Who Knows Servos?

[Noitoen]

Used to do some work at a plant with a few turret lathe type cnc's. All the axis except for the spindle were hydraulic actuated with glass scale feedback. The control card operated with step/dir signals and would give "excessive following error" if the hydraulic actuator for some reason didn't keep up the pace. I was amazed how precise the system was with only one hydraulic pump controlling various axis at the same time with precision.

[skunkworks]

Sure - On the emco lathe - each encoder count is .0000196" (these are pwm amps - the actual input is up/down pwm)

LOL - I just looked at the K&T - the lowest resolution axis is also around .0000194" per encoder count. (+/-10v drives)

The motion/hal layer of linuxcnc can compete with any industrial control. (the user interface? to each his own.. )

I have a video of the k&t (multi-ton machine) positioning to within .0001"



sam

[MaxHeadRoom]

With the Galil Motion cards, tuning the PID loop for optimum, (~20 pulses) guarantees minimum/acceptable following error.
When 0v-10v analogue Torque mode drives used

[Sparky_NY]

As to servo position error, for the DIY systems I put together, I used Galil Motion cards, that sit in a PC slot, these are about the closest you can get to a Industrial designed system.
As to PID tuning, you can achieve around 20 pulse error. The Minimum resolution encoders used are 1200 pulse/rev which translates to 1200x4 = 4800 pulses/rev when using the quadrature count.
This error is only meaningful when you then compare this to something Fanuc calls your 'Least Input Increment' or what you desire to be the smallest positioning value, e.g. 2 μm for example.

Positioning error is one parameter, following error is another and the parameter tuning is normally to optimize. (position error is static without movement, following error is while the axis is in motion)

Optimizing following error is what ensures your circles don't end up being ovals.

[MaxHeadRoom]

As to servo position error, for the DIY systems I put together, I used Galil Motion cards, that sit in a PC slot, these are about the closest you can get to a Industrial designed system.
As to PID tuning, you can achieve around 20 pulse error. The Minimum resolution encoders used are 1200 pulse/rev which translates to 1200x4 = 4800 pulses/rev when using the quadrature count.
This error is only meaningful when you then compare this to something Fanuc calls your 'Least Input Increment' or what you desire to be the smallest positioning value, e.g. 2 μm for example.

[RB211]

-I'm... not really following you.

One, virtually all the indexers you're going to find are "industrial". The 'home shop' stuff is manual- dividing heads and manual rotary tables and the like. You go automatic for production, and production means industrial.

And two... is there a reason you wouldn't want your employees operating the machine?

An indexer (as opposed to a true 4th axis) is generally a stand-alone, add-on device. Like the manual dividing head, you bolt it to the table and mount your parts in it.

The mill mills a flat or drills a hole or whatever, and then sends an "index now" signal to the indexer's controller. The standalone controller then turns the part- you tell it X number of degrees, which direction, how fast, etc. Once the indexer has finished moving, it sends a "I'm done indexing" signal back to the mill, which then proceeds on with the program.

All of this, save for the cheap Chinese "CNC router" axes/indexers you can find on eBay, are all going to be industrial accessories- meant to live in coolant, turn to sub-seconds-of-angle accuracy, and hold solidly enough for high-feed production milling. There is, near as I can tell, literally nothing between a $300 Ebay Special, or a $5,000 industrial unit.

Doc.

Yeah, I don't want anyone operating my machines if they all require knowledge of the inner workings. Industrial machines have enough polish on them that it shouldn't be an issue.

[Doc Nickel]

Oh sure. Something I can do in my spare time, with parts I already have laying around the house.

Realistically, the work I need to do is low-force, and I could probably get away with a cheap Chinese "4th axis" meant for a router or something- basically just a chuck, a bearing block, a stepper and a belt reduction.

And I [i]have[i] been thinking of maybe trying the same thing with something like a spindexer- I'd rather have a 5C than a chintzy 4" Chinese 3-jaw. But no matter what, kind of the last thing I need right now is another project. My "stuff to do" list just for the month of May, already extends into 2025.

Doc.

[elf]

Certainly wouldn't be very difficult to make one. Just get a servo, a brake, a couple of bearings, a few chunks of metal, and some control software. Shouldn't cost more than a few hundred dollars if you're a good shopper.

[Doc Nickel]

Looking at the HAAS stuff, it was very apparent they are designed around your employees operating the machines, not you. That is one reason you go with industrial controllers I guess.

-I'm... not really following you.

One, virtually all the indexers you're going to find are "industrial". The 'home shop' stuff is manual- dividing heads and manual rotary tables and the like. You go automatic for production, and production means industrial.

And two... is there a reason you wouldn't want your employees operating the machine?

An indexer (as opposed to a true 4th axis) is generally a stand-alone, add-on device. Like the manual dividing head, you bolt it to the table and mount your parts in it.

The mill mills a flat or drills a hole or whatever, and then sends an "index now" signal to the indexer's controller. The standalone controller then turns the part- you tell it X number of degrees, which direction, how fast, etc. Once the indexer has finished moving, it sends a "I'm done indexing" signal back to the mill, which then proceeds on with the program.

All of this, save for the cheap Chinese "CNC router" axes/indexers you can find on eBay, are all going to be industrial accessories- meant to live in coolant, turn to sub-seconds-of-angle accuracy, and hold solidly enough for high-feed production milling. There is, near as I can tell, literally nothing between a $300 Ebay Special, or a $5,000 industrial unit.

Doc.

[RB211]

-Yes, yes, I know, LinuxCNC is the end-all, be-all of CNC software, and it's only the short-sightedness of the manufacturers that anyone is still bothering with inadequate crap like Fanuc and Siemens.

But in this case, the control part is already taken care of. The project is to do some radial drilling in a round part. The Logan, with it's Centroid control, can easily run a "C" axis, the issue is simply the cost and labor of converting the spindle over to a decent servo drive.

The Omniturn control also already has the capability to use a C-axis, and all I'd need to do is buy the parts, off the shelf, from Omni. Unfortunately that's rather expensive at about $5,000.

The Trak mill can natively run most typical indexers, whose control boxes accept fairly standard "start indexing" and "I've stopped indexing" signals. Such indexers are available off the shelf and plug-and play... but also for about $5K, used.

Currently, my cheapest option, is that I have a nearly-completed CNC conversion of a little Grizzly mini-mill. That was a project from a friend, who was supposed to trade me some CNC training, in exchange for my fixing his badly-converted machine. Right now it needs little more than wiring, and I already have a complete, running Centroid PC for it. I figure I get one of those little cheap Chinese $300 "4th axis" units off eBay, and I could be drilling the radial holes for a couple weeks work and maybe $500 in parts.

I will end up with an 5C indexer for the Trak one of these days, I have plenty of other parts that could use it. But right now, funds just aren't there for it.

Doc.

Looking at the HAAS stuff, it was very apparent they are designed around your employees operating the machines, not you. That is one reason you go with industrial controllers I guess.

[Doc Nickel]

If you follow Skunkworks advice and go with Mesa and Linux CNC, pretty sure there is nothing you can't do, DIY.

-Yes, yes, I know, LinuxCNC is the end-all, be-all of CNC software, and it's only the short-sightedness of the manufacturers that anyone is still bothering with inadequate crap like Fanuc and Siemens.

But in this case, the control part is already taken care of. The project is to do some radial drilling in a round part. The Logan, with it's Centroid control, can easily run a "C" axis, the issue is simply the cost and labor of converting the spindle over to a decent servo drive.

The Omniturn control also already has the capability to use a C-axis, and all I'd need to do is buy the parts, off the shelf, from Omni. Unfortunately that's rather expensive at about $5,000.

The Trak mill can natively run most typical indexers, whose control boxes accept fairly standard "start indexing" and "I've stopped indexing" signals. Such indexers are available off the shelf and plug-and play... but also for about $5K, used.

Currently, my cheapest option, is that I have a nearly-completed CNC conversion of a little Grizzly mini-mill. That was a project from a friend, who was supposed to trade me some CNC training, in exchange for my fixing his badly-converted machine. Right now it needs little more than wiring, and I already have a complete, running Centroid PC for it. I figure I get one of those little cheap Chinese $300 "4th axis" units off eBay, and I could be drilling the radial holes for a couple weeks work and maybe $500 in parts.

I will end up with an 5C indexer for the Trak one of these days, I have plenty of other parts that could use it. But right now, funds just aren't there for it.

Doc.

[RB211]

Apparently no, I don't know servos.

All I do know is it takes a very, very minor error before the Omni control throws an "out of position" fault.

And either way, it's just not a feasible mod for the Logan right now (cost and time) or for the Omni itself (just cost- to the tune of $5K or more). An indexer for the Trak mill won't be cheap either, but that one would have a lot more utility- I have a fistful of projects lined up for that one.

Doc.

If you follow Skunkworks advice and go with Mesa and Linux CNC, pretty sure there is nothing you can't do, DIY.

[Doc Nickel]

Apparently no, I don't know servos.

All I do know is it takes a very, very minor error before the Omni control throws an "out of position" fault.

And either way, it's just not a feasible mod for the Logan right now (cost and time) or for the Omni itself (just cost- to the tune of $5K or more). An indexer for the Trak mill won't be cheap either, but that one would have a lot more utility- I have a fistful of projects lined up for that one.

Doc.

[RB211]

A servo worth a damn has an encoder with enough counts that a small error is of no real world consequence. Besides the servo will hunt the error down and not let it grow.

[J Tiers]

True if the system is a "feedback" system. The only thing that makes it move is an error between the "desired" position and the "reported" position from the position feedback. If an external force deflects it from the desired position, or prevents it from moving to the desired position, there will be an error reported, and the system will try to move back to the desired position.

If it ever reaches the desired position, the restoring force would be zero, since there is no error, and no reason to move. Any force applied would move to a position with an error in excess of the desired position, and it would move back

In order to resist a deflecting force, there must be some error sufficient to cause the system to develop the force needed to resist the deflection. In an analog system, the restoring force depends on the deflection. In a digital system, there theoretically would be full restoring force for any detected deflection. In practice, that could lead to an oscillation, so there is some means to avoid that, which may lead to more than one "count" of error, depending on how much force is required to move closer to the desired position.

But the error must always exist, or the system cannot operate.

You can see that actually moving to the desired position would lead to zero force. So then the external force could push the cutter (or whatever) back, at which point the system would see an error and try to correct it. If it did correct the deflection back to zero, the correcting force would go to zero, and the cycle would repeat.

While there cannot be a "resisting force" applied when in the desired position, there can be a "brake" applied, such as holding current in a motor, etc, to hold the existing desired position. That may or may not be sufficient to prevent moving out of the desired position. It is best viewed as "friction" and not as a force that is "actively resisting" any external forces, and has nothing to do with the feedback action, nor required errors.