Those of us who measure parts do not fully understand ( I don't anyway ) the theory of the uncertainty statement.

The theorists, if they have their way, will have us so boggled down with expanded uncertainty statements, that we will not be able to measure parts.

There must me a middle ground somewhere.

In general, the accuracy of your cmm is stated on the calibration report and that number is generally considered valid within the uncertainty statement that should also be a part of the calibration report.

The real problem occurs when we set up to measure a "tombstone" or round robin part. We get numbers that describe the deviations from the "accepted" or true size. The accepted numbers have an uncertainty associated with them also.

What it boils down to is the reported numbers from your "calibrated" tombstone are uncertain by some amount. When you measure that tombstone with your cmm, you also have an uncertainty with your measurement process.

The question is how to accomodate all the uncertainties and coverage factors so we can have a realistic view of how well we measure.

It is not easy ( for me anyway ) and we need a practical way to determine and correct for things ( variables) that affect our measurements.

I have not been able to find a really good way to do it.

I personally check my calibration sphere with 5 different probe angles and calculate the ranges and standard deviations of the X,Y Z and D values and plug them into control charts in Minitab. For each 5 X,Y, Z and D values, I get enough data for one plotted point of X,Y,Z and D.

I then RSS the errors of X,Y,Z and D and use that number as in indicator of when my machine changes.

I am aware there is no uncertainty and coverage factor with this method but what it does is tell me when my machine changes. That is what is important to me.

For example if the manufacturer tells me, after calibration, my machine is within +/- .00025 inch, and the numbers with my collected program come in at .00019 in. , I feel free to measure.

That .00019 in. becomes my baseline to check for change.

If I run another 25 point baseline in a year and I still get that .00019 in. number on my test, I am going to extend my machine and not recalibrate.

While my numbers are NOT definitive statistically, they WILL show a change if my machine changes.

HTH

H

[kite005]

I don't work in a production place, it's all r&d. So I might do a capability study on something and never see those parts again. I get all kinds of stuff blow molded, injection molded, machined, cast, fishing line, tissue samples, foam stuff, just about anything you can think of. Not all goes on the CMM though.


Also I get the occasional tour and they might be high school students, or suppliers, it might be new engineers or a group from China that does similar stuff that we do here, it could be anyone and they might know next to nothing about this (like me) or they might know a lot. They probaly spend 5-10 minutes in here and I give them an overview of the lab.

So I thought, even though it wouldn't be a real MSA study I could use an excel MSA worksheet and run a little gage block program 90 times plug the results into the spreadsheet and use the repeatability reproducability numbers and then average out the 90 readings and just kind of expain what I did and what the results were. I think there must be a simple way to say it like .0001 over a 20" distance or something like that.