Full 3D Metal (Jacket) Printing @HOME

3D Metal Printing for Under $3000; down from $100,000 and $500,000

3D Printing is revolutionising manufacturing, but 3D printing at HOME is still a novelty. Department stores now have low-end 3D printers in their PC electronics departments, yet their usefulness for making real “stuff” is debatable. Certainly these printers do not render the type of things we, ourselves, tend to buy in same said stores.

I aim to change that.

Executive Summary A.Game.Changer.

Common 3D printers have material limitations that prevent creation of parts, for example metal, that compete with parts created through traditional methods. Dedicated metal 3D printers do exist but are a) very expensive, b) slow and c) parts are difficult to process further for adequate surface finish. We demonstrate a process whereby common 3D printers can be employed using a filament type composed of approximately 90% metal powder and 10% binder material whereby parts are subsequently sintered in a kiln. The binder is removed in this step leaving pure metal. This is analogous to what laser sintering 3D metal printers achieve in a single step, but has several advantages: The intermediate product is softer than the final product, and therefore can be easier processed to attain an adequate surface finish in an intermediate step. All steps can take place in parallel. Hence part 1 can be sintered, while part 2 is surface-finished, while part 3 is printed. We call this a pipeline. We do not rely on a furnace for sintering but rather on a commodity glass fusing kiln (cheap). Commodity 3D Fusion Deposition Model (FDM) printers are much faster than Laser Sintering 3D metal printers. And finally, the combined cost for all devices in the pipeline is south of $3000. This means well over 30-150 such production pipelines could be built before we approach the cost of one metal 3D printer. This is the kind of game changer that can take a technology from being used primarily for prototyping to being used primarily for production.

The pipeline is shown here (also reproduced later in the article).

Let’s Start With Why…

Ever since I was a boy and got my first computer, I was fascinated with one thing: I could BUILD MY THOUGHTS – DIRECTLY. It was a tangible, highly tractable thing. In other engineering disciplines, say the automotive industry, engineers design and others build — in a long and protracted supply chain. IT has always been different. You, the originator of an idea, always had the power to make that idea come alive. Only one limitation: the type of thing you got back was the type of thing you put in. Or differently put, the input was “soft,” an idea, information. The output was still “soft,” the result of a calculation, an image, but still just information. Thus IT created a meta function of the type

information -> information

As simple as this looks, it’s the subject of two fields of mathematics called Category Theory and Lambda Calculus.

The Theory… Please Indulge a Digression

In a nutshell, Lambda calculus warrants composibility. If I take information, apply some function to it and I receive information, then I generally can, subject to adaptation, take the output from one function and provide this as input to another.

An example that people can relate to might be Google Map data being overlayed (composed) with search results. You searched for restaurants and Google presented you with restaurants in your area – overlayed onto a map of that area. What if mapping was only ever representable as something in a very different “category?” Say, sand… One might inscribe the search results on the sand at a nearby beach, but as for using your computer or smartphone… that would be impossible then.

Fundamentally, Lambda Calculus and Category Theory deal with making things composable. To date, these concepts have been mostly applied to information, that elusive “soft” ware. One of the pioneers changing this is Dr. Christian Schafmeister of Temple University in Pennsylvania, USA. Christian, is building a meta matter compiler called Clasp in Common Lisp. As you might imagine, the goal is to compile through not to information, but matter, here nanotechnology matter blocks that create other blocks of matter. His work has been funded by the United States Defence Threat Reduction Agency. You can watch his Tech Talk @ Google here.

Closer to home, we can apply similar concepts to 3D printing. Mat Keeter from the Centre for Bits and Atoms at MIT has created AO, a tool for programmatic computer-aided design. AO is written in Scheme, a descendant of Lisp. AO can be used to model shapes in a programmatic and algebraic way. As a Scheme, AO has the Lambda calculus at its core. Scheme has an excellent research community – see Racket Scheme. As with ClaspDARPA funding is never far…

Expanding on our Value Proposition…

We said earlier that common 3D printers of today do not make the things we ourselves buy. Web archives like ThingieVerse are full of items that occupy a spectrum of useful to merely novel. Part of this is rooted in material science. Common 3D printers utilise what is referred to as the Fusion Deposition Model. In a nutshell: layering. The methodology comes with a range of caveats. Cheaper 3D printers are capable of lower temperatures only, limiting application chiefly to cheap plastics such as PLA. More expensive 3D printers can handle better plastics as well as exotic materials like wood or bamboo and filaments infused with metallic particles. Hardened steel extruders are capable of extruding carbon fibre filaments, yielding considerably stronger parts. One common problem is warping of materials right on the print bed upon cool down — which impedes dimensional accuracy. Subsequent heat exposure may further lead to warping of the parts during ordinary use. The temperature at which this happens is called the material’s deflection point. Another problem is providing an adequate surface finish: fusion deposition results in lines across the model. Finally, tensile strength is limited by the characteristics of the material and the manner of composition.

3D Metal Printers, by contrast, commonly utilise laser sintering methodology on metal powders. This means that metal molecules are sintered “in-place.” Sintering means high temperature fusing without melting. It achieves tensile strength comparable to metal parts produced via a furnace. Some parts requiring high temperature resistance, as found in missiles, are printed this way. Resulting deflection points are very high. Metal 3D printers vary in price from around $100,000 to upward of $500,000. Many vendor web sites request that interested clients call in for a quote. CSIRO, the Australian research body, operates what is termed Lab 22 and offers access to 3D metal printing. The unit cost for their printer is given at $1 Million. Lastly, 3D laser sintering is slow. Print cycles of around 12 to 48 hours are said to be common.

All under $3000..

We aim to solve the above problems for under $3000. That is our value proposition. Speed! Tensile strength! Finish! All under US $3000. To this end we leverage a process pipeline that allows individual steps in the process to take place in parallel: while part 1 sinters, part 2 is surface finished and part 3 is being printed. Critically, surface finishing takes place on a shape that isn’t as yet sintered, so easier to handle. Tensile strength is created in the final step of the process.

Unlike the movie icon & title in the article header suggest, we will not be creating full metal jacket munitions. We will make an arbitrary metal part, say a cube. The only material connection (pun intended) to the movie is the material: typical munition jackets are made of copper. We will be using copper. To start with, here is some of the tooling: Lulzbot Taz 5: $1650. A glass fusing kiln: copper sinters at 750C-1000C. This brings it to within range of a kiln as used by glass fusing artists. You can pick one up on Amazonfor under $500. The kiln linked to in this article, reaches temperatures of up to 1200C, which is sufficient to sinter copper. I picked mine up “pre-owned” for about $300 with a firing chamber of approximately one cubic foot, which affords me the ability to fashion massive metal parts – right at home. Next, we will be using filament from the Virtual Foundary. Their filaments are almost pure metal and roll smoothly off a high-end printer like the Taz 5 — which has a metal extruder capable of temperatures of up to 300°C (572°F).

The Algebra

We will model our shapes using AO in a “live coding” way that connects you to your creation with instant feedback. As we define shapes, a view port display them.

What we are trying to build is this copper cube. The below cube is unprocessed and unfinished. For the purposes of this demo, we will approximate the below: a rounded cube with a cut-out and a drill hole.

We start with the following definition:

I like to work at “unit length” where everything is simply based on size 1 with the part being scaled to suit at the end. This gives the following in the view-port:

We now need the cut-out. We define a cylinder with a radius and a length, rotate it 90 degrees about the y-axis (in radians) and move it into position where we would like the cut-out. We then define a “cut-cube” as the difference between the original cube and the cut-out. This is the software equivalent of subtractive manufacturing techniques, such as milling, to remove material from a larger part.

This yields the following cube.

We now repeat the process for the drill hole: define cylinder, rotate, move. Done.

This yields the following cube:

The cube in the original picture had an additional drill hole and was rounded differently, but you get the idea.

We said earlier, we were working at “unit length.” This means we defined a cube of -1,-1,-1 to 1,1,1. This was a sleight of hand. We are actually using millimetres here. So right now our cube is 2mm by 2mm by 2mm. I would like my part to be 2.5cm by 2.5cm by 2.5 cm, so I define my scaled part accordingly. Each axis can be scaled separately. We scale all three uniformly.

What we have done is to compose our model algebraically, one feature to the next: moving scaling, rotating, extruding, differencing. The output from each function became input to the next. Traditionally, these functions would collectively be steps in a manufacturing supply chain. In particular, differencing corresponds to subtractive manufacturing techniques. More significantly, although not shown here, the homoiconicity of a Lisp and the higher order logic of Lambda calculus afford us the ability to reprogram all of our functions — essentially without touching them.

Our model can now be rendered as an STL file and imported into so called slicer software, like so:

We now have the following ready to go to the printer.

The print time, depends on a number of factors, chiefly fill density and print resolution, so we can vary this according to whether we are prototyping or producing a part for final use. At present, my printer tells me to expect 39 minutes and 7 grams of material to be used. It’s wrong about the latter because it thinks it has PLA. What it actually has is a material that is approximately 90% copper and 10% binder material.

Printing & Finishing

After printing and part finishing our cube looks something like this.

The heat has produced oxidation, but scraping with a ruler reveals the shiny copper beneath, ready for sanding and polishing. We also have a smoother surface than before.

The Missing Link

The smooth surface is the missing link that I’ve casually glossed over so far: how to produce a smooth finish. The answer is less obvious than it seems. Here is from CSIRO’s web site:

Credit: https://www.csiro.au/en/Research/MF/Areas/Metals/Lab22

Here is some copper jewellery that I have been printing, a small seahorse and a colibri pendant:

Common to all three metal parts is the surface texture. Why not simply sand and polish? Because not every part has regular surfaces. How might one sand the seahorse? One of the advantages of 3D printing is that convoluted shapes can be produced with the same ease as regular shapes. But how do we rid ourselves of those pesky lines and ridges? You will want to do this prior to sintering as the material is still softer!

Here is an older part, already oxidised – as copper does. This teddy bear has a far more convoluted surface than our cube, yet the surface is remarkably smooth.

How to achieve this in a systematic way (not manual) is what I’ve been working on. I have nearly perfected the technique and once I do there will be another blog article. The cost added by the technique is approximately another $400.

There is also far more here than was possible to demonstrate in this short article, owing to the homoiconic nature of AO and our process in general. The composability that has been shown is a very small slice of what is possible.


I am also researching non-planar infill patterns that maximise lateral tensile strength for parts printed in metal and carbon fibre composites. This expands on already existing infills that have non repeating patterns in the z-plane to study the effects of effectively interlocking infills that leave the shell undisturbed.

Meanwhile, happy 3D printing.