[peteeng]

Hello DK _ I'll answer yur Q's from the image:
1) curvature continuity is described by G1, G2, G3 and G4 continuity. G4 being the smoothest. When the cam system creates a spline it will not be initially smooth. It has to be analysed and smoothed. This is what adaptions are they are cycles in which the derivative of the curvature is looked at and then the curve is adjusted until it is to the required smoothness.
2) Just because we have a "toolpath" does not mean it is acceptable to the machine controller. The MC checks if the path is within its settings. The MC decimates the points that either create an out of tolerance path or can be combined into a longer path. A MC cannot add points.
3) It goes back two steps to the analysis block for more smoothing. The gcode tesselator can only tesselate continuous paths nothing more
4) The CAM has an adaptive loop that goes from toolpath creation to toolpath analysis and round and around until it meets the desired settings. Some software runs forward kinematics and reverse kinematics. This is good for crash detection.

Now I use rhino and make nearly everything from splines import into UCCNC and its quite happy to run stuff. Smoothness of a line is quantified via its curvature. Look up G3 and G4 continuity

You find the right amount of points by learning about curvature and continuity. For instance I loft boats and they need to be smooth & fair. Computers do a bad job of lofting boats you have to be careful how they are done.

CAM does not factor machine attributes that's why you need mechanical feedback for the "best" system. Otherwise you do motion trails ramping up accels until the machine complains then back down a bit. This is not an ideal way top tune the motion controller. Servos have tuning programs that do a good job.

Computers are fast enough these days to do all the work while looking ahead. That's how TinyG, your machine and all others do it. Post processing is relative to the CAM work. In terms of the hardware it could be called pre-processing depends on which side of the discussion you are on.

You keep talking about lines and arcs they are not smooth. They are discontinuous with infinite changes of curvature at each end. This creates infinite acceleration at each end, this is no good for smooth motion. finishing cuts can be lines and arcs as this is probably the geometry specified but these are done slowly and carefully to avoid the jerks. Roughing & HSM requires fast and smooth motion with no particular defined edge as long as it does not impose on the final surfaces... Peter

DK do you use a CAD/CAM system or are you programming directly to the machine in gcode? if you use CAD cam what do you use?

[deadlykitten]

When the cam system creates a spline it will not be initially smooth

hello pette but of course cam systems can output smooth splines, at accuracy beyond what a cnc system can handle

in which the derivative of the curvature is looked at and then the curve is adjusted until it is to the required smoothness

if machine is expected to interpolate nurbs, then if inital code :
... has nurbs g-code, and :
...... no more constraints are there, then those nurbs from g-code will simply pass through, no smoothness will occur
...... there are constrains, such as :
......... "alowable error", then only if "alowable error > 0" a set of nurbs modeling equations is being produced, and it's solutions will pop up within those initial constraints
......... "maximum feedrate", then solution from the modeling equations will be analyzed also cinematically, and variable feedrate will be mapped in advance
... does not have nurbs, then much more constraints will be used; its a more dificult case, because lack if initial data

is possible to use a full set of constraints even if there are nurbs g-codes, but this means that the cam software and/or programer can not be trusted, and in this case, there will be much bigger problems to worry about

The MC decimates the points that either create an out of tolerance path

if there is a point coming from the toolpath planner, then that point is there to be executed, not to be rejected as being out of tolerance; there is no time/no use to check if a point is somewhere where it shouldn't be ... this shows that the planner is screwed; there is no debate

what the mc does, is that it will try to reach that point closer or a bit distant, thus each point will be checked

A MC cannot add points

but the toolpath planner may add/shift/delete points, only if smoothnes algoritms are used ; is the mc job to be sure that what comes from the planner will be executed

The gcode tesselator can only tesselate continuous paths nothing more

it can handle also uncontinous paths ... there are tolerances for such things, as there are for arches g2/g3

as for nurbs, there are tolerances also for curvature g1 g2; g3 g4 comes out naturally after solving the associated equations

The CAM has an adaptive loop that goes from toolpath creation to toolpath analysis and round and around until it meets the desired settings. Some software runs forward kinematics and reverse kinematics.

only cam software with mcs algortims can do that, but again, is not round&round, is one way shot, and this is very very rare; common cam softwares can't do it

to simply put it, the cnc may toogle between a few cinematic modes, while a cam software is a bit lost, since it does not know how the cnc will behave; many people are satisifed if they have a working postprocessor, and very very few are into mcs algorithms; why should they be ?

This is good for crash detection

a cam software may detect a possible interference, but can not predict a crash; a crash can not be simulated, and is the mc duty to react if it senses a crash

Otherwise you do motion trails ramping up accels until the machine complains then back down a bit. This is not an ideal way top tune the motion controller. Servos have tuning programs that do a good job.

there are the initial factory settings, but also dynamic settings may occur on request, thus they tune themselfs

it is not about an ideal way, but about finding fast the settings that suits the machine for that instant only

they can ramp up at medium value, analyze data, then adopt a biger value that is safe for actual conditions, but still lower than full capability; thus it can go even faster

Computers are fast enough these days to do all the work while looking ahead.

no, such tasks are not done by computers, they can't keep up; is machine's electronics, that simply may use the computer as a side, to show data

those high frequency loops are not to be handled by clasical computers

Post processing is relative to the CAM work

post processing is an atribute of the cnc, without conections to cam

You keep talking about lines and arcs they are not smooth. They are discontinuous with infinite changes of curvature at each end. This creates infinite acceleration at each end, this is no good for smooth motion.

nurbs are even harsher, with more inpredictable changes of curvature, so to say; depends also on cnc dinamic profile, and how curvy the nurbs is ... but simply to put it, there is much more flexibiliy to the nurbs, rather than lines/arches, so the system will work harder

the frequency of the loop is similar for line/circle/nurbs/steady ( thus machine is not moving ), but it's reaction is, or diff control, is more pronounced in high speed nurbs or hsm; to generalize, is harder with higher speeds on low radius profiles, and this can be seen

That's how TinyG, your machine and all others do it.

i don't talk about systems that require a laptop i think i just had a revelation ....

DK do you use a CAD/CAM system or are you programming directly to the machine in gcode? if you use CAD cam what do you use?

i can create programing software, and 2d cad; also, i can record feedback loop data at each 0.008 - 0.015 seconds; if it matters, right now i am working on smoothing toolpaths from program files generated by cam softwares / kindly

[peteeng]

Well DK - set up a thread about NC code smoothing that sounds interesting. Peter

[dazp1976]

a cam software may detect a possible interference, but can not predict a crash; a crash can not be simulated, and is the mc duty to react if it senses a crash

My version of Autodesk HSM.
In some situations if I have something wrong.
As it runs a simulation the tool will turn red in the suspect areas.
In the timeline bar there will be red sections, when mouse hovered over it will say "collision with stock".
In the toolpath information section it will also say "warning, collision with stock".
Looks to be with Z G0 moves with suspect feed height settings and suspect rapid traverses, haven't really spotted anything else trigger it.


It still always depends on you inputting your drawings and stock info correctly though.

[peteeng]

Hi All - And welcome to 2023!! I'm building a 3D printer and came across the Voron using a bigtreetech controller board. This has the capability of placing an accelerometer on the printer head and feeding back accels to the controller hence the controller then adjusts its velocity/accels to smooth the motion. Smooth motion for printers is very important, they are relatively compliant structures and the head moves very fast resulting in head wobble which affects the print. Must be useful for machining as well... I'll get into this more and see if I can bring this across to the routing/milling environment. Peter

[mactec54]

Hi All - And welcome to 2023!! I'm building a 3D printer and came across the Voron using a bigtreetech controller board. This has the capability of placing an accelerometer on the printer head and feeding back accels to the controller hence the controller then adjusts its velocity/accels to smooth the motion. Smooth motion for printers is very important, they are relatively compliant structures and the head moves very fast resulting in head wobble which affects the print. Must be useful for machining as well... I'll get into this more and see if I can bring this across to the routing/milling environment. Peter

If you were using servos with 16 / 20 Bit Encoders smoothness / speed is not an issue, they are doing all these things for stepper use, at the end of the day they don't achieve very much, Steppers are not smooth running, no matter what you do.

[peteeng]

Hi Mactec - If the path is not "smooth" then whether its steppers or servos the motion is not smooth. By "smooth motion" I mean it has minimal or no jerk or snap. Peter

[mactec54]

Hi Mactec - If the path is not "smooth" then whether its steppers or servos the motion is not smooth. By "smooth motion" I mean it has minimal or no jerk or snap. Peter

No, the path does not matter very much when you have quality ac servos, which are the only way you are going to have smooth motion, at max velocity, acceleration / deceleration you should never compare Steppers with Servos

[servtech]

Hi Mactec - If the path is not "smooth" then whether its steppers or servos the motion is not smooth. By "smooth motion" I mean it has minimal or no jerk or snap. Peter

Accelerometers on high end CNC machines are there to protect delicate high speed spindle bearings, in some cases hybrid bearings with a steel race and ceramic balls.

When excessive vibration occurs during machining or bearing failure exceeding system parameters, the feedrate will decrease automatically to stop the vibration and protect the bearings. If vibration continues the machine should/ will shut down.

Yes, it may help the machining process by reducing the feedrate, but on a million dollar machine, it's not there for machining. In fact the reduction in feedrate often ruins the surface finish on expensive titanium parts.

[deadlykitten]

feeding back accels to the controller hence the controller then adjusts its velocity/accels to smooth the motion

yes pete that is finally, controls with feedback loop

having acces to such data, or seeing how the autocorection occurs, should give you a glance

Smooth motion for printers is very important, they are relatively compliant structures and the head moves very fast resulting in head wobble which affects the print.

is not only printers, or head woble ... the loop has no clue that there is a printer

if I can bring this across to the routing/milling environment. Peter

it does not matter, loop has no clue that there is a router, a mill, a lathe, etc ....

the loop can react on predefined paterns ( kind of adjust mode ), or dynamic ( real time )

is all about numbers, is not woblle, or whatever else ... is only numbers statistics

If the path is not "smooth" then whether its steppers or servos the motion is not smooth. By "smooth motion" I mean it has minimal or no jerk or snap.

a machine can perform smooth on a path that looks like /\/\/\; the toolpath from g-code, even if it is sharpy, will deliver increased smoother motion as droop increases, without altering other parameters; and here we go again, right pete ?

the cnc zone letter from last year, showed that this thread have had many views in it, but if i look at it, there is nothing constructive so far, only that you where keep going forward like other replies where not there ... this thread, was not a conversation, was only your stataments ...

---

When excessive vibration occurs during machining or bearing failure exceeding system parameters, the feedrate will decrease automatically to stop the vibration and protect the bearings. If vibration continues the machine should/ will shut down.

okumas can detect this at servo level ... thus, it simply knows the pattern for acceleration, and can display a message once it detects anomalies

Yes, it may help the machining process by reducing the feedrate, but on a million dollar machine, it's not there for machining.

some vibrations are too gentle to be sensed by the servo feedback loop, like those generated by small cutting forces; for example, finishing a large heavy expensive part, may come with vibrations becuase of high rpm, and or chip evacuation, but those vibrations occur near the cutting edge, with no diff for the servo; to protect the expensive part, senzors are there in the spindle head / kindly

[servtech]

yes pete that is finally, controls with feedback loop

having acces to such data, or seeing how the autocorection occurs, should give you a glance



is not only printers, or head woble ... the loop has no clue that there is a printer



it does not matter, loop has no clue that there is a router, a mill, a lathe, etc ....

the loop can react on predefined paterns ( kind of adjust mode ), or dynamic ( real time )

is all about numbers, is not woblle, or whatever else ... is only numbers statistics



a machine can perform smooth on a path that looks like /\/\/\; the toolpath from g-code, even if it is sharpy, will deliver increased smoother motion as droop increases, without altering other parameters; and here we go again, right pete ?

the cnc zone letter from last year, showed that this thread have had many views in it, but if i look at it, there is nothing constructive so far, only that you where keep going forward like other replies where not there ... this thread, was not a conversation, was only your stataments ...

---

okumas can detect this at servo level ... thus, it simply knows the pattern for acceleration, and can display a message once it detects anomalies



some vibrations are too gentle to be sensed by the servo feedback loop, like those generated by small cutting forces; for example, finishing a large heavy expensive part, may come with vibrations becuase of high rpm, and or chip evacuation, but those vibrations occur near the cutting edge, with no diff for the servo; to protect the expensive part, senzors are there in the spindle head / kindly

You misunderstood my point, most CNC machine axis drives can detect excessive peak current and high cutting vibrations and process issues, but accelerometers are often spindle bearing specific. Some spindles can cost $80k to exchange.

[peteeng]

Ahhh Deadly Kitten - As usual your word count is huge yet the message is blurry. Show me the math, the code or the hardware that helps with motion control. Put your typing to good use and as you claim competency in this area discuss smoothing in a clear explanatory manner. That would be helpful. Peter

[deadlykitten]

You misunderstood my point, most CNC machine axis drives can detect excessive peak current and high cutting vibrations and process issues, but accelerometers are often spindle bearing specific. Some spindles can cost $80k to exchange.

hy servtech, yup, i got that, you are right

but on a million dollar machine, it's not there for machining.

i only wished to underline that okuma uses senzors not only to protect the machine, but also to protect the part; even more, it has unique functions to protect the tooling, not available on other brands

in other words, if most senzors should be used to protect mainly the machine, okuma, as a pioneer, is looking also at protecting part & tools, and that is for the client intereset, it helps him there are also brands with agresive marketing/product ideas, that make the client be dependent somehow on their machines, that they do not serve the client, but restricts him somehow from going to other brands

this is under the umbrella of brand philosophy, or relation to client

Show me the math, the code or the hardware that helps with motion control.

hy pete all those are developed by normal people, like you, like him, it does not matter

don't look at math/code/hardware, because those only serve a purpose; i guarantee that cad formulas and cnc hardware is different, yet, in the end, final result is similar, because principles are similar

take a simple linear axis, when does smooth motion occurs ? i always try to make you reffer to the simplest motion, because actual cnc status is way more developed, being able to perform identical on different toolpaths, and you keep rushing into actual systems

from begining, i always sugets go to simplest linear motion, then imagine a feedback loop over it ... yet you go to cad, spline processing, etc

just like how the ffedback loop has no clue that there is a mill/router/lathe, also the feedback loop has no clue that there is a line, or an arch, because the loop is targeting somehow the actual moment, not the planning

things that you try to reffer are too far away one from the other

take it simple : you said you can program accelerations for a linear axis, you don't have the feedback loop, etc; stick to what you can do, then go one step forward, do not jump into modern cnc yet, making supositions, after which comes negative replies; you can not jump that much at this moment, pardon me, at least not last year

no more cad, no more splines discusions ... only a linear axis, motion control and feedback loop : at this level only, to proceed futher, let's clear all doubts / kindly

[peteeng]

Hi DK - Again a large word count and no forwarding of the discussion. I'm satisfied I've found solutions and directions for this issue from this thread and I'm happy for the thread to break. Peter

[deadlykitten]

Here is SoftServos from 2016 this covers some of what they can do at their control, there is no other control that comes close to what they have done, because most of their system they have is covered by Patents

hy mactec please, as for patents, etc, can you short highlight the advantages of such evolved systems ? i know you shared that *.pdf, also presented some stuff across here, but i would apreciate to hear more stuff from you, rather than scrolling throuhg those pages ... please ?

and on patents, how is that the competition always seems to find a way to achieve similar functionality, with a different pattern ? thus the patent covers/protects the way it's done, not what is done, so to say ...

Again a large word count and no forwarding of the discussion

Well DK - set up a thread about NC code smoothing that sounds interestingr

So if you have time it would be good to detail how you run a simple profile for instance from CAD to Gcode to controller to toolpath (and what those systems are) so I or members can get an idea of how these things really work. Seems there's a lot of smoke around some areas

my dear pete, so far, i have not yet started to present how such systems work on okuma, because from the begining, i felt that it won't resonate with you, the system motion control being pretty developed, way above average extra functions; if i will do it, i will hit into explaining details, at the cost of my time; i think it will be a nice to read story, but is not the moment yet

i preffered to stay near your answers, looking for common ground ... and it did not hapen; the common ground is the general loop behaviour, that was more or less presented by others, and at that level you did not conect, yet you jumped to the level of cnc's real spline capable motion, again, sticking to ideas that are not real; i simply noticed to many contradiction, in order to continue

the real issue, is that you framed the discusion from the start, by tring to discuss smoothnes, as a factor of cad/cam & g-code, while in reality those are minor, irelevant factors, because main core issue is the motion controller system, and only it, nothing else

systematicaly, always bringing back into discusions cad g-code and splines, you have somehow ruined it, showing a missconcept, to which you sticked from the start

smoothness can be discussed, but not on the background that you insist of keeping

i tried to find an explanation, as for why you see thing like that, and the only answer i found is that you use a hobby system, and also you have a lot of imagination, so to compensate for what that hobby system is missing i think i was pretty polite in saying it, without exageration

my recomandation, is to stick to a single linear motion, closed loop, and from there, begin to discuss the basic mcs, then progresivly add functionalities over it; to trully understand the mcs, is not needed to talk cad g-code splines math, but principles of how to make the system stable under a dynamic, unpredictible enviroment

perturbation factors are identified, quantified, and implemented inside the mcs as functionalities; a cnc system is more smooth as it can restore desired motion more quickly after a pertubation, with minimal deviation

in all these, jerk control has it's place, but not through splines ... for example, simple motions are actually not 3 stages ( acc constant decc), because deceleration occurs in 2 or more steps, in respect to accuracy class, and that targets different droop control, the system being more reactive at slower speeds, so to ensure desired accuracy

to really go into smoothness of actual cnc systems, you have to let go the things that you are very hard struggling to keep reffering to as important factors, because they don't matter as much as you believe they do, and on the other side, you missed the core; you need a fresh start, and small sure steps, not jumping to ipotesis that are not happening

if i share code for smoothing g-code, which is more complex than math thoeries, or spline related cnc functions, there is not yet a foundation in this thread, and those won't fit in here, and is a time loss ... focus on principles, not on tools / kindly

[peteeng]

Evening All – I'm drawing this thread to a thin end. Thanks to everyone who has contributed. This is a summary of the thread and what I have found due to the thread interaction.

1. Jerk control is done by various means in open loops and closed loops. They can be proprietary and open source
2. The aim of jerk control is “smooth motion” simple as that. CNC's behave better and produce better parts of they move smoothly
3. The middle of the thread got a bit lost but that's how threads go. My aim initially was to try to improve motion via mechanical feedback to the MC but that didn't work out
4. Its clear that servo's are smoother than steppers but that's self evident
5. What did open up was finding some controllers at Maker/Hobby level that had solved most of the motion problems already.

Examples
Tinyg controller – excellent motion control. Open source code that has been implemented in Linuxcnc by someone. This would be an excellent solution for small cncs as the output is limited to 2.5A but the control signals can be ported to use bigger drivers.
https://synthetos.myshopify.com/products/tinyg

Buildbotics– has S curve implementation of its controller
https://buildbotics.com/

I'm sure there are many more S-Curve controllers out there....

Latest information.
I have been looking at 3D printer controllers and they seem to have taken the lead in motion control at a very good
price. Many people have implemented 3D printing controllers on routers as they are cheaper and offer lots of axes and features.

1. 
   
   1. What I am looking at currently is “Klipper” which has “input shaping” via feedback from an accelerator placed at the printing hot end. This feeds back to the controller such as a BigTree Octopus or similar.

https://www.klipper3d.org/Resonance_Compensation.html this may or may not be useful for routers but I'll look into it. Early days in my printer build.

Thanks again – Peter

[mactec54]

Evening All – I'm drawing this thread to a thin end. Thanks to everyone who has contributed. This is a summary of the thread and what I have found due to the thread interaction.

1. Jerk control is done by various means in open loops and closed loops. They can be proprietary and open source
2. The aim of jerk control is “smooth motion” simple as that. CNC's behave better and produce better parts of they move smoothly
3. The middle of the thread got a bit lost but that's how threads go. My aim initially was to try to improve motion via mechanical feedback to the MC but that didn't work out
4. Its clear that servo's are smoother than steppers but that's self evident
5. What did open up was finding some controllers at Maker/Hobby level that had solved most of the motion problems already.

Examples
Tinyg controller – excellent motion control. Open source code that has been implemented in Linuxcnc by someone. This would be an excellent solution for small cncs as the output is limited to 2.5A but the control signals can be ported to use bigger drivers.
https://synthetos.myshopify.com/products/tinyg

Buildbotics– has S curve implementation of its controller
https://buildbotics.com/

I'm sure there are many more S-Curve controllers out there....

Latest information.
I have been looking at 3D printer controllers and they seem to have taken the lead in motion control at a very good
price. Many people have implemented 3D printing controllers on routers as they are cheaper and offer lots of axes and features.

1. 
   
   1. What I am looking at currently is “Klipper” which has “input shaping” via feedback from an accelerator placed at the printing hot end. This feeds back to the controller such as a BigTree Octopus or similar.

https://www.klipper3d.org/Resonance_Compensation.html this may or may not be useful for routers but I'll look into it. Early days in my printer build.

Thanks again – Peter

This is all well and good, if they fixed the mechanics most of this goes away, without any of this added to the control, the Tinyg you posted in another thread was a disaster.

[peteeng]

Hi Mactec - Why was it a disaster? Peter

[mactec54]

Hi Mactec - Why was it a disaster? Peter

If that was a Jerk control demo, it did a very poor job of it, the resonance from the motors did not help, it was painful to watch, everyone is looking for a solution for these low-cost low-resolution machines it does not matter what they do they won't solve this with software alone

[peteeng]

Hi MT - I agree that particular video was not good. But if you look at the Tinyg closer its algorithms are very impressive and can be implemented in other systems as they are open source. The TinyG approach is what I expect best practice open loop motion to be. To go further, closed loop approaches need to be used and these are appearing in 3D printing systems at the cost level that I operate at. The big machine players have proprietary open and closed systems that do amazing things in terms of motion control but they are outside the scope of my machines. Milli my first mill will use servos which is a new world for me.... Peter

In a slight tangent there are a couple of old threads in which flywheels and eccentrics were added to steppers to improve motion (by way of mechanical damping I expect). I can't find if anyone concluded that this was purely a flywheel effect and the eccentrics (small free weights in cavities in the flywheel) where not needed or the eccentrics were functional? Anyone comment on that? Peter