Saturday, March 26, 2005

Lay down white polymer powder for sintering as in SLS, but run an inkjet printer head over it as in the MIT system. In contrast to that, though, in the printhead there is a substance pretty rare in RP printheads, namely black inkjet printer ink. Behind the printhead you have a strip-heater the width of the scan. Now, the black printed-on powder will absorb more heat than the white, so you adjust the strip-heater temperature until the black powder just sinters and the white stays raw.

This is the opposite of Selective Inhibition Sintering.

A quartz-halogen heater would probably be best.

It might be a good idea if the ink did not wet the powder, as that way you would get a denser black because it would not soak in.

You may also have to flood the thing with CO2 or N2 to stop oxidation.

P.S. (6 April 2005): This is patented by Loughborough University, so we'll leave work on it to them.

Thursday, March 24, 2005

Neat suggestion from John Davis: use optical encoder strips from Precision Images to solve the measurement problem; they cost about $2 each and will resolve down to 0.05mm. Agilent make the sensors for around $10, part number HEDS-9720-P50.

Wednesday, March 23, 2005



This is the blog for the RepRap project. See that link for details.

This blog has a threefold purpose:

1. To solicit and to acknowledge contributions to the RepRap project from other researchers,
2. To get project ideas into the public domain as soon as possible, to ensure that they are unpatentable, and
3. To act as a project diary.

As a consequence of Item 2, in particular, some items are a bit scrappy and provisional.

If you have something to contribute, please get in touch. But understand that all solutions offered must be open-source and free (as in not costing anything, as well as in freedom...).


We may be using the results of the OpenRP project as a way of storing and transmitting RP designs. (Thanks to Sven Johnson for drawing OpenRP to our attention.) Check out that link for details.

Likewise, the RepRap RP machine will probably be using the software from the LinuxCNC project for its control. (Thanks to Josh Storrs Hall for drawing LinuxCNC to our attention.) Check out that link for details.

One thing we need is an open-source 3D CAD system that can output OpenRP format or STL files; we'd give it away with the machine. Our own Svlis geometric modeller is too experimental (i.e. fancy, but buggy...), so we're looking at BRL-CAD, which is a robust geometry engine. But its user-interface needs a lot of work... Another possibility is VTK (thanks to Deelip Menezes of the OpenRP project for the suggestion), and another is Blender and BlenderCAD (thanks to Michael van der Linden for that one).

We probably don't need anything with very complicated geometry (we're unlikely to need NURBS surfaces, for example). We could probably manage with just planes, spheres, cylinders, cones, helices, and tori. Those, together with sketch-and-extrude and sketch-and-revolve functions, and the ordinary CSG operators should just about do.

We might also look at the various Povray scene editors that there are out there. One piece of software (thanks to Vik Olliver for the suggestion) is Art of Illusion; we'll look into this.

Click here if you have another solution to our CAD problem.

Our current intention for RepRap is to put a microcontroller in the machine itself (probably a PIC) to control motors, temperatures, and timings, and to have all the smart stuff happening on the USB-connected PC. The code will be open-source, of course. And, though we'll almost certainly do all our development under Linux (probably in C++ or Java), we acknowledge that realistically it'll have to run on Windows as well.


We may use thermoset polymers, or thermoplastics. The big problem with thermosets is recycling (see Background to the Bath RepRap Project) , as they need a lot of energy to break down back into a monomer because you have to bust lots of covalent bonds. Thermoplastics are much easier, because they can be dissolved or melted and (more or less) you only have to overcome van der Waals forces. So we have a strong long-term incentive to go for the latter. However, the short-term get-the-thing-working incentive is towards the former...

For electrical conductors we may use Wood's metal, or conducting fillers in polymers.

Something that looks particularly promising is a mixture of bis-phenol-2 bis(2-hydroxypropyl) methacrylate [Bis GMA] and tri(ethylene glycol) dimethacrylate [TEGDMA] monomers with a camphoroquinone photoinitiator and a tertiary amine as a reducing agent . Added to that would be a filler (probably glass particles of a few microns in size). All this is a fancy way of saying dentists' white filling material, and it has the following advantages:

1. It is benign - dentists put it in your mouth...
2. It is dimensionally stable - it doesn't change volume when it sets
3. It is stiff - when laid down it retains its shape against gravity
4. It can be polymerised with light from blue LEDs - see this link
5. It is tough and hard-wearing.

We have an RP design for a syringe pump - see below - and this would be ideal for applying this material. We suspect that, if we use dendritic silver as an alternative filler, it will make a good electrical conductor too.

Other materials we are looking at are ABS, PVA (which is water soluble), and DuPont's Elvamide nylon multipolymer resin (which is soluble in alcohols).

Of course, if we decide to go with thermoplastics, we also get thermosets for free, as we can use RepRap to make a thermoplastic mould in which to cast the thermoset (or plaster, or ceramic slip...). All these latter can go up to high temperatures if need be, unlike the thermoplastic.


The vast majority of existing rapid prototyping machines work using Cartesian X, Y, Z coordinate axes. This is an obvious way to do things, but we think that it may be better to make a polar machine that has a radial arm, a turntable, and movement along the axis of the turntable. This would have a number of advantages, the two principal ones being that it would be easier to make the machine itself accurately, and that - when working - it would manufacture much higher-quality rotationally-symmetric parts. It is no accident that the lathe was invented before the milling machine...

At the moment we are looking at the possibility of such a machine that will build by depositing a thin stream of material from a syringe pump, creating the design up layer by layer. We will actually have two streams - one an electrical insulator, the other a conductor, plus possibly a third for support material for overhangs.

The RP syringe pump. The only non-RP parts are the syringe (which only costs a few pence), a length of M5 studding plus an M5 nut, four M3 screws, and the motor (245-6089; available from RS Components - click the link). We have a later versions with an optical counter on the motor shaft for precise metering, with heating jackets, and a whole range of different capacity syringes.

An alternative to the support material is to use the build material itself as a support, which then gives the problem of separating the two. One way to do this is geometrically - we would have the computer ensure that the points where the support touched the built object were thin and weak, and so it would break away easily. Another it to have a third deposition head that puts down a very thin layer of release agent on the top of any support, so the build doesn't stick to it. We suspect vegetable oil will work well for this.

Moving away from syringe-metered deposition, and slightly crazily, there's also the possibility of wiping a flat thin layer of thermoset paste over the entire build area, then polymerising the shapes needed in the whole layer at once by shining a (UV or visible) light through an image on a data-projector LCD to project that image on the layer. This would be very fast. Then we'd add the next layer, and so on. Problem is: the machine couldn't make a (doubtless expensive) LCD to replicate itself. We are happy to include cheap widely-available bought-in parts (screws, washers, microelectronic chips, and the odd electric motor), but not expensive ones.

P.S. 26 March 2005: There is such a machine already. The machine is pretty neat, and also low cost. Check it out at Envisiontec, who use a mirror chip (like the ones in cinema projectors) rather than an LCD.

A possible variation on this is to deposit a thermoset paste film, to set it by scanning a blue/UV LED over it, to wash away the un-polymerised paste, and then to go on to the next layer.

As to the base on which the build starts, the current plan is to have a perforated flat plate with a partial vacuum behind it holding down a sheet of ordinary kitchen aluminium foil upon which the build will be initiated. No need for a vacuum pump, of course - just attach a venturi to a water tap as chemists have been doing for centuries.

Measurement an calibration - we are confident that we can rapid prototype repeatable (if not accurate) linear axes. The repeatability means that - if we can calibrate them - the computer can keep a map, and thus make them accurate as well. But we need a cheap way to get a digital readout to 0.1mm (or better) of displacement over a range of at least 300mm. We are currently experimenting with Moire fringes (which one can print easily on an inkjet printer), with a modified design of an LVDT that we came up with in response to the problem, and the capacitively-linked "comb" patterns that are used in digital measuring callipers (**CHEAT WARNING** these are now so cheap - check here - that we may be tempted to use them in the first version of the machine...). But if you can think of a better way, that would be most welcome.

What's the LVDT idea? Well, conventional LVDTs are wonderfully accurate (and linear - hence the L...), and it's pretty easy to make one. But they take up much more room than the displacement they measure. Ideally, we need something not much more than 300 mm long to measure 300mm. So the idea is to take the LVDT core out and wrap it round the outside instead, and to have multiple coils instead of just three, switched in sequence by the controlling electronics. It would go something like this:

You could keep adding alternate excitation and sensor coils on the right-hand end to cover any length you like. Note that all the wires come out the left end, meaning that it's easy to assemble into a design because the right hand end is always free. When it's switched on, the microcontroller in charge can find the cylinder (which is the thing having its movement measured) simply by energising each excitation coil in turn. The whole device then acts like a chain of LVDTs that's as long as you like. Each separate section would probably need different calibration data, but remembering that sort of stuff is what computers are for. It may even be possible to wire it up simply without switching electronics and still pick up position accurately:

Click here if you have another solution to our measurement and calibration problem.

This page is powered by Blogger. Isn't yours?

Subscribe to
Posts [Atom]