SKR Mini E3 V1.2 wrong temperature measurement

[Magoo0876]

Hi,

I've played a bit with the new SKR mini E3 V1.2 board and liked it so far, but something was strange. After some time I've measured the head and bed temperatures, and they were too low. My sensors are 104GT-2 on the bed and B3950 on the hotted .

bed set to 60 measured 50 hottend set to 200 measured 160

I've played with the pull-up resistor setting and found that 6k almost compensates the error. Also different sensor type settings did not do the expected.

So I think the internal pulldown resistors of the STM32F102 are activated and this would cause a higher temperature reading for the software. I also tried to activate the pull-up resistors for the thermistor input pin, but this is not a possible setting in the config.

Thank's for your help.

klippy.log

[KevinOConnor]

Most of the time we see reports of a temperature mismatch it is due to an incorrect thermistor selection in the config. Another common issue is measuring the temperature at a location that has a different temperature than the location of the thermistor. Finally it is common for there to be manufacturing differences in the thermistors and/or pull-up resistors.

The stm32f1 chips automatically disable pullups and pulldowns when a pin is placed in analog mode, so it seems unlikely the issue is at the chip level.

It's not difficult to build a custom temperature profile for the thermistors in your particular printer. This can be done using the QUERY_ADC command (see docs/G-Codes.md) along with a custom thermistor config section (see config/example-extras.cfg). In short, bring the temperature to a stable (but possibly inaccurate) temperature in Klipper, obtain the resistance using the QUERY_ADC command, and then obtain the actual temperature via an external probe. Repeat this three times and fill in the new [thermistor my_thermistor] section with the values.

If you have a multi-meter (or similar) it is also possible to measure the actual pullup resistance on the board. To do this, make sure everything is at room temperature, power down the micro-controller, disconnect the thermistor, measure the resistance of the thermistor with the multi-meter, reconnect the thermistor to the micro-controller, power up the micro-controller, and then while everything is still at room temperature use the QUERY_ADC command to obtain the measured resistance. If the value differs significantly from the multimeter's measurement, then one can modify the PULLUP parameter of the QUERY_ADC command until a match is found.

-Kevin

[Magoo0876]

Room temperature during the test 23.8°C

Recv: ok FIRMWARE_VERSION:v0.8.0-132-gc188aa98 FIRMWARE_NAME:Klipper Send: M105 Recv: ok B:26.2 /0.0 T0:26.3 /0.0

measured thermistor 3950 102k2 measured pullup 4k69 Send: QUERY_ADC name=extruder PULLUP=4700 Recv: // ADC object "extruder" has value 0.952808 (timestamp 28.990) Recv: // resistance 94893.790 (with 4700 pullup) Recv: ok B:26.2 /0.0 T0:26.3 /0.0

measured thermistor 104GT-2 102k6 measured pullup 4k69 Send: QUERY_ADC name=heater_bed PULLUP=4700 Recv: // ADC object "heater_bed" has value 0.952686 (timestamp 53.790) Recv: // resistance 94636.774 (with 4700 pullup)

I also tried the pullup setting in the config and ended up at 5250 to match at room temperature. 200°C set I got 165°C at the hotend measured with meter and flir camera. As you see above, there is a needle in the haystack, the values differ a lot. My last idea was that the ref for the dac is wrong but I measured 3,306V for the analog part.

Thank's Kevin

[Rubber1Duck]

Hi, i have a problem similar to this toppic on an SKR Mini E3 DIP, but in the other direction! Roomtemperatur 21,8 °C. Klipper shows up 16 °C both, Extuder and Bed! For Testing i installed Marlin 2.0 and then with same Hardware ( board und termistors) 21 °C both extruder and Bed is showing up! In Marlin i configured Sensor number 1 with ist the EPCOS 100K termistor in Klipper i configured EPCOS 100K B57560G104F (i tried all the other ones, but no one has the target of roomtemperatur)

Send: query_adc name=heater_bed pullup=4700 Recv: // ADC object "heater_bed" has value 0.969231 (timestamp 924.890) Recv: // resistance 148050.000 (with 4700 pullup) Recv: ok [...] Send: query_adc name=extruder pullup=4700 Recv: // ADC object "extruder" has value 0.969109 (timestamp 935.800) Recv: // resistance 147446.245 (with 4700 pullup) Recv: ok

Messurment termistor extruder 112k5, Bed 116k1 at the same temperature

whats going wrong there????

[Magoo0876]

Hi, I'm happy that I'm not allone with this problem. In the meantime I did the calibration for two sensor types connected on the SKR mini E3 1.2, as a workaround and got the following values.

[thermistor my_3950] temperature1: 23.6 resistance1: 95000 temperature2: 108 resistance2: 3750 temperature3: 198 resistance3: 390

[thermistor my_104GT2] temperature1: 23.6 resistance1: 95000 temperature2: 100 resistance2: 3750 temperature3: 200 resistance3: 300

Further strange is the shape of the thermistor temperature curve during cooldown. There is a noticeable signal noise under 80°C.

I've heated the bed to 70° and switched off the 24V power supply. Only the raspberry pi supplied 5V to the skr board, but still a lot of bends in the cooling curve.

[KevinOConnor]

Odd. If you change the uint32_t aticks = 3; on line 64 of src/stm32/adc.c to uint32_t aticks = 4; and then recompile and flash the micro-controller, does the behaviour change?

-Kevin

[Rubber1Duck]

Hi Kevin, i will test that right now. Best Regards Peter

[Rubber1Duck]

Hi Kevin, i made this change and now the temperatures are to high

the resulting resistanc is to high

Recv: ok FIRMWARE_VERSION:v0.8.0-150-g43dfb3c9-dirty FIRMWARE_NAME:Klipper Recv: ok [...] Send: query_adc name=extruder pullup=4700 Recv: // ADC object "extruder" has value 0.955311 (timestamp 333.600) Recv: // resistance 100472.131 (with 4700 pullup) Recv: ok [...] Send: query_adc name=heater_bed pullup=4700 Recv: // ADC object "heater_bed" has value 0.955311 (timestamp 378.190) Recv: // resistance 100472.131 (with 4700 pullup) Recv: ok [...] Send: query_adc name=heater_bed pullup=4700 Recv: // ADC object "heater_bed" has value 0.955311 (timestamp 389.590) Recv: // resistance 100472.131 (with 4700 pullup) Recv: ok Send: query_adc name=heater_bed pullup=4700 Recv: // ADC object "heater_bed" has value 0.955311 (timestamp 392.290) Recv: // resistance 100472.131 (with 4700 pullup) Recv: ok

but the resistance (and adc value) is now stable ! (as you can see above), bevor it is going up and down more than 1000 ohm ....... This is the result for a SKR E3 DIP !!!!!

[Rubber1Duck]

The Steps at heating und cooling down are still there! And again .... with Marlin Firmware the temeratures are right Best Reagards

[KevinOConnor]

Very odd. I don't see any issues with my "skr mini" board (though it's not in a printer).

Could you try the code on the work-stmadc-20191126 branch?

cd ~/klipper ; git fetch ; git checkout origin/work-stmadc-20191126 ; sudo service klipper stop ; make ; make flash ; sudo service klipper start

The resulting Klipper log file will need to be attached here from any tests (with this branch or any other branch).

Another thing to test would be disabling heater_bed in the config to see if it impacts the results of the extruder measurements.

-Kevin

[Rubber1Duck]

Here are we go.... Branch work-stmadc-20191126 Roomtemperature 22°C Steps at heating und cooling down are still there. After round about 8 Minutes colling down both temperatures stall at 32,9 °C

klippy.log

to disable the bed heater, i have to try later, because i'm busy for the next hours

Peter

[Magoo0876]

It's not possible to simply deaktivate the heater_bed in the config because you imediately get: No option 'sensor_type' in section: 'heater_bed' But as you see in my post 2 days ago, heating up the bed only gave the same effect.

Which clock should be set for the STM32F103RTC6 firmware. 8/12/16/internal ? I thought the clock should be 72Mhz for that board.

-Robert

[Rubber1Duck]

Robert, it is possible, you have to delete the whole [heater_bed] section! # are not enough. Roomtemperatur between 22 und 23 °C Printer shows up 18,5 °C (19°C at the Display)

There is one Step at heating up und manny steps at cooling down. for more than 3 Minutes the Temperatur shows up no change !

klippy.log

Made the same check with Marlin, Same Roomtemperatur 22°C Marlin shows up 21 at the Display. But there are the same Steps at heating und cooling down !?

I am very confused and completely at a loss!

[KevinOConnor]

I suspect this is a board level issue, and not a software issue. (Specifically, I suspect the thermistor circuit on the board isn't well designed and it is therefore not able to accurately measure voltages around 3V. Thus resulting in inaccurate low temperature readings.)

Do you have the power jumper set to USB or 12V power?

-Kevin

[Rubber1Duck]

Hi Kevin,

the SKR E3 DIP dont have this jumper! I'm afraid you're right, it looks like a board level issue, but the temperature curves look better with marlin, even if they have these steps , but they are not so pronounced, and the Temperatur at Roomlevel differs only 1-2 °C. The result with klipper could be better! Unfortunately, my programming skills with Python and C are more than bad, especially C is not my world which would be absolutely necessary for the firmware part. To make sure that this possible board error only applies to my or possibly all SKR E3 DIP, I ordered another SKR E3 DIP yesterday, that arrives on Saturday and then I will test it again. Because there are no further error messages in this direction.

Thanks for your great work and your support! cu Peter

[KevinOConnor]

FWIW, in tests of my "skr mini v1.1" board, it also seems to show poor precision at low temperature. I spent several hours on Wednesday trying to track it down. My conclusion was that it was a hardware issue. If Marlin is somehow "less broken" then I don't know why.

-Kevin

[Rubber1Duck]

FWIW, in tests of my "skr mini v1.1" board, it also seems to show poor precision at low temperature. I spent several hours on Wednesday trying to track it down. My conclusion was that it was a hardware issue. If Marlin is somehow "less broken" then I don't know why.

-Kevin

Hi Kevin, if you have the same on your Board its rely looks like a board issue! So I have decided to send the new E3 DIP back and make no further tests. For my opinion this issue will be closed.

cu Peter

[Magoo0876]

HI, I have tried to solve the problem by adding some capacitors to the 3.3V. But nothing really worked.

This is my fourth board now on my ender and hoped to stay with it a while. From my side we can close here, if there is no genius idea anymore.

Thank's for your help.

[Magoo0876]

I turned on my printer and could not trust my eyes. The temperature on the display showed 33° on he and 32 on the bed. Room was at 23°.

So I updated Klipper and compiled the firmware, flashed the stm32, but still 32° I thought my board gave up, but finally I tried to flash the original firmware.

So it is a firmware problem, maybe someone could look at the marlin 2.0 code to find their fix.

--Robert

[KevinOConnor]

Alas, someone with familiarity with the Marlin code will have to chime in. (If anyone can even point to the location of the low-level ADC code that Marlin uses on stm32f103, that would help.)

-Kevin

[Rubber1Duck]

i would say https://github.com/MarlinFirmware/Marlin/blob/2.0.x/Marlin/src/HAL/HAL_STM32F1/HAL.cpp Row 305 and following ....

[KevinOConnor]

i would say https://github.com/MarlinFirmware/Marlin/blob/2.0.x/Marlin/src/HAL/HAL_STM32F1/HAL.cpp Row 305 and following ....

That code invokes other code (like adc.calibrate()), but where does that code live?!

-Kevin

[Rubber1Duck]

Hi Kevin, i am realy not a expert in C and C++ ............. but for my understanding thats part of the framework-arduinoststm32-maple and is inserted with #include in https://github.com/MarlinFirmware/Marlin/blob/2.0.x/Marlin/src/HAL/HAL_STM32F1/HAL.cpp STM32ADC.h is located here https://github.com/rogerclarkmelbourne/Arduino_STM32/tree/master/STM32F1/libraries/STM32ADC/src

This library is pulled by Platformio witch is needed to compile Marlin 2.0 and up.

That is what i can provide with my no knowledge from C and C++ ..... maybe that helps cu Peter

[Kirinkun]

Want to chime in to confirm that I've got a similar issue with my SKR Mini E3 on the Ender 3 Pro.

To continue the conversation of the code,

adc.calibrate()

is then calling adc_calibrate from the lib-maple which comes from this class: (at lest for the STM32F1):

https://github.com/rogerclarkmelbourne/Arduino_STM32/blob/master/STM32F1/cores/maple/libmaple/adc_f1.c#L177

Hope that helps somewhat, Kirin

[cmason1969]

Just adding another one with this issue. SKR mini E3 1.2 on an Ender 3. I am seeing temps around 5 degrees higher for room temperature readings. Sorry I am too ignorant to add anything to the coding discussion but would help in testing if able.

[Magoo0876]

I've tried to survive with my own thermistor settings but the printer needs very long to start after heating and quits if you change the temperature during printing. Also the display shows strange hieroglyphs sometimes during printing like issue #2227. In the meantime I forced a friend to run klipper on the same board, to see if it behaves different on his board. I also ordered a Cheertha 1.1b board with the same cpu to compare that to.

Hope to make progress, keep you informed. --Robert

[Magoo0876]

In the meantime I could compare a known working boad with my toublemaker. What I've found you can see on the picture. So maybe Kevin is right that it is a real hardware problem. So please check if you can also find spots like that shown in the picture. Funny that the connecterd pins via the solder blob are also connected via the board,

Stay tuned, Robert

[Coffee0297]

we really soon need some fix for this. marlin works.... heres a pic of some with the mini E3.

temps are not suposed to do this, and theres a lot of ppl that wanna get klipper and holds back bc this..

The same issues are at all 5 boards this person have. so its not only some E3's its all.

-Tonny

[vladbabii]

The source of the image @Coffee0297 posted was me. This is on a skr mini 1.1

I have two temp sensors on a single bed

• blue one - analog pin 13 on mks gen 1.4
• red one - PA0 on skr mini 1.1

There is a 1500w, silicon heater under the bed. It's a 450x350x8 mm aluminium bed with glass on top. There's no way there's such a big temp fluctuation at 5 cm difference.

Even my wife (which does not work/uses 3d printers and it's not technical) noticed somethings wrong by the smell of the hotend - "it smelled funny".

@KevinOConnor maybe we should add a warning about using these boards for extruders.

[vladbabii]

During printing... and zooming in...

[vladbabii]

blue - 60.4 red - 52.3 target: 60 printer would keep trying to heat, wich would be dangerous if skr would not read the correct value soon - my bed probably can go above 150C

-- edit

Values close after 25 minutes

[KevinOConnor]

FWIW, I'm aware of three differences between the Klipper adc code and the STM32ADC code (that Marlin apparently uses). There is a slight difference in how the adc calibration is initiated, there is a slight difference in the analog sampling length, and the STM32ADC uses continuous sampling with DMA. The first two are trivial and I'll make those changes to Klipper when I get a chance (unlikely for a couple of weeks). None of these differences explain the poor accuracy however.

-Kevin

[KevinOConnor]

FYI, I made the corresponding changes to sample time (bf3fa979) and calibration (502e8372).

If you do want to test this, grab the latest code (see the FAQ for details), rebuild the micro-controller code, flash it to the board, and make sure to fully power-off the board for 30 seconds after flashing (to ensure any new calibration takes effect).

I'm not expecting much difference, but if you do run a test let us know what the results are (success or failure).

-Kevin

[Rubber1Duck]

I will test this a.s.a.p. and let you know what the results are .....

[mental405]

Has anyone tried hooking one of these up to a scope to see if there is any noise on the analog reference voltage pin?

I had an MKS Gen L that exhibited similar behavior at higher temperature due to voltage instability on the 5V rail. After much searching someone had mentioned putting some capacitors across a 5V and ground pin corrected the issue. I later hooked the board up to a scope and confirmed the noise was there.

I have also heard around the water cooler than there are a number of the first revision Mini E3's which supposedly had crappy regulators on them, but that was supposedly fixed recently with the new ones.

[Magoo0876]

I also tried to put additional caps to all power rails. With supply from raspberry only, no fans or other rf generators, see post from 2 Dec, no noticeable change on the temp measurement. But had no chance do set up a scope so far.

[KevinOConnor]

Does the latest code not improve results?

FWIW, one thing you could try is connecting "analog_gnd" to "gnd" on the boards. (That is, use a jumper wire to connect the "gnd" pin on the thermistor plug to a "gnd" pin elsewhere on the board, and then wire the thermistors between that ground and the sense pin on the thermistor plug.) It's not clear to me that the analog circuits on the btt boards is correct.

-Kevin

[brianrjones69]

I tried the latest. Mine seems to be 4-5 degrees high. But when I set the bed temp to 60 degrees it goes to 60 degrees, as measured by my fluke IR thermometer anyway. It did the same with the previous version also. I am more worried about the hot end temp though. It seems that I have needed to lower my temps but I have no idea of how to measure it. I haven't tried Marlin, but i also compared the code and didnt really see any difference, other than DMA'ing the adc. I did increase the 'ticks' to 7 and it seemed more stable.

I also connected the gnd together as you suggested but didnt see any change. It wasn't the greatest test though because I connected them while it was already powered, but no change to the values.

[KevinOConnor]

Okay thanks. Probably want the ground connected prior to power up as the goal would be to have the "analog ground" truly grounded as part of the adc calibration routines. That said, this is a "long shot" anyway, so not worth spending a lot of time on.

-Kevin

[nitej]

I’m experiencing the same issue with an skr e3 dip v1.0. I’ve tried the standard thermister ports as well as using another analog pin and the normal gnd (not analog gnd) with an external circuit (the 4.7k pullup and a 10uF cap) but the temp offset and staircase like curves are still present.

Does the stock marlin firmware have the same “lumpy” temperature curve below 100C? Some have mentioned that the stock btt shipped firmware gave correct room temp readings but I’m curious if the strange heating/cooling are a firmware or hardware issue at the ADC level in the stm32.

[Coffee0297]

i ahve talkd with InsanetyAutomation, and he says marlin oversamples takes 16 readings and averages. that is fixing the problem at them. and looking at the diagrams of the E3 there is a 3.3v direct to the thermistor circuitry and not a filtered 3.3 as on the 1.3 1.4 and so on. if the board shuld be done right one shuld have 3.3v from after the LV1 coil..

i hope somthing like an over sampler can be implemented together with a faster temp update as described in https://github.com/KevinOConnor/klipper/pull/2538#issuecomment-589521416 to make us able to get more presice temp readings and fix the E3 issue.

-Tonny

[KevinOConnor]

This issue is not related to oversampling nor pid loop frequency. Klipper implements 8x smoothing on each 300ms sample followed by 2s (or see config for smooth_time) smoothing on top of that. Previous reports indicate the issue can also appears on Marlin (eg, https://github.com/KevinOConnor/klipper/issues/2194#issuecomment-559254285 ). My understanding of the issue is that the adc sensor on these BTT boards quickly lose precision when trying to measure voltages above ~3V. Thus resulting in "step transitions".

-Kevin

[vladbabii]

@KevinOConnor so would using a logic level convertor to drop voltage from to 0-3 range ( or 0-2.8 ) might fix this? This would involve also adding a new mapping for the new range for the sensor, right ?

[KevinOConnor]

@vladbabii - I don't know.

-Kevin

[brianrjones69]

Kevin, I did a little more looking into this issue. It seems that the adc is a bit sensitive to source impedance and capacitance. BTT probably should have put an op amp between the thermistor and the adc inputs. That being said, it looks like the only way to get better data is to slow the conversion rate down, especially at higher voltages. I found that using a ticks value of 6 brings the temp down, at least at lower temp levels. It does not eliminate the steps, but its a more believable number at room temp. The good news is that at lower voltages (higher temps), it seems that the adc is able convert the value much more accurately. I assume it is because it can charge the sampling capacitor fully at the lower voltage. At this point I have no choice but to believe that the issue is hardware related.

Is there an easy way to get debug messages from the micro to the klipper.log? I would like to see the adc values.

Brian

[vladbabii]

How hard would it be to support this board (I2C) https://www.adafruit.com/product/1085 ? This way we could have as many analog inputs as we want easily and cheaply.

Instead of trying to fix a broken product (and I'm sure it's not the only one), supporting a cheap analog board via SPI/I2C would solve the issue and also allow more versatility in klipper long-term

I would be willing to pay for this feature.

[nitej]

Another approach (the one I am using) is to use a second mcu to handle the analog sensors and use the multi-mcu support built into klipper. The second mcu could be another printer board or anything else already supported (I’ve tried an extra arduino uno and a arduino Leonardo with an external resistor/cap circuit to form the voltage divider for the thermister). It should in theory just need to provide adc channels. The thermisters should be relatively slow signals compared to the stepper signals, so I don’t anticipate problems. Plus it’s already supported and breakout boards with a atmega32u4 are very inexpensive.

A bit of a bandaid fix but if the skr e3 has a hardware fault, the alternative is using a different board altogether (or live with the wonky analog behavior).

[KevinOConnor]

BTT probably should have put an op amp between the thermistor and the adc inputs.

FWIW, an op amp isn't needed. The simple voltage divider circuit works fine on many many boards. As near as I can tell, BTT has done something dubious on these boards. In particular, some of the schematics show a resistor between the adc pin and the divider which looks wrong to me (it increases impedance for no reason that I can see). Some of the boards show an inductor between gnd and analog gnd which also looks odd to me.

Is there an easy way to get debug messages from the micro to the klipper.log? I would like to see the adc values.

If you add output() calls to the micro-controller code, the messages should show up in the log. See docs/Code_Overview.md for details. You can also add logging.info() calls to the host code.

How hard would it be to support this board (I2C)

As @nitej reports, using a low-cost off-the-shelf micro-controller is far simpler and cheaper.

-kevin

[l-Kage-l]

Ok, so since i discovered this post i wanted to chime in with a solution of mine.

I tried updating Klipper and then re-flashing the new updated version, which broke Klipper temps. Marlin works well. Thing is, when i re-flashed the old one, the temps are working perfectly fine which is intriguing. Im going to post the ones i have working and the one that doesnt work for my SKR E3 Mini v1.3

WORKING:

firmware.zip

Temps are reading fine, just like Marlin.

NOT working:

klipper.zip

Temps are just as stated above, hovering around 32/33 degrees.

The version that doesnt work has the uint32_t aticks set to 6, as suggested in the comments above. I tried flashing a version with those set at 4 and at 6, and neither works. Hopefully this helps find the problem.

[brianrjones69]

Yep. Your working version is working for me too. Thanks.

[l-Kage-l]

No probs! Hopefully Kevin seems the message and figures what was the problem between versions.