Hi,

EVERY system ever created has lost motion, and 18microns is NOT 'no backlash whatsoever'

18 microns is the potential inaccuracy, not backlash. I have the original manufacturers test certificates and all my ballscrews
had less than 12um error in 300mm and less than 0.6umm cyclic. You are right I do have backlash, the planetrary gearboxes
I use on my current mill have 2 arc min lash or 0.7um linear. There is even a larger source of lost motion,
the torsional flex of the coupler. The two combined is 4um. This is my current machines effective
resolution and repeat accuracy. Further this is my measured result over a range of reversals an load conditions.
The faster the reversal and heavier the load the lost motion is up to about 4um, under less challenging conditions its
somewhat less, I struggle to make any sensible measurements of such small amounts.

I do not have ball bar but I cut an interpolated circle in aluminum just before Christmas and could not measure
any out of roundness with a 0.01mm reading 75-100mm micrometer. This is consistent with my assertion of
4um repeat accuracy.

My new build will not have reduction boxes and therefore no backlash and I am thinking of bellows couplers
so it will have GENUINE sub micron backlash/lost motion.

I had an email from the Ebay supplier from whom I bought the Delta servos, and they have been able to secure an even
better deal from Delta. They offered me 750W B2 series (17 bit incremental) for $379USD and 750W A2 series (20 bit
and load sensing) for $439USD. I presume that does not include shipping but for genuine quality servo and drives, quality
to which I can personally attest, these are very sharp prices.

Craig

I don't doubt you have a good system, but I guarantee it's not as good as you claim it to be. Ask Renishaw for a demo of their Ball Bar system and you'll see the gulf between what you think you have and what you actually have. It will show you out of squareness as well as an absolute dimensional circular plot in both directions. Run that a few times at different speeds and then you will really know what you have.

What's the resolution of your system? Mine is 1micron steps in X & Y
I have 1 micron scales that I'd like to use.

Hi,

Ask Renishaw for a demo of their Ball Bar system and you'll see the gulf between what you think you have and what you actually have. It will show you out of squareness as well as an absolute dimensional circular plot in both directions. Run that a few times at different speeds and then you will really know what you have.

I would dearly love to do just exactly that but Renishaw measuring equipment is beyond my means. Of all the inaccuracies that I know of and probably those that I don't
know of out-of-square is one of the more difficult for me to measure. I have an AA granite square (only 150mm x 150mm x100mm) and using that I have identified
an out-of-square between X and Y of 8um per 100mm, between Y and Z of 6um and between X and Z of 11um per 100mm. Measuring such small quantities is a challenge,
they could easily be double my measurement but could be less too. All I can really say is that these are my best measurements but my 'confidence interval' is wider
than the absolute measurement.

A ball-bar measurement would probably be more conclusive and informative. To this end I have made an LDVT with a resolution of 10um, and with successive deign and
build revisions I hope to get to 2um. Amongst the things that I wanted to do with it was a ball-bar device.

My current mini-mill has a resolution of 1um. Note that does not rely on microstepping either. One full-step of a 5 phase Vexta stepper is 0.72 degrees and through
the low lash (2 arc min) 10:1 planetaries that is 0.072 degrees or 4.32 arc min per fullstep. Given the predicted lash of 2 arc min I thought it fruitless to go for even finer resolution.
In the event the lost motion due to the torsional flex of the couplers dominates the lash. As it transpires on the two occasions that I crashed badly the wee aluminum
coupler sheared off rather than even worse damage, the mechanical force when you crash, even on such a small machine are still in the tens of kN range. The couplers
act a bit like a mechanical 'fuse', and I have retained them because of it. This was not 'by design' but a consequence of buying el-cheapo couplers and funnily enough they
have an advantage I had never considered and will tolerate 3-4um lost motion they cause.

My new build has a 160,000 line encoder so I could have resolution as fine as I like within the signalling capability of the ESS/BoB/servodrive. My intention is however to
retain 1um resolution, if I can contain lost motion, out-of-square, ballscrew inaccuracy, linear rail inaccuracy to better than 5um I would be delighted. Finer resolution
is not required. I am trying to achieve a linear rigidity between the Z axis and the vice of 10um per 1kN. To obtain that sort of rigidity even in a small machine requires
thick sections of cast iron, and getting stuff cast is not a cheap undertaking.

Ballscrew inaccuracy, linear rail inaccuracy and flexure of both is beyond my control, all I can do is buy the best quality I can find, in my budget. The squareness and rigidity
of the beds/frame into which those components are placed are my province however.

Regrettably the gear needed to 'qualify' my machine could be as much as the machine itself. At this stage I have a 36 x 24 B grade granite surface plate, the granite square
I've already mentioned, micron reading Mitutoyo dial and test indicators. A set of matched parallels is still required.Still looking for a good height gauge I can afford.
I can borrow an A grade set of gauge blocks. I'd really like an auto-collimator but even second hand they seem to be in the $2000-$4000 range.

If my machine is inaccurate it will not be for the lack of me trying to achieve it.

Craig

I'm surprised that since you're clearly interested in making an accurate machine that you aren't using linear scales. Directly measuring the position of the table rather than indirectly doing that would certainly lead to a more accurate machine, and you wouldn't need the super-precision leadscrews either to acheive that. You can go to ever finer resolution on the drive system, but it won't help with reducing the lost motion in the ball screws, leadscrew thrust bearings and in the stretching and compressing of the leadscrew. All of these things vary with load and stiction,  and all of them are taken out of the equation by linear scales.

I don't know how you can claim sub-micron accuracy on a system that uses 1micron  resolution though, that's simply impossible. Checking it with a system that has less than 1micron resolution seems odd too. LDVT probes are available in much higher resolutions, that's what Renishaw used to use.

My system is as good as I can get it with precision components, but it's still a long way from where it could be if I could use the scales. I'm happy to spend another chunk of money on it, but I need to know that it will work.  I'm surprised that there isn't a motion controller that can take step and direction pulses from Mach4, interface linear encoders and output the required movements in step and direction pulses to the servos. A purely digital solution ought to be fairly easy to create, and that's how I'll do it if there isn't a commercial solution available.

Hi,

You can go to ever finer resolution on the drive system, but it won't help with reducing the lost motion in the ball screws, leadscrew thrust bearings and in the stretching and compressing of the leadscrew

The measured and published stiffness of my ballscrews is 1110N/um. It is for this reason that I am trying to aim for a stiffness
overall of 10um @ 1kN force.

Directly measuring the position of the table rather than indirectly doing that would certainly lead to a more accurate machine,

Without good and costly linear scales I rather think that angular position of the ballscrew is at least as accurate.

LDVT probes are available in much higher resolutions, that's what Renishaw used to use.

An LVDT's resolution is inversely proportional to its stroke, so yes sub micron resolution can be had but at strokes of mm's
at best. Longer strokes, +-50mm resolution of 10um is more realistic. My own design has a stroke of +-5mm and the best
available commercial units has resolutions quoted around 1 um. If I could achieve 2um resolution I would be well satisfied.
Resolution is probably a poor concept with an analogue device like an LVDT, resolution in fact infinite but there comes a point
where the uncertainty in measurement outstrips it accuracy.

I'm surprised that there isn't a motion controller that can take step and direction pulses from Mach4, interface linear encoders and output the required movements in step and direction pulses to the servos.

After a fashion....there is. You may have noted that feedback controllers like Gallil are getting fewer and fewer as manufacturers
fade away. This trend has been happening for twenty plus years as servo drives become ever smarter. With Ethercat,
Profibus, CANOpen and similar network strategies each servodrive becomes its own motion controller in a distributed motion
control solution. Who needs expensive feedback controllers with all the programming and tuning drama when you can buy
an off the shelf servo an drive which exceeds any feedback motion controller out there?.  Given that you have AC servos
you can only be aware how much more flexible they are compared to brushed DC servos of earlier years. Already the latest generation
servo drives, including such features as load sensing, are rendering the servos like yours and mine (Delta B2's) if not
obsolete certainly outdated. A load sensing servo and servo drive do just exactly what you want, that is take step direction
pulses from Mach4 and enclose a linear scale within its feedback loop.

Where an AC servo differs from a DC servo of years past is that an AC servo MUST have superbly accurate angular information
on the position of the armature to enact field oriented control. In absence of a rotary encoder the servo drive can only
estimate its angular position with the lost motion now corresponding to loss of certainty about angular position.
That would in turn reduce the current feedback loop bandwidth from kHz to hundreds of Hz and therefore the velocity and position
loops would be even worse affected, reducing position bandwidth to a few or perhaps tens of Hz, too low to be of any use for CNC.
A loads sensing servo must still have a direct coupled encoder to maintain its bandwidth but can also close its position
on a linear scale....is that not what you want?

Craig

No amount of stiffness is going to give you a result as good as measuring the position of what you're actually interested in with Linear Encoders. I got mine from Newall and they weren't excessively expensive. You can offset some of that cost by less expensive leadscrews if you don't need to indirectly measure the position. I don't know why you're so wed to the idea of indirectly measuring the position when you can directly measure it. That's clearly a more accurate option.

There may be fewer independent suppliers of add on motion controllers, but the trend in precision machine tool design is towards linear encoders supported by their own bespoke controllers. The improvement in accuracy and dynamic performance is significant, and you can also use linear motors for the drives if you want. Sure, if you don't need high accuracy there's a lot to choose from, and many machines are sold such as Tormach that don't even have backlash compensation. The hobby market probably doesn't need Linear encoders for most applications, but some of us still want to get the most from our machines, and Linear encoders is the ultimate way to achieve that.

You only need a very small travel on LVDTs when you're using a Ball Bar system.

Hi,

I don't know why you're so wed to the idea of indirectly measuring the position when you can directly measure it. That's clearly a more accurate option.

Rubbish!!! Zero backlash, high accuracy, high stiffness results in ACCURATE linear position from angular angular input.
Why do you suppose there is a market for high quality ballscrews? OEMs would use cheaper ballscrews if they could....
but they can't.

If you want to close the loop on a linear scale....go to it!!...there is a cost effective solution if you'll 'un-wed' yourself
from the idea of a feedback motion controller.

Craig

The market for precision ball screws is huge because on a lot of machines you don't need ultra high accuracy. However, when you do, measuring what you actually want to know the position of is obviously going to produce better accuracy than even the best indirect method. That's a statement of fact, not opinion. You have already agreed that every system has backlash and lost motion, so your argument simply doesn't make sense. Backlash and lost motion are NOT consistent and repeatable, they are dependant on load and stiction. Linear scales compensate for this.

Hi,

Backlash and lost motion are NOT consistent and repeatable, they are dependant on load and stiction. Linear scales compensate for this.

I agree 100%, if you are chasing the last micron then linear position sensing is required.

My contention is that with top quality ballscrews, rails/cars and a rigid bed/frame that micron level accuracy is obtainable, if not sub micron
without the complication of including a linear scale in the position loop. Having said that the latest generation of AC servos with load sensing
in addition to the regular rotary encoder have made including a linear scale within the position loop a very much easier and cheaper proposition
than it has been.

As you are probably aware the semiconductor industry have a great demand for ultra precision stages, they chase nanometers. They use interferometric
methods, an ultra precision version of more common linear scales. The position loop is closed by the servodrive and the servodrive itself handles its
own motion control as part of a distributed motion control system. A semiconductor production line could have tens if not hundreds of such stages, a
centralized feedback motion controller is totally impractical for such a large system.

Servo manufacturers are busy making network capable (Ethercat, Profibus, CANOpen etc) servodrives which are smart enough the conduct their own motion
control and closed loop control of the servo including load sensing.

My suggestion is to take advantage of that trend and use the hardware that is coming forward as a result RATHER than a feedback controller like Galill.

Craig

You can see the same trend in PCB drilling machines, but they also use Air Bearing slides and linear motors as well as integrated linear scales to achieve astonishing performance and accuracy. At that level of performance, you can't afford the inertia of a leadscrew, it limits the accelleration too much.

I doubt very much if it's feasible to break the 1 micron barrier without measuring the table position. There will always be stiction and the stiffness of the leadscrew is different when the ball nut is at the extremes of travel. When you get a leadscrew, you get a chart that shows the errors. If you're not putting that data into the control system, the table will track those errors. In metal machining, you're using Climb milling wherever possible, and the less stiction you have, the more the table will tend to be pulled in the direction of the backlash. The bottom line is that you can control the angular position of the leadscrew as accurately as you like, but you don't know the actual table position unless you measure it. It's essentially open loop on that last vital part of the control system.

I already have 1 micron linear scales fitted to my machine, so it makes sense to use them.