[slow-poke]

I need to make a decision on CNC software for the new mill. I have used Mach3 in the past, so I'm quite familiar with it. Some obvious competitors are Linux CNC , Acorn, UCCNC, others?

Perspective: This is for hobby use with an emphasis on low cost not necessarily plug and play. I will be using ac servos with step and direction input control. Will add scales eventually.


Would love to hear Plus and minuses, and experiences good and bad from anyone willing to share. Emphasis should be on low cost and reliability, I would rather pay $100 for a generic breakout board and have to play a bit to configure it than pay $1k for plug and play.

I think Linux CNC as well as Mach can be free, not sure on the cost for Acorn / others?

Should include hardware in the discussion. I have a spare via (EPIA) nano ITX single board PC that I could use that would require some form of breakout board(s)

If a breakout or controller board is required for you choice above, please expand on that aspect.

Much appreciated.

[joeavaerage]

Hi,
all development on Mach3 ceased eight-nine years ago. Not withstanding that it still works and has a loyal fanbase, its obsolete. For a new build use Mach4, its light years ahead.
There are enough differences that the set up of Mach4 that it is a learning curve, but operationally its is similar to Mach3. I have used Mach4 ($200USD) for 8.5years with an Ethernet SmoothStepper($225USD).
The ESS really requires a breakout board and despite the extra cost I'd recommend an MB3 ($200USD) by CNCRoom. You could economise on the breakout board but the MB3 is fully developed (51 IO's)
with well thought out balance of inputs and outputs, it has differential signaling for the servos, so you'll be able to run your AC servos flat stick, good PWM and relay outputs for the spindle.
The combination of the Mach4 license, the ESS and MB3 is just a little over $600USD.

UCCNC and one of their motion boards like a UC300 and an appropriate breakout board may a somewhat cheaper option. UCCNC is licensed one license per machine, whereas Mach4Hobby allows
five machines per license. The one-off cost for UCCNC is cheaper, but if you have multiple machines the price advantage evaporates.

Operationally UCCNC and Mach4 are very similar. Mach4 has better features for plasma tables and lathes, but for mills UCCNC and Mach4 are about a dead heat.
Mach4 has a very complete and powerful scripting environment and lends itself to customisation, whereas UCCNC is somewhat less amenable to customisation.
Small details really, unless you need a specific feature that one has but the other does not them UCCNC and Mach4 are very close competitors.

Centroid Acorn is a commercial solution and you are restricted to operate within the bounds of Centroid Acorns 'environment'. It is as near a turn key solution as there is,
and is pretty good value. You may rue the lack of flexibility of Mach4 or LinuxCNC, but for regular three and four axis mills Acorn is a good choice. It will drive servos
but not with the same resolution and speed as Mach4 or LinuxCNC, and about the same speed (max Step pulse speed) as UCCNC.

LinuxCNC with one of the many Mesa motion boards is a very appealing choice. LinuxCNC is free and open source. A very useful Mesa board can be had for $250USD or so.
LinuxCNC is very flexible, limited only by your programming skills. Many, like myself, could not really be bothered with Linux, and so I went for a Windows solution in the form of Mach.
I did have LinuxCNC running my mill for a couple of months, and I must say it works very well. I was using a parallel port and basic breakout board. Operationally its not disimilar to Mach. When you get to
programming closed loop controls and that sort of thing, that's where the programming skill comes in. If you are a fair to strong coder then LinuxCNC is a good choice, if you are a weak coder
then a Windows solution like UCCNC or Mach4 may be preferable.

There are very strong and vociferous adherents of each of these solutions, and I include myself in that category, but some respondents like to portray their favoured solution as the be all and end all
whereas all other solutions are conceived by the devil. A good discussion can and should be had about the strengths and weaknesses about and any all CNC solutions, but pure partisanship
is less than useful.

The four solutions I've mentioned...ALL.....would be perfectly good, but there are others that have gained a good reputation also. SimCNC by CSLabs, Edging, Masso etc.

Craig

[joeavaerage]

Hi,
I just re-read your post:

I will be using ac servos with step and direction input control. Will add scales eventually.

If you wish to have scales to close the position loop, then the Windows solutions like Mach4 and UCCNC are not suitable. Because Windows is not a realtime computing system the
computational delays enforced by the Windows OS all but preclude a PC from closing the position loop.

This is where LinuxCNC does shine, as it runs on a realtime distro of Linux.

Another alternative is to use servos and drives that have the normal rotary encoder (like all AC servos) but also have a secondary encoder input (eg linear scale) that can close the position loop.
This means that you get the 'full closed loop' goodness but do not require a 'full closed loop controller' like LinuxCNC, Gallil, Masso etc.

I use Delta B2 servos which have a single rotary encoder. A 750W servo kit costs $438USD. A Delta A2 750W servo kit is somewhat more ($634USD) but it CAN have a secondary encoder and can close
the position loop using a linear scale, without need for a closed loop controller as described above.

https://www.fasttobuy.com/ecmac10807...le_p27766.html

Craig

[slow-poke]

Craig,

Thank you for taking the time to post a fantastic and informative response. I particularly like this point:

There are very strong and vociferous adherents of each of these solutions, and I include myself in that category, but some respondents like to portray their favoured solution as the be all and end all whereas all other solutions are conceived by the devil. A good discussion can and should be had about the strengths and weaknesses about and any all CNC solutions, but pure partisanship is less than useful.

I guess the good news is if at the end of the day, I don't like my choice, I can always change later.

For the LinuxCNC option:
- what type of computer board is adequate?
- Messa board suggestions?

Thanks again.

[joeavaerage]

Hi,
I'm far from an expert on LinuxCNC.

I loaded a realtime distro of Linux onto a laptop, just a modest laptop at that, and then loaded LinuxCNC. It worked fine.

My understanding is that there is a test that is performed that measures 'jitter', and that in turn suggests how well or otherwise a particular PC approaches the ideal of 'realtime'.
Again my understanding is that 4us jitter, is considered good and suitable for use with LinuxCNC. The modest laptop (ex Windows) that I used had a figure of about 6us, and proved
perfectly good enough, at least for my then simple machine.

When I experimented I used a parallel port. I was very impressed as to how much better LinuxCNC was at extracting good performance from a parallel port that Mach3 always struggled with.
For a simple machine and low-ish signal rates as applicable for steppers then it may be that a parallel port is adequate. If you want higher signal rates then one of the Mesa motion boards
is appropriate. The Mesa boards have hardware pulse generators on board and can therefore generate high accuracy, high frequency pulses WITHOUT unduly taxing the computing platform.
As to which model is suitable, I'm sorry, I just do not know. There is an active forum on LinunxCNC, so you'll get all the advice you care to listen to there.

I don't like my choice, I can always change later.

That is certainly true, none of the solutions is that expensive that your choice is 'cast in stone'. Having said that you become very accustomed to a particular way of doing things
and will often struggle to change your CNC solution when you have great familiarity with another. I does make sense then to choose wisely, or as least as wisely as your foresight
permits, a solution that will accommodate the demands that you may place on it in the future.

As an example I use Mach4 and Mach4 is perfectly OK with five axis where the WCS is coincident with the machine axis centre. High end industrial five axis machines allow you to place
the material at will (with respect to the machine centre) and the machine will automatically apply kinematic corrections to suit. Mach4 does not have that feature....yet.
It is the first time however that I have encountered a demand that Mach4 cannot meet. I may well have to write my own kinematics scripts. In the mean time I setting my parts
relative to the machine centre, and so is adequate as is.....but there is a question mark about whether Mach4 will continue to be as a complete CNC solution as it has been to date.

Craig

[phomann]

I need to make a decision on CNC software for the new mill. I have used Mach3 in the past, so I'm quite familiar with it. Some obvious competitors are Linux CNC , Acorn, UCCNC, others?

Hi,

I’ve looked at all of the various CNC control software options.
I started with turboCNC (dos based), then Master5 that turned into Mach1, then Mach2, Mach3 and finally Mach4. They all worked well. In fact I still use Mach3 for some production stuff.
I’ve also used UCCNC’s control software and motion controllers and really like them. When giving advice to customers, if it’s their first CNC system I usually recommend UCCNC due to their very clean user interface, and ease of setup. If they have experience I point them to Mach4, UCCNC or LinuxCNC depending on their needs and what they feel comfortable with.
I’ve used Acorn but don’t really like it due to the User interface. To me it’s stuck in the DOS era.
It motion system is fine, so it’s just me.
In fact I have one of their boards with a licence for their lathe software. It was for a customer job that was reneged on after I started the build. This thread has reminded me to put it up for sale to recoup some of the cost.
The other option for people who are looking at their first CNC build is Masso. Very easy to use and works well.

I’ve converted my main mill over to PathPilot and absolutely love it. It just works. There are no issues with the probing and the UIF is intuitive. The down side is that it’s not simple to set up.

The latest system I’m working on is LinuxCNC running on a Raspberry Pi 4, along with a Mesa card. It looks to be a really nice turnkey solution. I still need to do a lot more work but it seems promising.

But as Craig pointed out, for 5 axes with linear encoder’s LinuxCNC is probably way to go.
Cheers
Peter


Sent from my iPhone using Tapatalk Pro

[Mecanix]

For the LinuxCNC option:
- what type of computer board is adequate?
- Messa board suggestions?

LinuxCNC is a "Real-Time" solution. And so it won't get closer as it get to a modern industrial controller(less their proprietary algorithms and compressors, perhaps).

Compute is irrelevant if using a fpga based control board e.g MESA 5xxx~7xxx, or the LiteX-CNC 5A-75B and 5A-75E. A fast network/switching will be required however (1gigabit ethernet).

You can also run your RT controls over a microcontroller 'if the tasks are simple' e.g "Remora" which uses an inexpensive NVEM (STM32) board. Some even run step & pwm generators using ESP32, although not entirely sure this can feedback real-time like the MESA and 5A-75x.

[joeavaerage]

Hi,
the modern paradigm is 'no central motion controller'.

Take Ethercat as an example, a central PC runs some CNC software and does the trajectory planning. It communicates the numeric trajectory data to the servos over a realtime communication protocol,
in this case Ethercat, and each servo performs its own slice of the motion control task. This is called 'distributed motion control' and many if not most machine manufacturers are going this way.
It means you do not need some central fancy motion controller, each servo is responsible for it own motion. Each servo can autonomously Home, it can accommodate linear scales etc and close its own position loop
accordingly....no need for the central motion controller to get involved.

The strategy places more and more of the motion 'smarts' in the servo drive and less and less in the central controller. This pays big dividends if you have a production line with many conveyors and manipulators between machine stations,
one PC controls the whole lot. The communication data is daisy chained from one servo to the next with a vast saving in cabling. Ethercat can handle 100 nodes, where a node might be a servo, but could also be a multi-IO node with switches and sensors
or some hydraulic machine itself controlled by a PLC.

The whole concept of some ultra smart and sophisticated central motion control is slowly going away for distributed control.

Craig

[Mecanix]

Hi Joe. Coincidence I was looking at the ClearPath-MCVC series a few hours earlier. Even the servo drives are getting ditched out for built-in logic apparently...

[joeavaerage]

Hi,
ever smarter servos seems the way of the future.
At the current time Step/Direction servos with control solutions like Mach, LinuxCNC etc are still very VERY good indeed and as cheap as they have ever been, ergo just right
for hobbyists.

I did consider Mach4/Kingstar Etherat/IntervalZero solution when building my mill. The extra cost of going that way amounted to about $600USD, plus buying Ethercat ready servos.
Not prohibitively expensive, but more than I wanted to spend. Thus I have Mach4/ESS/home brew breakout board solution, and it works really well.

Who knows, my next build might be Ethercat.

Craig

[Mecanix]

5A-75E ($20) + LinuxCNC2.9.1(free) = RT Feedback 5ax all-day.

[CNCMAN172]

Since you plan on using linear scales on your machine you might want to consider Dynomotions KFLOP which has that capability. It is a very stable system that just works

[joeavaerage]

Hi,
I found this at the same supplier (Fast-to-Buy) for $548USD. Its a 750W Delta A2 servo kit, including motor, drive and cables. Note that it is 'full closed loop' that is to say that you can hook a differential linear scale direct
to the servo drive to close the position loop. It does not require a controller that can close the loop eg LinuxCNC, Masso, Galil, Dynamotion, etc, the drive closes the loop for you. Easy.

https://www.fasttobuy.com/ecmac10807...le_p27771.html

Craig

[Mecanix]

...the drive closes the loop for you. Easy.

Useless, no offence. Won't deal with the mechanical (backlash, screw compensation, etc etc), which is what's all about when you talk "feedback" for machine-tools.
Linuxcnc has a robust PID that controls loops and errors. 1um scales interfaced into the control board's encoder inputs outperforms anything drive/motor feedback.

Edit: oh, I see. Missed the part where it says "hook a differential linear scale direct to the servo drive".

[joeavaerage]

Hi

Useless, no offence. Won't deal with the mechanical (backlash, screw compensation, etc etc), which is what's all about when you talk "feedback" for machine-tools.

BS, these servos use position feedback using a linear scale just for exactly for backlash or nonlinearity of the screws or racks or whatever. This servo drive has PID built into it. It can be tuned and programmed just like any other
feedback controller including all the tuning aids like an oscilloscope, tunable notch filters, programmable anti-integrator wind and programmable velocity estimators, lead/lag compensators....all the good stuff.

Just as a matter of interest have you ever programmed and tuned two, hell even one, notch filter in LinuxCNC? I have no doubt its possible, but it's not for beginners. Well, this servo has two auto-tunable notch filters.
This servo, well actually servo, drive and linear scale will have a higher closed loop bandwidth than LinuxCNC.

Further it's not just Delta, but all the top tier servo manufacturers have models that have this feature.....called 'load sensing'. That is how machine builders can claim 'full closed loop control' despite using a
protocol like Ethercat, CANOpen or Profibus and any one of a dozen proprietary protocols where the trajectory planning PC at the heart of it is open loop. If any of the servos cannot keep pace with the
trajectory commands they signal 'following error fault, and the central trajectory planner usually EStops.

This means that a simple open loop Arduino can STILL perform 'fully closed loop control', because that closed loop control is handled autonomously by the servo drive. All the 'fully closed loop
control goodness' without having to have some high-end controller or a realtime computer to do it all....that's why the servo manufacturers do it......to make money from people whom want or demand full closed loop
position control but do not have the skill to program a controller to do it.

Craig

https://www.youtube.com/watch?v=krhJwfVfwhY

[joeavaerage]

Hi,
as you can see the big green cable going to the servo drive is the (normal and obligatory) rotary encoder of the motor. That encoder supplies angular feedback for the torque control loop and the velocity control loop.
Field Oriented control servos require rotary feedback encoder with absolutely no backlash otherwise the high bandwidth torque and velocity loops fail.

In this video note how the linear scale can be used to jog the servo. This is because the servo monitors the position of the load by use of a linear scale, and is in fact closing its position loop
using that scale. This is called 'load sensing'. It requires that the servo drive have two encoder inputs, the regular rotary encoder and an optional load sensing encoder.

The best of both worlds....the backlash free high resolution rotary encoder is just perfect for a high bandwidth (ie high accuracy and high dynamic speed) velocity and torque control, and the
linear scale monitors the load with whatever backlash or nonlinearity it has.

In CNC machine glass linear scales are often used, but for smaller and much more precise machinery like optical stages Linear Voltage Differential Transformers (LVDT's) are used as are laser interferometers
for semiconductor equipment with feedback precision of a few tens of nanometers.

Craig

[joeavaerage]

Hi,
the 750W A2 servos (with load sensing feedback) are $100USD more than the 750W B2's (without load sensing) that I use. To be honest when I was looking for servos for my new build mill I did not know about
'load sensing'. Had I done so I might have been persuaded to spend an extra $100USD per servo.

I have my B2's direct connected to ground 32mm BNFN C5 ballscrews from THK. I can reasonably expect and have actually gotten in practice linear accuracy of about 10um . This I have found
entirely adequate. If I wanted to chase those last few microns down to under 5um or so then load sensing would be an obvious way to go. It would require good glass scales, I'm rather dubious that the cheap
Chinese scales have the quality required maintain better than 10um, could be mistaken of course, but I'd want to be sure before I parted with my money to some Chinese outfit with a dubious quality record. Then
I'd need the extra feature of the servos at $100 per axis. If I do all of that then I might reasonably get repeatable accuracy of a few um...to what advantage?

I seldom if ever have to get parts that accurate, and as good as I like to think my machine is, I'll be bound that it flexes more than a few microns when under moderate loads anyway.
As good as linear scales and full closed loop are, I can't see them correcting inaccuracies caused by flexure between axes....so would I end up with parts that are that much better than I do
now? I don't think so.

I have come to the conclusion that linear scales would be nice, but not a cheap undertaking, I'm guessing something like $500 per axis (including only the $100 extra for the load sensing servo).
It's not a priority. What I really want to do is get a decent post for my trunnion fifth axis, a new cast iron headstock (to replace the rather too flexy steel one I have now), a 3kW BT30 ATC spindle
for steel, and a 42000rpm 2.5kW HSK25 spindle for my PCB work, a new coolant pump, filtration and holding tank, through tool coolant, shrink tooling....and then maybe linear scales! It might be
a while before I get down to linear scales!!!

Craig

[Mecanix]

Yeah, read that's possible from the missed part where it read "hook a differential linear scale direct to the servo drive". See edit right after I wrote my initial comment. Fast-read and thought you were making ref to a servo encoder, of which too many calls and swear by the "closed/feedback".

And unfortunately yes, I had to swing a few too many scales and encoders already - someone really should do this for me but since I can't justify a helper/staffer for the home-shop I'm therefore on my own! All too familiar with the oscillating, resonance, chasing, following errors and what not. Good news is Linuxcnc embed a full featured signals/pins/parameters oscilloscope, and so it makes tuning those things quite easy and fun. I'd like to think I'm within +/-10 microns huge max, but that's because of the non-controlled environment e.g. the afternoon sun shining on my shop tools sends off my accuracy pretty damn quick.

All good fun. I like the servo/tech you've advertised, if I had to do it all over again I would consider the option.

[joeavaerage]

Hi,
the auxilliary encoder input is discussed at 53:17 of this video:

https://www.youtube.com/watch?v=etUo_etdgFo&t=3903s

Craig