Ah, so I misunderstood. You are using a constant acceleration setting just like Mach4? If yes, then you are still doing well at 30i/s^2. I missed your build log (I’ll have to go back and find it, I’m curious now :-)), so I don’t know what all you have hanging off the gantry besides the laser, but with my 4hp Hitec spindle and the laser + mount, I probably have around 0.01" p-p at 30i/s^2 when I do a rapid and then laser a line perpendicular right after the move. It settles out in about 3-4 cycles. That is ok for cutting cabinet panels, but I don’t like it for Vcarving or laser work, so I usually run at 15i/s^2 and that keeps everthing pretty smooth and doesn’t cost me much execution time.
1 Like
Yeah, linuxcnc has a fixed (in the config) max acceleration and uses that to do a trapezoidal velocity curve, so every move is 30 I/s^2 or less. I don’t know a way to modify the acceleration limits in gcode, although there are ways to modify the path smoothing to avoid the need for acceleration.
I have a 3HP S30C spindle hanging off Z with the laser, plus a pneumatic valve assembly. My Z also has a pair of constant tension springs offsetting the weight.
0.005" sounds about right. If you knock that speed down to a more reasonable G01 move then you will get the spec’d 0.003" accuracy these machines are advertised at.
The oscillation should “in theory” be around 734hz.
And the S30C does stick out a good bit further than the GMT spindles do from the Z carriage. I never measured it but that arm is larger for us.
Did they changed the specs in the advertisements? I remember 0.003" for accuracy. But 0.005" is what we all see with fresh bearings and pinion gears.
- Resolution:
- X/Y: 0.0005" (0.0127 mm)
- Z: 0.0002" (0.005 mm)
- Repeatability: +/- 0.002" (0.05 mm)
- Accuracy: +/- 0.005" (0.127 mm) or better
So your machine is considered in spec with the 0.005" oscillation.
Case closed! 
Your machine is as advertised.
Woot!
I’m pretty sure my servos are hitting the X/Y/Z specs, too.
1 Like
Replying to a dead thread here, hopefully you’ll see this.
A TMD is a good idea when some external excitation that you can’t control is pumping a structure near a resonant frequency. You can’t control the wind on a building or the tail of your F-15.
But you can control the stepper which is exciting your printer. Or at least, the guys that implemented Klipper introduced new hardware capabilities to make the control possible. The original stepper controls for a filament printer (“Marlin”) used a single processor to interpret the G-code and drive signals to the stepper. Klipper has two processors. The one directly controlling the steppers is running very fast and precisely timing the stepper control signals, and that’s it. If I understand correctly, each step is individually timed. The other processor interprets the G-code and sends some kind of command stream to the fast controller.
Klipper is not varying the maximum acceleration or speed. It has another parameter for each axis, which is the resonance frequency. The idea is to filter the path driven so that a fourier transform of the motion would show a notch at the resonant frequency. You don’t pump energy into the structure at resonance in the first place. The transformed path looks the same as the original, but with subtle changes in speed along the way.
Doing this on one axis isn’t too hard. Doing it in two axes is more tricky, since changes in speed in one axis affect the changes you must make in the other axis.
This demo shows it in one axis, though without an explanation it just looks like magic:
Input shaping demonstration
Once you have input shaping, it changes the constraints on your machine design. Without input shaping you’d like your structure to be as “dead” as possible, to reduce ringing, and have a high resonant frequency (light and stiff) so that for a given amount of energy the amplitude is low. But dead means low Q means a broad resonance, which means there’s more for the input shaper to notch out. And high resonant frequency means it’s pretty hopeless to measure it in real time and respond.
If you know you have input shaping (and you are really confident in it), you can design the structure to be high Q: long ring times over a very narrow range of frequencies, which you never stimulate. If you have feedback, then you can go further and design the structure to resonate at low frequency which you can measure and respond to.
I don’t think Klipper uses feedback, so the system design there is just stiff and light, with large separations between the various resonant frequencies (so that you don’t get energy sloshing back and forth between two different modes).
Klipper is unimpressive. I have people tell me how good it is and then I make them sit in from of my machines and explain why a $300 Amazon special can print just as good.
Sorry for calling it garbage earlier.
BUT, TMD will reduce ringing in CNC systems and is implemented in various ways. Here is just one;
What you are really looking for is the Teknic Systems Meridian controller;
Yep, I’ve seen that g-Stop video before, and had intended to use it, but my google-fu wasn’t working yesterday. It’s a better example than the one I found.
I was totally unaware of the Silent Tools product. That’s an awesome idea and well explained in the video. Thank you for sharing that.
I don’t know much about Klipper vs Marlin as systems, vs the software used by Bambu or Ultimaker. But the input shaping implemented by Klipper should be capable of lower total positional error than a perfect tuned mass damper. TMD is a way to lower Q without the need for a more stable reference frame, ideally getting down to critical damping. That means you still get at least one cycle. Input shaping eliminates the source of the vibration entirely, so you can get down to zero cycles… so long as you have control of the input.
Ultimaker is Cura and funds a huge portion of Marlin development.
I am a Ultimaker guy. I quit buying that other stuff years ago. My last “other brand” is a Lulzbot Taz 6 which is still producing gaskets in my shop because it can print the crazy soft TPUs.
Klipper will be something to look at in 10 years. Today I can just adjust the belts correctly on a $300 Amazon special and produce the same results as a Klipper machine.
But, things will change. They always do.
Yeah, this is where you and I depart. I cannot think of a single major manufacturer of CNC that does not utilize TMD in their precision systems. If you know one I would be interested in hearing about them.
Now, putting it into the tool is pretty darn cool.