Drew's Rostock: Replacing the J-head

Several months ago after slightly reworking the wiring to the J-head, the thermistor slipped slightly loose from its socket in the heater block.  The temperature monitoring code in the controller will detect a short or open connection on the thermistor and shut down the heater, but it can't do anything to detect a thermistor that is simply no longer thermally connected to the block, and the heater stayed on long enough to drive the heater block far above its normal operating temperature.  I knew something was wrong when the plastic coming out of the nozzle was bubbling.  Fortunately I noticed and shut the printer down before the plastic J-head barrel was completely ruined, but there was still damage done.  Ever since then the head's been prone to jamming, hard to feed filament into, and reluctant to let go of the filament when I'm trying to change colors.

I noticed a marked drop in reliability while printing the second version of my Tardis Transformer.  Trying to print faster than 30mm/sec inevitably led to a complete filament jam, often requiring a complete dis-assembly of the entire extruder from hot end to pinch wheel to fix.  Finally the entire thing jammed up with a plug of plastic in the barrel that just couldn't remove without destroying the J-head.  Fortunately I had already ordered another J-head, this one with a 0.35mm nozzle as I wanted to experiment with printing more detailed part.

Here's the new head, mounted in the plastic carrier that I use to attach it to the end effector on my printer.  The carrier is two pieces of PLA that clamp the end of the J-head, with guide holes for wires and a place for the push-fit connector to screw in.  In theory, I shouldn't be using PLA to clamp the hot end, since the hot end gets more than hot enough to melt PLA.  In practice, the top end of the plastic barrel never seems to get even close to being hot enough to damage the plastic clamp.

The new J-head seems to be using a different kind of plastic than the previous one was, dark-colored as opposed to the pale tan of the older one.  I'm not sure if this makes any difference, I suspect not.

Wiring the hot end is always tricky.  This part of the printer sees constant vibration, and the wires need to be able to flex freely yet not crack from flexing or vibration.  I've secured the heater resistor and thermistor into the heater block with fire putty, and additionally secured the thermistor cable to the J-head barrel with heat-shrink and copper jewelry wire.  This should help keep it from pulling out of the head again, as well as help protect the fragile thermistor wires from vibration-induced failure.

The final step is to wrap the heater block with insulation tape and Kapton.  This might not be strictly required, but it seems to me to help the head maintain a consistent temperature.  A few overwraps of Kapton tape also additionally helps keep the thermistor in place.

 The main downside to moving to a smaller nozzle is that it will take more force to push filament through the nozzle.  Rostock-derived printers already need a lot of force, with the long Bowden tube and (ideally) high print speed, and even with a 0.5mm nozzle it's not hard to exceed the torque your stepper motor and motor driver chips can deliver.  Fortunately I'd installed a geared extruder on my printer months ago, while trying to push the upper end of its printing speed.

I could have bought a geared stepper motor, or fit a gearbox on to my existing stepper, but to save money I decided to try and build my own gearbox first.  I got the idea after printing this amazing 6-speed automatic transmission model.  Printing and building it convinced me that the quality of parts from my printer were high enough to actually make working gears, so I set about designing my own geared extruder.   While the gear teeth were perfectly fine, the plastic-on-plastic bearings had far too much friction for an actual functional part, so my gearbox was designed to have R4RS bearings (chosen because I had 10 of them on hand) pressed into each gear.

I also replaced the plastic pins in the original transmission with metal shafts.  Here I'm using three 1/4" shoulder bolts (again, chosen because I had them on hand) with the heads cut off to couple the planet gears to the gear carrier, and a M5 machine screw as the shaft to connect the carrier to the filament drive pulley.  Using a machine screw as an output shaft is sadly not optimal.  I've managed to size it so the drive pulley is completely on the unthreaded part of the bolt (rather than resting on the threads), but the bolt isn't exactly the same diameter as the bore of the pulley, so there's a little bit of eccentricity as it turns.  That translates to a slight variance in punch pressure as the extruder runs.  This doesn't seem to be a problem yet, but I'd really like to replace that part with something custom-machined.

The extruder itself started off as a modified Airtripper design, but as is typical I've completely redesigned it from scratch.  Rather than two or four compression bolts I have a single bolt with a large hand-friendly knob pressed onto it.  This lets me easily and without any tools adjust the pressure on the filament.  It also lets me easily unclamp the pinch mechanism to completely release the filament, which I prefer to do when changing filament colors.  It's still not an ideal design - it takes a lot of turns to completely release the filament, and the threads on the screw strip after a few hundred cycles.  I've been meaning to redesign this with some kind of clever fast-release over-center clamp mechanism.

The completed extruder is admittedly huge.  The plastic gears work wonderfully, and have run for hundreds of hours of printing without any problems, but I did have to make them fairly large to handle the load.  A metal gearbox wouldn't have been any wider than the motor itself.  It's not really a problem since the extruder doesn't need to move or fit in a compact space, but can just sit out of the way at the top of the printer.

Looking up into the top of the printer.  The extruder mechanism is at the center.  The filament reel is on the top of the printer, with the filament fed down through a hole in the center of the upper plate.  Two small white LEDs are aimed at the point where the filament is pressed against the drive pulley, these cast enough light on that spot to let me see the filament drive is working properly.  I also have three fans arranged around the top end blowing air down into the print volume.  I'm not sure if these do much useful, but they look kind of cool.  You can also see my cheap little printed cable clamps, the heatsink for the 5V regulator for the fans and LEDs, and the terminal blocks which let me easily disconnect the cables to the end effector.

In addition to replacing the J-head, I've been experimenting with smaller layer height.  The default layer height setting on Slic3r is 0.4mm, which worked well enough for me when using my old 0.5mm nozzle.  The first few prints with the new nozzle were at 0.3mm, which along with the smaller nozzle made a noticeable improvement in print quality.  I tried pushing the layer height smaller, and discovered the microlayering function in Slic3r, which allows you to have different layer heights for perimeter and infill layers.  Unfortunately, when I tried printing objects at 0.135mm (with 0.27mm infill) I ran into an annoying bug where the printer would consistently freeze up mid-print when printing certain objects.  I've backed off to 0.18mm (with 0.36mm infill) which seems to be a sweet spot for reliability and detail on my printer.

Why those particular oddball heights?  My printer uses 36 tooth, 2mm pitch pulleys, motors with 1.8 degree step angle, and motor drivers with 16:1 microstepping.  If I've done the math correctly, that means that the physical resolution limit of my printer is 0.0225mm.  I had the idea that for consistent prints the layer Z-height should be an integral multiple of this resolution.  0.18mm is exactly 8 times the resolution limit, or equal to one half of a full 1.8 degree step of the motor, and the 0.36mm infill is happening exactly once per full step.  Given the way that a Delta printer works, where the physical Z height is determined by the combined action of three different motors, this may not matter as much as it would for a conventional cartesian axis printer.  I figure it doesn't hurt either, and with the 0.18mm/0.36mm layer height I'm getting some amazing print quality.

Drew's Rostock: The quest for perfect prints continued.

First layer adhesion seems to be the key to getting a good print out of this printer.  If the first layer sticks evenly and consistently, and stays stuck during the print, the print usually comes out well.  If the corners detach, they tend to curl upwards, deforming the part, and in the worst case interfering with the movement of the print head.  When a delta printer skips a step, it doesn't just shift the layers above the skipped point sideways as it would in a normal printer.  A skipped step on a Rostock printer both shifts and rotates the layers above the skip, often resulting in the head crashing into the print again resulting in more skips and a completely ruined part.

My printer was not printing well.  The plastic was not wanting to stick to the bed unless I brought the hot end down to the point where it was nearly touching the bed, forcing the plastic to smear out and flatten against the glass.  I was also having real problems with strings trailing from the nozzle whenever the plastic stopped extruding for a moment, which would sometimes yank the plastic free from the bed as the print head completed a line and moved to another spot on the first layer.  This was a real problem when trying to print multiple small parts in an array, the printer would lay down the first layer of one part and then yank it off the bed as it went to print the next.

One tip I'd read on the message lists was to paint the borosilicate glass with a diluted mixture of PVA glue in water.

This almost worked too well - parts became very difficult to remove, to the point where I was worried about damaging the glass when trying to get the parts off.

I could adjust the effective print height and platform level, to try and get that magic first layer height where the parts would stick reliably but not be impossible to remove.  Here I came up against the real problem.  My printer was acting as if the print bed wasn't flat.  If I adjusted for proper print height in the center, the edges of any print larger than a few inches wide would peel right off the bed.  If I tried to adjust so that the edges adhered properly, the center would be smashed so far down against the bed that the filament drive would stall and strip the filament when it tried to print.  The distortion wasn't consistent, either, being worse in some directions than others along the bed.

I knew that my borosilicate glass bed wasn't bowed.  The problem was that the head itself wasn't staying level as it moved across the bed.  I was also seeing some unusual dimensional inaccuracies, where parts were unevenly larger or smaller than they were supposed to be in different directions.

My printer's frame was off.  The wooden frame I had built was very solid, but the positioning of the rails wasn't as precise as it should have been.  A delta-style printer requires that the rails be aligned within a millimeter, or you'd see the kind of print distortions I'd been seeing.  Unfortunately, my wooden frame had no way to adjust the positioning of the rails, as they were solidly screwed in place to the wooden frame.

As I had been building the printer, I had been designing an updated set of parts with various improvements.  I had also been working on an alternate frame design, using threaded rods to connect the three rails in a way where it would be possible to precisely adjust the position of the rails relative to each other.

With my printer working just barely well enough, I printed an all new set of structural parts.

What had originally been simple blocks to anchor the motor mount screws to the rails became fairly complex structural blocks.  3/8 threaded rods anchor the three blocks to each other, and allowed me to very carefully adjust the positions of the rails to be exactly where they were supposed to be.  I also moved the bolts supporting the print bed from the motor mounts to the lower end mounts, so that adjusting the belt tension didn't also tilt the print bed.

Since the rails were no longer mounted straight to the wooden sides, I also printed feet which were slid onto the threaded rods and screwed into the printer base.  Even though the wooden frame was not redundant, I was still using it to mount secondary structural parts.

I have designed new motor mount blocks, but haven't printed them yet.  The old ones will still work for now.  I have slightly changed around the way the bolts connecting the motor mounts to the base blocks work, so that they can be adjusted from the top instead of the bottom now.  It's still fairly difficult to adjust the belt tension, I have to maneuver a small screwdriver alongside the belt in just the right way to reach the head of the tensioning screw buried inside the block, but it was nearly impossible to do so when the screws were adjusted from below.

The top ends of the rails also had new blocks printed, connected by threaded rods to allow them to be carefully adjusted as well.  I have also designed pockets for the upper end travel limit switches, with channels inside for the various wires and cables to the top end of the printer.

In the process of this complete structural overhaul, I've also replaced the printed plastic connecting rods with rods made from RC truck ball ends joined by carbon fiber tubes.

These are much lighter and more rigid than the plastic ones.  The rod ends are much smoother than the printed ones, but do have a little bit of looseness to them that translates to some free play in the head position. I'm not completely happy about that.

With the structural redesign I've moved the rails all in by an inch from their previous position.  They aren't anchored to the wooden side panels any more, anchored only at the tops and bottoms.  To keep the geometry of the delta platform the same, I have also redesigned the carriages and head.

The new carriages are much smaller, yet lighter weight and more rigid than the old ones.  The new carriage is a three-part assembly.  Side plates hold the V-groove roller bearings, now supported on both sides of the bearing with a complete thru bolt.  The center section clamps to the timing belt and also holds the adjustable screw that presses the limit switch.  There are a total of four threaded rods through the assembly, including the annoyingly hard to find M3 threaded rod that goes through the two rod ends.

With the rails moved away from the side plates, the carriage structures can now wrap completely around the rails.  There's a lot of careful adjustment of the nuts required to get the pressure on the bearings just right, to the point where the carriages roll smoothly but are completely rigid against rotation or sideways motion.

I've redesigned the head too, moving the rod ends a bit closer to the enter, but also massively simplifying the structure, reducing the number of parts and making it a lot easier to assemble and adjust the filament tube clamp.

The three tiny cooling fans are now more directly aimed at the hot end barrel now, no longer having to have the airflow routed through holes in the wooden J-head mounting plate.  The LEDs are also now mounted directly in the head structure instead of dangling on their leads.

Finally, while I had everything taken apart, I added some insulation - a layer of foil and some strips of fiberglass under the heated PCB.  It seems to help the bed get to the right temperature and stay there, and probably helps keep the bed heat away from the electronics in the base.

With the new structural design, the movement if the head seems to be exactly where it should be.  I finally have proper adhesion of the print over the entire area of the bed that the nozzle can reach, and object dimensions seem to be correct.  Still to do - proper mounting of the power supply and RAMPS board, and some way to support the filament reel.

Drew's Rostock: The long quest for a successfull print.

With a working extruder, and after MakerFaire a reel of black 1.75mm PLA filament, it's time to try printing.  So I drew up a standard 20mm calibration cube in AutoCad, sliced the stl into Gcode with Slic3r, and sent it to the printer in Pronterface.

It was, of course, an absolutely terrible print.  The block was spongey, with severe filament starvation, wavy and not even the correct size.  About what you'd expect of the first print of a scratch-built printer of a novel design.  I was just happy that it was printing at all.

The filament starvation was the first thing I tackled.  I noticed that starting about a minute into the print, the extruder motor would start stuttering, jerking back slightly about once a second.  ON a tip from the Deltabot message group I tried turning down the motor current.  There turned out to be a very narrow range where the extruder would work properly - too little current and it would stall, too much and it the stepper driver would cut out in thermal shutdown.  Keeping a large fan pointed straight at the driver board was mandatory.  I suspect my extruder might be a bit tight, taking more force to push the filament than it really should, or maybe it's the tightly coiled Bowden tube that's doing it.

The second calibration cube I printed was at least solid, but still had problems.  It was somewhat undersized, had bizarrely wavy sides, and a top that was concave and significantly upturned at the corners.  During bed-leveling tests I had noticed that my printer was acting as if the bed was bowed - the print head had more clearance over the bed in the center than at the sides.  I was fairly certain that my print bed - a sheet of borosilicate glass clamped to the heater with binder clips - wasn't actually bowed downwards, so I suspected something was wrong with my geometry.

When I had originally printed the six connecting rods, I hadn't checked to make sure their length was correct.  I removed and measured them one at a time.  Sure enough, all six of them were a few millimeters shorter than they were supposed to be.  I found I could mostly fix their length by slightly unscrewing the threaded rod holding the two halves of each rod together.  It wasn't quite ideal since I was still limited by pitch on the screw threads in how closely I could get the length to the required 250mm.  I am planning on replacing these rods with carbon-fiber ones eventually.

That fixed some of the distortion.  The test prints were still somewhat wavy and undersized.  At this point I switched to using a hexagon, designed to be exactly an inch across from flat to flat, with a three-eights hole through the center.  The part as printed was about 0.95" across, and had somewhat wavy sides.  The hole through the center was about 0.3" across, and distinctly oval.

While fixing the connecting rods, I had noticed that the carriages weren't very tight.  Whether through inaccurate measurement on my part or poor accuracy on the part of the Makerbot I had printed on the parts on, the V-groove roller bearings weren't pressing against the Makerslide rail rightly enough.  The carriages were actually shifting back and forth slightly as the printer ran, which translated directly to positional error of the print head.

I really didn't want to print any more parts at work to replace these without trying to fix the ones I had first.  I tried clamping the sides of one of the carriages with a big old C-clamp I had to press the bearings closer together.

This worked as far as eliminating the slack in the bearings - I was easily able to adjust the clamp to the point where the carriage rolled up and down smoothly but was rock-solid against rotation or sideways movement.  Obviously, I couldn't print with the clamp on there, it interfered with the movement of the rods and was really, really heavy.  I also didn't have three clamps like that.  I considered removing the carriages and bending them over heat to permanently reshape them, but then noticed that there was just enough room between the bearings, Makerslide, and belt, for a small threaded rod.

One trip to the hardware store, some cutting and drilling, and I had tie-rods that didn't interfere with the carriage movement yet held the bearings firmly against the slides.  It's not pretty but it works.  I have already redesigned the carriages to have adjustable bearing pressure, and will be printing out a complete new set of parts with that feature eventually, but this mod will get me printing for now.

With the tie-rods on he carriages my one-inch test print measured about 0.97".  The test block also had upturned corners and blobbing which suggested that too much plastic was being delivered - or rather, the head wasn't moving as far as it should for the amount of plastic being printed.

There didn't seem to be any looseness in any part of the mechanism, and even when I printed the test hex at agonizingly low speed the size was still off.  Furthermore, careful tests with commanding the head to move a known distance and measuring the actual distance moved suggested that something was wrong with the actual geometry of the printer.  A bit of measurement suggested that my Makerslide rails weren't located exactly right.  The wooden frame I'd built didn't have any way to adjust the rail position, unfortunately.  That will be something I correct in the next version, but I really wanted to get this one working without having to do a complete rebuild.

I took a look at the software instead.  The modified Marlin software has a handful of added settings describing the overall geometry of the delta platform.  One of these - DELTA_SMOOTH_ROD_OFFSET - described how far the drive axis were from the center of the print area.  With some trial and error I determined that if I changed this from the standard 175 to 178, the objects I printed would be within half a percent of their correct size.  That's good enough for now, but I still need to design a way to make fine adjustments to the rail positions.




When printing the test blocks, I had noticed strange, consistent wavy patterns on the sides of the blocks.  Turning the printer speed down below 30mm/sec eliminated them.  Initially I though that this was due to flex in the plastic rods, and expected it to go away when I replaced them with carbon-fiber ones.  While testing some larger prints and gradually increasing the speed, I noticed that when printing long straight lines the print head was actually stuttering, slowing down for a fraction of a second at certain spots, which was causing the wavy spots to appear as the plastic would be delivered unevenly across the line.

I had read on the Deltabot message group that other people had seen this too.  The modified Marlin firmware chops straight lines up into many small segments, since the amount of movement required on each axis is constantly changing as the head moves across the platform.  At higher speeds this can overwhelm the buffer which the firmware uses to plan moves, resulting in momentary pauses in the movement as the firmware struggles to keep up.  Changing a setting - BLOCK_BUFFER_SIZE - in the firmware from 16 to 64 fixed that problem.

The printer head seemed to be able to move in long straight lines accurately at a high speed now.  The next hurdle was getting proper adhesion to the print bed.  I found on larger prints that if any part of the print didn't stick to the bed properly, it would curl upwards as the part printed.  Eventually this would cause the head to jam against the part, causing one or two of the drive motors to skip a step, which would completely ruin the part.

When you have a motor skip a step on a delta printer, it doesn't just shift the layers above the skip sideways like it will on a normal cartesian machine.  It can result in the layers above the skip being shifted and tilted, making for some very strange-looking ruined parts.

One thing that helped is keeping the bed clean.  Any skin oil on the bed can result in failure to adhere.  I have very oily skin, so I've been cleaning the print bed with windex before every print.

Adjusting the bed height is critical.  The Rostock lets you adjust the bed level by turning the small screws on the tops of the carriages which press the upper travel limit switches.  This also seems to shift the center point of the print area, since you're effectively rotating the print volume around the home position.  I also have adjustable screws at the corners of the print bed which let me independently adjust the bed level and height.

The J-head seems to be very prone to making long, fine threads during a pause in extrusion.  Cleaning the hot end before every print seems to help.  I tried adding a silicone wiper and setting the printer to wipe the head before each print, but that seemed to hurt more than it helped.

Despite all this, I'm having trouble getting consistent height across the entire bed.  The printer is still acting as if the bed isn't properly flat, with parts of the print being too high and not adhering, and parts being so smashed down that the extruder stalls when trying to print the base level.  I think there are still alignment issues with the printer geometry, uneven placement of the Makerslide rails and uneven rod lengths, so I'll be running more tests to identify and fix those problems.

At the moment, the printer seems to work at least as well as the Makerbot Thingomatic at work, with a much higher top speed but adhesion problems with any part more than a few inches across.  I still have more work to do to get it working properly over the entire 200mm build area.

Drew's Rostock: Extruder

This started out as an Airtripper extruder, but I decided to tweak the design to better fit my printer and the parts I had on hand, and by the time I was finished it was nearly unrecognizable.

 Changes made from the original Airtripper design include:

Moving the mounting surface to be on the face that the filament enters the extruder through.  This was more convenient for mounting the extruder vertically, with the filament coming down from above.

Redesigning the extruder to use bearings I had on hand.  This included a 525 bearing for the motor axle support, and a 1614RSbearing for the pinch arm.

I replaced the Bowden tube clamp with a low-profile design I came up with that uses three bolts to compress a tightening cone around the tube.  It doesn't require printing threads and seems to take up less vertical space than other designs I've seen.

Finally, I'm a shoulder bolt and two R4RS bearings to support the pinch arm pivot.  This was an experiment I did only to use up some parts that had been in my junk bin for years.  In retrospect it was completely unnecessary, actually makes the extruder harder to assemble, and was kind of a dumb idea.

Plans for the extruder, and all the other parts I've printed for the printer so far, have been uploaded as Thing 31889 on Thingiverse.  I upload these only for reference purposes.  I don't actually recommend printing any of them as they are.  I've found design issues with all these parts as I've built the printer and am preparing a complete set of redesigned parts.

Drew's Rostock: the print head

For my printer's hot end I chose a Mini J-head Mk II, custom-made by hotends.com.  The standard J-head is designed for 3mm filament.  I liked the simplicity and light weight of the J-head, but my printer is intended to use 1.75mm filament, so I custom-ordered a mini J-head instead.  I had been considering an Arcol.hu hot end at first - liked the idea of an all-steel extruder - or possibly a Budaschnozzle, but the Rostock design really calls for a head that's as lightweigt as practical to make full use of the platform.  I won't be able to print the more challenging materials like teflon with this nozzle, but that's not really my aim anyway.

The J-head, like most hot ends, seems to be designed to mount into a slot in a wooden base plate.  Presumably the wood helps insulate the hot end from the plastic structure around it, and is itself sufficiently temperature-resistant to not be damaged by the hot end.  I used a bit of scrap leftover from the Makerbot, drilled and sanded into a mounting plate for the hot end.

Here you can also see the heater resistor held in place with fire putty and the thermistor taped into the other side of the block.

Here I've wrapped insulating wool and Kapton tape around the heater block.  This should make it a bit easier for the heater resistor to keep the hot end at the desired temperature.  I'm not completely sure this is necessary, but it seems to be good practice.

On the triangular black platform, which is normally connected to the six drive rods on the Rostock design, there are three mounting holes at the corner.  These were originally intended for adjustable screws which pressed the three lower end limit switches during the homing procedure.  The latest version of the Rostock software eliminates the need for the bottom end limit switches, so these holes aren't needed for anything.  On the Makerbot, it's hard to see the hot end nozzle when the print head is all the way down against the print bed, as the huge platform which supports the hot end blocks the light.  I'll be putting LEDs in these three holes to illuminate the hot end, to make it easier to align the head in the center of the print bed.

Here you can see the print head installed in the printer, with the three LEDs arranged around the hot end.  You can also see a piece of ducting I printed to go around the J-head.  The hot end is designed to have a continuous airflow across the midsection at all times.  Ideally you want the transition from ambient temperature to plastic-melting heat to be confined to a small area of the hot end barrel.  The transition region has cooling fins machined in it, and a constant airflow is recommended to keep the upper part of the hot end from getting too hot.  The original Rostock printer had a large fan placed next to the printer to blow air through the entire print volume.  I wanted this printer to be stand-alone, and ideally would like to eventually completely enclose it, so that wasn't an option.

There's not much room on the print head to mount a fan, and it's important to keep the weight of the print head as low as practical.  I used three miniature fans - 20mm square - and printed a structure to hold them on the upper part of the print head.  Channels inside the plastic support direct the air downwards, around the J-head, and out onto the print surface.

The fan support structure also holds the Bowden cable clamp and has channels for the hot end heater resistor and thermistor wires.  Power to the fans and LEDs is from the 5V supply lines on the PC power supply I was using for the printer.

Testing of the printer was going well up till now.  I could move the head around under command from Pronterface, and the temperature sensing of the print head seemed to work well.  I set the target temperature of the print head to 100C.  The heater came on, the temperature started to climb, and then my power supply blew up.

I was using an old 350W power supply recycled out of my wife's old PC (She has since moved onto a machine requiring a much larger power supply), using the 12V feed to power the RAMPS board (and through it the motors and heaters) and the 5V feed to power the LEDs and fans.  I knew that PC power supplies typically required a minimum load on the 5V line to be stable, which I thought the fans and LEDs would provide.  What I didn't realize was that cheaper supplies like the one I was using had some shared components in the 5V and 12V switching regulators.  The voltage regulator circuit was designed with the assumption that the load on the 5V and 12V lines would increase roughly proportional to each other, as would typically happen in a computer.  A massive imbalance between the 12V and 5V loads severely stresses the switching regulator.

I had some suspicion of the problem beforehand, though I thought the LEDs and fans would be enough of a load, but I wasn't expecting catastrophic failure.  This type of power supply is supposed to shut off when it detects an improper load.  Instead, something inside blew up and arced over for a few seconds, then the main fuse burned out.  I could probably repair the damage, but I've decided to do what I should have done in the first place, and buy a 12V 30A industrial supply designed specifically for this type of duty.  I'll have to add in a little 5V regulator for the fans and such, but that's no big deal.