RepRap: Blog

Just as a side note, these are the prototypes that evolved into the Maus C ('C' for Complementary) flexure system.

I started out with a couple a flat designs. Top left is mine, top right one I tried off Printables. The weight of the stage caused the flexures to sag, but I noticed they didn't deform vertically so I flipped the whole thing 90 degrees.

The latest one (but 10mm too short, dammit) is in the bottom right. That has the middle section of the flexures as a more solid beam, which significantly reduces the tendency for the platform to torque. It might be possible to make a print-in-place version of the completed assembly but I'm still fiddling with the overall design too much, so it can stay as a kit of flexures for now.

I'm reprinting the latest flexure panel 10mm wider and will rebuild Maus C with those. Probably have fiddle a few bits of the driver framework to fit. In parallel, I'm working on affixing the Z axis and probe holder to bring the thing into a single unit. When it has tested good I'll pop it on github as part of the Maus library.

Assembling this has also spurred further development in the Metriccano library, as I discover new beams and brackets that need to exist to make Metriccano fit together in useful configurations.

Labels: maus, reprapmicron

# posted by vik-olliver @ 10:21 PM 0 comments

The Maus C has acquired its second axis. Still fits under the microscope. The frame bends too much, so I'll reinforce that, add a base board and some cross bracing.

More torsion than I was expecting on the axes, giving only 50 microns positioning when combined with the frame flexing. I'll try out different flexures with the stiffer base, and if that doesn't work the axes will need cross bracing as well. At that point it becomes more complex than an XY Table with height tweaks in software.

Regardless, his shorter assembly has helped nail in the basic format of the device, and I'll continue with it to the point where I have figured out a good mounting system for the Z axis.

PS I've run out of M3 nuts, which is a good indicator that assembly has become over complicated :)

# posted by vik-olliver @ 10:53 PM 0 comments

The wire actuator was a fail, but fortunately the whole thing is built on Metriccano, so it didn't take long for me to strap a Maus axis driver onto the frame to form what I'm currently calling the "Maus C Prototype":

Movement looks smooth enough under the binocular microscope, though with one axis left waving in the wind there's only so much I can determine. Also that frame is finicky because I haven't left enough clearance on the sides of the Y axis flexure (they're rubbed smooth with a file...).

Adding the X driver will doubtless need a few mounting points and minor mods on the frame, so I'll reprint the frame parts with more clearance once I've try-fitted the X driver.

It is oh so nice having something that I can pick up, move around, and turn upside down, rather than an assemblage of parts screwed to a plywood slab.

# posted by vik-olliver @ 11:41 PM 0 comments

I made a quick hack of a job on making the direct wire actuator and connected it up to the new complimentary flexure table, not so much as to see if it worked, but how badly it failed. Interesting data emerged. Here's the test rig:

I  can't say how well the table works, but I can define a few characteristics of the stepper itself and I don't like them.

When microstepped, the motor would move a little bit and then leap by approximately 20-25 microns. Doing the maths for a 5mm diameter shaft (roughly 16mm circumference with the wire) it works out as roughly 800 usable steps per revolution - half-stepping in fact. I was hoping for much better than that, though with a better driver chip I might achieve more accuracy.

For now though, direct wire drive is out by a factor of 5 (I consider 5 microns just good enough). It is immune to the vagarities of thread and nut precision, alignment, coupling offsets and so forth so a hybrid approach might be possible. Needing a 5:1 mechanical advantage would mean spooling up 50mm of wire, 3-4 wraps on the shaft, which could present wire management issues.

Plugging the numbers back into the screw-driven axis with a 0.5mm thread pitch and 800 steps gives about 0.6 microns per usable step (plus the 3:1 advantage of the flexure arm) so that at least is comforting. Plus I finally have something I can fit under a real microscope!

# posted by vik-olliver @ 12:35 AM 0 comments

I've been searching for a way to make the XY Table shorter so I can stuff it under the binocular microscope. I looked at making a traditional flat XY flexure table, but the 3D printed ones sag in the middle too much (and forces on the X and Y axes are very different). The nice thing about the original Maus XY Table is that the flexures can actually take a lot of weight, so I took a flat design and made one set of flexures perpendicular to the other. The first attempt looks like this:

Those narrow flexures take a lot more weight in that orientation, which is a bit counter-intuitive. It needs a bit more bracing, and the yellow bits need to be replaced with a Maus slide holder, but it moves very nicely.

The clever part of using complementary flexures is that when one is "shortened" by flexing, the other one lifts it up. So I can use a shorter table, and not have the height changing much when it moves around. Time will tell if this is a good idea or not, but if it works I'll be able to make the mechanism much more compact. Assembly is simplified a bit too.

It's a bit too short for the current axis drivers, though it might be about right for the proposed simplified wire actuator. Tucking the UV source underneath it will be a challenge, however.

# posted by vik-olliver @ 8:17 AM 0 comments

As promised, printable STLs for the Maus axis, table, and probe stuff. Full source on github. More bits in the works.

Off to Everything Open to do speakery things.

# posted by vik-olliver @ 1:00 AM 0 comments

More and more, as one attempts to move down the Feynman Path it looks as if actuators are the real roadblock.

In retrospect, this should have been obvious if one pays attention to what one "does not see" as opposed to what one "does see". A casual survey soon turns up a dearth of viable millimeter scale actuator designs and yet people have been making millimeter scale devices for over 100 years. Yes, some actuators exist, but far less than one might expect. A quick search on terms like "millimeter scale actuators" soon reveals that many are only "milli" in one dimension – if that.

So what causes the problem? Why the shortage? Well, some of it is probably due to the fact that commonly available machine tools are quite accurate at the millimeter scale. In other words, if you need to move something 0.1 or 0.01 of a millimeter this is easily within capability of macro devices – so why bother creating hard-to-work-with millimeter scale actuators for the purpose.

The other reason is that there seems to be a gap between the scale at which electromagnetics stop working efficiently (centimeters) and electrostatics start working efficiently (microns). Simply put, finding a way to actually actuate anything at the millimeter scale is hard and so it is easier just to ignore the problem and find some other way of doing what you need to do.

Does this mean there is no point to milli-scale actuators? Why not just skip to the micron scale and have done with it? This is essentially what the MEMS people are doing, however, since there is nothing at the scale above them, things like assembly or complex multi-material manufacturing of devices is not practical. Machine tools and repurposed semiconductor fab equipment are suitable for the creation of parts but not really intended for construction, manipulation or operation.

The Feynman Path is intended to address this issue by working down the scales leaving behind a toolkit at each scale able to build and operate the tools and devices below it. Maybe milli-scale actuators really are impractical and a jump right to the micron scale is necessary. Nevertheless, an exploration of the millimeter scale actuation space would seem to be worthwhile - no need to give up on it without putting in some time.

So, having got the closed loop control thing operational, it is time to put that area of exploration to one side and begin work on millimeter scale actuators. I have some ideas…. more in future posts.

Labels: Feynman Path, FPath, millimeter scale actuators, Nanotechnology

# posted by OfItselfSo @ 12:10 PM 0 comments

Attending family matters away from the workshop. Had an idea for a vastly simplified axis driver. It wouldn't be quite as accurate but stands a chance of being significantly faster and having a greater range of movement. Plus it's linear!

The idea is to directly drive the XY Table with a wire that wraps around a 5mm NEMA17 shaft. One full turn of wire would give nearly 16mm of linear movement. No return wire would be needed, just a simple extension spring on the far side of the table.

Accuracy would suffer a bit, but speed would increase to the point where vibration would be a limiting factor. 1/4 stepping is reasonably reliable and would give 10μm/step. Of course, it could drive the XY Table at a distance from its pivot to give better mechanical advantage and accuracy but that's more complex.

As noted on the diagram, lots of tolerance and failure mode concerns, but if anyone wants to try it before I get back fill your boots.

# posted by vik-olliver @ 11:24 PM 0 comments

By carefully folding the edge of the aluminium foil contact plate on the slide, I was able to form a reservoir for UV resin. This is the view from the USB microscope. You can just make out the array of dots on a 100μm grid on the slide below the dyed resin in the foil.

These were deposited with the Nichrome 80 tip. Here is a rather grubby micrograph of a 3x3 area with a hair next to it for scale. Unfortunately the hair oil got on the slide, and it blurs out some of the dots. More unfortunately the slide was not recoverable.

The dots of resin appear to be roughly 10μm in diameter. The purple blotch is a marker so I can find the damn things.

I cleaned the slide and ran again. The probe was a bit closer, so the dots were bigger, but I seemed to pick up a bigger resin load. This allowed me to not only make an array, but to use the dots to build up a square 300μm on a side. Horizontal rows all deposited with one dip in the reservoir. Had to tweak the colours and contrast to make it visible.

This was all done by manual control of the CNC interface, because I wasn't sure of the locations of everything and don't have code to adapt to using a reservoir yet. Also that Y axis is playing up again. However, this is quite the step forward.

# posted by vik-olliver @ 2:17 AM 1 comments

My first physical step down the Feynman Path is a pantograph.

“What”, you say, “something that stopped being hi-tech in the 1700’s is somehow relevant on a path to Nanotechnology”. Well, no and yes. No, the pantograph is not really an optimal solution – I think flexures such as are being developed for the RepRapMicron are the more promising avenue… and... Yes. I do think machinery at the micro and nano levels will resemble massively parallelized versions of older solutions - maybe not the 1700’s but certainly the 1800’s. I will have more to say on this complexity vs simplicity issue in future blog posts.

Back to the point. My primary interest at the moment is how errors might be removed as big devices build smaller devices. The pantograph is a simple thing built out of LEGO bricks and has a reduction ratio of slightly over 5:1. The goal is to create a really nice, visible circle using closed loop feedback to iron out the errors as extremely inaccurate large scale actuators move the tool head about. The tool head has a barrel and lead scavenged from a mechanical pencil mounted on it to record the path.

The image above shows the pantograph. The macro end is on the left (red arrow) and the toolhead is in the center at the micro end (green arrow). 

Closed loop control was used to drive the macro and micro ends around a circular path. The image on the left shows a typical path taken by the toolhead when the control was applied to the macro end (it is also supposed to be a circle). This forms kind of a baseline case representing the inherent accuracy of the hardware. The image on the right shows the path taken when the control was applied to the micro end of the pantograph.

It is fairly clear from the image on the right that the errors attributable to the pantograph mechanism have been greatly reduced by applying closed loop control to the toolhead. Please see the webpage (FPath_Ex006) and short (8 min) video for more a more complete discussion.

Interestingly, both Heinlein and Feynman proposed using pantographs to have large machines make smaller machines. One reason I decided to use a pantograph as my “first device” is to acknowledge their ideas – a kind of homage if you will.  

Labels: Feynman Path, FPath, Nanotechnology, Pantograph, Stigmergic Path Following

# posted by OfItselfSo @ 5:35 PM 2 comments

I fixed the Y axis backlash, and put a temporary brace on the Z axis. Good move, must build a proper one. The brace eliminated probe wobble, though the touchdown point on the slide when the probe makes contact is still larger than I want it to be. That one is going to need some study, but the next major phase is going to be building with UV resin and for that the probe doesn't actually have to touch the build surface. I'm going to enlarge the hole in the XY Table under the build surface, because the aluminium foil contact plate can be used to shield the resin from the under-slung UV light.

I've tested out the latest Nichrome 80 probe, and I think an 8mm tip is too long. I put a dollop of hot glue on the shaft to help stop it wibbling around, but the slide is still dragging the point about. Again, future plan is not to hit the slide. Bear in mind that the probe isn't just touching a slick glass surface here, but trying to plough through about 5μm of solidified Sharpie as well. It has reasons to be behaving badly.

Good to see that the probe works well with the aluminium foil touch plate, and that the tip survived being accidentally over-driven into the slide surface by 50μm. I haven't tested it yet, but the surface of the 316 stainless steel probe is much rougher. That might actually help when it comes to transferring UV resin.

Finally, I've been talking to Eddie about doing the video, checking the AV equipment available. Sorting out the order of things, workspace, editing software, the logistics of file transfer etc. We have a plan. All considered, the first step is for me to come up with at least an approximation of the assembly instructions. I've done assembly manuals before, and there's a process to doing it right. The bad news is I don't think I can get it all down before the 16th, which is when the trip to Everything Open begins. I'll release printable STLs of what I have before I go, whether the instructions are ready or not.

# posted by vik-olliver @ 7:56 AM 0 comments

In an attempt to make a more rigid probe, I started with some 22 gauge (0.6mm) "Nichrome" wire. This took nearly a quarter of an hour to corrode away in the electrolysis cell, and when the end finally fell off it looked like this:

You do not have to be an expert metallurgist to see there's something ... interesting going on in the manufacture of this brand. There is a clear spiral pattern to the erosion, and fibres of incompletely mixed metal in the alloy. While sold as Nichrome, the packaging itself only states "Resistance Wire" and it appears that resistance is futile.

I'd recommend getting wire from a manufacturer that states the percentage of nickel and chromium, and make sure the numbers add up to 100.

# posted by vik-olliver @ 3:12 AM 0 comments

I found a limitation with jscut in that it always closes a path, so if you draw a 'U' it will join the two ends at the top. So jscut is not suitable. After looking for something that works, I found that PrusaSlicer imports SVG paths (though they need re-scaling for undetermined reasons).

So I wrote a RepRapMicron printer configuration for PrusaSlicer based on Marlin and told it not to use the extruder. It put 'A' commands and 'M' commands in the GCODE though, so I used a post processor to strip them out:

/bin/sed "s/ A[0-9.-]*//";grep -v "^M"

This allowed me to import an SVG line drawing and produce this:

 Due to the probe impact problem producing a bloody great horizontal line every time it hits the slide, some things are a bit obscured, but that's a 500μm line with end marks and the letters "500um" underneath it.

So now I can draw open-ended paths and can get on with creating some calibration patterns manually.

I also really need to print a microscope slide storage box to take to Everything Open. Must remember to set the printer type back...

# posted by vik-olliver @ 12:13 AM 0 comments

Happy New Year! Is there anything that I should put at the top of my New Year's Resolution list to release, document, or improve things that you want to see? 43 people have checked the new git repo, please do tell.

I wrote the GCODE for a calibration zigzag, 500μm lengths, 50μm spacing, repeated along X & Y with a staircase gradually lifting the probe by 2μm for each segment. I ran it twice, and took an image of the calibration slide with the same camera settings:

Immediate things, X scale is off, the lines are only 415μm long, not 500, fixed that in GRBL config.

Lighter lines are generally more consistent, which figures as less stress on the probe.

Y axis seems to have consistency issues being +/-30μm. Needs an overhaul. I'll probably do a diagonal calibration slide later to test the overhaul. Even so, that's literally within a hair's breadth.

[Edit: It seems one of the bands used for backlash has broken, which is going to tilt the drive nut...]

But the X axis, other than the scaling issue, seems to be positioning +/- 10μm absolute which I am moderately happy with for now.

Start point divots are reduced, possibly by a 5x reduction in plunge speed. I noticed a lot of vibration of the probe through the microscope. Not sure how much of that is the axis mount, and how much flexibility in the probe, but both need to be addressed.

I'll take a peek to see what's up with the Y axis. Now I have had experience in handling the setup, I think I might risk a decent probe with a more symmetrical cross section and see if that improves things.

This all brings to the fore the question of end stops, so that I can guarantee the same starting point every time and thus experimental consistency.

# posted by vik-olliver @ 2:33 AM 0 comments