a flat, modular and scalable light engine concept for stereolithography and its applications
A kick-off presentation of the project I put together on short notice (interview for PhD position)
Adobe Portable Document Format - 9.43 MB - 04/12/2018 at 22:46
While most of my projects currently suffer from extreme dilution effects (priorities something something), the closed-loop installation of the lenses is more or less clarified and just needs to be put together and executed... but the hinge mechanism with the piezo bimorph element is somewhat undefined and not super satisfactory.
The current design assumes two light-weight u-shaped brackets with bushings and dynamic pre-tensioning. When using four piezo elements, possible movements are
* vertical movement (both pivot points moving up or down simultaneously)
* horizontal movement (both pivot points moving left or right simultaneously)
* tilt correction (pivot points moving in opposite vertical direction)
* pinching / preloading (pivot points moving towards one another)
While this is nice to implement automatic calibration that compensates for creep, sag and wear of the bushings, doing away with the intricacies of multiple bushings would have its charms.
Watching the latest video from @Ben Krasnow as a horizontal slab is being printed got me thinking: if the material is too thin to peel from the window without support, maybe the drum geometry can handle this owed to the continuous, slow peeling process. If the peeling is very sketchy one might even consider printing perforated layers and filling them in 1-2mm behind the peel line. will require more degrees of freedom for the lenslet array though. Let's investigate that once the structural demonstrator has gotten somewhere :)
Here's the video:
We did away with the "lightweight and sturdy" aspects of CFK for now due to manufacturing difficulties and opted for FR4 parts ordered as conventional PCBs.
Three boards come together with a bit of epoxy glue and two 1.5mm dowel pins. Since I only have a pack of 1.5mm drill bits and a micrometer to check them I'm guesstimating the 1.5mm holes by feel to be 10-20µm oversize - 3 out of 5 drill bit shanks I used as ghetto pin gauges came out at 1500 +/- 5µm, one is about 30µm undersize and has a notably looser fit.
It's a pity they only provide 1.6mm FR4 so the three PCBs weigh in at 17.6 g (instead of 11 g) but we'll have to make do with that now.
The overall tolerances seem good, with excellent repeatability (no problems fitting the tight tolerance pins through stacked PCBs). The smaller radius isolation milling is tighter than the PCB contour milling so the outside edges are not in spec. Luckily only the inset contours produce reference faces (upper board left, lower board right) so it's ok.
The holes for the molded lenses are milled, nice and round with minimal chipping and for what it's worth, the stopmask is properly aligned.
I think I'll wait until monday and get myself a 12mm aluminium plate to help with the glueing so the sides don't end up crooked.
Trumpf has this neat video about laser processing of carbon fiber composites where they show cutting and ablation of the material.
It seems like some of their laser systems can just punch through the composite without making much of a mess (TruMark). Notice the distinct concave surface though as the laser ablates the material.
Since I need an array of holes drilled into a lightweight yet strong beam, I thought I'd give it a go. The starting parameter set was taken from a preset for ceramics marking, where more power equals more better.
The first image shows the result of a woven tube being "drilled".
This tube has a distinct weaving pattern and it feels like the top roving was a bit more resistant to ablation than the rest. Most importantly though, this is not a cylindrical hole by any standard.
The pultruded material behaves a bit more convincingly though the process leaves a stubborn fringe and exhibits a visible draft angle.
I also tell myself that the perpendicular cut creates more fraying. Decreasing the spot size to 20µm does not make things any better.
After looking at the machine scribing a circle for 15 minutes and not getting anywhere I had seen enough. As for the difference in line width for x and y directions: either something shifted on me (I did reproduce the linewidth directionality though), the material ablates anisotropically or the laser mode is unstable (it jumped from being bright do being much less so, then back to bright over a couple of minutes). Maybe it even has something to do with anisotropic heat transport.
And if that's not weird enough, some types of carbon fiber exhibit polarisation dependent absorption.
You know those wallpaper glue applicators?
They're basically a drum that picks up blue from the reservoir. Now imagine this for 3D printing upside down, on its side or even in normal orientation.
Since we've developed an essentially flatbed linear light engine, we might as well put it inside a precision glass tube (BK-7 or better) and expose some resin from the inside.
The glass tube would be moved back and forth, all the while being rotated synchronous to the motion to roll over the topmost printed layer, making contact, adjusting resin layer thickness and peeling off in one go.
The core of the design is the array of optical paths which have to be defined by precision pressing the laser diodes in a test jig. This should be easy enough since they can be powered and the optical performance can be monitored in a closed loop fashion.
On the left, the cooling profile, 8x laser diode tile and spring clip are shown, on the right, a CFK profile with press mounted molded aspheres is shown, along with an anchor rod used for steering.
Calibated assembly has to be done once and can be automated to produce the segment modules. Each module has 8 LEDs, temperature sensors and an EEPROM with calibration values and temperature coefficients.
the LD sub-assemblies self align via dowel pins and are held down with spring clips. The profile is designe for forced air cooling or water cooling with a small OD tube.
the anchor rods are located between two lenses and their beam paths. They are connected to a triangular U shaped segment (not shown). The upper point of the triangular piece connects to the anchor rods on both sides, its remaining to points are connected to piezo actuators. Their common motion moves the lens array up and down while their differential motion moves it around along its axis. There are two anchor rods and two pairs of piezo acutators respectively, also allowing a small dynamically controllable positive preload to avoid backlash.
I intended to use DRV2700 for the piezos (not suitable for piezo trimorphs without modifications...) and LDC1000 for precision position monitoring and closed loop absolute position control.
The position of the anchor rods is a thing in itself, see https://en.wikipedia.org/wiki/Airy_points but I guess it's more important to find points that don't dynamically coincide with nodes of the most easily excited beam vibrational modes.
Resilience and online calibration
Seems like I forgot to mention a few things.
Both the individual control over the emitters and the additional degree of freedom (note the two directions of linear motion and one tilt degree of freedom, which leaves another, more subtle one for clamping force / play) allow for a few gimmicks. Emitter grouping and sequencing however may be the single most important one.
Especially during prototyping not all laser diode module are guaranteed to be at the same height, maybe not even the single emitters. For maximum resolution, emitters can be calibrated and grouped by focus distance and then driven sequentially, re-adjusting the focus in between.
Emitters can also be grouped for simultaneous exposure when isolated emitters have failed. I've already mentioned that "dark strips" caused by failed laser diodes can be covered by their neighbouring emitters while increasing the stroke. This need not cause degraded resolution over the whole width when the scanning speed is decreased, allowing exposure to happen in a small stroke, small angle and high resolution phase, followed by a large angle coverage phase where exposure only happens locally where needed.
The fourth degree of freedom allows adjusting pre-load to adjust the backlash to zero as mentioned above. This allows to compensate for wear and avoid excessive forces on the bearing surfaces.
Single lenses cannot produce a flat field image. ( see https://en.wikipedia.org/wiki/Petzval_field_curvature ) Since we're using aspheric singlets for scanning, we can cheat by re-focusing during the scanning motion to absorb this error.
This makes it possible to get a sufficiently balanced spot size across the whole scanning motion
This OSLO PSF simulation has been performed to illustrate the spot shape over lens displacement. As the object distance (LD-Lens) is around 4 mm and the image distance roughly 40mm +/-0.3mm equals a 6mm scan line length which means that with a 3.3mm diameter LD every 5mm, sub-25µm spot size is achievable. If one emitter fails, the stroke can be increased to +/-0.5mm, covering the now empty segment with the neighbouring scanlines.
The working principle is rather simple, really:
The key enabler for this project is the combination of cheap molded acrylic aspheres and cheap low-to-medium power 405nm laser diodes. These lenses have the optimal shape to collimate the laser emission from a miniature semiconductor laser inside the diode housing. They really do great things for that special case but they still work ok when not focused to infinity but rather at 25 to 50 mm. In this finite conjugate setup, the laser spot is focused along an arc as the laser diode and lens are moved relative to each other which makes the spot blurry as you move it farther to the left or right. While this may produce sub-100µm spots one can still do better by doing a few microns of refocusing, doing the initial focusing along the way.