Solar 3D Printer: my take on Markus Kayser's Solar Sinter

Following up on this hastily edited post about using the Sun to extract some form of hydrogen and to decarbonate part of the construction industry, I present this Solar Desert Sand Vitrifier I have been working on.

MuJoCo simulation Link to simulation code and to the simulator: MuJoCo.

Briefly

My goal is for this machine to be able to print house walls by melting desert sand with focused sunlight. Requiring only highly available and renewable energy and material.

Building on Markus Kayser’s Solar Sinter I designed and sized a prototype which should…

  1. use a Fresnel lens ≥1.5m2 area (Markus Kayser’s lens that we know can melt desert sand)
  2. translate and rotate the lens within a print volume 100x larger than Solar Sinter’s 40cmx40cmx40cm
  3. achieve mean print speed ≥1cm/min which seems slower than Kayser’s machine (from the video)
  4. calibrate extra easily, with positioning tolerances in cm
  5. cost ≤1000€. Much cheaper than $28k QT10-15, $39k BigRep-ONE or >$1.5M VX4000 as well as orders of magnitude worse precision and speed.

Lens

Finding a large circular Fresnel lens was tough!

First I looked for cheap Fresnel lenses from (broken) 80s-90s-00s Rear-projection TVs on my local equivalent to Craig’s List. The few I could find ended up not large enough.

Here’s the result of around a minute of summer sunlight exposition around midday at Paris latitude with a AA67-00115B (Samsung SP43T8HPX/BOB) 43” (=0.63m2) lens: almost sintering

Then I went for the modern 65” (=1.2m2) flatscreen TVs but these turned out to only sport linear Fresnel lenses! Meaning they focus light into a line and not a single spot. Also, you wouldn’t believe how much money some people ask for a once expensive but utterly smashed screen…

On AliExpress I could only find lenses in the meter square area. To melt sand we must achieve temperatures ranging from 1000 to 1500 degrees Celsius and talking with different sellers (as well as a French one selling the cheapest lenses I found) all were confident temperatures upwards of 800-900C weren’t achievable…

In the future I’d like to come back to these cheap-ish Chinese lenses ($200-$700) and get a deal as there are a lot of competition there. Also, see these exhorbitant shipping fees lately ($600 for a 1.2m2 lens priced $250)? But first I need to replicate results!

I ended up finding CF1200-B2, a 1.5m2 PMMA rectangular spot lens with a smallish focal distance of 1.2m. Note that acrylic (PMMA) is as good as glass for lenses and is usually lighter (3.5kg here). This turned out pretty expensive: 250EUR/piece, 500EUR shipping to EU, 120EUR customs.

Again, bulk pricing is required here if I ever want to achieve less-than-a-thousand euros per machine…

Imaging lens

Let’s engineer a brick-printing lens: a Fresnel lens that produces a rectangular image so that the lens needn’t be moved on x-y (it only needs to be normal to the Sun).

ultra thin lenses

..WIP..

Large print volume

motors

spring+cable system x4 to hold lens down

cubic frame 4 motors

cubic frame 8 motors

cubic frame 8 motors identical mounts

Change powder delivery system

Improve on Markus Kayser’s Solar Sinter by replacing ③ & ⑤ with a powder deposition system inspired from FDM: Fused Deposition Modeling.

a sintering printer

vibrating tube

Cubic frame material

Curtain pole (interior diameters) (Castorama)

European Standard Rail Aluminum Extrusion Profile

Photocatalytic water splitting

Motors

Cameras

Vision things

Fibers

Name?

Crucibles

Toghrol Tower

So for my solar printer project I need to come up with the reverse kinematics so as to trace a light path on the sand surface
Meaning: I have to trace this path on the ground with sunlight, how do I move each of the 8 cables to achieve this path?
I'd like to learn that task instead of coming up with the maths. I have a camera looking at the ground + 3-4 reference points to correct the trapeze of the camera (since the webcam is a bit above the ground)
I found https://deepmind.com/blog/announcements/mujoco which I believe can help with data generation. I plan to simulate 8 rigid cables, one at each corner of a cube. Each cable is attached to a square plane in the middle of the cube (2 cables per corner of the plane, one coming for above one from below). Then I have two cones coming out of that square (ie. my lens). The cones connect at the lens' focal length. Then, from moving each cable I can look at the intersection of these cones with the ground (an ellipsis).
Now my learning task is to control how to move the cables so that the intersection is close to being a dot. Others have solved the maths before so I guess this should be learnable...
My question then: what kind of AI model should I be training on this synthetic data (8 motor input speeds [-1;1] + ellipsis shape and position relative to the center of the ground) ? An LSTM?

cdpr inverse kinematics

crazy art with CDPRs

large-scale seawater desalination plant

helical piling

dunes divert wind patterns

lens support structure

building strong is the backbone of society

regolight

emergency shutdown

lens top protection

Another, bulkier one here

numbers

moving frame

Making our own Fresnel lenses

the lens is a fresnel lens machined with a hand made jig from polycarbonate, and afterwards polished with methylene chloride.

pulleys

powering steppers off

rpi4 auto-join wifi list

heat reflector + windscreen

xy cable system