Final-year/MSc Geometric Modelling Course

Chapter 6

Additive Digital Manufacturing directly from Geometric Models

The phrase Additive Digital Manufacturing sounds very fancy and abstract, which it is - it's an attempt to be general enough to encompass all that is meant. But the usual names for this are Rapid Prototyping or 3D Printing.

The idea behind all the many techniques of Additive Digital Manufacture is pretty simple: move something about in a three-dimensional volume that leaves solid material wherever it goes; use that to build up any shape you want in layers. The material is usually a polymer of some sort, though some systems work with metals too.

Why bother, when we have normal CNC methods of subtractive manufacturing?

Well. There are two very big advantages to additive manufacture:

The forces needed to make things are very small, so the machine can be weak but still accurate.
Because everthing is built up in layers from the bottom, the moving parts have nothing to avoid.

The second advantage is particularly significant. Calculating a path for a cutter in a CNC mill to machine a single surface is fairly easy (we'll cover this later in the course), but - while the cutter is doing that - it must not hit any other part of the object that is being made. Solving that collision-avoidance problem is very hard (we'll do that later in the course as well). But additive manufacuring bypasses the problem all together.

History

People have built things in layers for hundreds of years, of course - think of cutting round contour lines in thick cardboard to make a relief map - but the first public appearance of the idea of additive manufacuring under computer control was as a joke. The extraordinarily gifted and prescient David Jones used to write a spoof inventions column in the New Scientist (and subsequently in the Guardian and Nature) under the pen-name Daedalus . In the New Scientist of 3 October 1974 he proposed a machine that would steer two laser beams using computer-controlled mirrors such that their point of intersection was in a vat of liquid monomer; one beam would not have sufficient intensity to cause the monomer to polymerise photochemically, but that would happen where the intersection point was. The computer would then be able to create a solid by moving the intersection point about.

The same idea was subsequently patented by Wyn Kelly Swainson in 1977, though Swainson had filed the patent (and had hence originally had the idea) in 1971.

Practical systems - SLA

Jones and Swainson's idea was developed, and now forms the basis of what are called StereoLithography Apparatus (SLA) additive manufacuring machines.

In practice, only one laser is used, and it causes photopolymerisation at its focus, which is made to coincide with the surface of the monomer. The object being made is lowered, layer by layer, as manufacturing progresses. Despite the name rapid prototyping, this takes several hours, even for comparatively small objects. The system also has to lay down a support structure that snaps off when the build is finished. This stops the manufactured part moving about and holds up any overhangs. Computing the support structure is much simpler for any object than solving the collision avoidance problem mentioned above, as we shall see below.

There is a variation in the SLA process which doesn't use a laser. Instead it has a UV lamp behind a conventional data projector LCD. This projects a pattern onto the surface, and so can expose a whole layer at once, which is much faster.

Practical systems - SLS

SLS stands for Selective Laser Sintering. A laser is still used, but this time to heat the material forming the object to melt it. The material is layed down as a powder, the laser then sinters regions of that together, the next layer of powder is added, and so on.

This process has the big advantage that the powder automatically supports the object being built, so no support structure is needed. Because the laser heats the powder beyond its melting point the whole process often has to be carried out in an atmosphere devoid of oxygen to prevent fire and charring. Nitrogen is usually used. An IR laser (usually CO₂) provides the heat; in some machines this can be sufficiently intense to sinter metal-alloy powder. Alternatively, a metal powder coated with polymer is used, the polymer being sintered together as before. The part is then subsequently placed in a furnace, which burns off the polymer and sinters the metal.

Here is a typical SLS machine (SLA machines look much the same, though their internal works are very different, of course).

Practical systems - FDM

Fused Deposition Modelling (FDM) is perhaps the simplest process. Here molten molymer is extruded out of a fine nozzle, which is moved over a surface to form a layer of the part to be built.

The surface is then lowered, and the next layer is added. The layers fuse together because the molten polymer slightly melts the layer underneath. FDM (like SLA) needs a support structure to be added to allow overhangs.

FDM is not quite as accurate as SLA or SLS, but the machines are cheaper. The parts made are also not quite as strong as SLA or SLS parts (especially in the layering direction), but are still good enough to use as working components in machines.

FDM is sufficiently simple to be installed in a desktop environment, rather than in a workshop.

Practical systems - powder gluing

This process is rather like SLS, but instead of scanning a laser over the powder layer to sinter it, an ordinary inkjet printer head is scanned over it. But instead of ink, the head contains glue. The printer prints the layer pattern, and another layer of powder is added, just as before.

This idea was originally invented at MIT. The parts produced are weak compared with the other techniques, but they can be soaked in a resin subsequently to give them strength. Also, as you can see, you can add colour to the glue...

The machines are roughly the same size as FDM machines, and don't need to be housed in a special environment, though the powder can be messy.

This and the LCD SLA are the fastest of all the technologies described here.

The standard additive manufacturing geometrical data format

All additive manufacturing systems use the same data format to describe the solids to be produced. You might imagine that, as geometric modelling was already well-understood when additive manufacturing started to take off, a geometric model would be used. But it's not. Foolishly, the industry decided to use an uncoordinated list of triangles covering the object, so there are no guarantees of solidity. The list is called a STereoLithography, or STL, file. Here is the start of one:

solid my_rp_object
facet normal 0.000000 0.000000 1.000000
outer loop
vertex 2.394684 0.987658 1.491202
vertex 2.492724 0.705324 1.491200
vertex 13.910414 3.727284 1.491202
endloop
endfacet
facet normal 0.000000 0.000000 1.000000
outer loop
vertex 2.319906 1.151621 1.491200
vertex 13.910414 3.727284 1.491202
vertex 12.471736 7.200560 1.491202
endloop
endfacet
.
.
.

That example is a text STL file; there is a more compact binary equivalent too.

The idea of just using triangles, as opposed to allowing curved surfaces, seems like a bad restriction, but it is not. It trades filespace (cheap) for algebraic thought (expensive) and assumes (correctly) that you can get any accuracy on any shape you want as long as you make the triangles small enough. But there is no Euler-type topological information stored along with the triangles, which was extremely negligent of the designers of the format - it would have been trivial to include and would have made everybody's life subsequently a lot easier.

Support and other calculations

Given an STL file of an object to produce, all additive manufacturing systems have to slice the 3D information into layers, calculate how to fill each layer, and (in the case of SLA and FDM) generate a support structure to fill-in under any overhangs.

The slicing calculation is straightforward: a flat plane either cuts a triangle giving a line segment, or it doesn't. All the line segments from all the triangles that are cut then have to be joined end-to-end to form polygons. There are various ways to do this - I have programmed it by superimposing a quad-tree with the terminating condition that a quad should contain either two different, or no, line-segment ends.

The polygons then have to be offset, to allow for the thickness of the thinnest line that the process can create (this is not needed for all processes). I have done this by storing the polygons in set-theoretic/CSG form (which makes the problem very quick to solve - keep the CSG expression the same, and just move the half-planes), but I don't think that is a common commercial approach. Polygons also have to be hatched, to fill them.

The support structure needed can again easily be calculated using set-theoretic techniques. If you start at the top layer (call the shape in that layer L₀) and go down to the next layer (L₁), then the support structure at L₁ is just L₀ - L₁. As you go down to the next layer the support is (L₀ U L₁) - L₂, and so on. The computer works down from the top recording all the results, then builds from the bottom up.

And finally...

This was made by a variation on the SLA technology described above. These pictures are taken by a scanning electron microscope - the wavelength of light is too big for the bull to be seen clearly under a light microscope.

And, at a rather larger scale:

There's a guy at the University of Southern California who wants to use additive manufacturing in house building.

You may be interested to see our own additive manufacturing research - we are trying to make a rapid prototyper that can make almost all its own component parts, and then give it away free to everyone... It is described here.