formlabs recoater chatter deep dive

similarly to an fdm printer, sls printers build parts layer by layer by fusing thermoplastics using heat. solid layers are built with a laser which sinters the powder together. resistive and quartz-tube heaters hold the powder just below the melting temperature and provide uniform temperature distribution, reducing warping. to build the powder layer prior to sintering, a "recoater" can be used, which moves across the print bed after the platform descends, creating a uniform layer of powder. the fuse 1/1+ uses a counter-rotating roller recoater which generally improves uniformity and print quality over other styles including blade recoaters, which are more economical. in the fuse 1 series, linear motion is lead screw driven, and a static gear rack creates counter-rotation to the direction of motion.

as the recoater moves across the print bed, any vertical oscillations will result in ripples parallel to the recoater on the top powder layer. you'll notice these ripples in large flat parts, and it's undesirable for aesthetics and functionality.

there are many causes to chatter, from mechanical slop all the way to how powder interacts and flows with the roller. these causes have unique visual characteristics which are huge clues in a root cause analysis.

an example of chatter in the next-gen printer is shown below. the recoater sweeps left to right. the visible ridges are regularly spaced, indicative of a mechanical root cause. chatter caused by powder interaction is generally more random. the period is a key insight into whether the source is related to the driving mechanism, the counter-rotation mechanism, or other mechanical compliance.

for each prototype of a printer, we print as much as possible, as difficult as possible to stress test our architecture and identify outstanding risks and problems to be resolved in future iteration cycles. during some prints with this prototype, it was observed that chatter was pretty bad. it certainly wasn't an architecture-killing problem, but would need to be resolved sooner or later to meet product-level print quality requirements.

before anything could be improved, the chatter first needed to be quantitatively assessed so progress could be tracked. the standardized test print is pretty simple; just a flat plate. for the purpose of iteration speed, stepped plates were also used, where each step corresponded to a different recoating speed. a colleague previously investigated quantifying chatter, and found that standard surface roughness measurement devices were too noisy and couldn't isolate the chatter. they devised a system using a bridgeport mill with power feed and v-block clamps to hold a usb-connected dial indicator, jogging the dial indicator over the entire width of the plate. they also wrote a python script to low-pass filter the result and translated the measurement into an average amplitude and period. i improved the script so results would better align with visual inspection.

in addition to the quantitative amplitude measurements, i used high-speed videos to observe the recoater movement, powder interactions with the roller, and flexing of motor mounts and gear rack mounts.

i repeated a ton of cycles of printing plates, measuring and recording their amplitude and frequency, attempting to analyze the causes (and getting a much better intuition each time), logging progress, and using the presumed root cause to make some mechanical tweaks.

some changes were minor, both in complexity and cost but had significant improvement. an example of this is that i noticed we were using polymer bushings (id tolerance +0.12/-0 mm). switching to bronze bushings (id tolerance +0.02/-0) gave less slop to the roller, which had an insanely good improvement already. other minor changes were higher quality parts, such as lead screws with a better straightness tolerance.

other changes started from a holistic inspection of the system. for example, investigating the lead screw motor mount, i hypothesized that the mounts being out of perpendicularity with the build plate could cause serious whip. i verified this through slow-motion video. a couple ways to solve this would be to improve the perpendicularity (for example, switching from bent sheet metal to machined), or to use a flexible shaft coupling between the motor and lead screw. the second option was preferred from a cost and lead-time standpoint. thinking long-term at BOM cost effects, a new machined part would risk BOM cost target overruns. for prototyping, the part would be several days, whereas we had plenty of flexible shaft couplings around the office for me to try same-day.

some improvements were found by trying crazy ideas. almost all chatter testing was done with 50-70mm/s recoater speeds with chatter getting worse as speed increased. it was hypothesized that higher speeds meant worse chatter. as early progress was made, i wanted to push up the speed. it was a pretty incredible change; at 70 mm/s the recoater sounded harsh and loud and produced terrible chatter but was much smoother and quieter at 130mm/s. thinking from first principles, i hypothesized that 70mm/s was a resonant frequency somewhere in the system. as a second test for this, i ran the recoater at the same speed, but increasing the mass. since the resonant frequency is proportional to the root of k/m, increasing the mass would decrease the resonant frequency, so i should expect similar results as the original recoater at higher speeds, which was proven through quantitative measurements.

a more mechanically complex change was to add an external linear rail to decouple the roller from lead screw straightness errors using an oldham coupling. essentially, an alternative to a standard anti-backlash nut allowing for a high degree of radial misalignment. since linear rail straightness tolerances are typically far better than lead screws, it pretty much guaranteed that linear motion would be true. however, it was additional complexity for a machined part, a more expensive lead screw nut, and an additional linear rail. after all, best part = no part.

the aforementioned changes had the most significant impacts on reducing chatter. Other minor changes which improved chatter were:

•powder "fences" on the recoater roller shaft, which contained powder and had an extra benefit of less waste.

•switching from straight to helical rack and pinion for recoater counter-rotation for improved gear meshing

•stiffening the gear rack mounting design, reducing vertical compliance (high-speed video was key for identifying this)

this project was a 3-week sprint to resolve this issue before my internship ended. all in all, through repeated mechanical iterations and testing, i managed to reduce chatter to almost unnoticeable, while also increasing the recoating speed as a bonus, which had a great effect on print times (projected 1.5 hour average reduction on every print!) i learned a ton about linear motion systems and how to make them buttery smooth.

above is the final status of the recoater when i finished this project. no visible or tangible chatter lines, and increased recoater speeds!