Jerk Control in machines

[peteeng]

Hi All - I'm interested in how machine controllers, trajectory planners etc and CAM programs control jerk. The machine controller uses position, velocity, acceleration and some use jerk to control the machine. Position is obvious but its still can be a moving target depending on what that position is. Velocity gets us around the table. The velocity profile can be simple (trapezoidal) or more complex (S-curve or parabolic). The more complex velocity profiles aim at minimising acceleration changes which produce jerk. This is done by creating smooth velocity profiles.

At each path inflection point eg a line to an arc the motion can produce infinite accelerations. As the machine moves it dithers, precesses and stick slips slightly producing micro or macro unsteady motion. This can produce "jerk". Stepper motors introduce vibration by their nature.

Smooth velocity curves can minimise accelerations. Accelerations produce inertial forces that contribute to machine vibrations. Large inertial forces can produce machine deflections. Jerk acts in the background as an overlaid agitation causing more vibrations.

Since jerk is a function of path and acceleration it would seem to me that to tackle this issue it has to start at the CAM with the path planner. Early and simple toolpath planners use lines and arcs as these are simple to offset and manipulate. But these have the inflection points. Then it seems they introduced transition curves (clothoids, cyloids) in between these as a subroutine to smooth the path. These days the path planner would use splines as these can be made entirely smooth. To see a little about this heres a vid. It discusses surface curvature but the commentary works for machine toolpathing. If we made the toolpath in G3 or G4 continuity (this G term is not related to Gcode) the paths would be very smooth and produce small accelerations. I presume this is how some HSM roughing codes are produced.

The other issue is that the path planner creates a continuous path and this is then decomposed into polylines. This is because Gcode is point to point. Each small line has an inflection at its start and finish. The controller tells the machine to go to each point and depending on the controllers path settings this path gets smeared back to near the original continuous path created in the CAM. So many issues to cover.

Continuity descriptions | Rhino 3-D modeling



So the smoothest paths would be produced in a CAM using splines, then the controller would inherit that code and use S-curve velocity control to produce smooth motion and toolpath jerk would nearly look after itself. It seems however that path creators are rough and controllers are left to smooth things out. Its even up to some motors to do the jerk control. Which is leaving the issue too close to the end for me. I expect some high end controllers to do the right thing from start to finish.

So any insight into how these things work in the real world, I'd be interested in a discussion here. Peter

Similar Threads:

• Need Help!- Reversing direction causes motors to jerk
• Need Help!- Reversing direction causes motors to jerk
• cnc jerk controller
• I+j arc jerk issue driving me nuts
• A few questions : jerk, offsets, hotkeys .

[dazp1976]

I have 'smoothing' checked in the cam, whether it makes a difference I don't know.

As I speed up the rapids I get more 'jerk' slightly wobbling the stand on direction changes.
As this happens all I do is slow down the accel/decel setting. Seems to do the job but the trade off is it can impact the machining time.

[mactec54]

Hi All - I'm interested in how machine controllers, trajectory planners etc and CAM programs control jerk. The machine controller uses position, velocity, acceleration and some use jerk to control the machine. Position is obvious but its still can be a moving target depending on what that position is. Velocity gets us around the table. The velocity profile can be simple (trapezoidal) or more complex (S-curve or parabolic). The more complex velocity profiles aim at minimising acceleration changes which produce jerk. This is done by creating smooth velocity profiles.

At each path inflection point eg a line to an arc the motion can produce infinite accelerations. As the machine moves it dithers, precesses and stick slips slightly producing micro or macro unsteady motion. This can produce "jerk". Stepper motors introduce vibration by their nature.

Smooth velocity curves can minimise accelerations. Accelerations produce inertial forces that contribute to machine vibrations. Large inertial forces can produce machine deflections. Jerk acts in the background as an overlaid agitation causing more vibrations.

Since jerk is a function of path and acceleration it would seem to me that to tackle this issue it has to start at the CAM with the path planner. Early and simple toolpath planners use lines and arcs as these are simple to offset and manipulate. But these have the inflection points. Then it seems they introduced transition curves (clothoids, cyloids) in between these as a subroutine to smooth the path. These days the path planner would use splines as these can be made entirely smooth. To see a little about this heres a vid. It discusses surface curvature but the commentary works for machine toolpathing. If we made the toolpath in G3 or G4 continuity (this G term is not related to Gcode) the paths would be very smooth and produce small accelerations. I presume this is how some HSM roughing codes are produced.

The other issue is that the path planner creates a continuous path and this is then decomposed into polylines. This is because Gcode is point to point. Each small line has an inflection at its start and finish. The controller tells the machine to go to each point and depending on the controllers path settings this path gets smeared back to near the original continuous path created in the CAM. So many issues to cover.

Continuity descriptions | Rhino 3-D modeling



So the smoothest paths would be produced in a CAM using splines, then the controller would inherit that code and use S-curve velocity control to produce smooth motion and toolpath jerk would nearly look after itself. It seems however that path creators are rough and controllers are left to smooth things out. Its even up to some motors to do the jerk control. Which is leaving the issue too close to the end for me. I expect some high end controllers to do the right thing from start to finish.

So any insight into how these things work in the real world, I'd be interested in a discussion here. Peter

Cam can do some smoothing in reality splines are not smooth moves, so you have lost before you start, controls are very limited as to how much Jerk control they can achieve, SoftServo have patented this process in more than one way, so are years ahead with their control, the user can adjust create their own acceleration / deceleration profile, to suit their machine, the control can never make it perfectly Jerk free just the nature of the problem, of acceleration / deceleration

[mactec54]

I have 'smoothing' checked in the cam, whether it makes a difference I don't know.

As I speed up the rapids I get more 'jerk' slightly wobbling the stand on direction changes.
As this happens all I do is slow down the accel/decel setting. Seems to do the job but the trade off is it can impact the machining time.

Build a better base fill it with concrete, then you will be able to go fast again, the machine will flex then the parts that flex may fail

[peteeng]

Hi Dapz - "jerk" is not unsteady motion. It is unsteady acceleration. What cam do you use? Jerk is often used to describe unsteady motion but at an engineering level this is not an accurate descriptor. Velocity, acceleration and jerk will be present in all motion. The aim is to minimise jerk to create steady accelerations hence steady velocities which we need for good machining. Machine deflections contribute as well but I'd like to discuss how CAM - machine controllers and motors deal with all these dynamics. Peter

As Mactec infers if we have a defined path like in pick and place applications we can tailor a velocity profile, accel/decel to minimise say the cycle time. The usual demonstration of this at trade shows is moving a glass of wine very fast from AtoB without spilling any wine or having the glass fall over. The issue with a machine toolpath is that it changes and changes and changes so you cannot optimise the inertial responses. You can only do the best you can.

Mactec G3 and G4 curvature spline paths are very smooth. If anyone is out there doing toolpath programming could you comment on this??? has CAM gone past lines and arcs? Peter

[mountaindew]

My router has a steel frame and cast iron base plate. I also have about 350lbs of tools, stored in base. This is a 2x4 router area with an estimated weight of 1300 lbs. Running at 120-180ipm tool paths cause the machine to jerk noticeably. I have requested the cam people add a function to slow speed down by a defined percentage when approaching and changing 90 and 180 deg directions. I can smooth corners and produce rounded tool paths of any size but you still run into jerk from direction changes. I noticed Vectric cad-cam provides more control like I mention above. I have never used the program, just read about it in manuals. It is a more router specific software so I kind of figured this was a common gantry machine problem.
With high speed router spindles turning 10k -24k even with single flute cutters you often have to run very fast feeds in materials like plastic or you get poor cuts. This forces user to just watch the machine jerk around ??

[peteeng]

Hi MD - By CAM do you mean the tool pathing program or the machine controller (MC)? Most MC have look ahead and do slow down at corners. Look into the MC manual. If this is done in the toolpathing software then it would be a post processor thing or a setting in CAM, but more likely its a motion controller function. So what CAD do you use, MC do you use as all of these influence the result. eg I use Rhino3D, Alibre for CAD, Mecsoft for CAM and UCCNC for the MC and UC100 for motion control. Depending on the CAD I use influences my results. UCCNC imports dxf for profile work and that influences what I get as well in terms of max speeds and cuts. Its all a mystery sometimes. Peter

re plastics cutting speeds - yes to get the correct chip thickness for the tool requires very fast feeds. On my machine with a 2F 6mm tool it should run at 4m/min (12000rpm) (158in/min) according to the charts for a 0.2mm chip thickness. My machine is limited to 4m/min at the moment. I have to realign the gantry rails a bit better to get the speed up it has a small tight spot at the moment that is an issue for high speeds. 6m/min would be 600rpm for my steppers so starting to lose torque by then. At corners and radii UCCNC slows it to 1200mm/min...

[peteeng]

Hi All - Although we speak about a machine being "jerky" this motion is not the engineering jerk that I am talking about. Lay speak, Jerky motion is due to unsteady accelerations. Accelerations create body forces called inertial forces. Jerk does not create body forces but it does create vibration at elastic connections like bearings and faying surfaces. Limiting jerk, limits vibration and large accelerations. Its also a chicken and egg problem did the large unsteady accel create the jerk or is it the other way round? Unfortunately the term "jerK' gets used incorrectly in common language, but there's no other word to replace it with. Some call it jolt.... Peter

https://en.wikipedia.org/wiki/Jerk_(physics)

https://en.wikipedia.org/wiki/Fourth...es_of_position

[peteeng]

Hi All - Here's a test track for interested people. The 3 tracks are the same length but the top one is a line and arcs, the middle one is an arc and splines and the bottom one is all splines. I get this into uccnc via dxf but it gets broken into small bits. I'm trying to get around it fast but so far I seem to be limited to 1000mm/min. Will program in mecsoft to see how that goes. The jpg shows the curvature combs. You can see the inflection points in the top image. Will read the uccnc manual again. I've attached the track as an iges file in the zip....Peter

[deadlykitten]

hy petteeng writing your nickname, i just wondered, is it your nickname kind of pette enginerring ? like you are an engineer in discuise ?

not unsteady motion. It is unsteady acceleration

not only acceleration, but nothing is steady, not even a simple g01, or even a static position; the corection loop is always active

At each path inflection point eg a line to an arc

yup, you are right, something happens also there, but don't think of it as being an effect of toolpath joint, but as being an effect of axis motion status change ( from forward to stop and/or reversal )

Smooth velocity curves can minimise accelerations.

of course, but then, you can change an acceleration parameter, and end back right where you started; for example, you can go smooth/bumpier, by runing same program with different motion dynamics parameters, like slow/higher acceleration, etc

Jerk acts in the background as an overlaid agitation causing more vibrations.

not always the amount of overlaid aggitation matters; you an add a drop of water to an empty, or to a full glass of water

The velocity profile can be simple (trapezoidal) or more complex (S-curve or parabolic).

trapezoidal for rapids, smoother for feed

Then it seems they introduced transition curves (clothoids, cyloids) in between these as a subroutine to smooth the path

transition motions they manifest less in exact stop mode/droop control, but usually, most of the times, they are there

I presume this is how some HSM roughing codes are produced.

because off all those variables flying arroud, the best hsm toolpath, for a given machine, is that one that will allow the machine to speed up freely, then cut at full power; how motion/arches are short, this is tricky, and one can fall pretty easy in one of these extremes :
... using the machine in a light mode, thus being too carefull
... using the machine in a hard more, leading to useless stress

i said tricky, because balance between axis speed up and cutting is dynamic, chip thinkess is variable, and is ok to be able to get things just right, not only gambling arround; for corners, this balance changes from one pass to another, and there are most complains : how do i manage my hsm toopath for this corner ? this is interesting to do in reality, and recent cam quick solution is to plunge the tool at corner, prior to finishing the toolpath : see, this is actually not a solution to the initial problem, but a way arround it, because solving those accelerations and toolpath arches, etc, involes too much time and a particular approach for each machine, but by plunging, one simply eliminates the problem

So the smoothest paths would be produced in a CAM using splines,

kind of, theoreticall, because is hard for most machines to follow a spline, or the segmentation/polyline asociated with it

but today, an okuma cnc can find a smooth path through a zig-zag line; for example, if toolpath is like a saw teeth, motion output by cnc may be /\/\ or ----; depends on what parameters are there

controllers are left to smooth things out

you can say that again if they can do it, or how ? hmm

is it ok ? what do you think so far ? kindly

[deadlykitten]

i just checked the video that you shared, pls check attached image : an okuma machine can take that corner in several ways, the classical fillet, or a polygon, with it's 2/3 sides setted from parameters

such level of control is rarely used, but it exists it's purpose is to protect or cut'of corners with full control, thus it handles issues asociated with exact stop mode

ps : i have a few spare hours, so here i am

[peteeng]

Hi DK - Not in disguise but I don't have a cape. I'm a mechanical engineer main jobs at the moment is doing stress analysis & detail design on very large mining equipment and designing medical equipment mainly patient lifting equipment. The other days I design/build & sell my cnc machines. In a past life I programmed robots in automotive industry. This thread is intended to figure out how the CAD, CAM, machine control and motion control works out in the field at a deeper level. Seems at the hobby level we are using disjointed systems. Some pro high level systems are longitudinally integrated right through from CAD to the motors. 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.

You mention its hard for a machine to follow a spline but at the Gcode level the machine does not know if its a spline or an arc. Its a series of point to point moves. My experience with UCCNC is that I need to make the segmentation as long as possible to get my speed up. Say 3mm segments. But thats fine for signage but no good for fine work. From CAD I do an dxf export and that seems to be one of the stumbling blocks as this segments the curves. I have started segmenting (polylines) in rhino first so I have better control of the export. I use mecsoft mill for surfaces but I'm starting to look at it deeper for profile work as it has more options to create code especially ramping starts. Its an area of interest that I want to understand better. Thanks for your input. Peter

[deadlykitten]

hy pete, nice background

. Seems at the hobby level we are using disjointed systems. Some pro high level systems are longitudinally integrated right through from CAD to the motors.

is not quite that fancy, only that most are kind-of satisfied with what they have

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.

is not needed to consider such a long chain as cad-cam-gcode-cnc-motion_control_system; last 2 are one and the same, while what's before them is simply the input

g-code is critical to run a machine

cad apeared separately, and only some cad softwares are for mechanics, and from them, some have cam as a module, thus they can be used to design whatever product for whatever industry, and by a single addon, they may take that product to futher metal cutting processing ( cam - postprocessor, etc )

point is that cam software evolves to keep up with machine, not the other way arround, so if you understand the machine, it will be easier to use a cam, but if someone only uses a cam, then there is no guarantee that he knows what the machine is doing

as an intermediate conclusion, is needed to know how the machine will react when g-code changes to put it simple, if a machine won't react as expected, people will start to have questions

Seems there's a lot of smoke around some areas.

okey, let's talk, let's begin with a single linear axis; if it moves too many times back&forth, it may no longer know where it is ?! where am i ? so it needs a reference :
... home position
... absolut encoder
... absolut scale
is that simple

ok, now it knows where it is and is put back to work, back & forth, forever .... so the big management arives, or the cam guy, and says that max speed/feed for that axis is 1000mm/minute, but it seems that it does not travel in reality 1000, but only 400mm/minute, so why ?

let's consider that travel is 100mm, so expectations are for 10 travels/minute ( so to reach the brochure spec of 1000mm/min ), while in reality only 4 travels/minute occur

if i may, what is causing the difference between 1000 and 400mm/min ?

[mountaindew]

Hi MD - By CAM do you mean the tool pathing program or the machine controller (MC)? Most MC have look ahead and do slow down at corners. Look into the MC manual. If this is done in the toolpathing software then it would be a post processor thing or a setting in CAM, but more likely its a motion controller function. So what CAD do you use, MC do you use as all of these influence the result. eg I use Rhino3D, Alibre for CAD, Mecsoft for CAM and UCCNC for the MC and UC100 for motion control. Depending on the CAD I use influences my results. UCCNC imports dxf for profile work and that influences what I get as well in terms of max speeds and cuts. Its all a mystery sometimes. Peter

re plastics cutting speeds - yes to get the correct chip thickness for the tool requires very fast feeds. On my machine with a 2F 6mm tool it should run at 4m/min (12000rpm) (158in/min) according to the charts for a 0.2mm chip thickness. My machine is limited to 4m/min at the moment. I have to realign the gantry rails a bit better to get the speed up it has a small tight spot at the moment that is an issue for high speeds. 6m/min would be 600rpm for my steppers so starting to lose torque by then. At corners and radii UCCNC slows it to 1200mm/min...

I use a Tormach 24r router stock out of the crate. I use Sprutcam for all tool paths and g-code generation. The 24r router uses path pilot for machine control and this is a linux cnc derivative.
I requested to cam people a couple years ago a way to slow the machine feed down by a percentage as it approached direction changes to help mitigate the jerk or shake I see the machine do when changing directions.
Maybe the machine control has the ability to mitigate this and or the cam software might also have strategies and settings that could help this and I have not found them.
The machine does shake a little when spindle changes direction and this has been something I would like to control.

[deadlykitten]

I requested to cam people a couple years ago a way to slow the machine feed down by a percentage as it approached direction changes

hy mountaindew, if you wish, i can make it to auto insert slow down movements, in a constant manner, or mapped, as you wish / kindly

[mactec54]

My router has a steel frame and cast iron base plate. I also have about 350lbs of tools, stored in base. This is a 2x4 router area with an estimated weight of 1300 lbs. Running at 120-180ipm tool paths cause the machine to jerk noticeably. I have requested the cam people add a function to slow speed down by a defined percentage when approaching and changing 90 and 180 deg directions. I can smooth corners and produce rounded tool paths of any size but you still run into jerk from direction changes. I noticed Vectric cad-cam provides more control like I mention above. I have never used the program, just read about it in manuals. It is a more router specific software so I kind of figured this was a common gantry machine problem.
With high speed router spindles turning 10k -24k even with single flute cutters you often have to run very fast feeds in materials like plastic or you get poor cuts. This forces user to just watch the machine jerk around ??

This is something that is normally done in the control, not so much in cam although it can be done in Cam as a friend had some custom software made that after his program was made it was run through this other software to do this optimizing

[mactec54]

i just checked the video that you shared, pls check attached image : an okuma machine can take that corner in several ways, the classical fillet, or a polygon, with it's 2/3 sides setted from parameters

such level of control is rarely used, but it exists it's purpose is to protect or cut'of corners with full control, thus it handles issues asociated with exact stop mode

ps : i have a few spare hours, so here i am

Not all machine controls are created equal, no exact stop involved, just smooth continuous motion

[mactec54]

Hi All - Although we speak about a machine being "jerky" this motion is not the engineering jerk that I am talking about. Lay speak, Jerky motion is due to unsteady accelerations. Accelerations create body forces called inertial forces. Jerk does not create body forces but it does create vibration at elastic connections like bearings and faying surfaces. Limiting jerk, limits vibration and large accelerations. Its also a chicken and egg problem did the large unsteady accel create the jerk or is it the other way round? Unfortunately the term "jerK' gets used incorrectly in common language, but there's no other word to replace it with. Some call it jolt.... Peter

https://en.wikipedia.org/wiki/Jerk_(physics)

https://en.wikipedia.org/wiki/Fourth...es_of_position

There are many articles on Jerk control, machine rigidity is very important, like I said and using S-Curve is the smoothest action for Jerk Control, Modern Machine Shop article I have pasted it here, the other is from Linear Motion Tips is a link to their article. Cam is very limited to what can be done to help Jerk Control

https://www.linearmotiontips.com/how...otion-systems/

Digital signal processing with computer circuits and software has encouraged control engineers to realize more capable control algorithms that today include jerk control. Today, "servo" can be defined as the use of negative feedback for the position, velocity, acceleration and jerk of inertial loads.

"Jerk" is the time rate of change of acceleration. It is the acceleration of the acceleration.

Jerk ramps the acceleration to smooth the velocity. Steps—sharp edges in command values—tend to excite mechanical systems to oscillate at their natural (resonant) frequencies. The bigger the step, the greater this tendency. If a system is not rigid relative to the performance that is expected of it, control over jerk can round the velocity corners. This reduces the amplitudes of the frequencies that excite resonance oscillation. As a result, acceleration factors can be set higher.

With higher acceleration factors, a tool can transition corners and traverse contours of constantly changing curvature at a faster rate, and do so more smoothly when the corners of both the longitudinal and lateral acceleration are ramped. So jerk is yet another aspect of achieving better performance—one that may work with other performance-enhancing control features. In the case of the 840D control from Siemens, for example, jerk control works with additional smoothing in the velocity controller of the servo drive (a feature called Advanced Position Control). However, these details, like so much practical detail of advanced servo control, generally should be transparent to the end user interested in fast, smooth, accurate stock removal.

Jerk control is less important if a mechanical system is very rigid. "Rigidity" here means that the system has high natural frequencies that are not easy to excite. These terms are all relative to the design performance. So, in the end, one can have a less rigid system that still offers good stability when the CNC is provisioned with jerk control. Many big machines are inherently unstable, meaning that they shake at very low frequencies. These machines are particularly difficult to stabilize sufficiently to achieve commercial production rates. Without jerk control and other filtering methods, they might not be practical in the context of today's shorter cycle times and tighter accuracy and surface finish requirements.

Normally, the choice of whether to have jerk control on or off is taken up by the machine tool builder when a particular philosophy is used to tune the individual servos. This is why we say that to a great extent many fine points of a CNC's servo algorithms are transparent to the end user. A factor that should be of more concern to the end user is the support the machine tool builder receives from the control vendor in the optimization of the machine as a whole.

On a machine that takes advantage of this capability, it's possible to see jerk control in action—and not just at corners. Some people have a hard time understanding why axes slow down at tangential transitions. The path seems as though it should be continuous. However, at a tangential transition, a change in curvature occurs. Such a change requires a step change in the sideways acting acceleration. Think about driving through the center point of an S-curve at a high speed. You are pulled first toward one side then the other. When this same effect on the machine tool threatens to exceed a limit set by the machine tool builder, the CNC lowers the feed rate to take the transition at a more appropriate jerk.

[peteeng]

A very good summary & overview mactec. I want to dig into details of how this is achieved though. I understand the math and logic. More interested in how various CAM. machine control and motion control etc systems achieve it (in principle). But as you say these are transparent and maybe proprietary. Peter

Hi MD - I'm not familiar with path pilot (machine controller) but a quick look at its info sheet says that it has a trajectory planner and this is where you want to look. It will have settings to play with to fix what you are describing. It is also wise to check the machine is mechanically correct. jerk can be created by way sloppiness or some sort of clearance in the systems. backlash etc Peter

[peteeng]

Hi MD - lets work with your problem theres enough of the brains trust here to solve it. You have a Tormach 24r in good condition. You use Sprutcam for toolpathing and path pilot is the motion controller. So if you make a square land (150mm x 150mm? say blank is 300mm square and the square land is in a 200mm dia pocket? pocket is 10mm deep? ) and cut timber or aluminium? you say it jerks in the corners. The CAD can't get that wrong so move to sprutcam and lets look at that. Or show us a shape that you do with the issue.... just looked at sprutcam and they do robotics as well. Interesting for me... Peter

https://tormach.com/machines/routers...nc-router.html
https://sprutcam.com/

do you use sprutcam for CAD or another CAD system. Exporting from CAD to CAM can have issues as well.... mainly model tolerances if you are using a nurbs modeller...