Hacker News new | comments | ask | show | jobs | submit login
Forget everything you know about 3D printing – the ‘replicator’ is here (nature.com)
186 points by chriskanan 15 days ago | hide | past | web | favorite | 55 comments

There are subtle differences between 3D printing types that seem at first to be the same.

Stereolithography is horizontal, one layer at a time, and uses photopolymerization (that's Formlabs). (Also Digital Light Processing is close but distinct: https://formlabs.com/blog/3d-printing-technology-comparison-...)

Continuous Liquid Interface Production (CLIP) also uses photopolymerization, but pulls the object from a liquid bath and uses a buffer zone. Still horizontal slices. The upshot is it's much faster. (Carbon 3D is the company behind this.)

The method in the article uses photopolymerization to solidify the object as a set of slices, but the slices are not horizontal.

The big drawback to photopolymerization is it only works on certain resins which can often have undesirable mechanical properties (high elasticity or brittlness, for e.g.) Potentially this method could be a way forward in that respect, because you might be able to put structural materials in the resin solution and end up with a composite. It seems easier to do this way than with CLIP or SLA/DLP, but I'm purely speculating.

Other well-known 3D printer technologies include:

- Fused deposition modeling (FDM): the typical plastic filament-extruding hobbyist printer. Low resolution, but fast and creates very strong parts in one of wide varieties of well-known engineering materials. Tricks (e.g., two extruders) can give limited multiple colors or materials.

- Selective laser sintering (SLS): a laser melts a pattern into one layer of powder at a time. Can make even stronger parts than FDM by using nylon, titanium, etc. Very common in industry, but usually too expensive for hobbyists. These are the printers used to make rocket engines.

- Stereolithography (SLA) as explained above works like SLS but cures liquid resin with light instead of melting powder. Has many subvarieties. Advantage is scary high resolution (certain engineering choices give 160 nm (!) feature size [1]), but at the cost of relatively limited material choices because materials need to be liquid UV curable (though Form 2 has a good library now [2]) and definitely no multi-material or color parts. I'd consider this new 3D rotation printer a variety of SLA. [Edit: for clarity, I lumped CLIP and DLP here with SLA]

- Inkjet-based printers: these are the _really_ cool ones. Objet makes a printer [3][4] that uses inkjet heads to deposit multiple colors and materials in the same part layer-by-layer (kind of combining FDM and SLA). Upside is multiple colors and materials and resolution (e.g., Lego uses these for prototyping), downside is ridiculous price and low speed. Other printers like HP's and Z Corp's combine inkjet heads with SLS powder instead.

There are also a few weirder ones, e.g., paper layering, but I don't think they're widely used.

[1] https://www.nanoscribe.de/en/products/photonic-professional-... [2] https://formlabs.com/materials/ [3] https://www.youtube.com/watch?v=M1sOdZqwn5Y [4] https://www.stratasys.com/polyjet-systems

One I came across today is SPD (Selective Powder Deposition). Powder is deposited for later curing, such as in a kiln e.g. http://www.iro3d.com/

Another con to resin based approaches that's not always obvious is that some of these resins are pretty nasty and dangerous to touch when wet or pour down the drain. That makes the cleanup process a pain, at least in the context of at-home or small-scale printing. Powders are aslo a big pain but they're too expensive anyway.

The big problem with stereolithography resins is that they tend to degrade with time, especially if left in the sun. Photocuring is carried out by creating free radicals to induce polymerization. Well it turns out that these same free radical producing chemicals stick around even after the curings done and can make more free radicals if left in light. These free radicals can cause some pretty nasty damage to polymer chains. The chemistry is improving though, at the very least to allow more interesting materials like silicones. One can mix in composites to stereolithography resins, but in general this isn't done. The problem is that whatever you mix in scatters light, so light doesn't penetrate as deep, so curing takes longer. Unless the composite used is transparent and has the same index of refraction as the resin, it would not be possible to use this tomographic approach to make parts.

There is one application of composites that's extremely disruptive though: producing ceramic parts. It has been found that you can mix in ceramic powder with the resin fire it in an oven to make very detailed ceramic parts[0]. Structural properties aren't necessarily that good, but that's ok because it's good enough to make very complex and detailed investment casting molds[1]. Investment casting molds for making single crystalline jet turbine blades are very complex and require a very complicated process to produce, with this process they can be made in one step. That's a huge disruption right there and investment casting is a pretty general process. The detail produced by this process is so high that the the triangulation of STL became a problem and a special file format developed for fax machines had to be used to represent all the slices.

[0]https://www.researchgate.net/publication/229293237_Photopoly... [1]http://ddm.me.gatech.edu/page8/page8.html

If I understand this correctly, there's nothing to stop them from...

* rotating the projector instead of the tube to reduce distortion from resin movement (like in a CT scanner)

* speeding this up, using higher energies to compensate

* using multiple projectors at e.g. 90degree angles, to reduce time (and thus probably distortions through movement in the resin)

* using other wavelengths with e.g. more energy, or just better absorption properties for the resin

or are there any obvious problems with that?

The other benefit with having multiple projectors/lasers at different angles are that you can use different wavelengths: one that causes polymerisation, one that inhibits it.

steriolithography[1] is neat, and i think older than deposition style printers.

This is kinda wacky. Rather than a conceptually simple layer by layer approach, this has this funky convolution. they're getting it for free with the simple rotation. You can kinda see how bulbous that airplane model is, because some finer details get 'overexposed'. i bet, if they could rotate 10x around 1 axis and 1x around another axis, that tumble would sharpen up some of those edges, since you wouldn't be baking the same adjacent space so much.

Think about a 6 sided pencil. if you just rotate around the main axis, it's probably going to look pretty round without the crisp hexagon sides. but if you can also rotate around a minor axis, you could also project just the hexagon shape for a while and still preserve the cone of the tip of the pencil since most of the solidification came from the sides.

There's an opportunity for deep optimization there. Seems tricky. but pretty cool.

it's almost like pca. what projection (or series of projections) maximize hitting the target, and minimize the overbaking, all while taking into account the characteristics of the resin. maybe you can let it cool for a bit, and have more freedom to expose without hardening.

also, not a replicator. this is just one _fabulously special_ super material you mess with, not metal and plastic and cloth and fur and chitin and whatever.

[1] https://en.wikipedia.org/wiki/Stereolithography

I don't think it would be possible to rotate around another axis, the printed object would start tumbling around the resin. If they rotated the projector instead of the resin it would be possible.


Their problem is that the projector they are using is projecting a 2D plane.

It would be a different story if it was a circular projector (think of a LED strip).

It has to be then rotated only along a single axis.

Turning the projector circle physically can possibly add additional precision.

Magnetic fields can be used to keep the container sphere in place.

I am now going to take the screenshot of this conversation and send it to my self in a registered mail.

Uh, that's a standard resin printer. Except horizontal plus rotation instead of vertical. Trades angular resolution for vertical.

Yes, they can be fast.

They should fix their headlines.

No it isn't. A normal SLA printer like a Form One prints a slice at a time and crucially it doesn't "print" in the middle of the liquid - only at the surface.

This works as the article says, like a CT scanner in reverse. The resin solidifies all at once in the middle. It's very different.

Could multiple projectors (at e.g., 120 or 90 degree angles) be used to speed the process further?

furthermore, multiple projectors could increase resolution.

if there's an energy threshold to overcome to cause hardening, and if you can arrange for 2 sources to just barely exceed that threshold, then you could have very crisp prints.

chemistry is statistical though. you can exceed the threshold, but nothing happens because you get unlucky with some fraction of the light. You don't exceed the threshold, but you catch a bad reflection and harden stuff you don't want. and there's probably not a crisp threshold for hardening. it's more likely just time under light.

But i think you're right. two projectors at half power (let's pretend it's linear), with the projections 90 degrees out of phase, i think the print quality would go way way up.


Except standard stereolithography only solidifies the outer surface layer.

These resin printers have been around for a decade. Multiple kickstarters, several commercial models on sale right now.

To classify as a ‘replicator’ it needs to be able to mix different materials (metal, plastic) seamlessly in one print as to be able to replicate itself. Not there yet...

This is not a traditional resin printer like the Form One.

My thoughts. I remember Autodesk made a resin print of the Eiffel Tower and some other object with plastic polymers catalyzed by light about 4 years ago, there is a video on Youtube of it somewhere.

This is much faster.

>> The device, described on 31 January in Science1, works like a computed tomography (CT) scan in reverse, explains Hayden Taylor, an electrical engineer at the University of California, Berkeley.

I have been wondering if something like that was possible. I've also wondered if something more akin to holography could be used to create a 3D interference pattern in a liquid to create an object. Both seem like they'd be interesting and have limitations.

From the comments here I think some folks don't realize the mathematical details involved in this - they are not projecting a "picture" of the object from each angle going around. Yes, it's an image, but not it's not what you'd expect. It's more akin to an x-ray of the object, and probably with some extra processing beyond that.

EDIT: Looks like I'm wrong. It says the create an image of what the object would look like from each angle. I believe that means their quality is lower than it could be had they actually used more sophisticated methods to create the images.

Holography actually sounds pretty promising but the issue is that most holography systems are phased (they trace the hologram with a single laser.) If a holography with hjgh energy could heat up the inside shape of a liquid in a beaker, it could turn it into a solid. The issue would be thermal diffusion but if a dense enough liquid were used, it could work!

So it can make you "tea, earl grey, hot" or anything else ... but made only out of unbcoloured acrylic resin.

Take a break and enjoy some scolding hot resin.

I thought the same when I clicked on the link with hopes high...

> The team realized that the process could be reversed: given a computer model of a 3D object, the researchers calculated what it would look like from many different angles, and then fed the resulting 2D images into a ordinary slide projector.

I'm pretty sure the researchers were using a normal computer projector and not a slide projector. But in my imagination, they printed a bunch of slides, fed them into a carousel projector, and pointed them at an old Smuckers jelly jar full of resin.

While you end up with an acrylic model, this should be suitable for lost-wax style casting. Being able to create an unpixilated identical or scaled copy of a model will be fantastic for fine art sculpture (or destroy the industry, if copies become too easy to make). Just take your acrylic model, attach casting channels, cover in ceramic, dry, pour in your bronze, cool, crack, trim, patina, ship. Avoid all of the fiddly mold making and wax finishing completely.

[disclaimer: I work at Formlabs, a 3D printing company]

This is already happening for jewelry, digital dentistry, and medical applications – you can buy specialized materials for investment casting.

For example, here's our blog post on casting jewelry from resin prints: https://formlabs.com/blog/introducing-castable-wax-resin-jew...

Looking forward to scaling things up beyond teeth, earphones and jewelry. All very promising though, and if the pixilization is solved that is excellent (one of the foundrys here invested in additive printing systems about a year ago, which then need the artist or laborers to resurface with wax before casting). I imagine cost will certainly be a factor for larger systems (say pieces 20-60cm height), both machinery and resins, because the profit margins are very tight until you hit the big time. Is the waste castable wax resin reusable?

> or destroy the industry

I don't think the fine art sculpture 'industry' is affected at all by the ease of making copies.

This is very similar to a patent I have filed on extremely fast 3D printing.

I have mentioned this method of 3D printing here previously. Transmitting data to a volume at high speed can be done in many different ways, at any scale, and with a variety of materials.

I hope to someday demonstrate the printing of a 2 story concrete house in an hour using my technique.

Has the application published? Would you mind putting the link here? I'd like to read it.

I'm not sure the application has been published by the patent office yet, but I'll see what I can do.

Edit: Here's a link to the application on google patents. https://patents.google.com/patent/US20180257307A1/

Same here, I hope he can post it. Would be really interesting to read what techniques are being used to accomplish the claims in question.

This is new to me and looks really cool despite the sensational style of journalism. I imagine in the near future we will have a variety of widely available methods for making physical objects from digital models, each with advantages and disadvantages for different usecases.

Something similar was proposed back when 3d printing as we know it today was getting started in the 1980s. Two lasers were shined through a photocurable resin so that the resin would cure at the interesection[0](see figure 9). In practice it never worked and the resin cured before intersecting leading to nasty blobs rather than parts. It seems that now that we understand the kinetics of photocuring better and that we have massive amounts of computer that we're able to use these approaches. Processes that cure resin have a lot of potential because resolution can potentially be on the order of that of the wavelength of light. [0]https://pdfs.semanticscholar.org/4716/c69f0b90a158589e54248a...

The speed is amazing but the amount of waste seems rather large as they most likely can't (re)use resin that has already absorbed (an unknown amount of) light before.

It doesn't work like that. It's like a match: put enough energy in, and it transitions to a new state. Put less than the activation energy in, and it doesn't. The energy you've put in will be lost as heat, but the resin is unchanged until it changes completely.

There's a minimum amount of energy required for the process to happen. Below that, it can not. Above it, things can happen but how often depends on the intensity of the light. Increasing intensity will make it happen a lot more often. Some parts will receive far more light then others, making the polymerization progress far enough to actually build a solid object while the rest of the volume is illuminated little enough for the resin to basically stay liquid. (Remember that you need to shine into the container, therefore illuminating liquid that you do not actually want to polymerize but need to shine through to reach the material that should. A sort of collateral-damage if you like). The remaining liquid material is removed but will still contain partially polymerized material that's not grown large enough to actually behave like a solid.

That can't be true, because then you would have to project light from lots of different angles into the resin all at the same time, rather than from one angle with rotation like they do here.

Sorry am I missing something? 15 years ago we had one of these in our engineering lab. 3d Printing wasn't a rage term yet, it was called "rapid prototyping resin machine".

Just as a computer tomography image is just a stack of 2D images, but can be taken a lot faster because the tomography machine and its software (sort-of) take all those images in parallel, this machine is just like a 3D printer that hardens the resin one layer at a time, but a lot faster because it prints all the layers at the same time.

Continuously illuminating the entire volume means you also illuminate parts that should stay liquid, but you prevent over-illuminating those parts by illum8nating from a lot of different directions, just as a CT image can see transparent parts of a volume that are hidden by more opaque parts in some views.

Another advantage I see is that this doesn’t need a start-stop motion that lifts the object being constructed by a layer at a time or that raises the surface of the liquid by the thickness of one layer, something that, AFAIK (I don’t followed developments closely, so I may not know much) the traditional printers need because they only can print at the air/liquid surface. Smooth continuous motion is easier to build and faster.

PR agencies are better now. I used one 20 years ago myself.

There is a more volumetric 3d printing method which prints all at once, no rotation required: https://www.llnl.gov/news/volumetric-3d-printing-builds-need...

This looks like it's wasting a lot of material comparing to DLP printers.

They could probably 3d print multiple parts inside the liquid simultaneously.

Then it becomes a "fill the cylinder" optimization problem.

Maybe they can 3d print a custom container first

Setup and fine tuning of such devices can take months so I highly doubt that.

I don't fully understand the process of creating 3D video from 2D slices. Is that some sort of holographic procedure?

Presumably such a video could also easily be computed from CAD models, right?

Reconstructing 3D models from 2D projections is a popular exercise in applied linear algebra courses (actually, 2D images from 1D projections are more common, because it's easier to visualize). The procedure is pretty easy to code up once you've understood the mathematics. Essentially you build a system of linear equations describing the influence of material density on the measured intensity, then solve it for density given intensity. https://scikit-learn.org/stable/auto_examples/applications/p...

Generating projections from a model is much easier, because you only need to write the rendering pass and don't have to invert it.

Producing 3D images from 2D slices is easy. The problem is that the 2D images that a CT machine makes aren’t slices, but projections of the 3D volume onto various planes.

For example, if your image is

an image taken vertically will count the x-es in each column:

and one taken from the left or right will count them in each row:

Those 2 projections (probably; I haven’t checked this image) aren’t sufficient to reconstruct the image, but they already add significant constraints. For example, those 5 X-es in the next to last row can’t be in the first or last column (if they were there, the first image would start or end with a non-zero value)

The more projections you add at other than orthogonal angles, the more constrained the solutions become.

This looks an awfull lot like 3d printing to me

In the sense that from a 2D representation a 3D object is created, yes, but that's it.

Could this be used to rapidly store data as a 3d matrix?

inverse ct scan ?


real fancy.

Applications are open for YC Summer 2019

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact