Machine design - optimizing for print quality?

Engineering is largely about picking your tradeoffs. The Mendel is designed to be self-replicating and require no complex tools, which makes total sense in some contexts. I want to build a lower-cost 3D printer which makes minimal compromises in terms of the print quality.

With machine tools (the area I'm most familiar with) it's all about rigidity, so mass is seen as virtuous. With a laser cutter, acceleration is a much bigger issue and there are no cutting forces, so mass can work against you.

I have a benchtop CNC lathe and mill at my disposal, so making precision metal parts is no big deal. I could build a Mendel or Huxley with aluminum brackets and 12mm/.500" drill rod, fully-enclosed linear bearings on precision shafts, and use leadscrews instead of belts. Or, I could just build the whole thing using T-slot extrusions.

Question is, is any of this worth doing? My sense is that print quality is a function of X parts positioning repeatability and Y parts extruder consistency. For all I know, X is 1 and Y is 100, so my effort might be better spent focusing entirely on extruder optimization?

I'd say extruder and x/y/z repeatability are pretty equal in terms of importance to quality.

The mass issue with a 3d printer is a tricky one, mostly because it interfered with the two things everybody wants; low cost and high speed.

While there are no significant cutting forces, the mass of the extruder needs to be supported as well as the forces to quickly move said mass.

People have tried (with varying success) to use an extruder motor separated from the extruder by the "bowden cable" method. Unfortunately the lag between motor motion and extrusion (caused by the elasticity of the filament) leads to problems.

I'm going to make an educated guess that stratasys has the right idea with the uber rigid frame, with extruder mounted in the head. It gives maximum flow control since you don't need to predict the elasticity of the filament. I think their print quality says a lot about the design...

I've probably said this before, but If I had all the CNC equipment and my only concern was quality, I'd probably use a t-slot frame, machined brackets, wide Kevlar/urethane belts with high tension, and MINIMUM 12mm OD ground shafts with linear bearings. If you can get adjustable linear bearings which allow you to have a high preload, all the better.

So, here is a question: is there a consensus on what is the best configuration in terms of machine axes? Looking at the Stratasys printers it seems like the extruder moves in X and Y and the table moves in Z, like Darwin, while Mendel moves the table in Y and the extruder in X and Z.

In terms of the Z axis, I would be tempted to have the table on linear bearings at four points, with a screw in the center below the table. The screw would be stationary while the motor spins a nut to raise or lower it. This would double the height of the machine relative to the Z motion but you could stick electronics or stuff in the base and if you're looking at say a 4-6" Z stroke, it's not that much. This is roughly how the spindle on my mill works and would seem to be very stable, if you can get everything aligned right--which should be doable if you start with highly-square T-slot tubes and machined supports for the bearings.

Also, I know it's easier to go fast with belts, but if the cost/power consumption of the motors isn't an issue, would screws be better? Ball screws can give very high speed and resolution and I also have some 1/2"-16/8-start precision leadscrews I've been looking for a use for. Screws certainly lend more stiffness and with a flying extruder that would seem to be potentially worthwhile....

I've been experimenting with different designs as well. I agree that a rigid platform is essential to good print results. My current project has high precision linear slides. I've ruled out lead screws due to backlash. Right now I'm using Kevlar belt drive system, however, I have some ball crews around ....which will be the ultimate drive platform. As stated before, the balance between weight/rigidity and speed is not easy to work around. For me the limiting factor is the electronics. My CNC machines, I can run high voltage for larger stepper motors and achieve good acceleration. The electronics choice for reprapers are much more limited. I'm currently using a gen 6 board, which can output up to 2 amps to the steppers (I think). This is not enough power to achieve my acceleration goals. I wish that there is a reliable way to run these machines using Mach3 or EMC. That would open up a whole different ball game.

If you have the screws I'm sure they will work fine. Belts can be just as good and for less $$, but since you already have the screws it really doesn't matter.

The axis configuration you use is largely decided by the machine's desired working area/footprint/simplicity. XY movement by the extruder works best for large machines, and separate x/y for the bed/extruder works well for smaller machines, but limits max height as the acceleration of the bed can wobble very tall parts off.

For up to about 12" I prefer the Mendel configuration. It is just convenient for mounting everything.

ball screws would be the way to go if we had a solid hardware that did not do micro pauses and had great interpolation. the belts allow the give needed for this, and microstepping allows the motors to absorb this a little better. sometimes x does not move for 2 steps between y but then it needs to move 3 steps a few times to keep the right angle. this is an example where the double precision float in software of the arm processor would come in handy. there would be the ability to time everything out to be spaced evenly. this way a motor would not stall out and inertia would not limit the speed or accuracy.

g0x0 y0
g1 x15 y40
would cause the lines to interpolate, but not very accurately.using belts allows for the micro-stops to not effect performance, and the belt absorbs this and has some give/take.

It wont bind with a belt,but you would need to get creative to get it to work on a lead screw of any kind.

stratys uses ball/screw setup in there 3d printers, but there hardware is a lot more advanced.

i have not looked into how interpolation is done in the teacup firmware but the regular reprap host used Bressenham algorithms.

Your quest for accuracy ultimately will lead you to the conclusion you need more processing power. if precision is the issue and you use ball/screw you will need to work on firmware...

At least the tabletop dimensions use a belt for the x/y axis, and something threaded for the z. Looks like a ~60mm/s travel speed for their low end printer, too.

[cdn.content.compendiumblog.com]

Edited 1 time(s). Last edit at 06/05/2011 05:39PM by Andrew Diehl.

EMC certainly can and has been used to run Reprap machines reliably. However, once you start using acceleration you'll run into print defects due to the fact that there is a significant delay between applying a change to the filament feed and seeing the change reflected in the extruder's output. Matthew Roberts has done some work to address this problem (See here).

I've found that having a short distance from pinch wheel to tip nearly eliminates the delay. aka why bowden extruders are horrible for accuracy.

I run my machine with roughly 2"" in-between, and will likely make it down to under 1.5" soon. Of course, active cooling of the extruder then becomes necessary.

[www.makergear.com]

The Mosaic is all the right compromises. Linear rail X/Y, multistart acme on Z, Geared stepper extruder, hybrid hot end, .35-.25 nozzle.

Not bad for $999

---

repraplogphase.blogspot.com

I think it's a little early to say how well or poorly it works. It is very new. Looking at the makerbot site you'd swear every print cam out looking perfect right after you assembled it, too

Anyway, back on topic...

there are some rail systems you could use, for example paint the back xbar on the top, and laser engrave lines 4mm wide, and along the length of the bar spaced .4 mm apart, use 4 optical encoders and you now have a laser accurate optical guide to reference in firmware for how many steps you take until where you want to be, and it is spaced at size along a linear rail.

since the entire bar is an encoder, and a laser machine in theory should be accurate to 1/1000th of an inch or 0.0254mm .4mm spacing with .4mm length of bar should work. and this is a common encoder design in machine equipment these days.

just make sure the encoder bar is placed where it will not be rolled over by the bearings....







IMHO.

Edited 2 time(s). Last edit at 06/06/2011 02:38AM by jamesdanielv.

jamesdanielv Wrote:
-------------------------------------------------------
> ball screws would be the way to go if we had a
> solid hardware that did not do micro pauses and
> had great interpolation. the belts allow the give
> needed for this, and microstepping allows the
> motors to absorb this a little better. sometimes x
> does not move for 2 steps between y but then it
> needs to move 3 steps a few times to keep the
> right angle.

How exactly is this any different than my mill, which has ball screws with 1 full step = .001" resolution and is powered by EMC?

emc2 vs reprap hardware. It could be emc2 is reprap hardware....
IMHO.

emc2 uses a more complex algorithm to determine motor positions. also it used double precision floats. (more detail data.)


read here as i can not simply explain how a float stores data.
[en.wikipedia.org]
single precision (arduino) (about 7 decimal digits).
double precision (emc2) (about 16 decimal digits).

Edited 1 time(s). Last edit at 06/06/2011 03:01PM by jamesdanielv.

Just to throw another possibility for X-Y out there - old thousand series Stratasys machines (not sure about newer P-class models) used a cable drum system for the X-Y stage rather than belts or leadscrews: [3d4u.org]

---

[haveblue.org]

Forrest Higgs is doing up a Darwin with Cables (here) on the Z axis, maybe this would work for all axis.

---

[richrap.blogspot.com]