Making Mendel

I aimed to build my Mendel in time to show it at the Makerfaire in Newcastle but completely failed. I had two weeks to build it, which I thought was plenty. In actual fact it took closer to three weeks before I got it printing successfully. I had no major problems, just a few snags here and there and a severe underestimation of how long it would take on my part.

Printed Parts

Unlike when I printed two sets of Darwin parts, printing the parts was the easy bit. This was due to three breakthroughs I had at the beginning of the year: -

• The heated Kapton bed removed the need for rafts, which not only take a significant time to print, but also can take a lot of manual work to remove.
• The extruder fast reverse got rid of all the strings, which also took a long time to clean up, especially from inside the Darwin corner blocks.
• The "no compromise" extruder is so reliable that I have the confidence to do multi-part, layer by layer builds, which gets a lot more on the table, allowing longer unattended operation.

I printed the parts with 0.4mm or 0.375mm filament and with 25% infill. For the larger parts I used two outlines for strength. Since the large parts don't need fine detail, I think printing them with 0.5mm filament and one outline would be quicker, but that would need a bigger nozzle.

The weight of the parts, not including the extruder, was only 730g. I printed the outlines at 16mm/s and the infill at 32mm/s, so it's hard to say the total time. Assuming an average speed of 24mm/s at 0.4mm diameter gives about 3 mm³/s. That would put the total time at about 65 hours. I did it as a background task over a few weeks. A lot of the parts were printed as experiments with heated beds.

Rods

I took me an evening to cut all the rods. The method I used was to nail a stop to my workbench to line up the rod against a metre rule.



I then lined a piece of masking tape up with the correct measurement and wrapped it round the rod to mark the place to cut. I also wrote the name of the rod on the tape to make it easy to identify later.



A Black & Decker workmate makes an ideal vice to hold the rods while sawing. I rotate the studding until the thread lines up with the edge of the masking tape. That guides the saw to start in exactly the right place.



I used BZP for all the studding except the z-leadscrews, for which I used A2 stainless steel because it is smoother and generally straighter. I bought the rods from Farnell and even the BZP studding was very straight, a lot better than the stuff you get in B&Q. I also used A2 for all the bars.

It was very hard work sawing the A2 until I switched to a new blade and used Trefolex cutting compound. I am not sure which made the most difference, but I could then cut the A2 much easier than I had been previously cutting the BZP. I wish I had done that earlier, it would have saved a few hours.

Thick Sheets

The thick sheet parts are not really suitable for making by hand, particularly the squashed frog. They have lots of slots, which are hard to make without a milling machine or a laser cutter, etc.



I am not sure exactly what the hole in the bed and the purge plate are for, so I made the bed a simple rectangle with four holes. I am using my own electronics, so I made the two circuit board plates to suite. I simply cut rectangles and I marked the holes and drilled them in the right place, so no need for slots. That just left the squashed frog.

I made a much simpler design with drill centres on it. There is no need for the bulging legs and sloping shoulders. I think they must be just to make it look more like a frog. Fine if you you are CNCing it, but a PITA if you have to make it by hand. Also the holes for the opto tab and the purge plate are mirrored for no apparent reason, so I made it chiral.



This just starts as a rectangle with some holes in it. Then the large slots are made with a saw thin enough to turn in the holes. The outer holes that mount the bearings can be round because they are in a a fixed place, dictated by the holes in the bed. The inner holes need to be slots because the bearings are adjustable. I just left them off the template and marked them with the bearings adjusted and in place.



I made the sheets from 3mm Dibond, which is below the recommended thickness, but seems stiff enough. It is also light weight and very easy to machine.

Thin Sheet

I didn't have any optos, so I used micro switches for my end stops, hence didn't need any thin sheet parts. I simply attached them to the bars of each axis with P-clips. A little RepRapped bracket would be better but I was building this in a hurry, so had gone into bodging mode at this point!







They seem to have sufficient repeatability and certainly will when I replace the electronics with my new design, which will know the motor phase, reducing the uncertainty by a factor of 32. It is the same switch that I have used on the z-axis of HydraRaptor, which has proven totally reliable. They seem to be this one from RS, not cheap.

Belts

These were easy enough to split but, because the reinforcing wires run in a spiral, the blade tends to follow one for a while before managing to cut through it. That leaves a ragged edge with a bit of wire sticking out.

I didn't understand the rationale for slackening the belts until you just don't see backlash when moving one motor detent. I am microstepping anyway, so a motor detent is not significant. I made my belts good and tight.

Snags

I had a few snags with the mechanical assembly: -

The x-axis spacers are too short. The STL files are 5mm shorter than the parts in the STEP assembly. That caused the motor to clash with the nuts on the 360 bearing.



The 180 bearing at the other end was about 10mm from where it should be.



A simple fix was to slide the axis along leaving a 10mm gap to the spacer, the only problem remaining is that the spacers rattle at certain step rates.



The STEP model shows this gap should be only 5mm, but I have been unable to find the discrepancy. My rods and inspection distances are correct and the ends of the rods are flush with the clamps, as they are in the model.

The bed springs seemed to be too long to compress to the length of the bed-height-spacer-31mm_1off, which is not actually 31mm, but 29mm, so I don't know what gives there, I just spaced them a bit higher.



The bolts in the z-bar clamps are too long to allow the bearing to be inserted. I replaced them with shorter ones.



Similarly the bolts in the x-carriage get in the way of the extruder I fitted.



The J3 jigged distance did not seem correct. The distance between the y-bars is set by the J2 distance and the 3 nut spacers.

Extruder

I used Wade's extruder design as I didn't have time to adapt any of my own.



The gears work well, with very little backlash, but the small one has some movement on the motor shaft. It is just a press fit with a flat on the shaft. I need to redesign it with a captive nut and grub screw.

I didn't have a suitable M8 shoulder bolt so I made one from brass by attaching a nut with a pin through it.



I hobbed it with an M3.5 tap. I haven't measured the grip, but I get the impression it is not as high as Wade gets, I am not sure why.

For the bottom half of the extruder I used some parts that Brian was looking for volunteers to test for him.



The insulator is made from PEEK with a PTFE liner. The idea being to get the strength of the PEEK and the slipperiness of the PTFE. It seems to work well with PLA, which is all I have run through it so far.

The barrel is long because it is designed to take nichrome, but I just screwed it into a block of aluminium with a vitreous enamel resistor in it.



This was left over from a previous experiment. I have now moved onto a smaller resistor size, so this block could be smaller. The barrel could be a lot shorter with this arrangement and that would give less ooze and less viscous resistance.

The extruder works well with PLA. The main problem with it is that it mounts at right angles to the x-axis, so the motor severely restricts the maximum height of the z-axis. Another issue is that to remove it you have to remove the motor to get at the bolts. To remove the motor you have to remove the big pulley to get at the motor's bolts, to do that you have to remove the pinch wheel assembly. I.e. to remove the extruder you have to completely disassemble it!

Electronics

To get up and running quickly I used the same electronics that I use on HydraRaptor. The only difference being that I used MakerBot V3 stepper drivers. These use the A3977 chip and give x8 microstepping. That gives an axis resolution of 0.025mm, but more importantly gives nice smooth running.

When the weather was exceptionally dry I found they are very sensitive to static. A discharge to any part of the machine would cause the A3977 to shut down its outputs and draw enough current from the 5V rail to cause the 100mA regulator to current limit. The red LED on the power rail goes dim. Powering off and on again fixes it and there doesn't seem to be lasting damage. I suspect that might not be the case if the 5V rail was not current limited. Apparently the only way to fix it is to add external Schottky diodes. That is very disappointing as one of the nice features of the chip is that it is supposed not to need them. I will investigate further to see if all eight diodes are needed before making my own board.

Firmware

I used the same firmware as HydraRaptor. I just added some compile time conditionals to cope with two pin outs and a different IP and MAC address for each machine. I also had to change from 16bit to 32 bit positional commands because the axes are bigger.

Software

I used the same Python software as HydraRaptor but I had to re-factor it quite a lot to support both machines. I added a class to represent the Cartesian bot which holds the axis resolution, direction, maximum speed and acceleration plus the IP address. I also added a class to represent the extruder controller as I have calibration values unique to each board. I already had classes to represent thermistors and extruders.

I can run both machines at the same time from one PC and, because I only use the Skeinforge output for the toolpath, I can use the same sliced files for either machine. This is despite the fact that they run at different speeds and are loaded with different plastic.

Results

So here is the finished machine: -



And here is a video showing it being tested: -


I am running the X & Y motors at about 0.75A and Z at about 1A. I have set the maximum XY speed to 100mm/s, but I think it could go a lot faster. Z only goes at about 5mm/s because not only is it a threaded rod drive, but it is geared down by the belt and pulleys!

I haven't printed a lot yet, but so far the results look as good as they do from HydraRaptor. The next thing to do is add a heated bed and try ABS.

PMMA on PI

Previously I described how I made some lamp shade clips for some friends from PMMA (Acrylic). Unfortunately the clips got lost in the post so I had to run off some more. I used a sheet of acrylic as the bed the first time, but since then I have developed my magnetic bed with PI (Kapton), so it was an interesting experiment to see if PMMA would stick to PI.

The originals were made with the bed at 100°C, but the insulating effect of 4mm of PMMA gave a surface temperature of only~85°C. That was sufficient to form a partial weld strong enough to resist warping. My first test was with the bed at 100°C, but that came loose after a few layers. With the other plastics I have found I need a higher temperature on PI because it does not form a weld. It sticks by some magic that I don't understand, possibly Van de Waals forces.

I tried again with the bed at 130°C and this time it stuck well but was easily peel-able by bending the bed. So another plastic that works on hot polyimide.



I have also been doing multi-part builds one layer at a time, something I hadn't dared to do until recently because my previous extruders were not reliable enough to risk a bed full of objects.

If you get the height just right on the polyimide bed the first layer comes out almost perfect. You can actually make one layer thick objects, i.e. 0.3mm in this case.



These came out very well.



I arrange multiple items by reading the STLs into AOI and orienting and positioning them. I then union them together in pairs and then union all the unions until I have a single object. I then convert that to a triangle mesh and export it as a GTS. If I don't convert first I only get one object in the GTS rather than the composite. I then slice as one object.

These springs show a very rare bug in Skeinforge or my code where it does a move with the extruder on. That is the only time I get any string.



Here is an assortment of smaller parts :-



It is best to group the taller parts together so when it gets to the higher layers the head does not have far to travel.



There were so many in this build that it overlapped the hot part of the table, so a few corners lifted a bit. I think ideally the table should be a bit bigger than the movement to keep some warm air around the objects. Perhaps a wall around the edge would keep some heat in.

I did have the crazy idea of building a wall around the objects as they are made so that there is a small gap of trapped hot air around them. I.e. the machine builds a disposable cocoon around the object to act as a heated build chamber. It would be a bit wasteful of plastic of course but it could be made from a cheap paste if we had a second head.

PLA on a vacuum bed

Having successfully made one Mendel z-leadscrew-base_2off as an experiment to try ABS on a vacuum bed, I decided to make the second from PLA to see how well that works on a vacuum.

Three of the corners lifted very slightly during the build (about 0.2mm) but not enough to matter except to a perfectionist.



When the object cooled it did not break the vacuum, unlike ABS. The part was still easy to remove though.

The base is flat apart from the corners and a few shallow dimples.



The quality was very good with no clean-up at all. I recently discovered a simple bug in all my builds after I moved away from 0.5mm filament. With 0.3mm and 0.4mm filament my layer height was 0.24mm and 0.32mm respectively. The problem is my z-axis only has 0.05mm resolution, so layers alternated in height, none of them being spot on. That caused the sides of my objects to not be as flat as they should be as the filament width varied from layer to layer. I now use 0.375mm filament giving a 0.3mm layer.

Suck it and see

The magnetic steel and polyimide tape bed works very well for manual operation but I am pursuing the vacuum bed idea for fully automated production. I went through a few designs in my head before actually making anything.

The first idea was to drill an array of holes through an aluminium plate and connect them on the back by milling a network of channels. I would close the top of the channels with some Kapton tape. The problem with that idea was there was then nowhere to mount the heating resistors unless it was on top of the Kapton sealing tape, which didn't seem ideal. My solution to that was to mill a channel into the edge of the plate and wind a coil of nichrome all the way round it. That was my plan until I realised there would then be nowhere to attach the vacuum hose.

A solution might be to use a Kapton or silicone stick-on heater and use it to seal the channels in the underside.

What I actually did was to mill a grid of very fine channels into the top surface allowing me to attach the vacuum hose to the side edge and drill a small hole down to meet it, leaving the bottom free for the resistors and thermocouple.



The channels are about 0.5mm wide and 0.5mm deep on a 5mm grid. I milled them with a 0.3mm conical bit that I bought for milling PCBs.



I used a feed rate of 2mm/s and 0.1mm cut depth per pass. My MiniCraft drill runs at about 20,000 rpm. The results were not very good. I have only ever milled plastic before with HydraRaptor. It struggled cutting aluminium such that the shaft of the drill was being displaced in the direction of the bed travel. It raised a burr about as high as the channel is deep. My friendly local milling consultant told me afterwards that aluminium does not like lots of flutes. He recommended a D-shaped cutter with a single cutting edge and a higher spindle speed.

I sanded the surface flat with 240, 600, 800, 1200 and 1800 grade wet-and-dry sandpaper and then polished it with metal polish. I did this to get as good a seal as possible with whatever was placed on top.

I attached some polythene pipe using an M5 copper welding nozzle screwed into a tapped hole in the side of the plate. I use a tapered tap so that the thread would bind to form a seal. I used Fernox LS-X jointing compound to make sure it was airtight. I think it is silicone, so should handle the temperature.

I was hoping to get a perfectly air tight seal and be able to use a static vacuum generated by a syringe. It doesn't seal fully though. I believe normal vacuum tables use rubber o-rings set into a groove to form a seal. I reasoned that would not work in this case because, whereas sheets of stock for milling are stiff enough to remain flat and squash the rubber, thin films would just bend upwards. My idea was that the thin film would be sucked into the channels and be compliant enough to seal it. I think it fails because the edges of the channels are too rough due to my poor milling.

I first attached a small vacuum pump that I made for my jukebox. It is just an aquarium pump with a pipe attached to the air inlet and the case is sealed with rubber glue.





It is not a very strong vacuum, but it is enough to pick up a CD with a suction cup made from the end of a child's rubber dart. I plan to use it for SMT pick and place soon. I measured it at 960 millibars, which is also the extreme low reading on our barometer. I knew that the vacuum it created was less than the atmospheric variation because I started off with an absolute pressure transducer on my jukebox to detect if a disk had been picked up. I had to change the trip point about twice a year because one setting would not work for both extreme high and low weather conditions. In the end I added a second sensor to make it differential.

I placed a piece of 0.075mm polyimide film on top. This is about twice as thick as the tape I use.



The video below shows the effect of the vacuum. It pulls flat and has some resistance to sliding but is not a very strong grip.



I built a Mendel part at 100°C for my first test. The film stayed flat during the build but a few corners lifted. When the part cooled it broke the vacuum and wrinkled the sheet. It was past the point where it had hardened so the base was perfectly flat apart from where the top corners had lifted slightly during the build.



I measured the temperature of the surface and found that it was 10°C lower than that measured underneath by the thermocouple. I raised the set point to 110°C and made another part. This time only one corner lifted (left side of the boss in the middle) .



Here is a speeded up video of the film releasing as the bed is cooled down to 40°C by a fan.



And here is me simulating removing the object by sliding the film. Ignore the ×16 annotation, I am not that slow!



The next thing I tried was a really big part of Mendel. I didn't trust my weedy aquarium pump to hold it down so I used a 1/4 HP 180W 3 cubic feet per minute pump rated to go down to 0.1 millibars. I bought it for £150 over two years ago to make a vacuum bed for milling but never got round to it, so it has been sitting on a shelf, like a lot of other parts and materials I have bought for experiments but not had time to use.



When connected directly to the vacuum gauge with a length of plastic hose it goes down to about 30 mb. I think I would need better quality fittings and pipe to get down to 0.1 mb. When connected to the vac table it gives 40 mb, so although it does leak, it still gets most of the available downforce from atmospheric pressure, i.e. ~15 lbs / square inch.

The part I made ended in disaster because the vacuum broke during the build. I think it was mainly because the object was not quite centred on the table so its outside perimeter was on top of the last vacuum channels. Before it failed some corners had lifted a little, early in the build, so it looked like the ABS does not stick to film as well as it does to tape.



I centred the table and made a slightly smaller piece. Actually this was the same depth as the last piece, so the perimeter falls about half way between the last two channels. Ideally I think you don't want to be that close to the edge.

I raised the temperature to 120°C, so the top of the film was probably ~110°C.

The film stayed vacuumed down during the build but still broke the vacuum and wrinkled during the cool down period. That means that even with close to the maximum vacuum it cannot hold the contraction force. Not a big surprise as I realised a long time ago the warping can generate a lot more than 15lbs / square inch of pull. The only reason I thought this might work is because the plastic does not warp while it is kept hot and indeed the vacuum holds during the build. The problem is that the object does not stick to the film well enough. One corner peeled early on and lifted further as the build progressed. The rest of the base is flat though, so the part is easily good enough to use. The higher temperature and vacuum meant that the grid lines are just visible on the object base if you get the light right.



So just to make sure I can make an object this size on tape without warping I made the larger part again on my magnetic bed. That failed because the bed slipped part way through. I knew that was a likelihood and that I need to add a couple of dowel pins in the corners, but I didn't want to do that while I was experimenting with vacuums. I gambled on it not slipping and lost.



It did build enough to show that it sticks much better though. The corners stayed down and the build lasted long enough to go way past the point where the corners lifted on the vacuum bed. The base was perfect.



I could only think of three possible reasons why the corners would lift on the vacuum bed and not on the magnetic bed.

1. The surface of film might be different to tape. After all, tapes have the magic property that the glue only sticks to one side and peels from the other without leaving any residue.
2. The film is thicker, so has more thermal resistance, which might have some influence.
3. Perhaps the film can lift a little in between the vacuum channels allowing the plastic to peel away and then be sucked flat again.

My best guess was that the third explanation was the most likely. Perhaps closer spaced channels or thicker film would solve it. I had a sample of 0.15mm film so that was the easiest thing to try next. It stuck much better but towards the end of the build I heard a snapping sound and saw that one corner had lifted.



More importantly though I could see that the other corners were deforming the film upwards as I had suspected. It is hard to see here, but the slightly raised blister of film was over a channel, so subject to the full vacuum force.



This shows that even with a heated bed, there is sufficient warping force at the corners to beat atmospheric pressure. The part ended up with one chamfered corner and dimples in the others.



So in general the experiment is a failure as it does not work as well as the magnetic bed. It does allow easy automated removal though. All you need to do is tape down one edge of the film. When the object cools it wrinkles the film and breaks the vacuum. A fence could then push the object off the bed with not too much force, as the film peels easily. When the object is gone the film springs back to being flat and the vacuum can pull it down again ready for the next object.

Although corners lift, the objects are usable for making a Mendel, it is only an aesthetic issue in this case. I think rounding the corners of the parts would fix it. I guess PLA might stay stuck as it warps less, and ABS only fails in extreme cases.

Thanks to Paul for providing the polyimide film samples and lending me the vacuum gauge.

Quick release bed

I am in the process of making a heated vacuum table to hopefully allow automatic ejection of finished objects. In a conversation with Laszlo he mentioned he was planning to use a heated steel bed and use magnets placed around the object to hold down a sheet of Kapton. I turned the idea upside down. Why not stick Kapton tape to a sheet of steel and clamp it to a heated aluminium bed using magnets underneath?

I found a thin sheet of bright springy steel that was part of an electric toaster. My best guess is that it is one of the grades of stainless steel that is magnetic. It is only 0.3mm thick so it is relatively flexible, but it always springs flat. It came from a Kenwood toaster that gave good service until our cleaner suggested to my wife that she should turn it upside down to get rid of some persistent crumbs. The next time it was used it burst into flames because a crumb got wedged between the element and the steel plate and burnt through the nichrome.

I made a tiny heated table from an off-cut of 6mm aluminium. It is only 105mm x 73mm, which is smaller than a MakerBot CupCake bed but I think it is just big enough to make all the Mendel parts.



I have run out of AL clad resistors so I made my own from vitreous enamel ones embedded in aluminium blocks with tin foil. I used two 6.8Ω resistors in series driven from ~ 26V AC. That gives about 50W and a similar warm up time to my larger bed driven with 200W.

I milled flat bottomed holes to within about 1mm of the surface and embedded five neodymium magnets which are held in with Kapton tape.

I used M3 threaded nylon stand-offs as insulated table legs and mounted it onto my XY-table using a sheet of 4mm aluminium / plastic laminate called Dibond. It is very nice material to work with.



The steel plate covered in Kapton tape then sticks to the top of the table. I heated it to 100°C and tried making some ABS objects.



This worked well and the objects were easy to remove by bending the plate and peeling them.



The magnets are strong enough to hold down even big objects. The only problem I had was that the nozzle snagged on the first layer of this object and managed to slide the steel plate, causing the first layer to be offset.



Contrary to popular belief, FFF does require significant force and benefits from a stiff extruder mounting.

A couple of pins in the corners to act as dowels would solve the sliding problem.

Here is a video showing how easy it is to remove the objects: -



It is still a manual process though, so I will pursue the vacuum table idea to attempt to make a bed that can eject the object itself.

Will it stick?

ABS sticks very well to hot Kapton, so I wondered what else would stick to it. The first thing to try was PLA. This sticks pretty well to cold masking tape and doesn't warp much, but large objects do have some warping. I figured heating the bed to around 50°C would fix that. Rather than changing from Kapton to masking tape I decided to see if I could stick PLA to Kapton and get a shiny surface as well.



The first bracket was made on cold masking tape so the base has a matt finish.

The second one is on Kapton at 50°C for the first layer, dropping to 40°C after that. My logic was to have the bed just above the glass transition to make it stick and just below afterwards to stop it warping. As you can see one of the hole outlines did not stick properly. The PLA was extruded at 200°C for the first layer and 180°C for the rest.

For the third one the bed was at 55°C falling to 45°C. The outline stuck properly and the base is nice and shiny. The surface imperfections you can see are from gouges in the aluminium bed caused by a slight accident with a decimal point. It caused the nozzle to be rammed into the bed and then the X-Y movement ploughed furrows. These show up through the Kapton tape.

The last one is my first ABS test for comparison.

It was looking good, so I tried something bigger, a Mendel belt splitter jig: -



The left hand corner lifted and the object ended up more warped than it would have been made on cold masking tape.

I tried again with the bed at 55°C all the way through the build. My extruder started jamming so I increased the PLA temperature to 210°C for the first layer and 190°C for the rest, the values I had been previously using on cold tape.

This time it was successful and stayed stuck down: -



The base came out perfectly flat and more transparent: -



The extrusion lines of the three solid base layers are less visible and you can see through to the sparse infill. This is only 25% but the object feels incredibly strong. I get the feeling the hot bed makes things stronger.

There is a bit of a meniscus around the edge. This is mainly because I had a bodge of a -0.1mm offset in the first layer outline to get PLA outlines to stick to tape reliably. I removed the bodge and made this object: -



The base layers are very transparent here, even more so to the naked eye than the camera shows. There is something a little odd with some of the extrusion lanes above the bottom left hole. I think those discontinuities must be the plastic squirming a bit while extruded, which is usually a sign of not being stretched enough.

The top of the object has a small defect: -



There is a small hole above and right a bit of the centre. I think this is because the plastic doesn't span gaps as well without a fan, so it fails to bridge the sparse infill properly. I wasn't watching so I didn't see exactly what went wrong.

The next plastic I tried was HDPE. Not surprisingly it doesn't stick very well to hot Kapton. With the bed at 130°C it stays molten but is quite rubber like. With the bed at 110°C it sets and turns white (because it crystallises I believe). I tried various combinations of these two temperatures but could not get it to stick reliably. I could lay down the first layer of a raft but then subsequent layers would rip it up as the adhesion is very low.



I think the way to do HDPE without a raft is to extrude it onto a thin sheet of HDPE, or maybe polythene, held down by a vacuum and heated to prevent warping. That will have to wait until I build a little vacuum table, hopefully this weekend.

Last on the list was PCL. That sticks very well to Kapton heated to 40°C but it never sets and makes a soggy object.



Before the heated bed I used to build with a fan, and at only 40°C the bed has no trouble holding temperature, so I tried with the fan next.



That worked OK and built a complete object: -



The infill did not stick very well to the outlines of the holes, especially on the downwind side. It probably needs a denser infill, and perhaps some overlap. 25% fill is not really appropriate for PCL as it very soft and flexible.



The bottom is smooth and shiny as expected and it took some effort to peel it off, so I expect large objects could be made. I couldn't experiment further though because the filament started buckling in my extruder.



I can't explain why it worked for a while and then stopped but I tried higher temperature and slower extrusion but could not get it reliable again. The pipe could probably be a few mm closer to the pulley but not much more because it would hit the pinch wheel.

I don't have a lot of use for PCL, other than using it up. Dropping it from the requirements for the extruder would allow me to use a smaller pulley. If you look at the table at the end of this article, you can see that it is only PCL that struggles for grip with a worm pulley. I think I could drop to half the diameter, which would just about bring the gear ratio into the range of a single pair of spur gears. I have a 4" Meccano gear that gives 7:1, so I might try that in my next extruder.

So hot Kapton works well for everything I have tried so far apart from HDPE.