Scanning speeds

[[Pa...]]

Hey everyone.

Piece by piece, I've been disproving anecdotal "rules" about using a CMM that have been passed down through programmers over the last 20 years.  As an aside, I'm really concerned about the way so many people just believe what the last guy said.

What I have not yet found time to test is the scanning speed limit.

If you had a workpiece made of 4140 steel with 4 concentric ground diameters (our grinders are damn good) at Ø18.800", 18.400", 17.900", and 17.350", and each has a diameter tolerance on size of ±.0002", what scanning speed would you use to evaluate the size?  This is a setting standard used to "zero" a 2-point measurement at the grinder.  Pic attached.

If I had the same piece but all cylinders were less than an inch in diameter, what speed would you use here?

If it matters, I've got the workpiece oriented so the normal vector of all cylinders is along the Z axis of the CMM.

I am not using Navigator.

The high limit scanning speeds used in my shop on an operation like this one is 0.590"/s.  As I type this I'm scanning at 1.0"/s just as a test.  Since this piece is a one-off and was ground on my day off, I don't have the luxury of looking for a method that best matches the 2-point measurement from the grind operator.  I don't know what his measurements were.

Feel free to comment on evaluation settings, stylus selection, etc as well.  I like what I have but am always happy to be proven wrong, especially if it makes me look good 😎

 

[[Ma...]]

Well it's depending on what head do you have.

I have tested same path with different speeds and forces. It is pretty obvious that for features of nice form you can go faster with same measuring force, but if you have bumps, then it can airscan. So if you know your surface, you can measure faster, but adjust mesuring force.

For our active head VAST XT i usually measure at 200mN and speeds are up to 10mm/s - on not so precise planes i go up to 15mm/s, but remember - you can have set 0,5s on begining of scan or even before end of scan of eliminating points - so you don't have data in that time - they are filtered and can not be brought back.
Thats why i keep curves without that option.

But doing your research is a best way to know what you can go.

 

[[Pa...]]

Shoot.  I knew I forgot something.

VAST Gold in all cases.

I always test.  I'm just curious what others have learned.

Also curious who else works weekends 😜

[[Ma...]]

Heh i am not working on weekends, but in some sense i work 7d/week.

Now i am figuring out how to construct a measuring device of bended shafts.
Problem: whenever is cylinder bended sideways - result is skewed.
Next problem: measure it in 90° to cover both forging directions, but there is residual flash.

I got into this stage ( i hope i'll find a missing formula to get correct radius ) ( ignore tolerances - not used here )
Key is to have same height in every angle ( constructed up to 2mm bended from axis )

[[Da...]]

I think and work in mm/s, but I will try my best and convert to inches.

The scanning sensors I know usually are capable of registering around 1000 points per seconds. So, there's one limiting factor right from the start. The slower the scanning speed and the smaller the probe diameter, the better the stylus will register form error. Then again, measuring is not all about form error. Most of the time you just want reliable information about a diameter dimension without the focus on form error. Depending on your needs, you decide to take less points and a higher scanning speed.

Up until now, it sounds all pretty random, but when you think about reliable measuring, you always think about filtering and outlier elimination. And effectively, you also think about memory usage. If you always took the maximum amount of points you could, and you measure a lot of parts, you will find that measuring data take up a whole lot of space. Why be wasteful? Identify your need on how important the identification of form error is and choose your filters accordingly. There are tables of experience recommending filter settings, and once you know what filter you want, select your point count accordingly.

A filter setting is defined by a cut-off wavelength, and every wavelength is mathematically defined by at least seven points. For circles, a filter setting is usually not given as a cut-off wavelength, but by the number of waves (or "undulation") per rotation, and a good example for a commonly used filter is 50 waves per rotation (or "undulations per rotation" or "UPR"). In this case, you'd need at least 50 * 7 = 350 points for the filter to work. If there are less points, the filter cannot even work. More points give you a little surplus for outlier elimination due to influences like dirt and such.

So much for the slowest needed scanning speeds and least point counts. But how fast can you scan if form error is a secondary concern? What if you have ten thousand parts to measure and you need only a diameter?

Well, if you go over certain speeds, you will find you will get inconsistent results and that is because with high speeds, the scanning sensor if influenced by mechanical effects such as centrifugal forces and momentum. Scanning sensors are intricate mechanical parts with moving parts and mechanical or electro-mechanical springs and will not behave in a linear way when moving at high speeds. But you can check a known diameter first at a reliable low speed and then at a high speed and use the Calypso gauge calibration to train Calypso to measure a very limited diameter zone to be measured reliably at that one high speed. If you use a different speed or a different diameter or you change from an inner diameter to an outer diameter, you will have to use different gauge calibrations in all those cases, but a least you can do it.

For example, I often measure parts with a diameter of 600 mm (23.6 inches) with a speed of 150 mm/s (almost 6 inches/s). With the use of gauge calibration, it works perfectly. I know that I will miss the finer details of form error, but I only care about the diameter.

And I also like your approach with the testing. No one keeps you from testing things out and learn from it. It's important to prove processes to yourself first.

[[Pa...]]

I can do both mm and inches.  All good.

Also, I'm not sure it's totally legit, but for the sake of consistency I like to collect 21 points per undulation.  I measure exclusively fine finish steel and carbide, so I like a low pass 50upr filter.

I haven't gone about calculating further, and I haven't found the easy button for wavelength conversion, but check this out:

If I multiply my diameter by .003 and use the result as my step width, I get 21 points per undulation given 50 upr.  Give or take a few.  Neato!

Gage correlation isn't really an option for me.  A typical inspection often collects some 20-200 diameters, their associated lengths and other details.  Diameters range from 1.5mm to 600mm; lengths are usually between 1 and 3 meters.  Tolerances are always under ±.1mm but sometimes get down to ±.005mm.  I couldn't possibly fit enough gages in the measuring volume, let alone mount each in the same orientation as the associated feature!

My VAST Gold visibly scans air even on fine finishes when I get to 60mm/s +.  My test above where I went from 15mms to 25mms scanning did affect results; I saw changes in resultant data as large as .004mm, which is half the tolerance.  But I'm unable to say which is the "incorrect" measurement and thereby cannot determine if 25mms was "too fast".  In both cases the points looked good and repeated fairly well.

Anyway, I'm only here to collect anecdotes and because I can't discuss anything about the work I love with anyone in real life so you guys are it.

 

Edited November 30, 2025
addition

[[Ja...]]

I let each individual CMM determine its’ own maximum scanning speed as the combination controllers and heads vary from machine to machine, and from feature to feature. As the old saying goes, there’s no “one size fits all” answer. For each feature in the program I simply set the speed in the strategy to an extremely high number, something like 9”/sec. I then set the number of points I want taken on that particular feature and the system will automatically determine the point width and maximum speed that it can collect the data with. I know some people don’t trust this method, but it works perfectly if you have all of your filtering set correctly. It takes longer to program doing this, but your program will run as fast as it can for the amount of accuracy you’re after.

So how do I determine how many points I want? Over the years I’ve developed a system that seems to work rather well. For circles (I use 365 degrees for a full circle), the formulas are a little different than for lines, planes, etc. For circles, the first thing I need to determine is the UPR. This is easily done by plugging the numbers to the ISO 12181-2 formula. That formula is UPR = ((PI x dia)/0.8) Keep in mind the formula is for metric use so you’ll need to convert the diameter to mm if you’re working in inches. In the case of the 18.800” circle, the UPR would be 1875. For setting pre-filters, I don’t have anything scientific for determining the numbers but if you’re interested in my formulas let me know. For this measurement I’d set them at 4 and 2946. For tight tolerance parts I get fairly aggressive on the sigma value because I don’t want any electrical noise affecting my measurements. In your case where the entire tolerance window is 0.0004” wide, I would use a sigma value of 2.40. This would eliminate roughly 1.6% of the outer data on a bell curve – Not much and it helps to ensure we’re looking only at data and not noise. Next, I would delete only the outliers and not any adjacent points. I’d set iterations to 2.

Getting back to the number of points – Now that I know how many UPRs I need it’s just a matter of multiplying that value times the number of points we want per UPR. Calypso mandates a minimum of 7 points. Keep in mind that’s after filtering. So I would recommend a bare minimum of 10 points per UPR. You can go higher if you like, but I would recommend not going real high as you’ll see it’ll increase the amount of processing time for each feature as the features get larger. For the 18.800” circle then, I’d want to take a minimum 18,750 points. When you plug that number into the strategy, you’ll see Calypso change the speed and step width. It knows what it can handle and what it can’t.

In your case where you’re trying to get collaboration with a 2-point mic, you’d probably be better off using a LSQ algorithm vs Minimum Circumscribed as the mic will always read slightly smaller than a CMM using Minimum Circumscribed, even though the CMM will give you a more accurate answer as far as what the mating part will see for a size.

Finally, I’d run the program 5 – 6 times initially on the same part (remove it from the fixturing each time) to see how it repeats. If I can’t get it to repeat within the 10% rule, how do I know which reading is correct? If any. If it doesn’t repeat, you’ll need to do some tweaking and retry the repeatability. On lengthy programs doing the mini-repeatability study can be quite time consuming, but if I can’t trust my measurements…..

Be aware if you’re measuring an ID to refer to ISO 12180-2 to make sure you’re using an appropriate size tip.