Introduction

For the past few days, I've decided to work on the Coaxial8or project rather than write about all the things I should work on afterwards. I will say that I'm now deep enough into the project that I'm looking at the number 8 a little differently in day-to-day life, and those feelings manifested as a parody to Oscuro - Shake:

My bones still shake... when I hear:
8...
8...
8...
8...
(du do du do)
8...
8...
8, the endless
8...
8...
8...
8...
My bones still shake.

New aluminium print

The part came in a needlessly large box, and it certainly looks brighter than my previous prints:

The input seal bumps came out excellently, but the output was a little tilted:

Another thing is that the 6mm cartridge didn't quite fit.

CAD model changes

• Changed the cartridge hole back to 6.3 from 6.2mm
• Increased the cartridge gap from 1.5 to 2.3mm so that 2mm-thick files could fit.
• Increased the amount of material around the output seal bump, as well as increase the height from 1.2 to 1.5mm.
• Decrease the emboss angle of my logo from 45 to 30 degrees.
• Rotate the M6 thread 180 degrees. 
  • For some reason, PCBWay printed with the logo facing down, which is opposite to the screenshot they showed me. I can only imagine it's because they thought the thread would print better

Plates and tapping

I asked for a new hotend holder top / bottom to be cut out, since those are the parts that changed when I removed the side panels. I also asked for the clamp plates to be made out of both steel and aluminium.

Tapping the M4 thermistor grub thread as a bit of work but, by imagining the forces when the tool cuts the adjacent wall and accounting for them, I was able to tap just fine:

The seal bump seems to work as intended, providing a nice flat face by using the stainless steel nozzle. This is much better than having to flatten one massive face.

It does seem that the aluminium plate slightly bends up. So does the steel plate, but to a much lesser extent.

The steel plate cleans off with a few quick brushes of a file:

I tried tapping it the same way I've been doing it since R1, but the thread came out bent. I also tried free-handed tapping and it came out about as bent.

My hypothesis is that the cloner plate can slide out of alignment enough that it taps misaligned. I had further evidence when I tried to do the same with an aluminium plate.

I even tried tapping a steel cloner plate with no success:

I then tried an idea, inspired by the M4 thermistor grub thread, which was to use the vice to keep the up/down axis aligned and then I visibly check the left/right alignment:

This worked, and I was finally able to get a straight tap through steel:

Unfortunately, that's where the road will end for the steel plate. The above plate was just to use as practice, and it turns out that there was a divot in the other two plates that completely stops the tap from turning:

Thus, I used the aluminium plate to flatten the remaining inputs, which caused it to deform slightly:

I then tried to enlarge the cartridge hole, failing with a 6mm drill bit but succeeding with a 6mm reamer the uni just happened to have.

Fitting into the cover and carriage

It looked to fit well in the cover:

Unfortunately, the CAD model of the cartridge I downloaded from the internet was wrong and so the frontback plate touched the standoffs before the to / bottom plates could touch the x-carriage as intended. However, it was very stiff and reasoned that it was probably better this way since the frontback plate would always be relatively flat compared to the relatively misaligned top and bottom plates. I redesigned the holder.

30mm 160W cartridge

All the steps I've had to do took so long that the 30mm cartridges from AliExpress arrived:

I'd imagine they're old stock since no modern-day hotend would use these. They're also unmarked, but I took one out and it was 3.3 ohms. The ends are kinda dented-looking but should be serviceable. I had to be careful when stripping as there's only 7 strands in this red wire. 2 of them broke off when stripping for the first time, and after a heatup test, I decided losing almost 30% of my strands wasn't good enough. It's a good thing I checked, because I took the wires out of the quick connectors and there were only 3 strands standing!

I had the idea to install the heatsinks and then assemble the holder:

As it turns out, there wasn't enough space for the wires so I couldn't install the new holder. However, the Coaxial8or was able to heat up from 20 to 220C in 110 seconds with an unrefined fan mount and uncalibrated MPC, so I'm onto a winner here.

I should mention that I had to completely replace the thermistor because the old one is effectively welded into R2. I took this time to also trim down the hook-and-loop cable ties to 75mm to make them much easier to put on and take off.

Recutting the hotend holder

For engineers, the lyrics of XO-IQ - Get It Right will resonate:

Get. It. Right.
Do it all again.
This time round, I'll be better than.

Or, for actually trying to manufacture anything, a parody:

Get. It. Right.
Not do it all again.
This time round, it is better than.

Anyway, I actually had a new holder design. The reason why I tested the hotend with the older one was because the waterjet momentarily stopped working and the technicians were pressing buttons like game combos trying to rectify the issue. 2 days go by and the issue was resolved, and my parts were the first to come "fresh out of the oven":

As you can see, this design has space for the wires, and now holds the frontback plates from both sides. I think it's pretty elegant.

I had the idea of screwing the bottom plate to the top plate to keep it from flying around.

Currently, the best strategy seems to align 2 heatsinks at a time. The first 2 are rather easy, the next two are doable, and the 3rd set was rather difficult until I had the idea to use needlenose pliers between them and the 4th set:

The same strategy can be done for the 4th set:

Unfortunately, because of the wires, it was difficult to slide the frontback plates in. Thus, I've reduced the heights of the tabs:

Filing the cover

I didn't want to reprint the cover because it takes 4+ hours to print and there are issues with the design. I tried out the fan duct and there is much less airflow compared to no duct, and the trenches I allocated for the fan wires are needlessly small. 

I had to cut the top a tad and then use calipers to measure how much I had eroded away:

It took like 20 minutes once I got the strategy.

With this, I was able to install the c8or onto the cr600s.

So first the parts got cut out of the waterjet:

I then filled off the edges (with a large file) and deburred (with a small one), which takes a multiple more time than it takes the waterjet to cut the parts in the first place:

When all together, I could test and deduce that the Holder Side plates were useless and didn't bear any load whatsoever. 

Thus, I've removed the side plates from the CAD assembly:

The outer cover setup and print

Surprisingly, this turned into a multi-hour ordeal. Some of that was just getting one of the uni 3D printers to stick to the bed, but it was mostly to do with setup. 

I don't usually design parts that need supports, so I had to be setting that up. But then I'd get supports where I didn't want them, and Cura really doesn't like reloading a multi-body .3mf, so I modelled the no-supports zone in fusion. Additionally, I added a thickened section. I should've just done 100% infill instead of the 40% I set. 

Furthermore, as I was looking through the gcode preview, it did seem like there was a lot of excessive material being used so I trimmed it down. It also allowed me to catch an issue in the design, which is that I wouldn't be able to install one of the fans without first threading the wire through and resoldering.

It has then been printed, and the tree support broke away easily.

It does seem I overestimated how steep an angle I could print, which was 60 degrees. Instead, it seems that even 55 is slightly over the limit. I've changed a setting so that Cura prints from the inside outwards in aims of improving this.

The hotend holder design

I thought through a mental simulation and, like I mentioned in the previous log, I didn't have the confidence that the way I designed the side plate a few days back was going to work as intended, due to material bending:

Thus, I've opted for a 5th M3x12 bolt and dual side plates:

Clamp plate of R2

In anticipation of R3 and so that I can test the fit of the 6028 fans in the newly-printed cover, I disassembled R2. The reason why I haven't done much testing is that all the inputs were slightly leaking, as can be seen by the multicoloured  blob that remained:

Additionally, it seems that the elevated temperatures cause the top surface to rust. I couldn't see any rusting on the sides or bottom:

Like with last time, the strategy to separate and clear the heatsinks worked just fine:

1. Unscrew the clamp bolts
2. Heat to 150C
3. Use spatula to separate
4. Cool to 130C
   1. this prevents long, thin wisps and also allows for a "warm pull" of the inputs for the next step.
5. Pull clamp plate from Coaxial8or
6. Manually extrude material from each input till it pops out, after which the buldge can be snipped off and the filament recoiled.
   1. I did have to use needlenose pliers on 2 of the inputs to help them pop off the heatbreaks. 

Thinking of reversing the extruders that have shorter bowden tubes

The DDEs, more commonly called "BMG Clones", should be able to work when mounted in reverse. The benefits include:

• Much easier access to the large gear, especially for extruder E6 
  • (remember it's E0 - E7)
• The short bowdens will be even shorter
  • The drive gears are closer to the handle end of the DDE.
  • The improvement would be about 50mm or so, which will contribute a larger percentage difference for the 30-35cm bowdens of E0, E1, E6, E7 than the 45 -50cm of the others.

I decided to just go in and scribble something up for the fan duct. In my years of FFF printer upgrading, I've always dislike having to design them since they usually need to be more organic. To my surprise though, this duct came out better than I expected. It's effectively invisible from the front, giving the coaxial hotend assembly a very minimalistic and open look:

Design methodology

So I first decided that I was going to go back to the symmetrical-bolt mounting of the 6028 fans, and with the hotend holder now entirely made of metal, it can also help with heat dissipation. I then noticed that there'd be more space above than below the cooling 6028 so I pointed it upwards.

Then, similar to how I'd go about solving a maze, I started thinking up ideas on how the duct would start and end. I knew that the rear of the Coaxial8or was the most suitable location for the duct output. It also looked like a revolve would be beneficial to cool from multiple directions. I gladly found out that Fusion supports lofting from such a curved face:

I went on to refine the geometry so that it was at least 3mm or so from anything else and didn't stick out beyond the bottom wheel bolt:

I was finally able to model the new hotend holder, which uses aluminium plates for its construction. No longer do I have to worry about the holder unintentionally melting, sagging due to the weight, or blocking the view when inserting the heatsinks into the couplers.

This was one of those things that is kind of hard to model because when there's nothing but an empty canvas, it feels like every plate depends on every other plate, thus a cyclic reference (aka catch22 / chicken and egg problem). 

I initially was planning on a jaws-based solution similar to the redesigned holder for R2, but had concerns with grip so I just went with grub screws. This design allows all 8 heatsinks to get a grubscrew.

Instead of saying something along the lines of "I expect the part will be printed in X orientation", I instead just specified "There shouldn't be any powder / supports in the internal channels" and the engineer quote was $47.69; almost 10 entire dollars lower than earlier quotes. Obviously, I wanted to find out why and the engineer sent this:

I thought the bridge overhangs (see below) weren't allowed and produce a coarse surface finish but the engineer informed me that it'll be the same as the forward tilt. 

The grand total with all the fees and shipping (that didn't budge from $12 regardless on if the part was 30, 29.9, 29.4 or 29mm high) was 4,777 pennies (£47.77) so I'll take that as a sign that I've "hit the jackpot".

I found Xometry had an article on DLMS design practices and it does mention that "unsupported bridges" can be 2mm long, and the channels are only 10% longer than that:

Essentially, I've been trying to model around a problem that doesn't exist. Thus, the orientation the engineer chose is indeed the better one when also considering the external features like the output port.

My main hope is that the port rings print correctly. I don't know what the shape would be called, but from looking at past prints, I'd guess that the minimum fillet is 0.25mm radius.

Then, by screwing in the stainless steel nozzle, it'll flatten out to a very narrow surface. This both reduces the chance for defects to allow material to leak through and places much higher pressures on the sealing faces.

Partially to answer @Trovoski, but mainly so that I can have confidence in the coaxialiser simulations before I spend another £55 (engineer quote + shipping) on the 3rd heatblock revision, I've been looking at the cross section of the extrusion.

0.6mm nozzle

While it's too small to see the shape of the inner core, I was able to at least see that there was an inner core with a 0.6mm nozzle. I did yellow : black and cyan : copper as a 2 : 1 mix and then cut the extrusion with the PTFE cutter.

With the cyan/copper extrusion, I could also see a streak of copper on the side. This seems to be disconnected from the inner core. I suspect that this is residual copper that has clinged to the walls of the internal channels when cyan came through.

Nozzleless

Unfortunately, the extrusion was very uneven and bumpy when I removed the nozzle. I hoped that it would just be a smooth, 2mm-or-so filament, but it actually came out 2.05 x 2.60mm. The good news was that I was at least able to see the transition from cyan to transparent-blue:

As you can see, one side transitions faster than the other.

1.2mm nozzle

I bought the largest stainless steel volcano nozzle. The reason I didn't buy a standard brass one is because I intend to use this nozzle to "stamp" the R3 input/output holes instead of using abrasion. 

I hoped to use the extrusion to push the material out of the nozzle but said extrusion swelled to 1.35mm so I didn't bother to try.

The image below shows the extrusion cut at 4 sections, going further back in time from left to right:

Now perhaps I should've extruded more than about 50cm of filament, but I was using the TFT24 and its "slow" speed indeed is slow, but it's "normal" speed seemed too fast for the R2 and it's inadequate meltzone. Out of the 4, the 2nd one to the left has the most consistent edge (hence the title).

I also cut one just before the nozzle to take a look:

It kind of seems that the inner core is mid-transition.

R3 simulation

I also questioned if there was an even better way to see the ParaView results than using cutting planes or streams, and I did:

Using the threshold settings, I was able to get this cool visualisation that looks like a Minecraft speed build:

Since the input is split in primarily perpendicular directions, compared to R2 (below), it seems that the 4 paths arrive at approximately the same time, hopefully meaning that the output core would be more squircle. 

Anyway, this is what the simulated core material looks like for R3:

[Nov 03]

I secured the 1.2mm nozzle to one of the aluminium plates and was trying to use the extrusion I left on it to push the material out of it. The PETG became much too floppy to succeed with that plan. However, the "task failed successfully" because the pushing and pulling caused the plastic at the nozzle input to flatten from the multicoloured cone it was initially, revealing the cross section:

Black area: 0.705 mm^2

Cyan area: 2.436 mm^2

Ratio: 1 : 3.46

Finally! I've been trying to solve a solution essentially daily since the 8th and I've finally arrived at one that minimises the drawbacks and incorporates the benefits.

The main issue was mounting the 30mm cartridge in such a way that it didn't block an allen key to the plate bolts and that the cartridge can be held in a clamping mount instead of grub screws. When I got to save v45, I more or less abandoned mounting the cartridge horizontally since I couldn't think of a solution that didn't look ugly (and likely use a relatively large amount of material). 

My ceiling for modelling was 20cm3, and the result is a respectable 17.2cm3. The PCBWay autoquoter says this print is $35.

To cut down on material usage while minimising radiation, I've come to this heatsink-fin-esque design. The cutouts are perpendicular to the X axis to minimise air moving through the heatblock during printing.

Additionally, measures have been taken to both try and prevent leaks and prevent difficult cleanup should they occur. All inputs and output has a print-in-place o-ring to encourage deformation at the contact points, hopefully avoiding the need for post processing like the previous revisions. Then, leak channels have been added to allow any leaks a low(er) pressure way out, instead of a leak deciding to show up wherever. Furthermore, 5mm steel plates are used, with the mounting bolts as close as possible to the inputs to minimise flexure. Lastly, the design has a gap between the input plates and the cartridge+thermistor as a last line of defence against rogue plastic.

This solution has been facilitated by a modified input shaft and coaxialiser geometry. In addition to the split (hopefully) increasing flow rate, it allows the inputs to be connected without a bridge overhang, which would be the case if a basic L path was rotated 45 degrees. The coaxialiser is slightly more compact now. I did try orienting the coaxialiser vertically, but it uses more Z height.

 From the top of the o-ring to the Y split is 19.5mm long uninterrupted, so I'm expecting that this heatblock can put the "ate" in Coaxial8or when it comes to consuming filament.

As for aesthetics, I think I've done a passable job with the limitations I have to abide by. Like previous designs, there's a main element of symmetry. Somewhat inspired by earbuds, I've used fillets to make everything seem more organic, like a pebble. There are also some sharp edges, a design tip I saw used for cars (and some earbuds). I like how my logo is somewhat obscured, peering out of the shadows.

I just don't like how flat the bottom of the heatblock is:

[Oct 21] - I've opted to go with a sharp circle for all inputs/output. I feel it will have more pronounced deformation.

I'm also trying a new variable fillet across the bottom to improve aesthetics:

[Oct 27]

I've added a cutout so that the allen key has uninterrupted access to the bolts, requiring a slight aesthetics adjustment, and I've made use of the internal space to add a 5-way intersection so that each collection of 4 inputs can change their resultant mix relatively quickly.

I'm also testing to see if there's a step-increase in PCBWay's delivery cost if the shortest bounding box dimension is 30mm or more. The R1 and R2 orders cost sub-$8 to ship, which both were 29mm or less on an axis. R0 was 35mm at its thinnest and was around $16. R3 was 30.0mm, but I've reduced it to 29.9mm in hopes of reducing its $12 delivery fee.

In Marlin, along with the standard way of retraction, there's a feature that will retract all materials instead of just the active one. I want to try this out in the future since, to depressurise the heatblock, each channel will only have to retract 1/8th as much as a single channel. For example, in perfect conditions, retracting 8mm from one channel would be equivalent to retracting 1mm from all 8.

However, reality is likely different. The length between the bowden tube means that the tip of the filament might not even move until the extruder retracts 2mm, for instance. For TPU, this is likely more.

Perhaps the firmware retract feature would be a good thing to look into. I've never used it before, most likely because Cura doesn't natively support it, and it just seems that it merely moves the already-basic slicer settings into the firmware without provisions for multi material situations.

Another thing to consider is Linear Advance. I haven't gotten around to testing it out, but I assume that it only works on the active channel. I'm wondering if the other channels could be used to assist in some way.

Ideally, there'd be a per-extruder toggle whereby some won't retract and others will instead, to make up for it (e.g. retracting all but TPU channels). This should mean that both retractions and LA can react faster to flow rate changes.

The main edge case I need to worry about is one or more channels slowly ejecting themselves, i.e. a few retractions or LA corners later, the filament in a channel slowly moves backwards, -0.3mm here, -0.9mm there, and accumulates to -30mm.

So, instead of trying to print ABS-core PETG to test before buying and trying a stiffer 64D TPU (compared to the presumably 95A TPU I have now), I've been wanting to implement some improvements into CAD:

• single-channel flow rate
• heater cartridge choice and mounting
• sealing
• post-processing without expensive machinery

I expect that this TPU would be most ideal to use as glue/core, but part of the reason for the higher melt time is ideally to make standard TPU more likely to print successfully. Additionally, I suspect that my flow rate is still limited at around 10mm/s/channel, which is not enough for my usual layer height at any reasonable speed, limited to under 40mm/s.

To increase "yields" of clamp plates (i.e. a plate where all M6 threads are perpendicular), as well as give space for the heatbreak tightening spanner, I've moved the inputs to 2 squares of 4. This way, I only need half as many threads per plate. It's better to have to redo the plate if the 3rd thread was bent instead of the 7th. I also hope this would make alignment to the couplers easier.

The plan is to then heat this with a 30mm long, 160W cartridge. Hopefully, I don't need to use a 47mm, 320W cartridge. I'll use a clamp design, which should also allow 20mm cartridges to be used.

As expected, the immediate 90-degree bend doesn't fair well when it comes to flow rate. As seen in the topmost image, my plan was to have a straight path to the bottom, then go back upwards with the coaxialiser before outputting to the nozzle. Additionally, I had recently seen the below video on tests done with multiple 2D path geometries:

He obtained good results with a straight-length distance of 11mm before the split. He also tried a 4mm length before the split and flow rate was worse than no split at all. The split distance in my design is 13mm. The drawback is that the high-flow columns alone will store 1250mm^3 (2/3rds total) of material in them. The R2 has less than 1000mm^3 entirely.

My only slight concern is the output directly after the coaxialiser since it's sloped such that it could become too steep an overhang when the part is printed at a 45 degree angle:

There was also a noteworthy comment left in the video above:

I think there is a flawed assumption by the community that smooth bore=better flow, that seems so obvious that nobody questions it, so nobody tries anything else.

I did an experiment polishing a 100mm long bore [...]. Instead of being shiny and polished, it was like the "orange peel" of a bad automotive paint job. I tested it anyway, and the results were a 30% improvement over any previous test!

Since then I have tried several similar strategies to intentionally "ruin" my barrels: scouring with stainless wire, running a tap into the bore, running a dull drill bit through it, etc.. All the tests either show an improvement or no improvement over smooth bore, but none perform worse.

I have several theories about why, but my two favorite are that the rough surface creates turbulent flow, mixing hotter fluid at the perimeter with cooler fluid in the center, and that a rough surface is more surface, and more surface = more heat transfer.

Thus, I don't think I need to concern myself about the surface roughness of DLMS parts, and just need to ensure that overhangs don't exceed 45 degrees.

I also like that this design is more accepting of arbitrary input amounts. The only thing stopping someone from a Coaxielveor (12-in-1-out) or Coaxihexor (16-in-1-out, not to be confused with a Coaxi6or) are other things like software / firmware / hardware support.

[Oct 12]

I decided to take a loft straight from the coaxialiser to the output pipe, which looks cool and seems printable. I've then simulated it and it takes 4.5s for the output to transition.

The coaxialised output conforms to the cross section. The below images show the output just as it leaves the coaxialiser (left), halfway up the loft (middle) and the end of the loft (right):

I notice that the midpoint is notably rounder than the simulated output of R2:

Thus, I've modelled an octagon output instead: