Beltless Drive X-Y Axis (Level 2 - Techie) (Page 1 of 5)

I had thought about this awhile ago, but there are some drawbacks discussed when this guy used it:

http://www.youtube.com/watch?v=FP2ciUBT6nY

I haven't really seen to many issues. The Tantillus has been using this system since they started. I notice that some people had concerns about stretching. As soon as you feel the spectra line, you will know that it is not a concern.

Also, the bobbins in that the SCAD file will create are designed to be the same size as the stock pulleys so your steps/mm will be just about dead on depending on your machines setup. I would still recommend calibrating. I was 1.1 steps/mm different than the stock pulleys.However, you DO NOT need to replace the pulleys. I did that on the X-axis simply because I could.

Not sure how relevant that is with our relatively short lines. The question for me is cost, longevity, quality of prints and ease of use.

Compared to the belts, the spectra line wins out in all those categories.

I am just about out of room. Looks like enough for Dual bowden but not a whole lot else.

Nylon/mono filament will stretch, braided - as you used will have zero stretch...  (using it just now in the inmoov robot )

Not to take away from the original post or anything, but I was wondering if using some anti-backlash lead screws would be more accurate. I've read that quite a few people have upgraded to acme lead screws for the Z, but why not do that to all the axis?
Instead of using a screw with such fine pitch as the Z axis, maybe one with the pitch more angled so that it travels farther with less turns?. (Sorry if my terminology is all "screwed," I don't know what I'm talking about most of the time.) But, even if it were slower, I would definitely sacrifice speed for accuracy if it worked. I would think using screws instead of belts would also eliminate all the belt calibration and tension fiasco we have to go through just to get more precise circles. All the corrections would be done through the firmware instead of fiddling with the belts and trying to figure out what an arbitrary word like "twang" means.

I think the fishing line idea is a great alternative to the belts, but it doesn't eliminate the problem of making sure everything is tensioned correctly. Although, it is much more economical to have a few hundred yards of fishing line to replace if anything goes wrong. I just think it's for someone that has much more experience with these machines than the average tinker/designer that wants things to just work.

In a completely different direction, I've often wondered about making the mechanical bits really really cheap, but having some sort of feedback system where you spend most of the money (lasers, diffraction gratings, optical sensors, who knows what all sorts of stuff :-) to measure where things actually are. Seems like that could be just as exact without needing precise mechanics and you remove all the tedious need for constant calibration.

ACME screws have more backlash than threaded rod, unless your ACME nut is a split delrin anti-backlash type. You guys should know that belts are inherently anti-backlash. What a screw would give you is more precision from giving you more steps/mm than the belt does.

But quite frankly, X/Y precision is not really as much of a problem. It actually exceeds the precision of the width of the extrusion by alot. Laymans terms: X/Y precision of the stepper is much much better than the plastic being gushed out of the nozzle to form perfect, consistent 0.42 mm width. Moire effect can be seen by the eye. Also, extrusion is not perfectly consistent during moves. It is not stopped during straight move ends-to-beginnings, causing that little extra roundness to the corners, and arc movements leave less plastic on the outside of the arc and more plastic on the inside. These issues have more bearing over precision than X/Y movement.

Also, while I agree that having mechanical feedback would be an awesome way to go, currently the controls are not setup for that (and may not be capable of it). The firmware barely keeps up with trying to calculate the movement for the steppers output pulses. It takes quite a bit more math processing to also read a feedback system and then re-apply corrections on-the-fly. This is the difference between raw stepper movement and servo movement. In order to accomplish this, alot of work would have to go into the firmware, and the current processors used by the open source community may not be adequate enough.

Can you link us some sources on that?
edit: as in links to the sort of math processing your referencing.
SD would love to help port some nice control theory to marlin when we have a second.

Yes, they are more precise. Precisely made to have backlash:

http://en.wikipedia.org/wiki/Trapezoidal_thread_forms

To put it simply, the cross section of the threads are triangles with the tips cut off. The threads on the rod will purposely have a smaller diameter than the threads on the nut to ensure that there will be clearance for the rod to smoothly screw in. This tolerance means you are guaranteed to have a specific amount of backlash (actually, a precise amount of backlash). If you have ever worked on a manual mill table, you would know this. Heck, my vice is the same way, it requires a little bit of reverse travel before re-engaging the threads.

http://lmgtfy.com/?q=delrin+acme+anti-backlash+nut

Thats what I was talking about. Not a standard ACME nut.

I have looked through the Marlin firmware and seen how the stepper pulses are generated. Although I am impressed that they were able to get the pulses derived from long integer  math for up to 6 steppers simultaneously using the hardware interupt, it is still an 8 bit processor running at 16MHz. Lets not forget it is processing USB serial commands and PID floating point math. Add SD card reading and LCD control, and the firmware even says it is pushing the limits.

In order to get a servo system, you need complex floating point math.

This would require constant monitoring of each axis with a comparison of its current actual position (read at the same frequency comparable to the stepper pulse frequency) with that of the expected position (something the firmware does not do, it just times out microseconds for each pulse in hardware interupt) and then compute not only the difference, but the required change in stepper pulse math to 'catch up' in a reasonable manner during the move itself. And be able to do this for a minimum of 4 steppers simultaneously.

Thats too much for an 8 bit processor. Most servo systems I have ever worked on have a central processor that communicates a 'move' to a drive. The drive controls a single motor and houses its own 32 bit processor to perform the feedback and math.

While it is viable for an 8 bit processor to have a constant feedback (possibly serial communication with the $10 china scale, or even an encoder strip used widely by professional inkjet printers (transparent strip with really tiny lines printed on it, and an mouse opto sensor reads how many lines go by)) and be able to adjust a ramp indexed stepper movement, it would be able to do so for only one axis while still trying to do other functions. Not 4 or more axis at the same time. Too many math cycles during each stepper pulse. Remember, it has to get the feedback value, do the math to figure out where it should be during the move, calculate the difference, then apply the floating point math to re-adjust the velocity of the move and apply the new stepper pulse all during the time it takes for one pulse. For 4 to 6 motors.

In order to get to this level, you would need a 32-bit processor, preferably one with an FPU to reduce math processing cycles. Then you may be able to get this working.

I personally think the X/Y precision is a big problem on my printer. All my circular objects come out as skewed ovals. Maybe yours was calibrated correctly from the factory or you know how to calibrate the belts yourself, but for people with less luck and/or knowledge, it's a huge nightmare. I've spent countless days trying to get the belts tensioned right with no real progress.

The extruder may not be perfect, but within 3 days of receiving my printer, I got it close enough to where I can say that I am satisfied with how the parts turn out. I don't think any FDM printer can ever have perfectly sharp corners or have perfectly precise walls. It's using melted plastic as a medium and not something more accurate like wax, resin or calibrated powdered materials.

Inaccuracies of the extruder are probably a few microns to a mm if dialed in well enough, but the slop from my belts make my parts off by 3-5 mm some times.