Hacker News new | past | comments | ask | show | jobs | submit login
MIT team develops 3D printer that's 10x faster than comparable 3D printers (3ders.org)
352 points by sswu on Dec 8, 2018 | hide | past | web | favorite | 77 comments

I wonder why they needed the laser pre-heater. I can't imagine it is that dramatically more effective than resistive heating. Even if it is, just having a longer pre-heater to make up for it would have to be cheaper.

EDIT: Did a bit of research, a major benefit of laser heating is direct control of energy transfer so you can be sure your heating the plastic to the right temperature regardless of how fast its moving through the heater. Still, aren't all these advantages eliminated by the traditional primary heater immediately after?

The laser penetrates the filament heating it internally instead of only heating the outer surface. It's near-infrared (808 nm) so it heats the small cylinder of filament essentially uniformly throughout its volume.

It's also nice because they can vary laser power much more effectively than you can vary the temperature of a conduction surface.


And it is low friction since it does not need surface contact.

I'm guessing here, but it could be possible that the actual heating of the filament is done entirely with the laser, and the job of the heated nozzle is simply to keep the filament at temperature all the way until it is extruded. It may have not been possible or feasible to combine the laser heater and nozzle, hence the need for both parts.

When the Laser heats the filament, it's 1.75mm in diameter, so to get it down to the 0.4mm diameter for extrusion a nozzle is still required which also needs some heat so the plastic doesn't resolidify.

The temperature needs to be consistent first. There is a PID cycle the nozzle goes through to keep it consistent. If you look at the temperature on a graph during print, it's not a flat line but rather a sine wave as the PID cycles. My printer has an LED that flashes with the heating element, and you can see it work more during heavy extrusion to keep the temperature up. The frequency of the cycle is related to the mass of the nozzle/heater and speed of extrusion. I can only imagine at that speed they could not get a consistent PID loop with a traditional heater.

FWIW, a sin wave is usually a sign of a simple threshold based controller, not a PID loop (or if it is a PID loop, it's very poorly tuned). A well tuned pid loop will not exhibit any periodic tendencies, though I would expect that it would exhibit changes as the filament extrusion is stopped and started.

The mode of heat transfer is very important for applications like this. Higher mass flow means you need a hotter heater which can get tricky for long term usage or you'll have to use some sort of refractory or ceramic based heater. Also with conduction, the biggest problem is heat loss either due to the environment of the system . It is almost impossible to direct all your energy towards a single source with conduction. Radiation on the other hand works really well for this type of application. Especially for lasers since the energy sources is highly polarized and condensed. The other method might be to use induction heaters or microwave, but plastics do not play well with that.

It’s definitely possible that you couldn’t reach the speeds that they are with any kind of traditional heater cartridge. You need a lot of power directly on the filament to melt is at quickly as they are.

Titan 3D Robotics has a extruder that claims to be able to print 1 pound of filament an hour which for ABS would be able 400 cm^3 (I think?)

I believe it's water cooled but still looks to be using traditional heater cartridges.

Specs: http://www.titan3drobotics.com/atlas/

Video: http://www.youtube.com/watch?v=mKPKIGcrlFU

Right. That's what you get when you build a 3D printer like a CNC machine. Titan takes in pellets directly, rather than bothering with filament. They probably use a heating system and drive screw like an injection molding machine. They can print with more plastics than filament printers. They end up with a lot of moving mass on the print head, but that's what big motors and controllers are for. The Titan unit uses 14KW of power.

I'd once considered heating the surface to which you are bonding with a laser, just ahead of the extruder, so you weld hot surface to hot surface, not hot surface to cold surface. That's why filament type 3D printers make such weak joints between layers.

I think the 400cm^3 figure was for the non-pellet version while the pellet version can use 10mm nozzles and extrudes a ridiculous amount more. I wish there was more published information about both of their extruders, they look quite interesting.

I'm guessing that absurd power requirement is for the 85C heated enclosure variant, otherwise I'm no idea where the 60amps are going.

I remember reading about a slicer with an option to use the hotend just as a heat source that traced the previous layer to smooth/blend it. Using a laser may be a good method of improving the surface finish as well.

What is the $ per kilogram cost difference if you can buy bulk PLA plastic pellets vs. buying filament on a roll? For a benchmark comparison figure the generally well-reviewed Monoprice filament ranges from $18 to $20 per 1kg spool.

it's about half as expensive.

What's the best way to keep the laser pointed just in front of the nozzle?

Not the parent commenter, but I imagine mounting it so it rotates around the nozzle’s axis would work. Fiber optics would help if the laser unit is too big to have swinging around on the print head.

You might also need to be able to tilt the laser up/down to vary the distance from the nozzle.

It should be pretty easy from there to have the computer keep it pointed at wherever the nozzle is going next.

I was thinking 3 lasers 120 degrees apart, turned on as needed, rather than a mechanical solution. 5W laser diodes would probably be enough power.

That is almost steampunk, we need cyberpunk :-P

> All technologies are improved with lasers.

I love this writing style, clear and informative but playful at the same time.

Sounds like a quote from Dr Evil.

I dont think I have ever read an article about MIT students that didnt start its headline with MIT, is this university policy of some sort?

It's the name that sells :)

Well, if it's done by professors at MIT, it makes sense.

I run a cannabis company (cbd). My cofounder and I were roommates at MIT. We rarely mention it in our industry.

10x faster is a substantial achievement for anyone waiting for 3D items being made.

For the technology as a whole we need something that changes more than a constant factor, i.e. better than O(n^3) because we're still using filament that's time proportional to volume of the object.

This isn't entirely clear cut though. For instance laser printers technically do trace a dot of light across a page, but a single raster scan is almost instantaneous compared to the time scales of larger operations. Not having to physically move a 'print head' seems to be the winning design.

The 'obvious' way to do that: parallel heads across the whole surface, so time is O(height). Feynman brought this up in https://people.eecs.berkeley.edu/~pister/290G/Papers/Feynman... (an 80s followup to "Plenty of Room at the Bottom"):

> You should have a thicker machine with tubes and pipes that brings in chemicals. Tubes with controllable valves - all very tiny. What I want is to build in three dimensions by squirting the various substances from different holes that are electrically controlled, and by rapidly working my way back and doing layer after layer, I make a three-dimensional pattern.

Added: since mechanical frequencies scale up as size scales down, this sort of design would ideally scale as O(1). That is, with smaller parts and increasing resolution, you have O(n^2) parts, each working O(n) faster, to produce O(n^3) units times O(n^-3) unit volume = O(1) volume per unit time.

This form of fabrication exists and is known as DLP. I have a DLP printer that cost me less than £500.

Neat, I didn't know of that!

Feynman was looking further out to a machine not limited to one material at a time.

10x faster is a bit exaggerated; maybe 2-3x faster than typical desktop FFF and still slower than HP or Carbon

Just at a glance it looks about 10x faster than my Prusa, but I agree it wouldn't be much more than 2-3x faster than larger scale printers for sure. I don't think Carbon would come under "comparable 3D printers" as it is resin based which is often faster.

Fused filament fabrication (FFF) is fighting against physics in the same way that O(n) vs O(n^2) vs O(n!) algorithms have wildly different performance at scale.

It's a point solution, depositing ~1 voxel per unit of time. Running print heads in parallel is still O(1). Speeding up the print head runs out of steam because you run into vibration limits for the machinery (you can hear the rattling in the audio for this article). To really scale you need to deposit ~n voxels per unit of time (HP's MJF technology) or ~n^2 (Carbon's CLIP).

You need lots of voxels for high resolution for most applications. There are certain exceptions like prototyping in PETG or 3D printing concrete houses where the speed limitations of FFF may not be a big issue. But for 3D printing to compete with many forms of traditional manufacturing, simultaneous parallel structuring of matter is key.

How does this compare to Carbon?


That's a good question because at the moment Carbon printers are the fastest and (most important) produce production ready materials.

It does not. That's like comparing apple orchards to single oranges. But if you want to just compare print volumes then the paper claims ~127cm^3/hr build rate in a prototype system with a peak extrusion rate of twice that.

resin can be much faster, but is not really that great for deasktop/at home printing.

Yes there are DLP printers for enthusiasts, but the multi-stage process is not that great, as the chemicals are pretty noxious.

There are a greater range of cheaper plastics available for extruding.

Fast and loud! Holy hell! I thought dot matrix printers were loud.

As hobby goes 3D printer really no good. Laser is fast and even some Cnc ok. But hours and hours ...

Hope this help as having a 3D model of your favourite item is great. Especially you can like programming iterate through it quickly.

Learning is part of the fun.

Hobby 3D is geat for me. Find some DnD minis I need, load 36 or so to a SD card, go feed into printer, and 24h later I have $100 worth of minis printed for $0.50 in plastic. A $200 ender3 pays for itself real quick.

Well if you order an object online, then you have to wait 1+ days for it to arrive ;)

More like 2 weeks for Shapeways...

From the headline, I assumed this was going to be an MIT writeup of Inkbit. Instead it seems to be another MIT group but at the opposite end of the quality spectrum (this is low-ish quality but super fast, while Inkbit is high quality and yet fast). Cool!

More info on inkbit please.

Regarding "normal" 3d printer technology, anyone who's thinking of getting a basic one to play around with, take a look at the Creality CR-10S. It sells for $369. There's lots of youtube videos of sample output from it and reviews.

The bigger version of the same thing which can print 50 x 50 x 50 cm volume, the CR-10S5 is $629.

I have no connection to the manufacturer or Chinese vendors, just throwing the name of something I'm satisfied with out there.

Ender3 can be found under $200 and is excellent. I think it is similar to the 10s

Sweet. Let’s see a benchy.

They should compare speeds printing benches.

These are some really cool and novel developments at improving the speed of 3d printers. While some of these might take a while (or forever) to come to desktop 3d printers its great that advancements like these are being made to push 3d printing forward.

I saw this 2 years ago. Long story short, you need a fiber laser and a driving circuit to do this.

Sure it's 10x the speed, but it's almost 100x the price.

A Creality Ender 3 is around $200. This printer, with fiber laser, is around $15-20k.

And the Ender 3 can't make moves that fast. The Atmega chip is just a 16 MHz chip. You can't generate the steps required even using Klipper firmware (which is a bitbanging firmware that uses your desktop CPU as motion planner). You could generate the steps using one of the ARM based boards, but you'd double your BOM - Smoothieboard and Duet both would be around the $150-$200 but can generate the required steps. The BeagleBone Black can generate upwards of 1M steps/sec, which is on the high end of pro-sumer.

It's awesome, but it's a pie-in-the-sky that most of us would never even have the source to buy, let alone approach.

Don't forget all 3d printers were 15k not that long ago.

Unfortunately, a large portion of that was due to patent interference from Stratasys. https://hackaday.com/2013/09/11/3d-printering-key-patents/

2009 was when the patent expired. And that's when RepRap picked up rather quickly. The patent was handled since 1989 reminded me the same way the Wrights brought avionics to its knees in the US until the US busted the patent for WWI.

To make a 3d printer, all you need is a slow controller for handing gcode, 4 stepper controllers, 4 stepper motors, thermistors, heated bed, and heated tip. Worst case, before being able to buy calibrated filament, you could use weed whacker line, and put in its equivalent diameter

However, one trend I see, is that optics does not lower in price. Sure, lasers have gotten cheaper. But when you talk about 200 laser diodes at 5w each and using a complicated assortment of lenses and glass fiber optics, that stuff's $$$$$. The costs can come down from $50k to $20k, but it's still way out of the reach of 'buy on ebay or amazon'.

Exactly. Just buy 100 Ender3s for somewhere around £15,000 and run them in parallel.

I like the last comment, that says the pic is not of the people named, and that one of the professors named is not a professor.

Who are those guys, then?

3ders is one of the best blogs in 3D - worth a follow

The threaded filament feed is a really nice touch.

Printer is coming...

>Grace wrote at 12/8/2018 1:45:34 AM:

>That picture is neither Jamison nor Professor Hart; and >Jamison Go is a PhD student, not a professor


'Cept the filament probably costs more per unit of weight than the printer...plus shitty plastic and not metal.

If you could have a liquid 3D printer below the starting surface and a solid PLA printer above, you could start in the middle of the object and print into two directions, at twice the speed. Of course- the basic question remains, what holds the middle? Retractable titanium bolt?

I've got a design for a printer that should be about 10,000 times faster than this.

It's really a game of data transfer speed. How fast can you transfer information about solid vs non-solid into space?

The velocity and acceleration of that printer is very impressive, but maxing out the movement speed of a single physical nozzle is not going to get us where we need to be in the future.

If anyone is interested in collaborating with me, I'd love to talk.

10,000 times faster at printing a single object?

The conceptual targets are printing a chair sized object from granular material in 30 seconds, and a 2 story concrete house in 1 hour, so 10,000 is a very rough "shock marketing number", but yeah.

You can't print a concrete house in a hour. It would collapse. Also, in general, the slower concrete cures the stronger it is - to the point where attempts to have it cure faster will destroy its compressive strength. Concrete walls should remain in forms at least one week and far better for two, and be wetted continuously during that period. 3d printing concrete needs a massive breakthrough in 2500 year old curing technology or it's a death wish.

With the concrete house in an hour, the idea is that the printing step takes 1 hour. The entire process would be performed in a massive temporary box, and an inert filler material would be placed at the same time as the concrete to fill voids and eliminate slumping. This would allow a second floor to be "printed" over the room voids below.

The printing step would be accomplished in 1 hour. The curing step could happen over a week if need be.

So you would have house specific forms requiring massive reuse? You can pour into a typical form in an hour too. The hard part is the forming and installation of rebar. It might work, but it seems tough to make it marketable.

I've built several printers (albeit FDM models) and am interested.

Email me at gmail, and we can talk

Can you give us an outline?

Sure, the basic idea is to separate data transfer from what I call locking. Data transfer can be done quickly, then locking can happen passively at whatever rate it needs.

Data transfer can take many forms. A simple example is an array of needles,pointing downward, forming a horizontal plane. The plane of needles is withdrawn upward through a granular material and dropplets of glue bind the material only where it should be hardened. In this example there is a resolution tradeoff, but you can see that the "printing" is basically completed in one pass of the plane through the print volume.

Holographic processes could transfer data to the entire volume at once.

The key is looking at the problem as a data transfer problem. We are very good at moving data very quickly.

The number of pins in such a design would dictate your surface roughness. Where would you be loading the glue from, if they're also being drawn through the binder.

Also, glue doesn't set instantaneously. I can see only problems with this approach?

That is the resolution tradeoff with this particular method. Each needle would be fed from a tube connected to its own control valve. Non bound material would act as support while the glue (or other binding agent) worked.

In my imagination I see a chair made from shredded tires, and bonded with a silicon caulking like substance. The shredded tires would be filled into a container between the needles as the printing array rises.

This sounds similar to HP's $100k+ solution. There's another similar attempt I've read about recently. It's a form of stereolithography resin printing.

Instead of multiple needles in a granular material it's using multiple fast-aiming lasers pointing into a tank of UV-setting resin. The lasers shoot through a tank of translucent resin from multiple angles and where he beams intersect there's enough energy to set the resin without setting the surrounding bulk of it. A bigger array of lasers means more points can be set at once.

Normal SLA/DLP resin 3D printers already work a plane at a time, drawing up through a tank and having lasers hit that layer right at the surface. A bigger array of lasers means each layer gets set faster up to the limit of the resin's minimum cure time.

Most professional and some high end hobby 3D printers aren't FFD/FMD any longer. Resin, metal sintering, and other alternatives are leaving fused filament to the hobbyists. There's no reason sintering couldn't use more lasers, speeding things up to a practical limit of the metal's time to cool.

Personally, although it's still mostly hobbyist size and speed, I just recently backed a ceramic extrusion printer. It can use a variety of cheap and readily available materials to create heat-resistant, durable, food-safe items. Items can be smoothed and some details added before firing.

For your solution, have you considered a two-part epoxy as an alternate to granules and glues for finer resolution?

Have you seen the subsurface laser engravers that create those 3d acrylic point cloud portrait things? I looked into them while doing some research on this. They basically use a focusing lens to create the hotspot at a point within the block of acrylic. They can do up to 20,000 points per second in a plane by aiming some galvos, and the material is moved slowly to create the 3rd dimension.

The granular material plus binder combinations are essentially unlimited. I even thought of making 3d printed treats by using sugar with water binder, or rice crispies with a food safe binder.

Cool thinking! The linked printer can be seen as optimized for serialized throughput, whereas in the end the total amount of data transferred is what matters.

So parallelize!

Are there FDM printers with multiple nozzles extruding in the same plane? That seems like an obvious incremental step from regular single nozzle printers.

I've got an unfinished 6 head printer that was intended to print 6 identical objects at a time.

Could also be used to accelerate the print of a single item by having different nozzles work at different parts of the design at once. Would need lots of updates to the slicer and solving some tricky optimization problems.

Autodesk's "Project Escher" was intended to do just that. The machine was built by Titan Robotics and uses Smoothieboards but the main issue is getting all the controllers to recognize where the other heads are in real time. Setting the CAM system to attempt to keep them all sync'ed doesn't always work as intended due to the way the firmware on the control board processes the incoming gcode.


Here's a video that I like that kind of demonstrates how fast things could go if you optimise for speed this way...


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