Here's the three versions I when changing my bed.g for my delta printer. They use 32 points and it generates a column that I can cut and paste into my bed.g It works for me, my calibration runs in the 0.01 mm range. Feel free to make a copy and do whatever you want with it, I'm just sharing in hopes of it being helpful to someone. Please watch the printer carefully until you're confident it's working for you, while it works for me I'm just throwing it out there to help and don't make any claims to it working flawlessly although I can't see how it'd go bad if you use values reasonable to your machine.

Variables to change are:
-bed size (the outer ring of probe points)
-Calibration Factor
-Oversize factor (only on the oversized one where I use it to dial in my arm lengths)

The first one is my default arrangement
12 points at the bed size distance (perimeter), 12 more at 2/3 the bed size distance, 7-points at 1/3 and one for the center.
https://docs.google.com/spreadsheets/d/1trfFyw2LqvRUeq3WPKtEaKaRGZOVwk4034zErVPo3jE/edit?usp=sharing

The second arrangement I get better calibration numbers from
16 points at the perimeter, 12 inside (2/3 distance), and 3 points (1/3 distance) and a center point
https://docs.google.com/spreadsheets/d/17S-CsZrUSqeNFooCKBznFLhmKTVUeivGca3umMWm86s/edit?usp=sharing

The third is what I use to test / calibrate my arm length. I calculate the oversize factor as the distance to just inside the maximum bed radius over the bed size parameter that I can test within the limits of the towers. Please be careful and watch closely if you use this one.

https://docs.google.com/spreadsheets/d/172jtcU9UQyQi1yUPvm9tagQ7Fl1T8Yxcv_s8HN2PATI/edit?usp=sharing

I will try to update if there is a major flaw, otherwise please treat it as a starting point for you play with generating your own test points.

depending on the repeatability you want, you may want to slow the feedrate which is really because of the worst part of inductive probes. Most inductive probes switch at 500hz until you you break the ~$60 retail price point for industrial sensor to get 1khz to 3 khz ( next price point is like ~$150+ and specs can get to 5khz). So the faster you're probing the larger the distance variation between trigger intervals. i use a ie5390 i grabbed off of ebay for like $30 and it operates at 1khz and i go with 60mm\min to get .003mm repeatability which is only a microstep and it levels (delta printer) to .01mm over a 300mm probe area

You can get the mathematical differences with one of the calculators ( i like the openscad simulator). The optimal arm spacing is dependent on your print diameter, but you also have a trade off with printable area since you'll have to move the balls further from the center of the effector. Since you can't really go further than having the joints be over lapping, that's kind of the concept to ride that optimal point at least math wise. As for how it works out in practice, not clue but there's so many little variables in the whole system that I don't see it helping that much when there's enough knowledge about the normal configuration you can easily tease out what is probably off when your troubleshooting.

@Danal I use the same Motors on my Rostock max for years now and they've held up quite well for the abuse I did on them. I have looked for lower inductance / inertia 0.9 steppers with good torque but I haven't found one I'm confident will perform better for a decent price (I'm looking at you moons' industries...)

If you get a chance, test out my settings I scope tuned for the motors. Mine run 1500ma without getting hot and are as silent as I can expect of them.

https://forum.duet3d.com/topic/13028/my-tuned-tmc2660-settings/2

Following the trinamic tmc2660 manual and spreadcycle guide I busted out my o-scope with a current probe and a helping hands on the sense resistor. The motors I tuned were Wantai 42BYGHM809 and a PG35L-048. I run a delta, and my steppers are running at 1.5 out of the specced 1.7A and they are staying significantly cooler than when I tuned by ear and they are behaving much better for me, it even got my past the .02 bed leveling barrier I was at.

Anyways, thought I'd share my efforts in case it might help someone.

Wantai 42BYGHM809
M569 P0 D2 F2 B2 Y01:09:02

PG35L-048
M569 P3 D2 F3 B3 Y03:07:02

I prefer the contactless approach and just wire in (directly with no extras has worked fine for me) A3144 digital hall effect sensors (~$1 ea.) for my endstops and slap a magnet on the moving part. You can home fast and do a slow back and return and the repeatability has been hands down better over time than contact switches I've tried.

I finally cleared this up on my E3d titan, had moire patterns like the second picture first post. I eliminated the software/ electronics ( changed stepper driver settings, each value individually although it was on the fly during 10mm test towers, and even purposely bad ones), motor current, microstepping high and low, thegolden ratio layer heights, extrusion/flow rate,slicers, motor cables. It confirmed what I suspected in my case which was that it was hardware and top of the list was the titan itself.

I don’t know what fixed it specifically since at that point I wasn’t doing that many disassemblies. I reprinted the motor mount and sanded to try to keep things perpendicular. I drilled out the the old titan cover plate and added a copper sleeve to the motor shaft intending to constrain shift alignment and any deflection from the tensioner. I had lots of spare parts since I swapped everything on the titan but the main body. When arranging the gears, I adjusted the delrin gear on the hobb so the tensioner arm was as close to perfect feeding filament into the center of the hobb. When I was doing that I noticed the both delrin gears had slight issues, one didn't sit perpendicular on the hobb when looked at with an engineer square ( I used it a on anything I could to see who my culprits could be for not being perpendicular when they should) and the other was not truly flat when viewed against the straight edge. Fixed one up and put it in and noticed when I was trying to get the gears flush to a razors edge (literally) that the hobb gear wobbles weird in it's bearing. So I swapped out the hobb's bearing in the body of the titan, (the one in the cover was checked too). I also added so kapton tape where the v6 mounts into so there is zero play even with effort. I finished putting it back together and one or a combination of things fixed it. Going to a hemera was my next move if I didn't get rid of it here, if only for the on motor mount points for my probe and and flex filament performance.

My gut says based on the pattern, it was likely the fact that the gears were not perfectly in line (seriously, I couldn't tell other than the bearing until I got out my good square) and wobbling during rotation either due to the motor shaft not being perpendicular for the metal gear, the delrin gear not being truly flat / crooked on the hobb, or less than stable bearing the hobb gear rode in. With the hemera you have a few parts you out together that could be the source, and it makes it even harder to identify since a few critical ones are small and short so perpendicularity of shaft to gear and round and centered of the rotating parts will Tedious. Still you could get by simply with A printed jig and dial indicator along with some squinting at the part on a engineer / machinist square or at least straight edge in front of a light. .

"I am guessing 1.3V is still too high for it to be considered as low?"
@likevvii

This is true, I went through a process to hand select a protection diode out of a set for my induction probe to get the off down to 1.1V for a constant responsive low reading (Duet Wifi V1.2)

@fcwilt I got a few of the 10 to 12 mm ones off of amazon and they did fine, until I went down the number chasing rabbit hole. The LJ8/10/12's are all 500hz switching, so you can expect them to vary about +/- 0.033 mm in their response (not including things like microstepping) if you run the feedrate at 1000mm/min. With that though, you'll at least know when it's the switching time or mechanical play which is perfectly usable and still better than I've gotten out of physical switches. For physical switchs, oddly the best repeatability I tested was actually a switch I pulled from a spare sanwa arcade button I had, much better than a limit switch although definitely not as rugged. Makes me want to test a cherry MX red mechanical keyboard switch as a probe when I think about it... I think playing with delta's for 7 years is starting to show....

note: the field coming off the induction probes are typically two lobes (oblong spheres) which isn't labeled in any way. The detection pattern for most of the useful distance would plot in an oval or even a peanut shape, which isn't an issue for their typical use (and why the detection object is usually specified as at least the diameter of the probe). This means the probes will be more sensitive to tilt in some arbitrary axis (well, where ever the center line between the two lobes is) compared to a corresponding perpendicular axis. TL;DR being do you best to mount it as perpendicular to the bed as possible it'll carry the tilt into the measurements like any other probe (and if you're a number chaser like me, incorporate a locking ball joint in the mounting).

You said it's directly powered from the power supply, which makes me picture it hard wired to the +/- rails of the supply and just one wire to the Duet connector. My guess would be that it's likely caused by a weird signal level difference without the directly common ground level leading to either a distorted PWM waveform (if your PWM isn't nice and crisp 0.4 to 5v signal you get "erroneous fan behavior") and the other is just reference level differences so the fan just isn't reading it right, that one is easy to test with a multimeter. Just double check that the ground/neg of the fan's power to the PWM voltage should be 5v when your running at 100% and under 0.4V at 0% which is the standard noctua fans run.

It could clear up with connecting the GND and PWM to the Duet and just wiring the power wire to the 24V supply, which solves either problem but does open your duet to damage if the fan ever shorts out (I've never seen them fail this way and it'd be really unlikely, but that doesn't mean it doesn't happen). That should let you run it down to 20% speed before shutoff.

Pretty much that. It's still an open loop system at it's heart so why not if it makes thing better. Under ideal conditions sure it'll be the same and it's worth a best effort attempt to make it so, but at certain point you gotta make a call how far down the rabbit hole you wanna go. Throw on closed loop servo-like stepper control boards? Characterize a bunch of stepper motors and their performance so all three axis match? There are lots of timing belt alternatives too if you want to go industrial solution (ie, priced for a business). One sanity check i would suggest though is to measure carriage movement per tower without arms attached. That would at least let you isolate whether the discrepancy is originating in the towers or the arms and effector.

Thanks, I'll try again with some of the cloth covered wire and see if that won't get stuck. I used jig's and spring tube benders which worked ok for the aluminum but that brass I tried before was just determined to fold in on itself.

@Surgikill I'm curious what size brass tubing you use? I've tried using small hobby brass tubing and even packing the tube with salt powder it would still collapse in on itself in bad ways when i'd try to bend it. I'd much rather use brass than aluminum since the brass would deform without tearing/ breaking as easy as the aluminum, and the akuminum gets more break prone after exlosure to hotend buuut i have not been able to shape it without issues.

I'm about to dive into the details soon, I just want to confirm that its possible to brute force through iterative changes to individual delta arm lengths with 8 factor calibrations results. I figure i could let something like this run and come back to find improvements until it hits a local minima in results.

When I think about it, I think the last time I cleared up the ripple as you see in the height map was actually solved by swapping my magball arms around. Accumulated errors between carriage + arms+ effector made measuring arms with magballs in them unattached to the actual build a trap, I had to measure from the rail to the ball end, then from ball end to ball end, then put the arms in the right position to get most uniform pairs on a tower as well as best average length. For my magball arms (360mm haydn huntley magball arms), what threw me off was the effective pivot point. For my set at least, If I measure the arm with a 9.5mm ball stud on each end from the edge of a ball to the other edge of the ball, the firmware measurement is that distance minus 7 (plus corrections for effector/carriage errors). Why 7? Because (once again for mine at least) the 9.5mm ball is inset about 3mm into the arm, and the effective pivot is 3mm further in from the edge of the arm, leaving 3.5mm from each ball end to subtract on each side. My desire for things to work neatly kept driving me to think it'd pivot in the center of the ball stud which is not the case. Figuring that out was so frustrating I think I want to forget it again.

@Veti It is, calibration only compensates for tower angles off of a perfect equilateral, not tower distance from center or individual arm length short/long. Shim the joints on the carriages of the two good towers to compensate for one that is inset, or shim the offending tower if it is too far away from center. As long as the lengths of the rods are equal per each tower (it twists the effector if not) you can compensate that manually with M665.

I'm there with you on being stumped by it. I finally got the last of my surface patterns gone (I've been chasing it down a rabbit hole for a LOOOONG time, but existed running under 45 mm/s print speed. I ended up swapping the bearings in a linear rail carriage a third time and it just disappeared completely woo woo!) I just called it good enough though on the height map after reaching -0.045 / 0.042 max deviation and .017 mm RMS error shown below. So I don't know how to get rid of the alternating pattern with my similar build.

If I was to guess though from the map, I'd say you're Y tower ballends+arms (basically the sum of everything from effector to carriage) are either a hair longer than the other towers or the tower is slightly inset in respect to a perfect equilateral triangle. I don't believe individual arm tower distances are handled by the auto-calibration yet, although you can manually adjust them with M665 ex: M665 L260.1:260.2:260.0. If you see how individual tower arm lengths affect your map by +/- .1 to an arm length and compare your map results if you really want to check that (I haven't figured out a good setup to do this measurement manually to that precision). If you find the offending tower you can just leave it compensated for software or shim it for a mechanical fix. Luckily tower angle issues though should be corrected by a 6 factor calibration.


edit: correction (the thought came in soon as I quit thinking about it) arm length differences can be handled in software, tower position from center needs to shimmed, still the process to determine the offending tower should be possible with small adjustments the individual arm lengths.

@zapta I second the sunon fans. They last, and come in high CFM or low CFM for quieter.

I personally think the best way is to put wire two 12V in series though more fans! ( Edit: OK ok, I was being facetious there and I'll clarify this is NOT the way to do it, but the responses did bring me a smile )

@DaveA As a C & C++ programmer from around 1976 I'd love to see the Kernighan and Ritchie style with { } syntax. I hate Python indention style.
Just my opinion since you asked.

What? But python style works great in editors like EMACS! Fancy IDE's highlighting { } groupings is for schmucks right?

You can get the mathematical differences with one of the calculators ( i like the openscad simulator). The optimal arm spacing is dependent on your print diameter, but you also have a trade off with printable area since you'll have to move the balls further from the center of the effector. Since you can't really go further than having the joints be over lapping, that's kind of the concept to ride that optimal point at least math wise. As for how it works out in practice, not clue but there's so many little variables in the whole system that I don't see it helping that much when there's enough knowledge about the normal configuration you can easily tease out what is probably off when your troubleshooting.