V4??

[barry99705]

https://youtu.be/T_347m_lxes

[Eric]

http://hackaday.com/2017/01/19/3d-print ... ilted-bed/

[Qdeathstar]

Now that is a nifty idea. Gotta imagine the processing for that is someone complicated...

[626Pilot]

Like earlier efforts based on a Stewart platform (six axis motors tilting the effector to arbitrary angles), except this way, gravity is always on your side. Quite brilliant. Shame you have to have such a huge frame to get a build volume that's only moderately large.

The slicer is obviously custom. I'm guessing they are doing CAM-style setups in AutoCAD or similar, where you have to tell the machine details about how to do certain things specific to the part, as opposed to a regular slicer that doesn't care and figures everything out layer by layer. Parts probably have to be subdivided into components that are optimized for this production process, e.g. each individual side of the bowl they printed. (Unless it's extremely simple, and can only do bowls and vases, things where there is just one contiguous hull.) It wouldn't be as simple as handing an STL or other mesh to a slicer. It would need a parametric, history-based modeling approach, like STEP files. The whole toolchain could probably be built around Fusion 360, and perhaps that's what they did, although they could've used another CAD package.

[tjschweizer]

Like earlier efforts based on a Stewart platform (six axis motors tilting the effector to arbitrary angles), except this way, gravity is always on your side. Quite brilliant. Shame you have to have such a huge frame to get a build volume that's only moderately large.

The slicer is obviously custom. I'm guessing they are doing CAM-style setups in AutoCAD or similar, where you have to tell the machine details about how to do certain things specific to the part, as opposed to a regular slicer that doesn't care and figures everything out layer by layer. Parts probably have to be subdivided into components that are optimized for this production process, e.g. each individual side of the bowl they printed. (Unless it's extremely simple, and can only do bowls and vases, things where there is just one contiguous hull.) It wouldn't be as simple as handing an STL or other mesh to a slicer. It would need a parametric, history-based modeling approach, like STEP files. The whole toolchain could probably be built around Fusion 360, and perhaps that's what they did, although they could've used another CAD package.

I agree a CAM program would be necessary (I've used MasterCAM's multi-axis before, pretty fun), but I doubt that's what they did for that video/demo part. More than likely they made a simple 3d program, and added in custom gcode to move the table. The Hackaday post says they haven't gotten the print head to move at the same time as the bed, so they aren't doing true multi-axis printing. Not that it isn't still extremely impressive, but from everything I've learned from CNC machining is that the hardware for multi-axis setups is easy, the software is the bottleneck so far.

In all reality though, if you just modeled an inverted trapezeoid, sliced it in Cura or something, and copy-pasted the g-code 6 times with a simple rotate command inserted in, that'd be a pretty easy demo program. Super impressive though!

[dc42]

... The Hackaday post says they haven't gotten the print head to move at the same time as the bed, so they aren't doing true multi-axis printing. Not that it isn't still extremely impressive, but from everything I've learned from CNC machining is that the hardware for multi-axis setups is easy, the software is the bottleneck so far.

I think the real challenge is to write a slicer than can work out the optimum way to move the bed and the effector during a print. The firmware to control the movement would be easy. There are some shapes for which it is obvious how the motion should be done - for example, an inverted cone (overhangs all around) - but for more general shapes there are some complex decisions to be made in the slicer.

[tjschweizer]

I think the real challenge is to write a slicer than can work out the optimum way to move the bed and the effector during a print. The firmware to control the movement would be easy. There are some shapes for which it is obvious how the motion should be done - for example, an inverted cone (overhangs all around) - but for more general shapes there are some complex decisions to be made in the slicer.

The more I think about this, the easier I actually think it would be. If you ignored parts that had severe overhands - limited to 80° from vertical or so, and only along the outer perimeter. So most (all?) slicers already calculate the overhang angle. All you need to do is add a bit of code so if the overhang angle is larger than 30° or so, rotate the platform.

This gets rapidly more complicated if you are talking about multiple part prints, or doing anything other than external perimeters, but for the majority use case (single part, external perimeter overhangs), it should be pretty simple. Just calculate the normal vector to the face being sliced - or rather find the triangle facet that the current edge lies on and find the normal to that - and move the bed to limit that angle.

The actual math would work out to literally only two matrix multiplication steps (per edge, it adds up but still wouldn't be terrible). If you made a reference frame on the edge being sliced, with the y axis normal to the face, x axis parallel to the face and the XY plane of the bed - solve that reference frame so that the X and Y rotations relative to the machine's reference frame - there you have a rotation about the X and Y axis. Multiply by the positional offset from the machine's origin to the edge's reference frame, and boom - you have your XYZ coordinates as well as angles required to rotate the bed.

Sorry that got long winded, I took a kinematics class back in college and haven't really found a place to use it yet. If anyone tries to open-source on of these beds, I'd be glad to help with the kinematics!

[626Pilot]

I wonder how the slicer would handle a complex object with individual protrusions that go off in random directions and possibly curl around each other, like a giant squid-monster with its tentacles wrapped around a multi-spired castle. Easy to do with a normal 3D printer and supports. Not that easy to do with a six-axis printer. You have to make sure the effector doesn't bump into a tentacle while it's printing a spire, etc. Could involve an immense amount of repositioning of the bed, and due to this limit, there could be sections that have to be printed with supports because there's just no way to get the tool where it needs to go without ruining something nearby.

I suppose the process would work the same as it does for machining sculptures like that on a 5-axis CNC. I've never set up a job like that, but from what little I know, a human has to get involved to some extent - you aren't just handing a mesh or a brep to the software and saying, "okay, have fun". The process gets more complex because it depends on (and leverages) gravity, whereas the 5-axis can cut material from any angle and it doesn't matter.

I don't know if there is a straightforward mathematical solution (like Newton's method) that can figure out how to "grow" the castle-and-squid model off the plate, but it seems like some sort of heuristic could do it. Perhaps a genetic algorithm. I don't know enough about those things to have a clear picture.

I agree with dc42 that the motors and firmware are the easy part. Like a humanoid robot with many dozens of axes, the mechatronics are very easy compared to figuring out how to tell them all what to do.

[dc42]

Sorry that got long winded, I took a kinematics class back in college and haven't really found a place to use it yet. If anyone tries to open-source on of these beds, I'd be glad to help with the kinematics!

If you write the slicer, I'll write the firmware!

As 626pilot says, one of the main issues would be to prevent the head bumping into parts already printed when tilting the bed. Perhaps the head could be modelled as an inverted cone, and this model used to calculate the maximum safe angle of tilt in a particular direction as a function of position and tilt direction?

[Eric]

What about tilting the entire printer instead of the bed, leaving the bed-printhead relationship intact? Then the slicer doesn't have to change dramatically. Mainly a matter of deciding what tilt is required to avoid support and adding that information to the stream. Also you'd need a bigger table for the printer.

[Minim]

Sorry that got long winded, I took a kinematics class back in college and haven't really found a place to use it yet. If anyone tries to open-source on of these beds, I'd be glad to help with the kinematics!

If you write the slicer, I'll write the firmware!

As 626pilot says, one of the main issues would be to prevent the head bumping into parts already printed when tilting the bed. Perhaps the head could be modelled as an inverted cone, and this model used to calculate the maximum safe angle of tilt in a particular direction as a function of position and tilt direction?

If I'm not wrong Fusion 360 handles this. I use the CAM there for my CNC router and it has collision warning when it hit my wice or workpiece when not in a milling operation. There you can design your own head and parts on the table to avoid stupid crashes.