[joeavaerage]

Hi,

The reason why I prefer to have LinuxCNC close the loop is that it can see what all 3 axes are doing. Correct me if I'm wrong (I may very well be), but it's my understanding that having a closed loop back to the controller will enable the controller to limit all axes to whatever axis happens to be the weakest link for a given operation.

Sorry to say it but that is old school thinking and went out of fashion over ten years ago.

The modern paradigm is that each servo drive is responsible for its own motion control and there is NO CENTRAL motion controller. This is called
'distributed motion control' and there are a number of popular technologies that do it, among the most famous are Ethercat, Profibus and CANOpen..
The PC then becomes trajectory planner and realtime communication master only.

All this is because servo drives have been getting ever more intelligent and better as the years go on. Even without introducing distributed motion control
the AC servos we are talking about are still way WAY smart.

One of the drives from either of the brands we are talking about (Delta and DMM) accept pulse commands, be it step/direction, CCW/CW etc.
At all times it has a cumulative record of all the commands that have been issued and so at any instant knows exactly where it is supposed, or commanded to be.
At any given time it also knows the position of it own rotor and therefore at any given instant knows the error between where it actually is and where its supposed to be,
called the 'following error'. If that error increases beyond what you program as acceptable following error it will fault out and signal your controller. In a mill you might for
instance program 20 encoder counts as the max permissible following error, which may equate to 5um say.

Each servo in the machine will be monitored in the same manner, so that if any one axis deviates by more than 5um all axes will stop.

Its up to your controller settings to ensure that each of the axis servos can keep up. For instance if the motion controller commands an X axis acceleration
of 15m/s² but the servo and axis can at best manage 10m/s² then naturally the axis will lag the command and will in short order fault
out 'following error'. You must inform your controller what the servos are capable of. The trajectory planner decides what the axis movements are and will limit
the X and Y axis accelerations to accommodate the very heavy and slowly accelerating Z axis say.

In this way its not necessary for the controller to close the loop, the individual drives do that, and signal the controller only if it can't keep up.

Modern AC servo drives have tuning aids in the setup software, I bet you don't get an oscilloscope view of following error from LinuxCNC, but you do with
Delta. Likewise you get one or more notch filters with modern servos that allow you to tweak the response in remarkable ways. While I have no doubt you could try
the same thing in LinuxCNC, but how much of a realtime programming master are you?

The modern way is to allow the drive to close the loop and take advantage of all the smart stuff the manufacturer provides and you use a modest motion controller.

You may be interested in this solution:

https://www.automationtechnologiesin...-3-axis-110vac

This kit has 3 750W Ethercat servos and drives, Mach4 license and the Interval-Zero and Kingstar runtime licenses all for $2850.
Note that it does not have a motion control board, it does not need one!!!!

I mention this solution more as an example of the modern design paradigm than a solution for your consideration....but it may appeal to you.

Where Ethercat (and other similar strategies) shine is that you can have up to 100 Ethercat slaves on the network. So you could have your mill, joined by a coordinated
conveyor to a second machine, another coordinated conveyor to a third machine and so on. They can all be controlled by the one Ethercat Master. Each of the servo drives (and other
Ethercat nodes) are joined together in a daisy chain....gone are the days of a multicore cable to each drive, just a whole bunch of Ethernet jumper cables. Large manufacturing plants
with many machines benefit particularly with this strategy, simple two and three axis machines can certainly use it but the advantages over conventional solutions are less pronounced.

Craig

[strantor]

Hi,



Sorry to say it but that is old school thinking and went out of fashion over ten years ago.

The modern paradigm is that each servo drive is responsible for its own motion control and there is NO CENTRAL motion controller. This is called
'distributed motion control' and there are a number of popular technologies that do it, ...

Ok, well I guess it's time to try something new! Thanks for the info. This little hint combined with the knowledge that the A2 has glass scale feedback, and the fact that two glass scales just happened to spontaneously jump into my shopping cart and pay or themselves with my card, has me leaning a certain direction.

How ironic is this?

https://www.cnczone.com/forums/attac...d=456506&stc=1 https://www.cnczone.com/forums/attac...d=456508&stc=1

Only $4 difference in price?? With shipping the deltas work out to $45 more. Small price to pay for a 10-fold improvement over my stated minimum accuracy.

Since I'm divorcing myself from the notion of controller closed loop and embracing the idea of distributed motion control, is there a better controller suited to this? Because TBH LinuxCNC is a bit of a pain. I've been eyeing the Centroid Acorn but until a few minutes ago didn't seriously consider it because it doesn’t have encoder feedback. Other than that it seemed like actually a pretty professional system on a hobby budget.

[joeavaerage]

Hi,
I spent big dollars to get preloaded ground (C5 rated) 32mm doublenut ballscrews to have a zero backlash solution . Why would I need glass scales?

I'm presuming that you will go for ground ballscrews, you did state at the top of the post that you wanted an accurate machine and that perforce means
C5 or better....Sort of makes glass scales redundant.

At the time I was looking to buy the servos for my new build I was unaware the the A2 series had that dual loop ability. Had I known I might have been tempted
to get them. In retrospect I'm glad I didn't because I would be looking to get glass scales....just because I bloody well can! My budget has taken a real hammering
as is.....I don't need more temptations.

I believe where dual loop really comes into play is in the semiconductor industry where the use of LDVTs and interferometers is common, that's a factor of 50 to100 times
the resolution and accuracy that I need.

I think there are really four CNC solutions that offer genuine performance and reliability affordably:
1) LinuxCNC
2) Mach4
3) UCCNC
4) Centroid Acorn.

Who cares about encoder feedback? Get over the idea that you need positional feedback.....the controller tells the axis where to go and the drive and servo
go there in exactly the specified manner, and only ever let you know if for some reason that it can't do as its commanded.

Mach4, UCCNC and Centroid are all run on Windows PCs are are perforce NOT realtime computing solutions and therefore all their motion controllers
are known as 'buffered'. There are some implications about that which could be considered disadvantages. LinuxCNC however is run on a Linux distro
with RealTimeExtensions (RTE) which gives a pretty fair realtime (4us jitter with common PC hardware) performance.

Any of these solutions will work well, each has their advantages and each their disadvantages, but you would be happy with any of them.

I use Mach4 and have done for six years and it works well.

Craig

[strantor]

Hi,
I spent big dollars to get preloaded ground (C5 rated) 32mm doublenut ballscrews to have a zero backlash solution . Why would I need glass scales?

I'm presuming that you will go for ground ballscrews, you did state at the top of the post that you wanted an accurate machine and that perforce means
C5 or better....Sort of makes glass scales redundant.

Let's say a guy had some cheap China rolled ballscrews with plain (non-anti-backlash) ballnuts laying around, some glass scales, and some Delta A2s with glass scale feedback, would that make the C5 ground ballscrews less important? I know the answer for a 3D mill that's doing all kinds of dynamic stuff, lots of back and forth, cutting on all sides, but for a lathe that turns simple shapes and always cuts on the same side? Might this be "ok"? It seems pretty close to "ok" for my manual lathe with acme leadscrew with 10 thou of slop.

This would only be a plan B. I want to try my hand at making anti backlash rack & pinion drive as I mentioned, so I don't have a goofy looking servo sticking way out of the side of my machine. A rack & pinion could be concealed behind the column (under the bed, for a typical horizontal lathe).

[joeavaerage]

Hi,
backlash is a killer, no matter how good the servo system.

Lets imagine you have an axis with a small amount of lash. The axis gets commanded to a given position and the glass scale closes the position loop so we know
that the axis is in position. Then cutting forces 'pull' the axis and it moves ever so slightly because of lash. The glass scale will detect that and try to close the error,
but no matter what the error HAD to occur in order for the scale to measure it and THEN the servo system can correct it.

In Control Engineering terms that is the 'principle of casuality'.In practice the amount of error is determined by the sensitivity of the glass scale AND (most crucially)
the servo system bandwidth.

The bottom line is get rid of the lash.

I suggest you start your search here:

https://www.ebay.com/str/industrialp...cat=4322381011

You can get good C5's and C3's for the same or less than new C7's. I got three of these indentical units:

https://www.ebay.com/itm/THK-USED-Do...sAAOSwPWRZSm4G

for $1000USD including three day shipping to New Zealand. They are superb.

Second issue with rolled screws is less about their average accuracy, C7's are 50um/300mm, but the cyclic error which can be up to 30um per rev. In order for
a load encoder to close the loop on an error that occurs that frequently it must have mega-bandwidth. So why are you spending big bucks on high bandwidth servo systems
merely to cover (at best) poor mechanics? You should have spent more money on the mechanics but less on the servo system.

Craig

[rodw]

Hi,



Sorry to say it but that is old school thinking and went out of fashion over ten years ago.

The modern paradigm is that each servo drive is responsible for its own motion control and there is NO CENTRAL motion controller. This is called
'distributed motion control' and there are a number of popular technologies that do it, among the most famous are Ethercat, Profibus and CANOpen..
The PC then becomes trajectory planner and realtime communication master only.

All this is because servo drives have been getting ever more intelligent and better as the years go on. Even without introducing distributed motion control
the AC servos we are talking about are still way WAY smart.

One of the drives from either of the brands we are talking about (Delta and DMM) accept pulse commands, be it step/direction, CCW/CW etc.
At all times it has a cumulative record of all the commands that have been issued and so at any instant knows exactly where it is supposed, or commanded to be.
At any given time it also knows the position of it own rotor and therefore at any given instant knows the error between where it actually is and where its supposed to be,
called the 'following error'. If that error increases beyond what you program as acceptable following error it will fault out and signal your controller. In a mill you might for
instance program 20 encoder counts as the max permissible following error, which may equate to 5um say.

Each servo in the machine will be monitored in the same manner, so that if any one axis deviates by more than 5um all axes will stop.

Its up to your controller settings to ensure that each of the axis servos can keep up. For instance if the motion controller commands an X axis acceleration
of 15m/s² but the servo and axis can at best manage 10m/s² then naturally the axis will lag the command and will in short order fault
out 'following error'. You must inform your controller what the servos are capable of. The trajectory planner decides what the axis movements are and will limit
the X and Y axis accelerations to accommodate the very heavy and slowly accelerating Z axis say.

In this way its not necessary for the controller to close the loop, the individual drives do that, and signal the controller only if it can't keep up.

Modern AC servo drives have tuning aids in the setup software, I bet you don't get an oscilloscope view of following error from LinuxCNC, but you do with
Delta. Likewise you get one or more notch filters with modern servos that allow you to tweak the response in remarkable ways. While I have no doubt you could try
the same thing in LinuxCNC, but how much of a realtime programming master are you?

The modern way is to allow the drive to close the loop and take advantage of all the smart stuff the manufacturer provides and you use a modest motion controller.

You may be interested in this solution:

https://www.automationtechnologiesin...-3-axis-110vac

This kit has 3 750W Ethercat servos and drives, Mach4 license and the Interval-Zero and Kingstar runtime licenses all for $2850.
Note that it does not have a motion control board, it does not need one!!!!

I mention this solution more as an example of the modern design paradigm than a solution for your consideration....but it may appeal to you.

Where Ethercat (and other similar strategies) shine is that you can have up to 100 Ethercat slaves on the network. So you could have your mill, joined by a coordinated
conveyor to a second machine, another coordinated conveyor to a third machine and so on. They can all be controlled by the one Ethercat Master. Each of the servo drives (and other
Ethercat nodes) are joined together in a daisy chain....gone are the days of a multicore cable to each drive, just a whole bunch of Ethernet jumper cables. Large manufacturing plants
with many machines benefit particularly with this strategy, simple two and three axis machines can certainly use it but the advantages over conventional solutions are less pronounced.

Craig

Craig, you really should acquaint yourself with Linuxcnc before you make uninformed statements about what it can and can't do.

Its all very well to have a feedback loop attached to a motor via its encoder but there are many other sources of position error. Linuxcnc can assist in this error with dual loop feedback which is not uncommon in the industry. You let the drive close the velocity loop and allow Linuxcnc to close the position loop (possibly by taking its feedback from a linear scale).

And you lost the bet about the oscilloscope.

Linuxcnc includes a tool called halscope, which is a very capable software oscilloscope that can monitor any system variable including the following error when tuning a servo. I have used it extensively.

The features of Linuxcnc , its open architecture, the huge range of available motion interfaces including Ethercat and the ability to write custom components which once installed with a single command line are treated as if they are part of the linuxcnc core makes it an unbeatable proposition for complex retrofits without emasculating the machine as I see done with some of the far less capable Windows solutions. I read threads where users throw out perfectly good motors because they are not step and direction when Linuxcnc can control them in its stride.

You may choose not to learn about alternative operating systems to Windows. I use Windows and two other operating systems every day. To me, the OS is just a tool to support an application and for me, Linux is the best tool for CNC because of LinuxCNC. If you truly wish to position yourself as a CNC expert, you should acquaint yourself with what else is out there or restrict your comments to your sphere of expertise.

[strantor]

Craig, you really should acquaint yourself with Linuxcnc before you make uninformed statements about what it can and can't do.

Its all very well to have a feedback loop attached to a motor via its encoder but there are many other sources of position error. Linuxcnc can assist in this error with dual loop feedback which is not uncommon in the industry. You let the drive close the velocity loop and allow Linuxcnc to close the position loop (possibly by taking its feedback from a linear scale).

And you lost the bet about the oscilloscope.

Linuxcnc includes a tool called halscope, which is a very capable software oscilloscope that can monitor any system variable including the following error when tuning a servo. I have used it extensively.

The features of Linuxcnc , its open architecture, the huge range of available motion interfaces including Ethercat and the ability to write custom components which once installed with a single command line are treated as if they are part of the linuxcnc core makes it an unbeatable proposition for complex retrofits without emasculating the machine as I see done with some of the far less capable Windows solutions. I read threads where users throw out perfectly good motors because they are not step and direction when Linuxcnc can control them in its stride.

You may choose not to learn about alternative operating systems to Windows. I use Windows and two other operating systems every day. To me, the OS is just a tool to support an application and for me, Linux is the best tool for CNC because of LinuxCNC. If you truly wish to position yourself as a CNC expert, you should acquaint yourself with what else is out there or restrict your comments to your sphere of expertise.

I chose linuxcnc the first time around because of the reasons you listed. My mill was a real CNC when I got it; an old one, but born to make money. It is a 1988 model and had a Dynapath controller with a CRT. It came with NSK ground ballscrews and was originally equipped with DC servos but the previous owner had upgraded to AC servos. The controller had cancer and I had the option to either spend time and money fixing old stuff or designing new stuff. I chose the more exciting option but wasn't about to cripple it with steppers and Mach3. The LinuxCNC learning curve was steep but I wanted it to happen so it happened. I never did figure out how to use halscope though. I could probably get better performance out of it by figuring that out, but as is, it performs as I think a professional CNC should. I think I will stick with LinuxCNC but I was just putting feelers out to see if anything has overtaken it in the few years since my retrofit, as technology changes so fast.