Input Shaper graph comparisons between EBB36 and Nozzle based ADXL probes

[noble night]

I'd heard many people talk about getting better quality input shaper measurements by mounting the sensor as close to the nozzle tip as possible. I have an EBB36 CAN toolhead board with an ADXL on it, and also picked up a Provok3 ADXL sensor that mounts on the end of the hot-end.

I ran back-to-back tests, and thought I'd share my graphs and my interpretation of them with you all. Please feel free to challenge my assumptions, I don't know much about these graphs, and am always happy to learn more!

[noble night]

Here are the X and Y graphs - The first is measured using the EBB36, the second at the hot-end:

The charts...

From what I can tell, it would appear that the nozzle based approach seems to pick up some additional resonance that the EBB36 didn't, however the peak resonant frequencies are largely the same when measured at both points.

The Y recommendations were the same

As for X - it looks to be detecting some additional resonance in the higher frequencies causing it to recommend 2Hump

I've got a CPAP coming in towards the end of the month, at which time I'll get rid of my dual 5015s, and can rebuild my EBB36 mount - I hear tons of vibrations from those fans so I'm hopeful that eliminating them will help reduce some resonance.

Time to run some test prints and see if the new recommendations for X result in better prints!

[fleet hemlock]

they do seem pretty similar

are you willing to run a test print with the recommendations from both as well?

that would give some more insight whether those extra frequencies detected are relevant

[noble night]

Good idea. Maybe I'll print the manual input shaper tower, unless you have thoughts on a better model?

[noble night]

Dare I say my first ghosting test has come out purrrfect. 🤔

Lol

[fleet hemlock]

the manual one seems like a good idea

[noble night]

Don't like my ghost cat? :p

[fleet hemlock]

I do but something where it's more obvious would be better tp compare

[noble night]

Clearly my 'dad' jokes aren't working well over discord. 😆the print is a cat in a ghost costume... and we're talking about a ghosting test, get it?! 😂

I crack myself up..

Ok, I'll get down to business with an actual test... 😂

forgive me 🙂

[glossy cobalt]

Hey there. It's nice to see someone taking tests seriously 😀
First, I don’t consider myself much of an expert on the topic, but I could observe something quite obvious - the power spectral density. In both axes, one can’t miss the resemblance of the graph, but it is as if the resonance is dampened on the measurement done by the EBB compared to the measurement taken at the nozzle.

Otherwise, judging by the way your printer and prints look, I am convinced that you have a very well-built and properly tuned machine, making the difference between these measurements somewhat of an “extra 5% bonus in quality” and not earth-shattering difference as if it would be for a machine that never had IS calibration.

[knotty dagger]

Where did you get that STL? I need that ghost cat in my life 😺

[noble night]

https://www.printables.com/model/298644-cute-cat-ghost-remix-hollow

[knotty dagger]

❤️

[noble night]

I ran a few tests, photos forthcoming (when it's daylight).

• 6k Accel - 400mm/sec perimeters PA / No PA
• 10k Accel - 400mm/sec perimeters PA / No PA

For both Input Shaper profiles (EBB36 and Nozzle)

I also ran some follow up tests:

• 6K / 10K Accel - 400mm/sec perimeters No PA / PA
• 6K / 10K Accel - 400mm/sec perimeter - Nozzle ADXL config - IS Damping Ratio - No IS Damping Ratio

My takeaway so far is - if you ever get to meet a developer of the Input Shaping or Pressure Advance algorithms - you should probably give them a ~~high-five ~~ high-ten. That software is magic good applied math!

[fleet hemlock]

forgive me as I was completely oblivious to the joke

[fleet hemlock]

waiting on photos

[noble night]

I'm finding it quite difficult to capture the ringing...

[fleet hemlock]

well that's actually good

[noble night]

well, here are some pics

I ran a few tests - EBB36 ADXL recommendations, Provok3 ADXL recommendations, and Provok3 ADXL recommendations with no input shaper smoothing... For each I ran 2 iterations 400mm/sec @ 6k accel, and 400mm/sec @ 10k accel

For reference, from left to right, the tests are

• EBB 400mm/sec 6k accel

• EBB 400mm/sec 10k accel

• Provok3 400mm/sec 6k accel

• Provok3 400mm/sec 10k accel

• Provok3 400mm/sec 6k accel - no smoothing

• Provok3 400mm/sec 10k accel - no smoothing

At 6k I was only hitting 400mm/sec very briefly - at 10k, I was hitting 400mm/sec on the longer straight sections

here are a few more angles - I kept the test pieces in the same order as above

There were some cooling issues that showed up in the feautre with the small gap, especially at 10k accel - this was PLA and I didn't have the fan on

I'm not too sure what to make of it - ghosting seemed very similar and very light across all tests, I can see a little bit on the parts printed at 10k accel

[noble night]

Now I'm starting to think I need to revisit this entire thing... 😂 given that the gap is so wide, I'm trading accels/speed at the expense of excessive smoothing.

Maybe I should run another test at 3k accel...

[noble night]

Meh, I think i boned this entire test - I hadn't read the Resonance Compensation test documentation in a long while https://www.klipper3d.org/Resonance_Compensation.html

my SQV was not set to 5 as recommended, and I had pressure advance enabled

the SQV has likely contributed to the increased gap and smoothing... I'm going to run a few test again!

[noble night]

OK - I've started over, here's the new testing methodology...

1. Print the entire Ringing Tower - Start at 3k accel, increase accel 1k, test ends at 14k
2. Set Square Corner Velocity to 5
3. Set Accel to Decel value to 16k

I've also had to essentially disable Acceleration control in Orca by setting Normal Printing Acceleration to 0

The models should show a few things - the affect Input Shaper has on reducing ringing/ghosting, and smoothing.

[noble night]

These are with input shaper disabled. You can see ghosting increase as velocity increases...

You can also see that the gap in the first picture remains consistent as velocity increases. I think that's because there's no smoothing imparted by input shaper because it's disabled.

With recommendations from the EBB36 adxl readings. you can see the ghostiong is gone, but you can also see the effects of smoothing as velocity increases. first bump on the right starts at 3k and increases by 1k

[noble night]

recommend values from nozzle based adxl. Ghosting is eliminated, smoothing effects are similar to EBB36, maybe a little more evident at similar accels

[glossy cobalt]

I love your dedication for testing the topic! Hat off to you! Beautiful looking test pieces, too!
So, summarize the verdict? Toolhead board ADXL - yay or nay?

[noble night]

Thanks! To me it seems to be a tie, I'm sure the values measured at the nozzle are more 'accurate' but the end results in terms of removing ghosting seems to be roughly equivalent (at least to my eyes)

[glossy cobalt]

Coming from you, with your high standards, I'll take that as "Toolhead board ADXL probe is viable".

[noble night]

I think so. ☺️

I'm still learning for sure

[glossy cobalt]

Very well 😀

[noble night]

There may be better models that highlight ghosting better

I was also interested in seeing the effects of smoothing, that's definitely something that can be tuned for

[glossy cobalt]

I don't know anybody who will do a head-to-head comparison, just for the sake of science 🙂
Plus, you didn't even do it for likes and channel traffic, so... super awesome 😎

[noble night]

Here are the actual recommendations (pretty similar):

## PROVOK3
[input_shaper]
shaper_freq_x: 63.8
shaper_type_x: mzv
damping_ratio_x: 0.075
shaper_freq_y: 44.4
shaper_type_y: mzv
damping_ratio_y: 0.0585

# EBB36
shaper_freq_x: 62.2
shaper_type_x: mzv
damping_ratio_x: 0.1114
shaper_freq_y: 41.6
shaper_type_y: mzv
damping_ratio_y: 0.0713

[glossy cobalt]

Uh, looking back at the graphs you posted and the input shaper recommended values, I think your case is very pretty much an edge case where the two measurements are similar.
I did some tests on different machines with different ADXL setups and the graphs were amazingly different. Admittedly, I never actually made any test prints...

The major conclusion I have is that your machine is very well-built, and there's not much you can do (unless you try very hard) to get bad looking prints 😂

[noble night]

😂

I was wondering too, how much this just applies to my machine and how much is universal... in the end I suppose your individual milage may vary! 😜

I'll post up the latest input shaper graphs...

[glossy cobalt]

...like almost everything 🙂

[noble night]

I just moved over to CPAP, so these graphs will reflect that, if anyone was curious to see if CPAP has less resonance that dual 5015 fans shaking around

[fleet hemlock]

yes

[noble night]

First, lunch 😆

[fleet hemlock]

whenever you can, I just had dinner

[noble night]

EBB36 Charts:

PROVOK3 - Nozzle based ADXL Charts:

Bonus charts! PROVOK3 - but increased belt tension by 20mhz - was 110mhz went to 130mhz:

These charts are measured after converting to CPAP - original charts above are with dual 51015 fans for cooling

[glossy cobalt]

Looking good.
On the topic of cooling, have any comparison of prints done with CPAP compared to those made with SCS?

[noble night]

well, funny you mention it - yes, indeed I do! 🙂

was going to write up another thread on that

I did a speed boat on dual 5015s, 5015s + SCS, and CPAP

[glossy cobalt]

Looking forward to see the results!

https://tenor.com/view/popcorn-entertainment-gif-9904298

[noble night]

I'm still working on the CPAP cooling amount - it's hard to push the cooling and keep bed adhesion! :p

[glossy cobalt]

Wait, you thought you can solve all the problems and not create new ones?

[noble night]

lol - for reals

not to mention I totally jacked my P-Series build plate, it's ruined - forgot to save my z-offset at the end of the night, sent a print the next day and totally killed it - just glad it didn't take my nozzle with it :/

that p-series has some significant adhesion power

[glossy cobalt]

😲

That's how I ruined an almost new wham bam plate, long time ago. Not buying expensive plates anymore!

[noble night]

I hear you... I've instituted a new policy - whenever working on toolhead - use the old build plate until everything is dialed in 🤦‍♂️

[knotty dagger]

That's a pretty good policy

[noble night]

I did order another p-series, but until then I now need to bump build plate temps by minimum 10c to offset CPAP cooling :0

[knotty dagger]

I've been using my OG ender 5 plate thingie when I do tinkering

It wasn't great ever since I sucker-punched a hot nozzle into it at full speed 😄

[noble night]

gah! and been there!

[noble night]

Lol, great minds think alike @reef whale - nice video!
https://youtu.be/ufmUmeDnr0Y?si=HW_rxWjiBSjjACVV

I'd be curious to see your results printing the tuning tower with all 3 recommendations. See if you can spot the difference IRL.

In my tests of the ebb36 vs provok3 it was pretty inconclusive

[undone oxide]

sorry to be crass but im starting to think its more fuck around and find out then chasing numbers and nice lines on graphs