An empirical insight into a major flaw of PINDA (v2) bed level probing… and why 7x7 makes things even worse  

RE: An empirical insight into a major flaw of PINDA (v2) bed level probing… and why 7x7 makes things even worse

@anachronist

I didn't do a deep read - the notorious [TL;DR] - plus I am still wondering where he is getting the numbers for his spreadsheets. A lot of time and effort for sure, but:  how does anyone else reproduce his results or even do a sketchbook methodology check?

The idea of interpolation is problematic because the single pass drift isn't linear and is almost small enough to be ignorable. All a person needs do is add a bit of bed level correction to the rear and it's taken care of. 

The time taken to measure more accurately as a factor in thermal induced error is within the range of measurement noise, too. So it's really not a factor. Sure, with copious samples, and unchanging ambient conditions, you can ferret out the changes, but it won't be something you'd see on a daily basis (and it is easily fixed with BLC).  It's the old "This new 7x7 5 sample cal takes too long! It's a minute more than the 3x3 was!" Then the user waits 12 hours for his print to finish.

RE: An empirical insight into a major flaw of PINDA (v2) bed level probing… and why 7x7 makes things even worse

Posted by: @anachronist

I hope this gets the attention of Prusa.

I find it curious that your colleagues at work got frustrated with the first layer calibration to the point where they didn't want to use the printer anymore. My experience hasn't been like that. I bought the Prusa i3 MK3S because I wanted to spend more time printing with it than tinkering with it, and I feel it was a good purchase for that purpose.

One question I have: While the printer is warming up to prepare for printing, I like to have the print head raised fairly high up so I can wipe the nozzle clean when it's hot, before the print head starts moving. From your work, I suspect this may be harming my bed leveling calibration because the PINDA probe is far away from the bed during warm-up, and the probe starts warming up while it's doing the calibration. Would it be better to have it down near the bed while it's warming up, so the probe temperature could be somewhat more constant during calibration?

Hey first of all, wouldn't have expected this thread is still alive 😛

I personally let the PINDA heat up just above the bed via start G-code to 45°C, but not primarily to get a more constant temperature, but more in order to get the right height values (which works ok for our pinter btw). PINDA definitely heats up faster during a ABL run when it has a temperature around 25-30°C, but we're talking only 4°C max ime which seems not that concerning to me. So I think you could do it when you get good first layers with a cold PINDA 👍  This is my start G-code:

M83 ; extruder relative mode
G28 W ; CUSTOM! home all without mesh bed level
G0 X0.5 Y50 Z0.15 ; CUSTOM! good PINDA heating position
M104 S[first_layer_temperature] ; set extruder temp
M140 S[first_layer_bed_temperature] ; set bed temp
M190 S[first_layer_bed_temperature] ; wait for bed temp
M109 S[first_layer_temperature] ; wait for extruder temp
M860 S45 ; CUSTOM! wait until PINDA is >= 45°C
G28 W ; home all without mesh bed level
G80 ; mesh bed leveling

In the meantime, I got also my hands around two other MK3S with v2's and made some quick measurments with them, one of which showed that behaviour only ever so slightly while the other one had no obvious temeparture drift, not nearly as extreme as our PINDA v2. So it seems we somehow got a very sensitive one.

Didn't have the chance to measure a SPINDA yet, but I guess they will be good. Unfortunately they are always sold out when I check it.

RE: An empirical insight into a major flaw of PINDA (v2) bed level probing… and why 7x7 makes things even worse

@this_dude_jay

Many of us are using preheats for our PINDA's and have been for years. This isn't exactly new. lol.  And 45 is way too hot imo ... you'll be waiting a very long time for that preheat unless you are inside a heated chamber. I've waited over ten minutes just to reach 40 in summer. 

You still haven't answered where you are getting the data.

RE: An empirical insight into a major flaw of PINDA (v2) bed level probing… and why 7x7 makes things even worse

@tim-2

Procedure is described at the end of part 3. Temp. compensation is always turned off and values are not normalized, if I understand it correctly that you mean only "corrected" values are outputted. Compare temp. compensated runs in part 2 and it is clearly visible.

Compensation/interpolation is always indicated by odd numbers instead of ending with 0 for the last digit. 

The manual calibration of the temp. comp. values didn't work for me, as for reasons seen in part 2 & 3.

Of course, 7x7 is potentially improving things, but I don't get what you mean by red herring (from what?). If I get less interpolated values and replace them with actual measurments, thats defintely a benefit, but doesn't work when the outputted values get increasingly (and in my special case majorly) shifted and wrong within one sequence.

Part 3 TLDR: there is no simple temp -> offset correlation, there's a time element in it. This means values still shift within the ABL progression, albeit PINDA temperature being stable across the whole run (+-1°C, which was noticed for nearly all 35°C to 55°C runs).

If we got a PINDA which isn't very sensitive to temperature, this won't probably be noticeable (as I saw on the other two MK3S).

As for linearization, it will be always an approximation of natural behaviour, but is commonly used in engineering if the margin of error is acceptable. The only thing that matters is if the result is sufficiently covered (and the results in part 3 are more than promising in my case). Can you backup your statement that a single pass drift isn't linear (besides that obviously nothing in nature is perfectly linear)? As mentioned, it fits pretty nicely with our PINDA, just take a look at the outcome in part 3.

Why is 45°C is too hot? It takes for me ~5 min at constant room temperature of 23-26°C, 1-3 minutes longer on really humid days. I get decent first layers across the whole build plate, which I don't get with 35°C.

So why I chose 45°C? This is the temperature that comes closest to the results of my new compensation method and for me it kinda confirms, that these are the actual correct height values. Nevertheless, this is for me no more a workaround than a fundamental solution.

I personally guess my solution will not be regarded, since not all PINDAs have the need for it as they aren't as sensitive to temperature, and SPINDA is their latest fix to it. But users with PINDA v1 and freaky v2's like outs would benefit from it I guess 

RE: An empirical insight into a major flaw of PINDA (v2) bed level probing… and why 7x7 makes things even worse

I've been using 35C for my set temperature for mesh bed leveling with my PINDAv1. I haven't worried about cool-down as the bed cools quickly even on my hottest prints. I switched to 7x7 leveling when it became available. First layer consistency has been excellent. I've read through most of this thread, but I'm left with a feeling of "... and so what?"

1. On reading the dramatic subject line, I was expecting some finding that would make a major change in how we approach leveling, but I'm understanding that the discussion is more or less down to what target temperature to level at. Is that correct?
2. How much actual real-world impact does any of this make?
3. Is this an issue that will primarily affect the ever-shrinking number of PINDAv1 and some PINDAv2 users only?
4. How much do these dramatic new calculations vary from existing 35C PINDA warmup results? Are existing procedures close enough to not matter? So long as the sensing is done at roughly the same (e.g., 35C) temperature, does it matter?
5. Will implementing a new procedure significantly reduce start times or otherwise benefit the average user?
6. Looking at your CUSTOM! start gcode, it looks very similar to the original recommendations posted back in 2018.

I appreciate your enthusiasm and you've done a lot of work on this, but I'm not seeing an actual actionable finding.

RE: An empirical insight into a major flaw of PINDA (v2) bed level probing… and why 7x7 makes things even worse

@cwbullet

yeah it's absolutely odd that PINDAs seem to have such different behaviour, maybe we just had bad luck. Regarding the filament, I did not encounter any differences, for the most part I stick with Prusament PETG.

Concerning the enviromental factors I cannot give a valid statement, but with the PINDA heated up to 45°C I get pretty consistent first layers, which needs only a slight adjustment of 20-50 microns once in a while. Nothing major and the printer is definitely usable now 👍 

RE: An empirical insight into a major flaw of PINDA (v2) bed level probing… and why 7x7 makes things even worse

@this_dude_jay

 if the margin of error is acceptable

Not sure this is a correct assumption. What is most important in using interpolation is not accuracy - it is repeatability. They are different.  We don't care if the probe senses 1mm or 10mm- as long as it always repeats the same value : e.g.,  3.451mm +/- some small number. If the measurement is repeatable, we can do all sorts of good things with averaging, interpolation, offsetting, etc.

I heat my PINDA to 35 sometime 37. I never need to do L1 adjustments. I just heat, bed level, print. Been with the same number for Live-Z for a year...

As for why PINDAs so widely vary -- there is nothing saying Prusa isn't sourcing PINDAs from 10 different vendors. The cheapest this week is >>>>

RE: An empirical insight into a major flaw of PINDA (v2) bed level probing… and why 7x7 makes things even worse

Posted by: @bobstro

I've been using 35C for my set temperature for mesh bed leveling with my PINDAv1. I haven't worried about cool-down as the bed cools quickly even on my hottest prints. I switched to 7x7 leveling when it became available. First layer consistency has been excellent. I've read through most of this thread, but I'm left with a feeling of "... and so what?"

1. On reading the dramatic subject line, I was expecting some finding that would make a major change in how we approach leveling, but I'm understanding that the discussion is more or less down to what target temperature to level at. Is that correct?
2. How much actual real-world impact does any of this make?
3. Is this an issue that will primarily affect the ever-shrinking number of PINDAv1 and some PINDAv2 users only?
4. How much do these dramatic new calculations vary from existing 35C PINDA warmup results? Are existing procedures close enough to not matter? So long as the sensing is done at roughly the same (e.g., 35C) temperature, does it matter?
5. Will implementing a new procedure significantly reduce start times or otherwise benefit the average user?
6. Looking at your CUSTOM! start gcode, it looks very similar to the original recommendations posted back in 2018.

I appreciate your enthusiasm and you've done a lot of work on this, but I'm not seeing an actual actionable finding.

Hey @bobstro, thanks for your appreciation. It started all out of curiosity what's going on with our printer and as I got more into it, I thought why not share some of this stuff, maybe some find it worthwhile as well. If there would be any necessitiy I didn't know, but working as an engineer I guess I simply like trying to understand problems and get my head around possible solutions for an easy user experience 

1. yeah the headline is overly dramatic haha. But in essence that's what has been turned out for our printer. Current temp. compensation is rather pointless and Prusa knew that well as the dev states in the thread I linked in the introduction, so why not try to improve it. Prusa's answer to this issue is their latest SuperPINDA, but I tried to improve functionality with existing hardware and an closed loop approach.

I lately used it to pick a the right PINDA temperature, but that is not the actual point of this thread.

2. If there's no problem, none. If you have trouble with inconsistent and varying first layer heights, I'm convinced it provides a troublefree solution, especially for users who don't want to spend their time with researching and tinkering. I guess that's why Prusas are bought in the first place.

3. Hadn't got the chance yet to use a SuperPINDA, but yes I think so. Further, as far as I can tell, a lot of PINDAs aren't as temp. sensitive as ours seems to be, so I don't know how many would benefit from it. Good thing is, it wouldn't affect or mess up well working ones as far as I can tell.

4. For me it tunred out 45°C is the way to go, but no generalization from my side here. Everything that works for somebody is fine I would say.

5. Yes, in my eyes it would, both starting time and ease of use. No wait for preheating, no corner adjustments, no minding of any temperatures, no researching on how to approach anything. Simply press start and watch a nice first layer right away. That's what my company expected as many experience this convenience, but reality was different with our freaky PINDA. Might be the minority who struggle with it, but again, why not if it makes life easier for everyone.

To put it in simple terms, I think of it as a SuperPINDA upgrade without needing one, no matter how odd one's PINDA is behaving. At least that's what I tried to achieve with this thread, don't know of course if there are no issues somewhere ahead. But I'm confident it's a good approach 🙂 

6. yup, nothing new here from my side

Hope you see my point. The preheating thing is fine for now and I'm glad I can print decently at all with it, but waiting 5+ minutes before every start is nevertheless a little bit annoying. Can't wait to get a SPINDA when it is available.

By the way, may I ask how you figure actually out with your PINDA v1, when it has reached 35°C (since it has no thermistor inside)?

Cheers Julian

RE: An empirical insight into a major flaw of PINDA (v2) bed level probing… and why 7x7 makes things even worse

@this_dude_jay

No such thing as a PINDA V1 V2 V3 - that's jargon.  All Mk3's and Mk3S's came with a four wire PINDA that includes a thermistor. Mk2's came with a 3-wire uncompensated PINDA sans thermistor, and the Mk3S+ has the new compensated 3-wire model also sans thermistor.

RE: An empirical insight into a major flaw of PINDA (v2) bed level probing… and why 7x7 makes things even worse

Posted by: @tim-2

@this_dude_jay

 if the margin of error is acceptable

Not sure this is a correct assumption. What is most important in using interpolation is not accuracy - it is repeatability. They are different.  We don't care if the probe senses 1mm or 10mm- as long as it always repeats the same value : e.g.,  3.451mm +/- some small number. If the measurement is repeatable, we can do all sorts of good things with averaging, interpolation, offsetting, etc.

I heat my PINDA to 35 sometime 37. I never need to do L1 adjustments. I just heat, bed level, print. Been with the same number for Live-Z for a year...

 

As for why PINDAs so widely vary -- there is nothing saying Prusa isn't sourcing PINDAs from 10 different vendors. The cheapest this week is >>>>

 

Yeah you're right, repeatability is key. If there hadn't been any (which I guessed in the beginning), every approach would be useless. But I later recognized, that our PINDA is definitely consistent, only within its set of own rules I might never fully understand. If you take a look at the 4th last plot in Part 2 (falsely named part 1 🤐, the one with the thousand data plots), we can see that across more than 100 measurments over 4 days with varying weather conditions, all front left points have been within 30 microns, which astonished me. What differed was how the drift for the latter points took place, on one day it drifted more than on others. Nevertheless, everytime it turned out, that the rear of the plate was measured higher @35°C than @50°C whilst the front left point was poured in concrete figuratively speaking. That's something current compensation method couldn't take account of, that's why I came up with my solution.

Yeah I guess 35°C works well for a lot of people, that's great. I'm also quite happy with the preheating atm, over the time I adjusted it also only minorly. 

Haha that might be true. No luck for us 😬 

RE: An empirical insight into a major flaw of PINDA (v2) bed level probing… and why 7x7 makes things even worse

@tim-2

Ah thanks for the clarification, wasn't exactly sure what PINDA the MK3 had. I thought v1 was simply the one without the thermistor, v2 with one and the SuperPINDA now again without one since not needing it

RE: An empirical insight into a major flaw of PINDA (v2) bed level probing… and why 7x7 makes things even worse

Posted by: @this_dude_jay

@tim-2

Ah thanks for the clarification, wasn't exactly sure what PINDA the MK3 had. I thought v1 was simply the one without the thermistor, v2 with one and the SuperPINDA now again without one since not needing it

I don't know if Prusa ever issued a formal naming convention, but you can find a lot of confusing posts on the topic from 2018. The "V2.x" naming is probably the most correct. It doesn't help that a lot of clone PINDA probes were sold with vague naming schemes. There have definitely 3 variants of the Mk3 PINDA probe over time.

The PINDA warmup routine was originally for the earliest variant, but I and others find that overall warming of the printer is a good idea, so use the warming procedure for consistency regardless of PINDA flavor. Interestingly enough, I've been contacted by at least 2 SuperPINDA users asking for variations on the startup routine to pre-heat the bed to 65C in order to get consistent prints with the SuperPINDA. I can't test this myself.

I've got an early 2018 configuration with the early mushroom gray tipped PINDA (v2.1 I guess). Are all of yours consistent?

RE: An empirical insight into a major flaw of PINDA (v2) bed level probing… and why 7x7 makes things even worse

Posted by: @this_dude_jay

we can see that across more than 100 measurments over 4 days with varying weather conditions, all front left points have been within 30 microns, which astonished me.

And had you selected to normalize them all to start bed level at 35c, they would be even closer. And not just the front corner - all samples would be consistently better.

RE: An empirical insight into a major flaw of PINDA (v2) bed level probing… and why 7x7 makes things even worse

@bobstro

I'm not privy to pre-2019 lore ... friends have the Mk2 & 2S ... but we never talk bed level. I just go by the Prusa store and what little documentation they have on the variants. I wouldn't be surprised to find Prusa buys from the cheapest vendor, so there might be 20 varieties of sensors in circulation.

RE: An empirical insight into a major flaw of PINDA (v2) bed level probing… and why 7x7 makes things even worse

Posted by: @tim-2

[...] I'm not privy to pre-2019 lore ... friends have the Mk2 & 2S ... but we never talk bed level. I just go by the Prusa store and what little documentation they have on the variants.

As I recall, the store just showed "PINDA" with no notes on any differences between them. Unfortunately, the store link in the old thread is broken. You had to pay attention to the color of the tip to notice any difference between them. In hindsight, v1 for Mk2 family, V2 for Mk3 and SuperPINDA for the latest make sense. There was just a lot of uncertainty as to what to call the different flavors "back in the day". I've got my old mushroom gray version but a black-tipped version in the spares box. I've felt no need to change even if PINDA v2.2 technically didn't require it since warming the entire beast up at the start of a print is a good idea, especially on cold mornings.

RE: An empirical insight into a major flaw of PINDA (v2) bed level probing… and why 7x7 makes things even worse

@bobstro

Didn't I read somewhere the new SPINDA has a black tip?

https://help.prusa3d.com/en/article/p-i-n-d-a-superpinda-sensor-testing_2091/

Mentions V1, V2, and now Super...

But with Prusa also mentioning PINDA, MINDA, PINDA< and SPINDA all in the same breath... I just have to chuckle.

RE: An empirical insight into a major flaw of PINDA (v2) bed level probing… and why 7x7 makes things even worse

The new SPindas I have do have a black tip.  

RE: An empirical insight into a major flaw of PINDA (v2) bed level probing… and why 7x7 makes things even worse

@this_dude_jay

The bottom line is the Spinda is different and it does have superior performance.  I have been using one for a few weeks.  I was going to hold off installation until one crapped out.  I am glad I did not.  I am getting a better first layer with more consitance.  

RE: An empirical insight into a major flaw of PINDA (v2) bed level probing… and why 7x7 makes things even worse

One thing I have noticed is a tendency for my PLA prints to adhere with more difficulty to the top right quadrant than the rest of the sheet. I have one of those powder-coated textured steel sheets, though, and PLA doesn't stick well to it in the first place. However, this thread makes me wonder if I'm experiencing the effects of temperature changes in my PINDA (the gray-tipped one) as the 7x7 calibration progresses.

If I do the z calibration, I can look closely at the line printed in that quadrant and I can imagine that it isn't as squished as the rest (suggesting a calibration error), but that could easily be my imagination.

It's hard to say with that powder-coated steel sheet.

When a print comes loose from the top right quadrant, I know it's time to wash my sheet in warm soapy water. That fixes it for quite a while... the interval between washings has increased since I got the printer in 2019, I assume because my sheet is getting "seasoned" like a well-used cast-iron skillet (and I print only on one side of the sheet). Oddly, 99% isopropyl alcohol does nothing to improve adhesion in this event. Only warm soapy water does the trick.

I guess the best way to test whether the weaker adhesion in the top right quadrant is due to calibration, is to rotate the sheet 180 degrees and print with the sheet upside down and see if adhesion continues to be weak in the top right (suggesting the problem is the PINDA and not the coating on the sheet) although that redistributes the steel on the top and bottom edges, and I don't know if that messes with the PINDA sensing.

RE: An empirical insight into a major flaw of PINDA (v2) bed level probing… and why 7x7 makes things even worse

Posted by: @tim-2

Posted by: @this_dude_jay

we can see that across more than 100 measurments over 4 days with varying weather conditions, all front left points have been within 30 microns, which astonished me.

And had you selected to normalize them all to start bed level at 35c, they would be even closer. And not just the front corner - all samples would be consistently better.

What you mean eaxactly by that? Normalize to what? And why would it better the consistency?

All 35°C measurements experienced the 30 micron variance for the front left point as well, like with all other temperatures.