CMM Arm Accuracy

[geesamand]

Hello all.

We have a few CMM arms from a major manufacturer in 4', 6', and 8' lengths.

We use them primarily to measure gear housings that range from a refrigerator to a lunchbox in size. Due to the way the bores are machined, the arm(s) must articulate all the way around one side of the part to the other to reach the important features. We typically compare the XY positions of the bores to determine the alignment of shafts/gearing. We need to know position and feature size within .001". The arms have accuracy ratings of under .001", depending on the arm size.

It seems that the more the arm needs to articulate, the more erroneous the feature position turns out to be. So I devised a simple experiment using a plate with a single precision through bore. (Picture an 8x10x.5" plate with a 6" bore, mounted vertically on the table) We measured the XY position of the bore reaching from both sides of the plate to touching on the same points. The location varied by .005"!. This discrepancy appeared when the bore position was calculated from both a cylindrical feature or circular feature.

What is the real accuracy of these arms? Is this type of thing normal? Our arm vendor has been happy to exchange our arms for recalibration, but unhelpful about discussing the true accuracy of their arms. I am still trying to find what method their accuracy claim is based on to see if it's realistic and/or applies to our usage.

I would like to hear your experiences and how you manage and measure the real-life accuracy of your CMM arms.

David

 

[MikeHalloran]

A few years ago, my then boss bought a toy-like arm that was supposed to be good to .009", because it was within our budget. It wasn't even that good, and displayed the same sensitivity to articulation as you report, only much worse. It's gathering dust in a closet somewhere; nobody would bother to use it twice, because you couldn't get the same measurements on the same object twice, even without disturbing the setup.

Subsequent to that, I interviewed at one of the major arm makers, and came to appreciate the challenge of inferring linear dimensions by making angular measurements to extreme resolution. During the interview, I think I got close to conjecturing the operating principles they were using, but they were very reluctant to reveal anything, and I didn't get the job. Their loss.

They did show me their calibration facility, comprising long stacks of gage blocks on rotatable arms, big surface plates, pretty much everything you'd need. The facility was clean and impressive and all that.

At that time, they were claiming resolutions around .001", and that seemed credible based on the little I had seen and/or guessed. Repeatability with angulation was clearly an issue, sort of skirted by the way they stated their accuracy claims.

Since then, they have begun using lasers within their arms, and also claim resolution at least ten times better than when I spoke to them.

They also have one or two major competitors who still claim .001" accuracy, who clearly have not made the leap to lasers, and who also seem careful to sell in markets where precision and repeatability are not such an issue.

.....

Now I work in a place where the standard tolerance is the width of a dull Sharpie, and the measuring instrument is whatever tape rule they had on sale at WalMart when the tech went there to buy his...

Anyway, back to CMM arms...from my perspective, it's a wonder the damn things work at all. I'd humbly suggest that you not be real quick to throw away the big surface plate and the gage blocks...




Mike Halloran
Pembroke Pines, FL, USA

 

[geesamand]

Well if I were the one buying them I would have approached the transition away from our large B&S Validator much differently.

What you mentioned: "Repeatability with angulation was clearly an issue, sort of skirted by the way they stated their accuracy claims." seems to be the crux of our problems. Our technical committee required better than .001" absolute accuracy. The sales people came in and parked their arm right on the granite table of a Validator and received questions about accuracy. The arm had a spec with the word "accuracy" and less than .001". It also had a spec stating precision to be less than .001".

To me, if the accuracy and precision are sub-.001", then I should be able to consistently locate the coordinate location of a feature to .001", and the use of extra points in defining the feature will ensure that outcome. As I mentioned before, the reality is between .001" and .010", depending on how differently the arm must articulate compared to the origin features.

Hopefully this vendor has not used selective accuracy measurement methods to mislead those of us with real accuracy requirements.

While I have the utmost respect for the technology at work here, and the technical difficulty of the job, I would be disgusted and angered if the capabilities of this machine are really as poor as I've seen at times. SOMEONE knows the answer or at least how to expose it, and I must find it.

 

[MikeHalloran]

Hey, I'm on your side.

It seems to me that you've found a simple, repeatable way to get the answer you seek, you've gotten it, and you don't like it one damn bit. I wouldn't either.

Now, see if you can explain it to your company's lawyer.

That's all I can think of.



Mike Halloran
Pembroke Pines, FL, USA

 

[geesamand]

Sorry, I hope that didn't read like I was snapping at you.

I've read elsewhere that some people have found the accuracy claims (as I evaluate them, robustly) to be off by an order of magnitude. But people don't seem to be too put off by that fact.

I'm thinking now about getting a piece of metal machined in the shape of a prism to very high degree of accuracy with several bores located on the various faces. Then this can be a test piece for any CMM equipment that we as a company use. I can see no better test for accuracy than that which exactly mimics the inspections that we conduct.

Dave

 

[MintJulep]

How does the prism solve your problem though? All it does is confirm that your CMM really isn't accurate.

Could you perhaps apply the techniques used by differential GPS? That is, build some features of KNOWN location into your surface plate. By measuring the locations of those features with your CMM you can establish the error of the CMM in given regions of space, reached by various arm articulations. From there you could interpolate or extrapolate the real positions of features of interest based on CMM reported positions.

Maybe.

 

[geesamand]

The prism does nothing to solve the inaccuracy problem. However it answers the most critical question, which is "how accurate it, really?"

I found that we have a housing of appropriate size which was recently validated with a small Mitutoyo bridge device that has an accuracy of .0002". I will set up and repeat that exact validation with our gear and see 1) how repeatable our results are and 2) how accurate our results are.

David

 

[geesamand]

Just a follow-up. Our sample piece has been done at a couple of off-site CMM vendors. There is finally good agreement between the units.

We just received our 8' arm back from the factory and it is producing results consistent with the bridge CMMs. For bore diameters it's about .001", and bore locations are within .001" also. If we were to invest equal (or even much greater) resources in more accuracy it would cost us more because we would be able to CMM fewer parts.

A couple of lessons learned:
Arm CMMs are definitely fragile and SPAT tests are not sufficient to ensure accuracy. SPAT can catch some problems but not all of them. So we will keep this uber-validated sample piece around and re-run that part whenever we need confidence in the accuracy of a component.

If you are using or interested in an arm style CMM, and you need all of it's claimed accuracy, put your own measures in place to verify that. Have a sample part ready with a complete validation and let the arm repeat that validation letter for letter. Let those results be your basis for accuracy performance.

Finally, after all of this I will disclose that our arm CMM vendor is Faro. Their telephone tech support is rather feeble and does not do well with answering questions that didn't occur in their training program. You have to get their most experienced people to get much done, not surprising for a rapidly-growing company. We still have yet to receive the details of their B89 "subset" accuracy test (our most responsive company reps are now taking up that task, so we shall see). They are funny about talking accuracy in general. So they know their accuracy claim is a bit, uhm, assertive and whenever we poke at accuracy questions they stall and/or get evasive. Fortunately we are getting sufficient real-world accuracy.

 

[larrylcoyle]

David,

Have you thought about running down a former employee for this company and trying to ply the information from him? I know that FARO has been involved in several patent disputes with another company regarding some of their technology. I considered purchasing one several years ago for reverse engineering cylinder heads but even then they had trouble explaining some of my questions to my satisfaction. Personally I am glad I chose to stay away from this technology altogether with the exception of a Microscribe that we use for minor measurements.

Good luck my friend...

Larry

 

[geesamand]

While I am sore about our collective experience with their customer support, I am not disappointed with the technology. I am now dispositioning inspections that are consistent within .001", which is in the range of accuracy advertised in their literature. So it is delivering the accuracy we were promised. The speed and TCO remains very good, and fortunately we have 3 arms so that if one goes a little off our production is not severely affected.

We have been invited to visit their factory and observe their implementation of B89.14.22. This test requires the arm to measure the length of artifacts throughout the arm's reaching volume. There is latitude in the form of the test artifacts in the standard - I am keenly interested to see if their implementation of the test artifacts might make their accuracy results more optimistic than the B89 suggested artifacts.

We are definitely using our own accuracy tests on a regular basis to ensure the arm is performing at it's peak. This is where end-users have to look out for themselves.

Dave