What's new
What's new

Established Methods For Sub 5 Micron Positioning Accuracy?

apt403

Cast Iron
Joined
May 24, 2015
Location
Yelm, Washington
Been a minute!

I've been handed a linear motion project that's got me down this rabbit hole. My application is far less intensive, but these questions remain.

What are the established methods to achieve high-ish (sub 5mu/2 tenths) positioning accuracy and repeatability?

Figure, industrial servo w/ a ~6000 count encoder has an angular resolution of .06° (1.05x10^-3 rad). Through a 5mm pitch screw, that's .83 micron (30ish millionths) linear positioning resolution. Let's assume it's also got that level of accuracy. So, servos would appear to not be the limiting factor.

But, ballscrews aren't nearly as accurate. A C7 class ballscrew has an accuracy rating of +/- 50 micron (.002") over 300mm (12in). Even fancy C0 grade screws are only good for around +/- 4-5 micron (2 tenths) over 300mm. If your application calls for say, 1000mm travel, that cumulative error starts getting pretty bad. Screw mapping obviously sees quite a bit of use, but screws wear. The error map is only good for so long. Especially when you factor in nut preload, etc.

Okay then, the servo encoder is fast and (hopefully) accurate, but can't compensate for the screw. The screw can be mapped, but the map has a life time. Feeding the servo PID loop the output from a linear encoder (glass/magnetic scale) can compensate for lead errors, and something like a non-contact incremental magnetic encoder is a non-wear item. But, while the resolution and repeatability of linear encoders can be fantastic, +/- 10 micron accuracy seems to be a pretty typical figure.

Best thought I've had so far is to map an absolute linear encoder with some arbitrarily accurate measuring system. No wear, and the absolute encoder always knows where it is, so the map can be applied with confidence. It would appear that's how Renishaw is doing it with their RESOLUTE series of absolute encoders, and they're hitting +/- 1 micron (40 millionths) over a meter (3.3ft).

What other solutions exist?

This is all obviously assuming environmental factors (temp/humidity) are accounted for, motor-to screw couplings are sized correctly to minimize torsional deflection under load (perhaps it's warranted to treat the whole system like a big spring?), some method exists to compensate for thermal growth during operation, a sacrificial goat has been offered to the appropriate deity under a full moon, etc, etc.

Appreciate it,

- Apt
 
Laser interferometer in the feedback loop. Renishaw RLE for eggsample.
"low precision" version is 0.02 micrometers over a meter and sub-nanometer resolution if needed.
Luckily you didn't ask for cheap :D
 
We usually used Heidenhain LC glass scales, they are +-3um or +-5um, so not as good as the Renishaw. I think we used to hold drill countersink depth to a few tenths, but that was more of a repeatability test as the depth was dialed in with comp.

When you're working at these levels, everything matters. Straightness and twist of the rails/cars matters a lot.

What type of travel are you looking to get this accuracy over? That changes things a lot.
 
The other alternative, of course, is a stupidly accurate leadscrew. Moore's book "Foundations of Mechanical Accuracy" is a fascinating read and tells you everything you need to know to build a ridiculously accurate machine, just add decades of testing and millions of dollars.
 
AS I learned from replacing my umpty thousand dollar heidenhain absolute encoder, it is only absolute on startup. It ignores the absolute section when it is running.

I think you set the thing up with an accurate encoder, and commission it with a lookup table or however you keep track of inaccuracies.

My machine doesn't have a ballscrew, so all its accuracy is in the encoder.

https://www.heidenhain.com/fileadmi...Tool_Inspection_and_Acceptance_Testing_01.pdf
 
Get rid of the ball screw and find a linear motor. Sodick EDM machines have used them for years. Use laser feedback. You could also use air bearings for less friction if that's a problem.
 
The most foolproof method is linear motors reading off of glass scales, sodick uses this system in their new edm machines, there are no moving parts, so no backlash. It is very limited in torque and susceptible to thrust. Cost is probably prohibitive for a project like yours.

As you already stated, mapped ballscrews are the current standard. If the machine you are looking to create does not see any cutting forces or high loads, these will stay accurate for a long time. I have worked on a 20 year old mitsubishi edm that is still capable of holding plus or minus 2 tenths. The ballscrew has never had to be re-mapped, no glass scales on the machine.
 
First get a handle on abbe error and pitch, yaw and roll.
If you have not explored this you are fighting a loosing battle. The best feedback loop in the world will not help until this is contained.
Have you considered dual glass scales on a axis along with the motor/screw feedback?
Bob
 
To what degree are your environmentals are controlled?

1 degree C increase in temperature will cause encoder to grow by nearly 10um/m. Don't forget about temperature gradients caused by human presence, air flow patterns, heat from motors/electronics etc. So different parts of encoder will have different temperature which is also different from the temperature of whatever you are positioning. A lot of care needs to go into design to account for all of that.

Repeatable and verifiable system accuracy of 1um/m is nearly impossible to achieve outside of a metrology lab environment, regardless of measuring device used. If you look at some of the encoder calibration certificates, they typically show an uncertainty value, something like 200nm +600nm/meter. And that is uncertainty of measurements performed by a world class metrology lab at 20+/-0.01 degree, controlled humidity, pressure, vibration isolation, etc. And using multi million dollar custom designed machines with all possible factors accounted and compensated for.
 
Last edited:
Is this a single axis system? As CarbideBob suggests, your error budget includes many more terms than just the precision and accuracy of your ballscrew and/encoder.

Alexander Slocum’s Precision Machine Design textbook would be very useful to you here. It provides a manual for how to do this and what terms you need to cover. It’s pretty involved but the textbook covers basically the entire trade space. Foundations of Mechanical Accuracy is a very cool read but far less useful for practical engineering problems like this.

A common solution is linear scales closing the position loop with an interferometer to map errors, but without knowing more details it’s hard to say if that is the right answer for your application.
 
Been a minute!

I've been handed a linear motion project that's got me down this rabbit hole. My application is far less intensive, but these questions remain.

What are the established methods to achieve high-ish (sub 5mu/2 tenths) positioning accuracy and repeatability?

Figure, industrial servo w/ a ~6000 count encoder has an angular resolution of .06° (1.05x10^-3 rad). Through a 5mm pitch screw, that's .83 micron (30ish millionths) linear positioning resolution. Let's assume it's also got that level of accuracy. So, servos would appear to not be the limiting factor.

But, ballscrews aren't nearly as accurate. A C7 class ballscrew has an accuracy rating of +/- 50 micron (.002") over 300mm (12in). Even fancy C0 grade screws are only good for around +/- 4-5 micron (2 tenths) over 300mm. If your application calls for say, 1000mm travel, that cumulative error starts getting pretty bad. Screw mapping obviously sees quite a bit of use, but screws wear. The error map is only good for so long. Especially when you factor in nut preload, etc.

Okay then, the servo encoder is fast and (hopefully) accurate, but can't compensate for the screw. The screw can be mapped, but the map has a life time. Feeding the servo PID loop the output from a linear encoder (glass/magnetic scale) can compensate for lead errors, and something like a non-contact incremental magnetic encoder is a non-wear item. But, while the resolution and repeatability of linear encoders can be fantastic, +/- 10 micron accuracy seems to be a pretty typical figure.

Best thought I've had so far is to map an absolute linear encoder with some arbitrarily accurate measuring system. No wear, and the absolute encoder always knows where it is, so the map can be applied with confidence. It would appear that's how Renishaw is doing it with their RESOLUTE series of absolute encoders, and they're hitting +/- 1 micron (40 millionths) over a meter (3.3ft).

What other solutions exist?

This is all obviously assuming environmental factors (temp/humidity) are accounted for, motor-to screw couplings are sized correctly to minimize torsional deflection under load (perhaps it's warranted to treat the whole system like a big spring?), some method exists to compensate for thermal growth during operation, a sacrificial goat has been offered to the appropriate deity under a full moon, etc, etc.

Appreciate it,

- Apt

This is a good description / outline ^^^.

Also wondering about pre-tensioned ballscrews (depending on application).

Also wonder how double nut symmetric pre-tensioning combats wear/ degeneration of pitch error compensation on a ball screw over time ?

Hollow core cooled ball screws ?

Cooling of other friction related elements ?

Linear motors build up a lot of heat + magnetic 'Suck down affect" with ferrous components (like a table). So need to design and deal with that in a compensatory way.

Some linear scales have temperature sensors so you can map out the linear expansion of the scale with the machine. Better if they are compatible.


Makino D 200Z can position to 1 micron nearly volumetrically on a 5 axis machine with regular ball screws but with higher count encoders and DD motors for rotary axes.

Can map out environmental effects but better to control environment.

^^^ Random list of points.

Ohh yeah pre-loaded "spring" on pillow block thrust bearing to maintain tension over thermal expansion changes of your base / frame / machine.

Try Granite / synthetic granite lower coefficient of thermal expansion. + Vibration absorption (if relevant to application). .

Kern on some of their machines switched to Aluminum + granite due to thermal properties.

Laser interferometers are a must to map things out / calibrate for your systematic errors / distortions.

AND as CarbideBob said ---> Straightness + squareness , planarity (depending on application). (as others have mentioned hydrostatic / hydrodynamic systems can "float" straighter/ flatter depending on loading conditions if you need it ? Same with comparatively light loads and air bearings ).

Straightness over what distance and loading can be a big deal if other orthogonal or oblique axes are involved.

Is a breaking system required (also) ?

There are special machines that do nothing else BUT map out ballscrews… I'm thinking Okamoto semi conductor division and related products , I'll dig about time willing see what I can dig up on pre-mapped ball screws.
 
Been a minute!<snip>

Best thought I've had so far is to map an absolute linear encoder with some arbitrarily accurate measuring system. No wear, and the absolute encoder always knows where it is, so the map can be applied with confidence. It would appear that's how Renishaw is doing it with their RESOLUTE series of absolute encoders, and they're hitting +/- 1 micron (40 millionths) over a meter (3.3ft).

<snip>

- Apt

I'm familiar with some of the Ring encoders from Renishaw + Resolute.

For rotary approaches of higher accuracy one can use multiple read heads at 180 degrees from eachother to compensate for bearing runout. High accuracy 5 to 1 arc second instruments like surveying instruments and Astronomy type applications use multiple read heads.

Basic principal of reversal methods...

Sampling techniques and statistical methods also creep in especially for encoding scheme and multiple read heads . (also moiré techniques for some read heads.)

Scratching my head on single linear dimension "reversal methods"... Would be different if there were more variables like height over distance then reversal methods could be applied. (but that's more metrological than "machine") as others have said.

Mathematical techniques that lean towards least squares adjustments and weighted errors and linear regressions rather than just "averaging stuff".

+/- 5 micron over a meter not impossible at all, (as per the title).

but like what @Zinstrumnets said 1 micron over a meter is super difficult without direct interferometric results + control of air currents and environment.

From machines and machining I tend to look for straightness over 100 mm 150 mm maybe 200 mm... (just our applications).
 
An extreme example of what is possible can be seen at the Precitech or the Moore Nanotech websites. The products are lathes with nanometer accuracy. The machines at both companies were designed and built by the same person. The machines use hydrostatic slides with linear motors and interferometer position measurement. The slides are attached to a granite base that is supported by air piston optical bench stands. The work holding spindle is supplied as a air bearing type. Precitech uses a custom designed Unix based control.


The other good example would be the Hembrug hard turning lathe. Hembrug is now owned by the Danobat company. It uses diamond or CBN insert tooling to replace operations that would normally be done by grinding. The lathe uses oil hydrostatic bearings for both the slides and spindle. The control is a Siemens 840D. Hembrug does not provide any details on the scales. My guess is that they are using a magnetic scale built into the linear motor assembly. The positioning accuracy is 1 micron and the repeatability is .1 micron. The Z axis travel is 350 mm.

The design of the hydrostatic ways becomes complicated if the travel is much longer than this. The clearances on the hydrostatic box ways are roughly .0002 inches. If there is any thermal warping of the machine frame the ways will jam. The Landis tool company uses a large number of small hydraulic pistons along the length of the box way to adapt to the thermal warping on their cylindrical grinders. Other companies use hydrostatic flat and V-way designs which are not fully constrained. The carriage can follow the thermally deformed way without jamming. There is a loss of stiffness and accuracy with this method.


There are a number of patents that can be downloaded to provide more insight into the machine design.
 
Don't take all this as bad news
5 microns over a meter is not super high precision or needs crazy methods.
I can do this with plain ole screws, medium class bearings and normal motor feed back with nice couplings.
When you want one or 1/10th of a micron.... different stuff and that is a different build.
Add the errors screw, bearings etc. Assume worse than published (I use 2X), give your motor or servo loop some realistic room to work and go from there.

Not a hard target to hit at all on final position and somewhat like falling off a log. Different if you need to contour.
Bob
 
I find it strange that no one has said a single word about what is being positioned here. If we are talking about making something like a diffraction grating, then only a few ounces of tool and holder need to be moved and positioned. And the forces involved, gravity and resistance to the "cut" are going to be relatively constant. On the other hand, if we are talking about moving a 10,000 pound hunk of metal under a milling spindle, then we have a whole "nother" thing. I don't care what accuracy you achieve on the various axis, any but the stoutest machines is going to bend as it moves around. And that is with perfect accuracy on the axis and 100% temperature control.

In the first case, the diffraction grating, DanielG's "stupidly accurate leadscrew" has been made to work with a relatively light frame. And I am not even talking about a ball screw, just a very, stupidly accurate standard screw. But with that five ton hunk of metal, it is going to be super hard to prevent errors that the screws and their super accurate scales will never even see. And you can put any ball or other type screw in there and any encoders with any level of accuracy. If the machine frame does not stay rigid, all is lost.

Any serious discussion of this must start with more information about just what is to be done. Otherwise, you are just whistling Dixie.
 
I worked in a powertrain plant once.

They used air gaging for measuring stuff that was really small.

I don't know enough about it to know if its capable of doing what you require for gaging position relative to hard stops, but if it is, its just a matter of using that feedback signal to direct your servo to make little incremental moves and get in the "acceptable window".
 
An extreme example of what is possible can be seen at the Precitech or the Moore Nanotech websites. The products are lathes with nanometer accuracy.
There is a difference between resolution and accuracy. There is no such thing as nano-meter accuracy over 1m length. I checked Nanotech specs and way straightness is specified at 1um/0.5m which would be 2um/m. Sounds reasonable for the ways them self, need to add encoder errors, Abbe, thermal, etc

Hembrug does not provide any details on the scales. My guess is that they are using a magnetic scale built into the linear motor assembly

I would rule out magnetic scale at micron level accuracies. You would be happy to get 30um/m out of a typical magnetic scale, plus they have a higher thermal expansion coefficient than glass. If the scale is built into linear slide, very likely it is a non-contact reflective type glass. I am a little skeptical about interpreting 1um spec correctly, as they don't state for what axis and over what length, and it is the same for all machines regardless of size.

System accuracy is never the same as encoder accuracy, so even if encoder is accurate to 1um/meter (e.g. Heidenhain LIF series), the actual positioning accuracy will always be worse then that.
 
<snip for claity>

I am a little skeptical about interpreting 1um spec correctly, as they don't state for what axis and over what length, and it is the same for all machines regardless of size.

System accuracy is never the same as encoder accuracy, so even if encoder is accurate to 1um/meter (e.g. Heidenhain LIF series), the actual positioning accuracy will always be worse then that.



Along those ^^^ lines @apt403

quick one ;-)

With Renishaw you have to dig a lot deeper, they have most of the information published online but if you are going for higher precision and accuracies systems it's best to contact one of their application engineers and then they generally give the under the table documents for "How to make it go lovely " (or our systems really are that accurate but you are going to have to do this, this this this and this and thiiiiis; in terms of actual application realization.).

There are a lot of different error even on the spec sheets that "stack up" like tolerance stack up. Even their stated errors have errors (in their tables). It's all cumulative.

Might be worth cracking open the Renishaw docs/pdfs for people to read.

DMGMori Heidenhain scales and magnescales use a magnetic encoded moiré effect and read resolution of the order of 0.01 micron (ten nano meter ) but with a control system and real stuff to move + just physicality of the scale itself, and numerous thermal compensations including the scales, not just the machine - over one meter +/- two tenths +/- 5 micron is not atypical (as multipostional bidirectional full travel moves and associated tests). Tighter tolerance relative moves can be achieved if it's just unidirectional more localized moves over 100 mm or 200 mm or so (most machinists know that or have practical experience of that. ).

Higher resolution is helpful for natural smoothing and signal processing for things that actually move ~ as they move. It's like the "way of going" within the commanded move and the quality and smoothness and accuracy of that move itself. That's where the "Art" comes in. And for a machine tool executing that "artfully" under brutal cutting conditions and forces in some cases. ~ "In Cut".


https://arnd-sauter.com/wp-content/cnc_datenblatt/Matsuura/VX-1000.pdf

PDF ^^^ for Matsuura 3 axis c frame vertical VX 1000

page 5 they show laser plots for X, Y and Z axes.

so in this case full travel Y axis is the most accurate and stable (seemingly), stays within a couple of micron travel over 400 mm (impressive , does it actually do that shrugging shoulders )… X axis has more variation along it's length but stays within a couple of micron bi-directionally over 200 mm runs (hence the envelope of multiple readings / traces bi-directionally) but over 1 meter is getting closer to 5 micron spread. Still impressive.

Z axis is always the most tricky as you can see from the plot much wider tolerance band and variation.

Interestingly Matsuura present the Z axis as having a 7 micron spread but seems to have the highest/ "tightest" repeatability. (maybe that was cherry picked or maybe not ?)

With all these machines there is a "sweet spot" for straightness, sometimes the straightest axis is not always the one you assume it is or would be (on a number of different machines (from different MTBs) I'm surprised to learn their Y axis is disproportionately straighter than their X axis travel.)


With some builders over a number of years or decades of machine building they artfully tune in the sweet spots (even on 5 axis machines) to best fit most likely use case scenarios of angular and linear positing, especially for most typical part sizes and functional geometries vs. envelope. Or tweaking / tuning stuff so that 0, 90, 180 , 360/0 and (60, 120, 240, 300 degree) rotational positioning is a little "sweeter" than positioning 34.982 degrees.
 








 
Back
Top