Best posts made by bondus

I love strange kinematics, and when I discovered the parallel scara I had to build one.

The Duet did not have support for this kinematics in stock firmware but @JoergS5 had an almost complete implementation that we have developed further. The firmware works very good now.

The machine is made mostly of printed parts and items you can easily get hold of, and a hacksaw and hand drill. No machined parts. The precision is not the best but it still works pretty good (just a nightmare to calibrate). It's still a prototype/proof of concept so it does not look very good and many parts are stuff I had laying around.

The body rides on 3 12mm rods with LML12UU-like bearings, and has one leadscrew. The inner joints have two 6806 bearings per arm. The elbows use two 8mm thrust bearings each, preloaded with an m8 rod, a crude solution that works very well. That joint is very stiff. The hotend joint has two 6704 (20x27x4) very thin and light bearings inside. Lots of bearings.

The motors driving the arms are 48mm 0.9degree NEMA 17 with a 20t pulley leading to a 200t printed pulley. This gives a gearing of 1:10 which is very much on the low side. 1:30 would be far better. But upping the microstepping to 64 or even 256 saves the situation. A problem with 256 microstepping is that the Maestro starts to get stepping rate problem over 400mm/s.
Using a normal gearbox is not possible due to their backlash. A machine like this needs a gearbox with almost no play, like a harmonic drive. A belt gearbox works pretty good but the gearing is limited due to size.

With the 180 and 216 mm long arms the printing area is approximately 300x200mm. The actual print area is actually half-circleish shaped and quite a lot bigger.

The printer can move at blazing speeds and accelerations without any drama and shaking. The kinematics is similar to a polar delta but just 2 axis instead of 3. Currently it has a fair amount of ghosting at high speed/acceleration, but prints very good at a more reasonable 60mm/s and 2000 acceleration. I think the ghosting is due to a far too wobbly tower, or possibly flex in the belts. I have got some metal clamps for the 12mm rods and some more 3060 extrusion in the mail, it might help.

I made a little video showing it moving fast https://www.youtube.com/watch?v=MW8HApFoy38.

Here is an earlier version doing the first reasonably successful benchy print.

Those arms were made of small U-beam and far too wobbly, changing to bigger 20 mm square beams made a massive difference.

The very first little prototype to get a feel for the kinematics, a very useful little toy

I'll bring my 5 bar scara that was supposed to be shown two years ago.

Just have to get the thing moving again..

After some quick tuning of the hotend and more mechanical calibrations I managed to print this stunning benchy, at 120mm/s and 2500mm/s^2 acceleration. I'm surprised!

@iamthebest22 No updates I'm afraid. It works perfectly fine, I use it to print various small parts for other projects. The accuracy is not good enough for big mechanical parts, we need a calibration method for that.

It's actually a very simple mechanical construction. Some parts from my CAD files:
https://a360.co/3dNKGjZ
You have to make your own Z axis

Is there any way to build the firmware without eclipse? Good old make?

I just updated my firmware sources for the first time since january and getting it to compile and link again was the same nightmare it was back then. Hours of randomly clicking around on "Build clean", "refresh", "build all", restart eclipse, delete directories, pressing F5, sometimes getting one step closer to a successful build.
The problems are all in configuration and/or dependency management. Or my total lack of understanding eclipse.

I finally got it building again, slightly frustrated.

I am still shocked with the quality of the prints this machine can produce. It beats my CoreXY machines and is almost on par with the delta.

Apart from the need for better cooling and extrusion control (not the focus of this project) this dragon looks amazing. I hold it in a light to show the flaws as much as possible.

There is a very small amount of ghosting and some strange visible lines.

I think these lines are an stepper artifact, the gearing of the arms are still on the low side.
My delta machine (not duet but TMC drivers) have similar patterns but not as visible.

I managed to break my last ceramic heatblock, a fragile construction. And they stopped producing those They were amazing, heated up in 20s and could handle high flows. The RRF PID tuning used to warn that "ultimate temperature" was unsafe when auto tuning

Sorry for my rant before, I was far too agitated to post on forums.

The problems had nothing to do with my configuration and setup. I didn't change a thing between starting to try to compile and successfully building a working firmware. No build configuration changes, no source file changes, none of the files in git were changed. The issues I had was everything from strange compilation errors in CoreNG, something about a Pin class, to linker errors where it was missing syscalls.o or symbols defined multiple times,

I think one of the keys to why it got resolved was a project refresh (F5) in eclipse.

I noted the change of GccPath and the need for an updated compiler. The old q2 compiler crashed on the new sources, "../src/Storage/FileInfoParser.cpp:764:6: internal compiler error: Segmentation fault". Scary.

My dislike for eclipse is now even stronger than it was before

A hint for everyone building a parallel scara:
Before mounting the head move the arms around in all kind of strange positions and make sure that the distal arms are at the same height (or at the offset you designed them to have). Especially try to reach the same spot with different arm-angles. If they do not line up everywhere it will be impossible to get flat XY movement in the z-plane.
It can be quite a lot of work to get it all right but it is at the same time a kind of self-verification of the machine that all joints rotate perpendicular to the same plane. A very nifty feature.
I had to use a mixture of gentle filing on some plastic pieces and brute force bending and twisting the aluminium tubes to get this machine straight. Alu-extrusions from the local hardware store are not made for precision machinery.

It looks like typical delta waves on your printed parts. Bent diagonal patterns on flat surfaces. That's a sign that something is not perfect with the linear motion. It could be the pulleys, idler, linear bearings, ...
The bend in your patterns is not very sharp, the predator is a very large machine. A smaller machine will get sharper bends. If you move the model to the edge of the platform the angles should change, and the patterns will get tighter or more spread out depending on what/which towers is the cause.

When it is motor/driver related you usually get a tight, highly repetitive patterns. If you turn off microstepping it will show up very clearly. That's not a suggestion to solve your problem, but a fun experiment to see what it looks like

I'll bring my five bar scara and possibly some other fun stuff.

Mr Joerg himself

Updating looked easy, little has changed. Just one new parameters to a few methods. We will see if it works once I hook up the machine.

Homing is a bit of a challenge.
My proximal arms can swing over 180deg, but they can not go the backmost position at the same time, the distal arms will hit the center.
I guess if I rotate the arm the same direction and have the homing sensors at asymmetrical points it should work. Just like your example for work mode 1. Mine is built to be in work mode 2.

The machine is far from being a complete printer. I have to verify that 5 bar works fine first. It looks very good so far, light and stiff.

The DICE is a piece of art when it comes to packaging. It's a beautiful machine. I'm sure they made many tradeoffs when it comes to serviceability when they designed it. That's your problem now

I'm waiting for the same crowdfunding, and also reused the Maestro for other projects.

@matt3o said in slant lines:

could it be that my motion system is so well tuned that it emphasizes the extruder "signature"?

I think you might be right. In my experience many extruders have a very small periodic error. In most cases it is not visible. It depends on what hotend you have too, I would guess that if the "melt-pot" is big it will provide some elasticity inside the hot end.

My standard test object is a square cone in vase made. It is very good for showing periodic extrusion errors and mechanical errors. The varying of the extrusion length over height of the cone will make periodic extrusion errors show up very clearly as interference patterns.

Even a BMG through a 50cm bowden tube shows extrusion errors (ignore the slant lines in this case, they are mechanical). The errors in this case are so small that in any normal case it is totally invisible, this test provokes it.

I made the face interactive using some JS and html: https://fivebarscara.000webhostapp.com/test.html.

It's a little web page where you can input arm lengths, actuator distance and work mode. And interactively see the work areas, and the arms moving. Move the mouse over the work area and the arms will pop up. Different colours for different work modes.

It's quite mesmerizing.

I might add the angle limitations to the interface. It could be a useful tool for anyone brave enough to build a five bar scara. It's still useful to understand how fire bar scaras moves.

@JoergS5, I really think that some critical parts of the arm should be machined. Tiny errors in precision in the parts holding the proximal arms are amplified further out. Possibly you could make some clever parts where you could tune it by tightening/loosening different bolts. This is not unique for this kind of robot arm, any arm moving by rotating a join will have the errors amplified. Linear rails are far easier to use.

My design with the whole arm assembly moving up and down on the z-axis is tricky to design. There is very little room to place screws from the lower part to the upper part, to preload the bearings. The large pulleys and the goal to keep the arms rotation range as big as possible does not make it easy. The current version is far from ideal.

I am working on a better version, but every time I start fusion I lose a day. It's very mesmerizing to 3D CAD.

I installed an old precision piezo to measure the Z error properly. I had one of the the old crummy versions with a drilled piezo element laying around. It still works fine.

I can see that the arms lifts the actuator the further out the actuator is. It's not due the tower being at the wrong angle, or the weight of the arms bending it down. Some other angle(s) are wrong. It's a complicated machine

It's officially spring here now, at least on my balcony. The bulbs from last year are flowering.

After poking around a bit in the source code of the firmware it looks like G29 mesh bed leveling is based on a grid. Points are filtered out if you select a round area.

Looks like I have to put on my c++ hat and do some coding to get it to do what I want. Love open source!

@arhi If you do not return true from IsContinuousRotationAxis() that is exactly what you get. You can still rotate multiple rotations, it only affects the automatic wrapping.

@JoergS5 This version has two ball bearings too, behind the pulley. I preload the poor deep groove bearings by tightening a lock nut a little bit. They are really not made for that kind of axial load, but it works. I have some angular bearings that I might use further on, but I save them for now. They are prone to fall apart and were pretty expensive.

A very simple construction. Two plastic parts, two bearings and three nuts.

One idea to avoid the soft plastic could be use let the axle run through a drilled hole in a 3060 extrusion or other metal part, push the bearings against that and then push the arm directly onto the bearings. Some washers and spacers will be needed. The printed plastic parts would just hold things in place.

The thin walled aluminium square arms I use are actually very stiff. The U-beams in the drawing are just for testing, they are not good at all, they twist very easily.

@thwe those features are not available in 2.0 firmware, deprecated.

@dc42 It's my own homemade kinematics. And I have not implemented the IsReachable() function, yet.

Sounds like implementing IsReachable() will solve the mesh leveling problem.
And it will prevent the small accidents I have all the time, banging the hotend against the center body of the printer or trying to reach mechanically unreachable points where the math of the kinematics returns NaN.

I may get other problems though. When calibrating, and doing other other moves, the printer may have(edit) want to make moves from one side of the unreachable center area (point A in the picture) to the other side (point B).
If the motion planner selects the shortest straight line in stepper coordinate space it should work fine, but if it works in cartesian space it will try to travel though the unreachable area. Non-cartesian printers are fun

The printer has a rotating tower with an arm that moves in two dimensions:

I'll bring the printer to 3D Meetup Sweden 2019 this weekend (April 27-28) if you want to see it live.

I finally started working on a calibration method again, my arm produced horribly twisted prints. It works now.

I did something similar to how klipper does delta calibration. You print a model, measure a lot of distances and feed that into a program that figures out what arm lengths (4 of them) and homing angles you have.

It's doing a pretty brutal search looking for a solution with the least sum of square of errors.

It took a few models and measuring methods to get it working. Too few or not properly selected measurements made it find strange solution, the resulting printed models where right but other prints were twisted.

I could probably steal the model and method used by klipper delta calibration. It looks like they have a clever method to make it ignore the errors introduced by printed walls being slightly too big or small.

I'm not sure what to do with the code. I could port it to javascript and integrate it into the little tool I made before.

This calibration tool is pretty necessary if you build the printer with printed parts and hand tools, it's hard to get things exact and it's hard to measure the built printer.

Before calibration to the left: P161.40:161.40 D207.00:206.00 B227.00:151.50
After calibration to the right: P161.18:162.43 D208.28:209.97 B227.84:152.16

I did some experiments with crazy print speeds and accelerations. Printing at 300mm/s and 12000mm/s/s acceleration works fine mechanically. The extruder complains and the ghosting is horrible, but the arm keeps up fine https://www.youtube.com/watch?v=MW8HApFoy38
It's quite fascinating running with very high acceleration. I wish my machine was stiffer.