Locked Dovetail Triangle

When I posted my Dovetail Triangle one user suggested adding a locking mechanism while another argued against it.  So let's have it both ways: I built a new dovetail mechanism that has a locking mechanism, shared here at Thingiverse, to complement the one that doesn't have a locking mechanism here.

As before, I aimed to keep overhangs within the buildable range of Makerbots and other hobby printers.  That imposed some interesting design constraints, which forced me to distribute features of the locking mechanism between the base and lid.  But I filleted the internal corners to 0.5mm purely for my own convenience: it's a bit easier to peel unused material out of SD300 models if there aren't sharp internal corners.

At first I built a transparent model and tested it with two black triangular pieces so I could look inside to see how well it was working.  It's hard to see, but the picture below shows how the black triangles slide toward the outside of the puzzle when it's lying flat and thus lock the joint.

The black triangular pieces (dark shadows) slide toward the outer corners.

When it's flipped upside down, the contours in the lid cause the black triangles to slide toward the outside edges, locking the joint like the previous picture.

When flipped over the triangular pieces still slide toward the corners.

It worked, but those triangular pieces just didn't slide very well.  So I tried using round discs instead.  That worked much better because the discs easily rolled around when the puzzle was tipped.

Here's a video showing how the parts work and how the puzzle opens.

I designed the mechanism so it would only unlock when one particular corner of the triangle was pointed downward, because I thought that would be the least intuitive way to hold the puzzle.  As it turns out I'd guessed wrongly about that: testers instinctively hold the puzzle with one of the dovetails facing up (to look at the joint, I guess) and hence they naturally hold it with a corner pointing downward.  So it isn't especially challenging, but it definitely works!

Return Tool Error

While the SD300 was building a model on Tuesday night it made a distinctive 'click' noise, just once while it was using the Anti-Glue pens.  It didn't sound bad, but it was an unfamiliar sound I'd never heard before.

It never made the click noise again, but a few minutes later it began exhibiting problems putting the Anti-Glue pens away.  Eventually the machine stopped with a "Return Tool Error" message.  In SD300 lingo the Anti-Glue pens are named Tool #4, Tool #5, and Tool #6 so this message means it was unable to return a pen to its holder.

That wasn't too alarming.  Maybe that click noise had been a pen or holder had been knocked loose.  I removed the pen from the plotter and carefully returned it by hand.  The machine resumed building for one more layer but the error happened again as soon as it tried to use the Anti-Glue pens.  I installed a brand new set of Anti-Glue pens, but the error kept coming back.

When the machine had logged 24 consecutive errors I decided to give up.  It was late, so I dashed off an email and went to bed.  By Wednesday morning I already received two helpful replies from the folks at SolidVision, which recently merged with FISHER/UNITECH.  I appreciated the very prompt response but I had to go to work because I only do 3D printing as a hobby, so I didn't have an opportunity to do any troubleshooting until the following evening.

Jason Harris offered me several options, but he suggested the solution might be as simple as squeezing the spring on the affected pen holder.  The pens holders are spring-loaded clips, likely the same ones used in Roland DXY-series pen plotters.  A spring is supposed to ensure an empty pen-holder is narrower than an occupied one.  Could it really be this simple?

Curious, I brought out the calipers and measured the pen holders both with and without pens in them.  The openings on the working pen holders were all about 7.9mm wide, but the malfunctioning pen holder was open about 8.4mm.  That didn't seem like much difference, but I carefully applied pressure to the spring until the bad pen holder measured less than 8mm, then I restarted the machine.

It worked!  Then it worked again!  Just to be sure I repeatedly pressed Pause/Resume to make the machine take and return the pen a dozen times, and the pens returned to the holders every time.

This week my SD300 has logged over a thousand build jobs.  I've occasionally had issues that resulted in canceling a job or damaging a model, but this is the first time the machine itself had exhibited a fault.  In hindsight I consider that a very good record, especially considering how easily the fault was resolved.  Admittedly there are challenges cleaning the support material from some models, but the machine itself has never demanded anything more than clearing a jam or replacing spent consumables.

Peppermint Candy puzzle refined

I took another stab at the Peppermint Candy puzzle which had been vulnerable to coming apart in unintended ways.  I carefully refined the curves so each piece would 'reach around' to brace itself against the back side of the puzzle and made the walls thicker so it would be more rigid.

Honestly, I wasn't sure it would suffice.  But it worked!

Here's the side with the ball sticking out.  The cut in the ball is lined up with the cuts in the shell but it won't open in this position.

The other side of the puzzle looks about the same, but it has a white dome instead of an opening so the ball can't be seen from this side.  The parting line forms an S-shape where it passes through the domed part.

Here's a video explanation of the puzzle in detail.

For comparison I ordered a test part built in SLS nylon.  It's stiffer than the part built on the SD300, a good thing, but it lacks the contrasting colors.  But the colors aren't just cosmetic; my testers thought the colors contributed to the challenge of solving the puzzle.

I built the ball pieces in a very specific orientation in order to put the colors exactly where I wanted them.  Here's how the ball looks outside the puzzle shell.

Since the part required a specific build orientation I couldn't really position it for optimal post-processing.  Each piece had a thin edge that would have been particularly vulnerable to chipping during cleaning, so I isolated that area in a little 'box' of peeling cuts.

Most of the layers in the isolated box fell away when the surrounding areas were cleaned, so it didn't make the cleaning work any slower.  Except when I got to the thinnest part, the isolated box allowed me to gently peel away the last bits of material...leaving the thin edge smooth and intact.  As usual, I solvent-dipped the parts to give them greater strength and a slick finish!

MINI Cooper Grocery Bag Hook

MINI offers a "Grocery Bag Hook" accessory that's supposed to prevent shopping bags from sliding around in the trunk.  But bags don't slide around in the tiny trunk of a MINI Cooper, so their accessory seems unnecessary.  What I really wanted was a hook that would keep a shopping bag from sliding around in front of the seats, so I designed this simple hook.

This hook attaches easily to the dealer-supplied cup holder accessory.  Shopping bags slip neatly into the deep slot so they won't slide around.

I put it online at Shapeways and built a sample in their Red Strong & Flexible material.  Don't expect it to be a big seller!

I had replaced the material kit partway through the build, and the tints didn't quite match even though both kits were Amber Transparent material.

Antichron

A year ago I probably wouldn't have attempted to build something as twisty and complex as Bathsheba's Antichron model.  But someone sent me a reduced-size model of it, and I confidently devised a small set of peeling cuts that should have made it an easy build on the SD300.  Not so easy, as it turned out.

I got it mostly right.  I'd divided the model into four zones in each of the upper and lower sections, so the support material peeled away effortlessly at first.

But when I got near the middle section I found layers of support material permanently wrapped around sculptural features I'd misinterpreted.  If it had been properly planned, there should have been additional cuts to divide the support material but I hadn't instructed the SDMove software to make the proper cuts.

Was it necessary to discard the model just because I hadn't set up the build software properly?  Probably not, but I decided to free the model so from the rest of the material to get a better view of the trouble spots.

The support material was thicker than the walls of the model so I couldn't just remove them by brute force.  So I decided to separate the support layers and snip them one-at-a-time with a small knife.  This was tedious but it would free the model after cutting through dozens of layers on each side.  That would take a while so I put on an old Edward G. Robinson movie (Scarlet Street) while I worked...and it took most of the movie to free the two models I'd built.

Because I'd scaled down the model by 50% the walls were dangerously thin, about 1.4mm.  Somehow I managed to free the first model without any damage, but cracks appeared in the second model while I was cleaning it so I stopped several times and applied Weld-On to strengthen its weak spots.  When I finished cleaning the second model (below right) I dipped it in Weld-On 2004 to heal the cracks and give it a nice sheen.

Test Model vs. "Clean Bubbles"

In Monday's post I concluded that whenever the SD300 has been idle for a week or more that it's wise to either build a Test Model or run the Clean Bubbles maintenance routine to clear bubbles out of the glue delivery lines.

But which method is better?  Is there a difference?

I ran some tests, reviewed the logs, and weighed the glue cartridge: a Test Model takes 15 minutes, uses 8 grams of glue, and consumes some PVC material. The Clean Bubbles routine takes about 1 minute, uses 5 grams of glue, and doesn't consume any other consumables.

So Clean Bubbles is better, faster, and more economical.

From now on, when I want to build a model after leaving my SD300 idle for a week I will 1) first clean the glue trap with a paper towel, 2) run the Clean Bubbles routine in SDMove, and 3) confirm liquid glue has been deposited in the trap.  (Repeat Clean Bubbles if the trap is dry.)

Bubbles in the glue lines tend to resolve themselves, but it's nice to discover how trivially easy it is to prevent the problem in the first place.

Another mystery solved

I think I solved another mystery or two.

Jason Harris at SolidVision told me when I was setting up my SD300 that it might give a false 'replace cutting knife' error around layer 4 or 5 if the build material wasn't stacked tightly at the edge of the build table.  It happened to me once in June 2010 the first time I used my SD300 after a long vacation.

A British SD300 user once told me that he tended to build several batches of models near the end of each month.  His first build job would usually be sticky and the support material would be hard to remove, but all subsequent jobs would be normal.  He worked around the problem by always building the built-in Test Model if he hadn't used the machine that month.

Today I built my first model since 17 December 2011, so my SD300 has been sitting idle for two weeks.  I noticed something strange immediately.  Normally the SD300 starts by bonding a layer of PVC onto the build table with an even layer of SolGL adhesive, but this time the adhesive was spotty.  The adhesive appears here as dark spots underneath the red plastic--it should have been spread under the whole sheet, not just little spots as here.

Then I noticed a shiny droplet of SolGL glue near the edge of the table, on top of the PVC sheet.  I'll bet it dripped there when the SD300 measured the material height with its micrometer, which is attached to the same unit that spreads the glue.

Ordinarily the machine's next step would spread anti-glue (masking fluid) onto this surface with a device like a felt-tip pen, but the droplet of SolGL glue would obviously contaminate the pen so I paused the SD300 and wiped up the glue with a paper towel.

Then I let the SD300 continue building to see what would happen.  The next layer had fewer bubbles than the first, and the following layer didn't have any bubbles at all.  I'm guessing air had seeped into the glue delivery system while the SD300 sat idle, but the air had all been cleared out now.

I wondered if I should cancel the job and start over now that the problem had apparently cleared itself up, but out of curiosity I let it keep running.  Lucky thing, since what happened next was a revelation.

Starting at layer 4 in every job the SD300 automatically verifies the condition of the cutting knife by cutting a test pattern at the edge of the table.  When this build reached layer 4 I heard a loud noise from the media feeder clutch which usually signals a cutting-knife failure.  I watched closely as the SD300 automatically  repeated the knife test a second time: the knife was good but the material was too loose.  The machine raised the table and repeated the test again: the test succeeded this time, but the leverage from the knife dislodged the material so the PVC sheets loosened from the table.

The SD300 started building layer 5 by ironing a new sheet to the top of the model.  Ordinarily it measures the stack with its micrometer to confirm the thickness of the last layer, but this time it started taking repeated measurements over and over.  The red light came on and the SD300 displayed "HEIGHT CHECK ERROR" on the panel.

In my experience, the SD300 tries not to abandon a build if it can possibly continue so I tried pressing the material back onto the table and pressed the continue button.  The SD300 obediently measured the model several times more, apparently trying to get a usable measurement, but stopped again with the same error.

The logs demonstrated the SD300 had made a good-faith effort to continue, but with the material flopping loose from the table it couldn't get a measurement within 446 microns.  This was an error of over 100% because the last layer had only been 188 microns, so there was no sensible way the machine could have compensated.

So the only solution was to cancel the job and start again.  Then it worked just fine, which makes sense in hindsight:

• The air bubbles in the glue lines had formed while the machine was idle.  The bubbles would have cleared themselves within a few layers, but I could have cleared them manually by using the Clean Bubbles function in the SDMove maintenance utility.
• The droplet of SolGL glue was also caused by the air in the glue lines.  It could have contaminated the  anti-glue pens, but the contamination would have eventually resolved itself.
• A height check error within the first few layers of a job probably indicates there were bubbles in the glue lines, but the problem has already resolved itself by the time it's been detected.

All these issues are self-rectifying within a few layers, so I accept my British friend's solution as the simplest: just run a Test Model build job whenever the SD300 has been sitting idle.  A test model uses very little material and it gives the machine an opportunity to work these issues out of its system without needing to understand them.  Or I could just ignore these potential issues, since the worst that seems to occur is that I might have to re-run a job that fails within the first few layers.

Rapid Evolution of "Candy Ball"

Over the weekend I built a simple test model that inspired me to work on a new (but familiar-looking) puzzle design that rapidly evolved in the next two days...

As an experiment I recently built a two-part shell that fits over my existing Green Marble puzzle, following exactly the same contours when the two pieces are precisely aligned.

I added a dash of green color to the model while it was being built, just to help identify the parts.  Serendipity!  The green calls attention to a swirly shape like a yīnyáng on the surface of the outer jacket.

The marble sticks out of an opening in the jacket, opposite the green yin-yang, and it can turn freely inside the jacket.  By coincidence the swirly shape can be replicated by moving the marble around but doing so causes the marble to block the whole puzzle from being disassembled.  Could I design a new puzzle that deliberately includes such a hint at a false solution?

Next I built a whole set of parts based on the same curved geometry as Green Marble with a 2-part inner ball and a 2-part outer shell.  The shell includes a round dome centered on the swirly shape, opposite an opening where the ball sticks out.  I switched between red and white material, hoping the coloring scheme might emphasize the illusion.  The colors remind me of peppermint candy.

The ball pieces fit inside the shell pieces, then the whole thing screws together.  It looked cool, but the screw action was a bit too pronounced for my taste.

 The color coding worked perfectly.  The white dome formed a ball-like shape with the same swirly shape as the green puzzle.

On the opposite side, the marble could be turned until it also exhibited a similar swirly shape.  The contrasting white color on the ball further emphasized the swirly shape, but the puzzle won't open when the swirly shape is exposed.  The camouflage works very well, but there were plenty of other details I disliked: the correct solution required too much unscrewing, the jacket pieces were too flexible, and the ball was too hard to maneuver inside the jacket.

Applying what I'd learned, I designed a whole new model from scratch without using any of the model data from Green Marble.  It looks very similar to the last model but there's less twist and the geometry is better optimized for the ball to move inside the container.

As it turns out, the newest design has an easy-to-find method of taking it apart without aligning the pieces as intended.  If the ball is turned so the white portion sticks up at the top as shown here the outer shell can be gradually withdrawn by pushing it to the left while the ball rotates and unscrews itself within the jacket.

Looking back, the previous version didn't have that problem because the red section was wider and the twists had overlapped so there wasn't as much freedom to move the jacket independently of the ball inside.

This design evolved rapidly until it reached a dead-end, so now I have an opportunity to go back a few steps and see if I can develop it into a fresh puzzle.

Hands free memo holder

I had intended this memo holder to use a marble to grab papers behind a wedge-shaped panel.  But the marble didn't work very well so I 3D printed a plastic cylinder that worked better.  It works by sliding papers up into the slit behind the cylinder, whereupon the force of gravity traps the paper between the cylinder and the back plate.

There are two versions, one plain and one with slots added for hanging lightweight articles.  After a little more experimentation I modified the cylinder into a barrel with a fat middle section.  That held papers better than the cylinder.  I rebuilt the memo holder with side walls oriented for better transparency to illustrate how it grips papers by the wedging action of the red barrel.

Here's video that illustrates how it works...

This particular note holder was attached to the wall using a small Command brand adhesive strip.

I also made a refrigerator-magnet version by trimming a square from a worn out magnetic pad from my SD300 and gluing it to the back of a memo holder.

STL and STP files for this model are shared here.

Helical Burr Followup

After my Helical Burr Experiment had required sanding to make the pieces fit together, I refined the data and built a second model just to fix the problems with the first.  The new one (on the right) fits together smoothly, but it still has impractically-thin walls so it's just another experimental prototype.

I'd intended to design it so the parts could only be put together in one specific sequence, but testing the new model revealed a slight flaw: one of the pieces could be put on (or taken off) out-of-sequence because I'd only blocked movement in one of the two directions it might move.  I probably wouldn't have discovered the oversight without building these test models.

Here's a movie demonstrating how the puzzle works and revealing the design oversight.

One of the pieces had some exceptionally thin edges, which came out kind of wispy the first time so I tried building it in a horizontal orientation for comparison.  The new orientation is probably a bit stronger, but the difference isn't compelling.  Both parts work about the same.

I will probably start over from scratch because of the thin walls and un-blocked movements, but I consider it a very successful experiment nevertheless.

Adapting the Magic Screw for Makerbots

Last month I shared my Wrong Way Bolt model, but home users found themselves unable to build functional copies of it using hobby-type 3D printers like the Makerbot Thing-o-Matic.  The model's geometry seemed to fall within the rule-of-thumb limits, but I looked closely at individual 'slices' of the model and thought about what would actually occur when it was built using a hobby extruder.

Here's what might have happened.

At left is a slice of the trick nut, color-coded with shades of red to indicate the region to be built for the present layer.  Bright red areas are supported by the layer below while dark-red areas are unsupported overhangs.  When an hobby-type 3D printer tries to build the overhang marked with a circle its extruder would follow the path in white but the thread of extruded plastic won't have any support.  As a result, the thread will be pulled into a straight line between the two points where it's supported by the layer underneath.  The Makerbot would probably build some of the overhang, like the diagram at right.  It would probably need multiple layers before it could build that whole overhang, but each layer moves the overhang to a different place (because it's a screw) so the hobby-printer wouldn't have a chance to catch up.

The model could probably be adapted for hobby-type printers by designing the threads with a goal of eliminating the concave shapes.  Overhangs could probably be built with very high accuracy so long as they didn't have curves or corners in the unsupported regions.  So I designed a trick thread with a cross-section whose overhangs form straight lines, like so...

I don't own a Makerbot (nor any other hobby printer) so I don't really know if this will be any more buildable than the model I'd originally designed for my SD300.  But it's a starting point, in any event, as the trick threads function correctly and I'm sharing the source data so other users can adapt it if needed.

This new model and its source data are available from Thingiverse here so anyone can download it and edit it.