Jerk Control in machines

[peteeng]

Summary 12-2022 till 02-2023 "Jerk" control. So there has been a lot of input into this thanks everyone. The outcome has been to find all sorts of approaches to having a machine behave smoother. So I'll summaries here:
1) Trapezoidal velocity planning is simple and common. Motion is limited either by set Vmax or set Accel max. Unfortunately this strategy has high jerk at its "knee" when velocity changes instantly producing mathematically infinite accel and large inertial forces creating vibrations. Since gcode is feed to the machine in strict time segments we can calculate acceleration as we know Vstart and Vfinish at each segment and we know the segment length. So accel=(V-u)/t this is calculated at each segment so the controller knows when to limit the accel. It also limits velocity. The accel is created by the torque capacity of the motor at any point of time reacting to the inertial load upon it.
2) To get rid of the knee "s-curve" velocity profiles are used. These place a bezier curve at the knee to soften the curve. The S does not mean sine or anything special , It just means the curve could look like an S. The length of the curve can be adjusted in some systems. Its an improvement over Trap
3) Input shaping - is a damping technique that uses an added inverse signal to cancel the vibration of the system. Firstly the system is excited and measured to determine its major vibration modes. Then an algorithm is selected to run alongside the velocity profile. This dampens the vibrations. This is the same process as a shock absorber on a wheel or noise cancelling headphones use. 3D printing has used it for many years to improve their machine motion. Should be used on CNC more often. Superior to S curve as it includes the actual machine responses vs guessing what that response is
4) Active feedback - This takes a signal from an encoder, usually acceleration (can be displacement as well) and uses similar math to the input shaping to reduce vibration by smoothing velocity. This is usually by a PID controller in the loop. This is common in servo systems. It is better then input shaping as it is working with the inertial conditions in real time.

For shaping and active systems to work there has to be co-operation between the machine controller developers and the motion control developers. Since these have been divided tasks in the hobby CNC market "advanced" motion has been impossible. Companies like UNCNC could integrate their Machine control with their motion control and implement shaping. 3D printer controllers are all on one board so can do this easily which they have done. Many printers now have on head accelerometers and its logical to expect that they will implement active systems very soon as shaping has become the norm.

Thanks again hopefully this has been useful for someone. It has for me and I shall be using a 3D printer board on the next build and implementing input shaping. Probably Marlin controller and Duet3D board. Peter

[deadlykitten]

ploters operate in controled enviroment, while cnc machines are for hazard conditions

a controlled enviroment requires only initial calibration, while hazard requires the ability to react to unknown i think this can be seen as the line between a hobby and a profesional system

Q is block compression. best to set it to 0 always. What it does is collapse lines of code within a specified angle of each other, so you get less lines, longer segments and faster motion. sounds fine in theory but in practice you can get very out of square cuts.

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.

there is something to add here; i remebered it when i was away from pc, but forgot when i was posting

about cnc's capable of geometry manipulation; a cam system has it's own aproximation methods, so adding another aproximation at the cnc is a bit useless

for example, if you wish for 0.5 precision, is nonsense to use cam with 0.4, and cnc with 0.1, is like having to split/balance errors between cam and cnc

one good method to adress it, is to keep all math inside the cam, and generate a toolpath with a minimal number of elements ( like to avoid a too crowded code, only to put useless stress on the cnc )

for example, if there are only lines, then the cnc should have a larger ( maxxed ) value for roughing, and technical drawing tolerance for finishing

thus cam to handle only static, only shape related things, and the cnc to handle only motion related things

in practice, a common aproach with splines, is to segment the toolpath way too much ( very very dense code at 3-5um), then cnc it with ~0.01 deviation, thus to be sure that the solid gets recreated almost 'identical' in reality; this aproach can lead to increased cycle time( even if buffer size is bigger, machine is doing too much math, and advancing too little ); a better aproach, is to filter the toolpath inside the cam, just how the cnc does, so to have a g-code as simple as it could get, but this requires a skilled programmer

to meet this demand, more like to meet this situation off too dense codes, the cnc algorithms got even better at being able to ignore useless points, and aproximate only what remains ...

so, if cnc has such numerical aproximations, one can dump in it whatever trash code it has; otherwise, achiving short cycle times, requires skills / kindly

[JR107]

I noticed this thread specifically a while back and wanted to throw some insight into the forums.

It is hard to pinpoint a good response since the initial disscussion/s is a more generalized question & theory of how the large CNC's controlled and accounted for jerk. After reading the entire thread, it shifts toward the hobby-oriented side of what is possible. I bring this up because apart from theory, both are somewhat different in the practical implementation of error corrections. It should go without saying, but I'll reiterate that when all is said and done, a 60-year-old manual Bridgeport will still give a higher-quality surface finish than a benchtop CNC imported from China. There is only so much customized software + CAD/CAM can fix; you quickly realize the sunk cost fallacy.

The above is essential when looking at large "commercial" CNC machines. The kinematics behind their base operation is already good enough for 70%+ of their work potential. You would be surprised, because they do not require many advanced adjustments to motion. Any higher percentage value becomes a cost/complexity tradeoff that increases exponentially. 10-15% additional percentage points on that arbitrary value is simply in the form of adequately configured - closed loop; error correction routines. For brevity, I am placing servo motor motion profiles in this category. Pick your poison, as the velocity and its derivatives within the servo profile, can be optimized for different factors. Adding ACTUAL linear encoders on all the axes may yield an additional 5% to the potential accuracy, even within a simple PID loop. Someone reading this will instantly jump to comment, "but my servos already have an encoder with xx-bits or x-million PPR of positional accuracy," or (my favorite), "glass scales are only for GD&T / checking thermal growth," failing to realize that a servo is essentially a DC motor with an encoder stuck on the back; once you figure in torque is applied perpendicular to an axis of rotation: regardless of ballscrews, way type, etc.. roll, yaw, and pitch errors are introduced. As the bed moves further from the servo, the probability of the reported position being exact decreases with simple physics. Speeding up latency on the control network with something like EtherCAT will yield an additional 2-3% of potential.

Now comes the fun part. There is an inherent disconnect between CAM software and commercial CNC controllers. Many (so many it will make your head hurt) mathematical methods exist to optimize motion and tool path planning. All require access to the control loop of the servos. In the first step of tuning the servo profiles, errors along each respective axis are often parameterized and then transformed into a 3-D vector model in joint space. An equation is estimated from this three-dimensional space as an error mapping function. This approach is as "black box" as it gets; low cost and is processed entirely behind the scenes by the central controller. G-code passes through, and the machine passively corrects the errors. *errors as in erroneous values of the derivatives of position, pre-determined by whomever.*

The next question is how spindle load and feed rate factor into the equation. The most recognized approach is constraining the jerk term as a function of path displacement while still within a servo control loop. This method uses quintic splines to smooth consecutive steps, regardless of the toolpath from the CAM software. Keeping the step size constant (if the base servo profile is configured correctly) allows the spline chord lengths to be recursively updated using the previously known position for the following position. A 10th-order polynomial is solved in real-time. This is all happening within the servo control loop. So instead of move here -> accelerate -> decelerate etc., every consecutive movement is fitted to a spline for smoothing.

Without direct access to a machine's servo controller loop commands, there aren't many things to "optimize." Using glass scales along each axis is the easiest way to overcome this pitfall in commercial machines. Because if your linear encoder along any given axis has roughly 5-10x the resolution of the tolerance you are trying to hold, the general philosophy is that the manufacturer's controller should have enough sampling bandwidth to adjust the motion profiles for smoothness.


I tried keeping all of that as non-mathematician-friendly as possible for other readers. Still, I can easily derive almost any significant motion planning optimization algorithm you care to ask about. It all boils down to implementation, and once again, there is a massive disconnect between CAM and Machine. 5-axis also gets interesting since all of the above no longer works.

[peteeng]

Hi JR - Thanks for chipping in. The Maker/Hobby and to some extent the commercial suppliers do have an issue with the disconnect as you say of the CAM SW, the machine controller, motor drivers and the motion planner. This is a function of how the industry has developed over the last 40 years as each component was and is still seen as a specialist area. Only the large commercial players have integrated the functions. Contrary to this the 3D printer market promptly integrated all of this onto one board 1) as it makes sense 2) costs less to make one board vs 3 or more and 3) they require the performance as its a developing performance driven/ cost down market vs maker level CNC which is small time slow moving tech. So I think the printer tech has gone past the CNC tech and it will become the defacto stuff we use very soon... I was in an electronics shop the other day and there were some school kids there discussion an arduino project. If they were to build a CNC they would not use Mach4/MC and and motion controller etc. They'd dive into big tree or similar. ... I'm helping a guy in Cairo build a router and he's the same, very comfortable with 3D printer stuff so is going that way with the control and input shaping. Peter

[peteeng]

HI JR - This disconnect you mention reminds me of when I first got involved in CNC stuff some 35 years ago. I was making some components in a project via water jet cutting then finish machining. I came across our factory neighbor one day and found he had a CNC mill so I said can you make these? He said yes but he could not program the part as it was complex (complex for then even though it was 2.5 axis work). So I found a programmer but he wanted a CAD file or a point table. CAD programs were really $$$ pro-engineer cost over $30000AUD at the time and the computer was nearly as much. So I sat down and manually calculated to 4 dec places all the inflection points on the part. Took this to the programmer and he charged $250 (a weeks wages for me) for the program on a floppy disc and I took that to the machinist at which point he said what's this? So he got a reader installed and off we went. I did a few jobs like this for the machinist after that then bought a wireframe cad and made the front end a bit easier. Now you can download Fusion360 hobby and get 1000x more stuff for free! We have come along way but we still have a long way to go. Technology does not stop. Peter

[JR107]

A critical thing to note is that as 3D printer motion controllers become more and more advanced, their effectiveness with CNC machining work will diminish. The math involved in control planning additive-based manufacturing differs from subtractive-based, primarily due to loads and tool displacements.

I agree that somewhere in the future, there will be an inflection point where companies realize they can likely produce better parts with a custom machine that can control every last parameter than one purchased due to convenience and distributor relationships.

Worth looking into is Granite Devices and Trinamic. Both companies are working towards more user-defined functionality within motion control. ICAM also gets added to the list, but they are focused on the CAM/post-processing side of things. Supposedly they offer software to build your own post-processing from scratch, though I hesitate to know the cost of that one.

I do a lot of research with LabVIEW, and can vouch that the functionality is there. However, I would not consider that practical in any working industrial environment. If you think AutoDesk nickel and dimes - N.I. is worse.

[peteeng]

Hi JR - Trinamic is already well embedded in printer tech, they particularly were embraced due to their resultant zero motor noise. This is important for printers because they are domestic appliances. I've had a quick look at Granite Devices and will investigate further. I think once printers go active we will be able to tune both ways. My planned printer will have an integrated accelerometer at the head so it would not know if its a printer or a spindle... time will tell...Thanks Peter

[JR107]

HI JR - This disconnect you mention reminds me of when I first got involved in CNC stuff some 35 years ago. I was making some components in a project via water jet cutting then finish machining. I came across our factory neighbor one day and found he had a CNC mill so I said can you make these? He said yes but he could not program the part as it was complex (complex for then even though it was 2.5 axis work). So I found a programmer but he wanted a CAD file or a point table. CAD programs were really $$$ pro-engineer cost over $30000AUD at the time and the computer was nearly as much. So I sat down and manually calculated to 4 dec places all the inflection points on the part. Took this to the programmer and he charged $250 (a weeks wages for me) for the program on a floppy disc and I took that to the machinist at which point he said what's this? So he got a reader installed and off we went. I did a few jobs like this for the machinist after that then bought a wireframe cad and made the front end a bit easier. Now you can download Fusion360 hobby and get 1000x more stuff for free! We have come along way but we still have a long way to go. Technology does not stop. Peter

Funny you mention this; quite a few years back I bought an old machine with an Heidenhain controller for the sole purpose of trying to hack it. It was one of the only controllers that I could find with the ability to actively read / write data via rs-232. The internal memory was so small that you could only realistically fetch data via rs-232. The solution was to trick the controller into using the floppy drive read commands from a different data line as the overwrite. So I was retrieving real time data directly onto a pc -> attempting to calculate things (5 axis stuff)-> drip feeding the updated G-code to a virtual machine running DOS -> which then drip fed back to the controller. Making use of some clever if & then commands in the G-code as well as temporary stops allowed a semi working controller. I quickly realized how much of a train wreck the idea was and abandoned the project.

[deadlykitten]

4) Active feedback - This takes a signal from an encoder, usually acceleration (can be displacement as well) and uses similar math to the input shaping to reduce vibration by smoothing velocity

he pete that is true for a 3d printer head, going way too fast, and needed to be slowed down ... while most common cnc systems, don't sense vibration, for example drilling with a big tool will push the machine to it's limits, and reaction loop will simply make it vibrate more and more, and machine won't stop unless the overtorque is reached ( if they have overtorque, otherwse, it keeps going ... damage may occur ): this aplies to almost all cnc machines, unless expensive ones ( protected like servtech described; idea is that they have more than the basic loop )

so, there is vibration when initial machine is calibrated ( like video you shared with rextorh ), and also from the hazard enviroment ... they are both vibrations, but they are sensed different

thus, reaction is not always to reduce speed ... depends on what the loop is there for

So I think the printer tech has gone past the CNC tech

with printers demand growing, some solutions arrived; new market; as how those may be better than those for an equivalent cnc ?! i can't see this through ...

i mean, there is like this :
... plooters :
...... edm : repetitive fast motion; same thing over and over
...... wire : slow motion
...... routers : fast motion, all over the place
...... printers : even faster motion
... cnc machine : enviroment bites/fightes back

i mean there are some similarities between systems with kind of identical requirement, but still, the range is soo large, so to say that printer tech has gone past cnc tech

for example, high end machines have aditive and laser heads, but if their cnc system would be developed only for those heads, that machine would not be able to cut metal anymore ... it would loose it's primalry function

I first got involved in CNC stuff some 35 years ago. I was making some components ...

good memories

My planned printer will have an integrated accelerometer at the head so it would not know if its a printer or a spindle...

i can't believe you just said that i read it twice .... indeed, it has no clue; zero clue; zeroooooooo

mcs is numbers, algorithms, has it's own mind whatever the g-code ( or the input is )

i am sure you remebet that page you shared, where they described how to adjust the printer, based on vibration pattern found on the printed part, and mactec said that mostly it will go away if mechanical structure is rigid; indeed, the jerk control, the pure control, has .... how to say it ... does not care much of g-code; and is not jerk control, is called motion control system, mcs ... ( handles also jerk, if you wish )

most programmers focus only on cam & gcode, while the critical things are inside the machine, and they are indeed 'invisible'; only a few programers actually deal with cnc motion parameters ( need to understand 1st, then the cnc to allow such acces, then they to have tools to acces those ... is not easy, is not common, and is mostly an off-topic, off-limit subject for usual shop guys; is more relevant to machine builders )

time will tell

the idea with the accelerometer, for a fast moving printer ( or heavy cnc machine at high feed), etc, is that it can tell if the system has stabilized so the next move to begin, or there are still vibrations manifesting

in order for the basic cnc to detect inertia, it must occur at a very high value, so to trigger a big reaction in the system ... it can not, while such an accelerometer, being placed on the head, is way closer to the source of vibrations

how to join accelerometer data into the allready existing system ... i believe this is maybe what you wish for; i don't know

, a 60-year-old manual Bridgeport will still give a higher-quality surface finish than a benchtop CNC imported from China. There is only so much customized software + CAD/CAM can fix;

hy jr yes, you said it ... indeed, a controlled cnc system is not a guarantee; thus its presence ant the fact that it is working as designed, does not mean that it meets client expectation, so to say

so 1st step is data from feedback loop analysys, but also what is that loop monitoring is also important ...

Someone reading this will instantly jump to comment, "but my servos already have an encoder with xx-bits or x-million PPR of positional accuracy," or (my favorite), "glass scales are only for GD&T / checking thermal growth, ... As the bed moves further from the servo

that is so true; i had the 'honor' to have to explain to such persons why the new expensive cnc machine dispplaying 0.001 accuracy on screen, is not stable within 0.01 for some dimensions; once you understand such things, life with cnc machines is a bit easiear

There is an inherent disconnect between CAM software and commercial CNC controllers.

please develop on this a bit; for example, what should a cam do more, or what kind of functions would you like to be there ?

This method uses quintic splines to smooth consecutive steps, regardless of the toolpath from the CAM software. Keeping the step size constant (if the base servo profile is configured correctly) allows the spline chord lengths to be recursively updated using the previously known position for the following position. A 10th-order polynomial is solved in real-time. This is all happening within the servo control loop. So instead of move here -> accelerate -> decelerate etc., every consecutive movement is fitted to a spline for smoothing.

there is a big difference between 'move here -> accelerate -> decelerate' and 'spline smoothing capable machines'; i mean this spline ones are top tier, and next motion is indeed fitted to an allreadt existing one, high degree poly are there ... yes, that's true such spline machines are good as long as the final shape, the object to be machined, is shapy, curvy, otherwise, having splines control active for only lines toolpath, with arches here and there ... there is no reason

5-axis also gets interesting since all of the above no longer works.

if i may ask, please, what is with 5 axis ?

So I was retrieving real time data directly onto a pc -> attempting to calculate things (5 axis stuff)

please, can you develop ? thank you

[peteeng]

Hi DK - I think most of what you ask has been covered before in the thread. My aim is to have smoother motion than is available with Mach4 and UCCNC and similar machine controllers. I believe input shaping is a viable and economic way to achieve this at the Maker/Hobby level and indeed the printer controllers have achieved this via integration of the machine controller/motor drivers and motion control. And they will get better whereas the CNC tech has been stagnant for a long time due to the disconnected components (or toolchain as it is called) and their developers and the lack of "push back or forward" from the cnc users to achieve better products. All of the technicalities of how to do it are out there. Peter

[JR107]

please develop on this a bit; for example, what should a cam do more, or what kind of functions would you like to be there ?

if i may ask, please, what is with 5 axis ?

please, can you develop ? thank you

To best clarify the comment about a disconnect between the software and machine code, I should explain where my mind is at. If, for example, Haas (or any other brand) decided to open their machine control system up for "developers," I would go tomorrow and drop $250,000 on one of their VMCs. You can tune and optimize a ridiculous amount of things if you know what you are doing on the software and math side. As it stands, that opportunity does not exist.

A machine builder goes to AutoDesk/Creo/Dassualt/Siemens and requests a new post-processer development, and between the engineering department at the machine builder and software engineers at the CAM company, they develop a system that roughly optimizes the g-code for their line of products. Give or take, based on conversations I have had with others in the field. I would love input from an expert on the exact process.

But there is no per-part optimization, and there is 100% no way to bake every possible part a machine can make into a general model. And at the end of the day, there is no direct communication between the machine and the model in CAM. So you are left to implement and optimize motion control algorithms on a machine controller(which is off-limits on commercial machines), and you are left to optimize your CAM/post-processing for that machine. The word adaptive was mentioned a lot in this thread. I prefer to use the word dynamic, as solving 5-axis motion in real-time is a large complex dynamic system problem. It is possible. Every industrial robot arm does this, but you need direct access to your control loop to implement it.

While technically, what I initially said does work, you run into more difficulties; that either become intractable, undifferentiable, or non-optimized. When modeling five axes, you are no longer dealing with singular prismatic joints since you introduce two or three (depending on the type of machine) revolute joints. These rotational axes present inflection points that must be solved or smoothed, just to name a case. *Do note that I am talking about proper 5-axis movement and not 3+2 work positioning*

If you want me to go into more detail about something specific, I am okay with it. I have been avoiding math equations since, without LaTeX, it quickly becomes unreadable to a normal audience.

[See attachment] is a screenshot of pages 203-204 (211-212 in Adobe PDF) from AutoDesk's HSM Post Processor Training Guide. I would be confident in wagering; most have never even looked through it, but this shows an example of what I have been mentioning. Ignore the fact that the documentation is poorly explained; instead, take note of the phrase/s "minimizes the effects" and "tolerance is optimized for." One should ask: on a machined part, what effects are minimized, and what is the algorithm optimizing?

Well, I will save you all the trouble. These are the two implementations offered as simple on/off options->
-Cone method with linearization turned on:
Affouard, A. & Duc, Emmanuel & Lartigue, Claire & Langeron, Jean-Marie & Bourdet, Pierre. (2004). Avoiding 5-axis singularities using tool path deformation. International Journal of Machine Tools and Manufacture. 44. 415-425.

-Massage method with linearization turned off:
Tournier, Christophe & Castagnetti, Cedric & Lavernhe, Sylvain & Avellan, François. (2006). Tool Path Generation And Post-Processor Issues In Five-Axis High Speed Machining Of Hydro Turbine Blades. Fifth International Conference on High Speed Machining.

Believe it or not, that specific inflection point instance CAN be solved AND smoothed if you constrain the jerk term of the servo's motion profile. Neither of which the two methods above can make use of.

[deadlykitten]

You can tune and optimize a ridiculous amount of things if you know what you are doing on the software and math side. As it stands, that opportunity does not exi

hy joe indeed; more or less, one would have to fight it's way in ... since you mentioned you 'locked' to that heidenhain and got real data back made me curious, like what exactly did you got, and what did you do next ? such things opens many posibilities, at least at theoretical level; if you wish, i would like to chat more with you on wathsap/telgram/icq, etc, thus a place where the conversation could occur faster

it should be interseting, because many only think what would they do if they could get in, while you did get in

A machine builder goes to AutoDesk/Creo/Dassualt/Siemens and requests a new post-processer

you put it very gentle, but in reality, a postprocess developer may have to actually do sometihng to grab the atention from the cnc producer, like they can always chose another, have preferences

i know situations, with reasons from both cnc rep, and cam rep, cutting one loose without aparent explanations things happen

and most clients are satisfies with vanilla versions ... basics

whatever this direction is, in the end, the postprocessor may be ... $%^&%$#, at least it does spin the spindle

I would love input from an expert on the exact process

i am much into okuma machines, and i have had discusions with diff posptrocesor agents, and when discusios shifted towards special functions, things prety soon hit into a wall as years have passed, i started being able to manage better such things on my own, thus reducing dependency to a postprocessor, yet being able to handle a larger amount of different softwares

But there is no per-part optimization, and there is 100% no way to bake every possible part a machine can make into a general model

yes ... most optimizations can occur at subgroups of operations for example, to define 3 sets of opertaion groups, and part A requires set 1 + 3, while part B requires set 2 +3

such things can and are implemented easy, while tring to optimize at 100% is ... i think is possible somehow to a specific family of parts

The word adaptive was mentioned a lot in this thread. I prefer to use the word dynamic

onestly, i also did not like 'adaptive', from the 1st time i hear it it sound too technical ...dynamic, live, reactive; at some point, does not matter anymore how you name it

One should ask: on a machined part, what effects are minimized, and what is the algorithm optimizing?

singularities

congratulations for breaking down to the core of such 'generalistic' informations, as shown into that postprocessor guide; very few go that path, from various reasons, and the one that do, may have different aproaches to it

requires critical thinking, and tools to push it, and a favourable background ... what could do a simple operator that deals with financial strugles ? imagine, he can not find the peace of mind to be able to questions such things, while maybe he could if only in a more relaxed enviroment

Believe it or not, that specific inflection point instance CAN be solved AND smoothed if you constrain the jerk term of the servo's motion profile. Neither of which the two methods above can make use of

i think you reffer to the machine having a different cinematic only arround the singularity, then going back to normal, like to avoid a too fast spin

most machines only use one profile, and some advanced ones may be able to calibrate on demand, while actually changing the profile among a toolpath, is hard; i tought of it, i have done such things, but methods are not common

i will send you contact data on a private message / kindly

[JR107]

since you mentioned you 'locked' to that heidenhain and got real data back made me curious, like what exactly did you got, and what did you do next ? such things opens many posibilities, at least at theoretical level; if you wish, i would like to chat more with you on wathsap/telgram/icq, etc, thus a place where the conversation could occur faster

it should be interseting, because many only think what would they do if they could get in, while you did get in

Yeah feel free to send me a PM.

On a few of the older machines you can get some of the "real-time" positional data as well as spindle speeds through the RS-232 connections, and then you essentially bootstrap into the PLC logic through the EEPROMs. Although there isnt much to be gained from the PLC since you are still left to reverse engineer the data.

I am surprised nobody has grilled me yet on the fact that Fanuc offers direct communication through FOCAS / C - Executor and an OCP connection. As someone with a decent amount of experience with it; it is a hot mess. And you still dont get full control of everything. 10/10 times id rather start from scratch.

[peteeng]

Hi All - This makes using BTT boards very easy. Will research this a bit more.... Peter

BigTreeTech to Release Open-Source Klipper-Based Tablet | All3DP

[peteeng]

Hi All - I have cleared out a space to rebuild one of my small cnc platforms into a test machine for a high voltage high speed control system. I was to use the BTT stuff but one of my clients uses the Nighthawk controller and so for his 8x4ft machine I was going to go with that, so he would be familiar. So I spoke to the Nighthawk people and they are implementing input shaping plus they are releasing a 6 axis version of their controller. This lines up really well with a couple of future projects. So they are only 1 hours drive away from me so I shall obtain one of their controllers and get familiar.... that's very exciting in a way, they seem really focused on maximum product development, great to see. This is also good as their controller is for CNC vs printing so it has the full gcode available. It also supports plasma, laser and printing so many opportunities. And you can run your machine from your phone, wireless! bargain...You can use their one box controller or buy the board only and run high voltage drivers or any driver you like. Many many options... Peter

https://www.cnc3d.com.au/nighthawkcnc-controller

[deadlykitten]

hello i was on batteries today, and i remembered this discusion

i shared okuma's documents for spline's, as a general info ... but such things are so 'unique', that most okuma users that try to use them, may end up with some average machine settings, or give up, or mess up the controller parameters in a way that requires servo retuning by specialized personal, as some parameters are not to be messed up with

some okuma employes may reveal some technical informations, a piece here, another there, but for regular shops, the knowledge is hard to get, as it requires math skills, good understanding of the controller, ways to input logic in the postrocessors, debuging skils, quantify change-effect, etc ... that's not a regular combo of skills

in short, is a top specialized function, developed from other functions, and it shares parts with other servo tuning functions, and very few persons are into cnc functions good understanding

implementention is different in cnc brands, different extent / kindly

[deadlykitten]

a little tip : if you wish to cut a curvy shape, but your machine has no special functions, then is needed to break the toolpath in small segments g01s

the smaller the segment, the harder for the cnc to follow, so is up to you to find the breakpoint of your machine for okuma, is 1/7500, thus a 1mm segment can be executed at F7500mm/min max speed

such a trick allows you to machine within balance, thus at the limit where cnc still delivers; you may go beyond that point as long as your part will look ok, but the cost is cycle time, that will start to increase pretty much

another thing to consider, is that toolpath segments should not be equal, but dependant on curvature and desired tolerance, and implementing this should deliver a much smaller program size, thus the buffer won't be overloaded, so again, machine limit will be reached later, as computational time will be able to stay in front of execution

this requires machine behaviour tests, and software logic / kindly

[jackjr-123]

Hi Peter,

I think this is relevant to what you're looking for:

[peteeng]

Hi Jack - Thks for the link .I understand this at an "in-principle" level. The receding horizon approach I have not seen before. The maths have to be run in loops each loop gets better until a converged answer is achieved at the desired tolerance.

This work attempts to do the calc in real time. This is possible these days but it could be done as an adaptive post process before it gets to the machine. Its a trajectory planner and I'm tending to think that it has to be coupled with the machine mechanics (maybe the mechanics are more important). No good having a wizz bang jerk free trajectory on a wimpy noodle machine. I'm setting up a machine with a Nighthawk (NH) controller. The nighthawk people are implementing input shaping so I wait for that. In the mean time I can run Klipper into the NH controller and have a play with input shaping (IS).. When I get IS running I'll report here. Thks Peter

[jackjr-123]

This work attempts to do the calc in real time. This is possible these days but it could be done as an adaptive post process before it gets to the machine. Its a trajectory planner and I'm tending to think that it has to be coupled with the machine mechanics (maybe the mechanics are more important).

Actually their trajectory planner (it's open source, you can check it out) has no realtime requirements. As long as you provide the machine parameters (max velocity, accel and jerk), the trajectory can be computed entirely offline.

I'm also curious about these input shaping algorithms used in 3D printers, but I fear they are not up to the task for a machine tool where cutting loads are constantly changing. Please keep us posted.