Quieter motors¶
Testing
This is in testing on the Kalico phase-stepping branch and needs its
mainboard firmware. It isn't in a release yet.
The Creator 5's mainboard drives the X, Y and Z motors itself, through FlashForge's MCLib motor control (see VFA compensation). OpenCreator changes two things in it to make the motors quieter, mostly at printing speeds up to about 100 mm/s:
- Step interpolation follows the real move between steps.
- Ripple correction cancels the motors' own wobble, calibrated per motor with the toolhead accelerometer.
Step interpolation¶
Klipper sends the mainboard a step for each 1/16 of a full step. Between steps the mainboard moves the motor smoothly on to the next one, about 20,000 times a second. Stock guesses when the next step comes by assuming it takes as long as the last one. That is right at a steady speed but wrong whenever the speed changes, so the motor runs ahead or falls behind and then snaps back on the next step. Speed changes are constant in a print: acceleration, input shaping and pressure advance all change it.
On OpenCreator the mainboard's step queue already knows exactly when the next step is due, so the motor is moved on at the real rate. After the last step of a move it stays on that step, instead of running nearly a step past it like stock does.
Nothing changes on the Klipper side: homing, endstops and positions work as before.
Ripple correction¶
A stepper motor doesn't turn perfectly evenly. Its detent torque and the shape of its current give it a small wobble at fixed multiples of its electrical cycle (four full steps). On the Creator 5 the strong ones are 2, 4, 6 and 8 times per cycle; 4 is once per full step. The wobble shakes the toolhead at those frequencies, so it moves with speed: at 10 mm/s the once-per-step wobble is at 49 Hz, at 20 mm/s at 99 Hz. That is most of the hum you hear at printing speeds.
Ripple correction adds a small opposite current to each motor at those multiples, timed to the motor's position. The right current depends on the speed: for the once-per-step wobble it shrinks as the speed nears the motor's resonance, flips over, and grows again above it. So each multiple has a table of corrections at five speeds, and the mainboard picks the value for the current speed from the step rate. Just above the fastest and below the slowest calibrated speed the correction fades out, as nothing was measured there.
It is calibrated per motor, and per direction of travel, with the toolhead accelerometer.
Calibrating¶
Dock all the toolheads, then send:
It homes, picks up the first tool and parks in the middle. Then, for each
motor, it moves along the diagonal that turns only that motor and measures the
vibration at each multiple. It takes about 10 minutes a motor and ends with
SAVE_CONFIG. Add AXIS=X or AXIS=Y to do one motor, or SAVE=0 to leave
saving for later.
Each speed it tries prints the vibration before and after:
Run it again after changing motors or belts.
How it finds the correction¶
At one speed, the vibration's strength at a multiple is a simple curve (a quadratic) in the correction's size and timing. Measuring it with no change and with three small test corrections fixes that curve, and the best correction is at its bottom. One more move checks it, and it is only kept if the vibration went down. It does this twice, the second time with smaller test corrections. Both directions are measured on the same out-and-back move.
MCLIB_RIPPLE_MEASURE STEPPER=stepper_x SPEED=30 shows the vibration at each
multiple at one speed, and MCLIB_SET_RIPPLE STEPPER=stepper_x shows the
saved tables.
With VFA compensation¶
VFA compensation also corrects the 1, 2 and 4 times ripples, but tuned near the frame's resonance, so mostly at higher speeds. The mainboard hands over by speed: where a ripple table covers the motor's current speed, it replaces the VFA value for that ripple; outside the table VFA takes over again, blended in between. Keep both calibrated.
Fans during calibration¶
Calibration holds the mainboard and power supply fans at 50% and waits for them to settle before measuring, like input shaper and VFA calibration do. Otherwise they speed up as the mainboard warms during the test, and their changing vibration throws the measurements off.
Measured¶
Creator 5 Pro, toolhead accelerometer, board fans held at 50% throughout. The new firmware (step interpolation plus calibrated ripple correction) is the average of three sessions, against the firmware before these changes. Numbers are how much less the toolhead shakes, in three frequency bands: 10–100 / 100–300 / 300–800 Hz. "Diagonal" moves one motor only.
| Move | Change |
|---|---|
| Y 5 mm/s | −24% / −39% / −58% |
| Y 10 mm/s | −37% / −38% / −52% |
| Y 20 mm/s | −42% / −54% / −45% |
| Y 80 mm/s | −4% / −30% / +1% |
| X 5 mm/s | −40% / −15% / −17% |
| X 10 mm/s | −34% / +18% / −4% |
| X 20 mm/s | −23% / −42% / −36% |
| X 40 mm/s | +2% / −27% / −4% |
| X 80 mm/s | −12% / −22% / −9% |
| X 150 mm/s | −11% / −6% / −4% |
| Diagonal 5 mm/s | −34% / −29% / −16% |
| Diagonal 20 mm/s | −32% / −62% / −14% |
| Diagonal 80 mm/s | +2% / −14% / −13% |
| Slow speed changes | −2 to −17% / −6 to −10% / −5 to −14% |
Repeating the same test 15 minutes apart moves single numbers by up to about ±30%, so read small changes as no change. At the speed each point was calibrated at, the wobble it targets drops by 75 to 95%. The rest of the vibration is the frame and belts, which this doesn't touch; input shaping handles that.
Inspiration: Prusa's phase stepping¶
This work builds on what Prusa Research published about phase stepping on the Original Prusa XL (firmware 6.0) and later the CORE One. They showed that every stepper motor turns slightly unevenly in its own way, which causes both noise and vertical fine artifacts, and that correcting each motor's drive signal, calibrated by moving the head diagonally at several speeds while recording the toolhead accelerometer, removes most of it. Prusa reported over 96% less motor vibration on the XL. Their help article covers the calibration.
What we took from it: per-motor correction of the motor's own ripple, calibrated with the accelerometer on moves that turn one motor at a time, at several speeds. No Prusa code is used. What differs on the Creator 5:
| Prusa phase stepping | OpenCreator | |
|---|---|---|
| Motor driver | Trinamic stepper drivers | The mainboard's own motor control (MCLib), adding a correcting current |
| Steps | Replaced: the firmware drives the motor phases from the motion plan (hence the name) | Kept: Klipper still sends steps, the mainboard follows the real move between them |
| Correction | Per motor and direction | Per motor and direction, with a table of corrections per speed |
| Crash detection | Turned off while it's on | Not affected (skipped step detection is the separate, optional motion guard) |
Why not full phase stepping¶
Prusa's phase stepping has the printer's computer send the motor's exact path instead of steps. Step interpolation already follows the path to within a fraction of a 1/16 step, and the measurements show the remaining noise is the motors' wobble, not step size. So OpenCreator does the part that removes it, the ripple correction, without changing how Klipper talks to the mainboard.