Comparison between temperature sensors used in 3D printers

Hi everyone!

I've just started writing a blog post about temperature sensor used in 3D printers.
This first part is very introductive, part 2 will be more technical. Part 3 will be about our 500°C thermistor and general question we have regarding sensors.

I'd like to hear your opinion about it!

Here is the link: [dyzedesign.com]

---

DYZE DESIGN
Hotends, Extruders, Liquid Cooling and Accessories.

Quote
Dyze_Design
Hi everyone!

I've just started writing a blog post about temperature sensor used in 3D printers.
This first part is very introductive, part 2 will be more technical. Part 3 will be about our 500°C thermistor and general question we have regarding sensors.

I'd like to hear your opinion about it!

A nice article, but inaccurate in places:

- The electronics table is not right, because Duet electronics can use RTDs and thermocouples as well as thermistors. The new Duet WiFi has purpose-made 2-channel daughter boards to support a total of 4 RTDs and/or thermocouples. See [duet3d.com] and [duet3d.com] for details.

- You illustrate the classic thermocouple with a bare welded tip, which is not easy to use because it needs to be insulated form the hot end heater block. However, you can also buy insulated cartridge-type thermocouples, for example from E3D.

- You state the resolution for a thermocouple as 0.5C, and for an RTD as 1.2C. But the resolution depends entirely on what electronics you use to read it. I suspect you are assuming that an amplifier board is used and then the ADC in the 3D printer electronics is used to convert the amplifier output. A better solution is to use a purpose-made chip to read the thermocouple or RTD and send a digital readout to the main electronics. This is what the Duet and Duet WiFi do. The resolution obtained is 0.25C for a thermocouple, and better than 0.05C for a RTD. The quoted error for the conversion chips over the whole temperature range is 2C for the thermocouple up to 700C, and 0.5C for the RTD.

- An important factor you missed out in the table is accuracy without calibration. I know you are trying to sell thermistors, but IMO this is where thermocouples and RTDs really score (as well as handling higher temperatures than traditional thermistors). A thermistor has a tolerance on its resistance at 25C and on the resistance change with temperature (often summarised inaccurately as the B value). For example, the popular Semitec GT104 has a R25 tolerance of 3% and a B value tolerance of 2%. Thermocouples by their nature vary very little between samples, and RTDs are made to much tighter tolerances. It's not easy for most 3D printer users to check the temperature reading accuracy at high temperatures, and this is why I believe thermocouples and RTDs score over thermistors, especially in high temperature situations.

---

Large delta printer [miscsolutions.wordpress.com], E3D tool changer, Robotdigg SCARA printer, Crane Quad and Ormerod

Disclosure: I design Duet electronics and work on RepRapFirmware, [duet3d.com].

Quote
dc42
A nice article, but inaccurate in places:

Many thanks!

You've made great points and I thank you for sharing this information with us. I will take the time to improve this blog based on your comments. Below are my explanations regarding your remarks.

Quote
dc42
- The electronics table is not right, because Duet electronics can use RTDs and thermocouples as well as thermistors. The new Duet WiFi has purpose-made 2-channel daughter boards to support a total of 4 RTDs and/or thermocouples. See [duet3d.com] and [duet3d.com] for details.

I wanted to compare boards as they are, without add-on electronics. As cost is a factor for this blog post and in the following, I want to compare the boards alone.

Quote
dc42
- You illustrate the classic thermocouple with a bare welded tip, which is not easy to use because it needs to be insulated form the hot end heater block. However, you can also buy insulated cartridge-type thermocouples, for example from E3D.

Their new sensors are quite nice! I've seen them a few days ago. Most MakerBots clones uses a thermocouple nicely integrated inside a threaded housing or a tab (flat or bended sheet metal), these are other options. I wanted to keep these picture to avoid any copyrights actually.

Quote
dc42
- You state the resolution for a thermocouple as 0.5C, and for an RTD as 1.2C. But the resolution depends entirely on what electronics you use to read it. I suspect you are assuming that an amplifier board is used and then the ADC in the 3D printer electronics is used to convert the amplifier output. A better solution is to use a purpose-made chip to read the thermocouple or RTD and send a digital readout to the main electronics. This is what the Duet and Duet WiFi do. The resolution obtained is 0.25C for a thermocouple, and better than 0.05C for a RTD. The quoted error for the conversion chips over the whole temperature range is 2C for the thermocouple up to 700C, and 0.5C for the RTD.

As very few boards include a specialized IC (covered in my part 2 ) such as MAX31855, AD595, MAX6675 for thermocouple and MAX31865 for RTD, I wanted to compare based on a common microcontroller ADC resolution basis. By having a better ADC, as the DUET and most 32 bits boards have, each sensor benefits, rather than comparing the specialized IC with 14 bits or 15 bits integrated ADC.

Very few boards have integrated IC for other temperature sensors. Also, most boards have SPI or I2C available for any breakout board. This is also a reason why I didn't want to compare add-on boards.

Quote
dc42
- An important factor you missed out in the table is accuracy without calibration. I know you are trying to sell thermistors, but IMO this is where thermocouples and RTDs really score (as well as handling higher temperatures than traditional thermistors). A thermistor has a tolerance on its resistance at 25C and on the resistance change with temperature (often summarised inaccurately as the B value). For example, the popular Semitec GT104 has a R25 tolerance of 3% and a B value tolerance of 2%. Thermocouples by their nature vary very little between samples, and RTDs are made to much tighter tolerances. It's not easy for most 3D printer users to check the temperature reading accuracy at high temperatures, and this is why I believe thermocouples and RTDs score over thermistors, especially in high temperature situations.

You are right, thermistors aren't the most accurate between the three options. However, the repeatability is very good. As any filament require a lot of fine tuning, the key is to get the same temperature each time a target temperature is set. For scientific experimentation and laboratory measurements, I do think it is very important to get a reading without any offset. For 3D printing, I think repeatability is the key.

---

DYZE DESIGN
Hotends, Extruders, Liquid Cooling and Accessories.

Quote
Dyze_Design
You are right, thermistors aren't the most accurate between the three options. However, the repeatability is very good. As any filament require a lot of fine tuning, the key is to get the same temperature each time a target temperature is set. For scientific experimentation and laboratory measurements, I do think it is very important to get a reading without any offset. For 3D printing, I think repeatability is the key.

Good start but personally I'd like to see the table include the 32 bit boards separately, as it stands it tells the Newbie that a Duet board falls into the Ramps 1.4 basket due to the way the pictures sit above the table....... not even close

DC42 is right in that calibrated accuracy is important but in my humble opinion more for that ability to provide accurate temps to the rest of the community.

I quoted the passage above to comment on the filament fine tuning. I agree that jumping from PLA to ABS to PET requires you to learn what temperatures to use I find that the fine tuning is more about tuning the temperature to the print speed whereas the general "print abs at 230" type guides get you in the ballpark at sub 40mm/s speeds.

I changed the mk10 flashforge setup to all metal at the beginning of the year and upped the ABS temps from 236 to 242 but unless i'm planning a slow print that 242 is used across all the brands/colours of ABS I have as I print nearly everything at 55/60 mm/s and this works at that speed.

I have PLA to be a little bit more sensitive through the dyzdesign hotend on the Delta but have settled on 210c as a normal temp for 90mm/s prints, again fine tuned to the print speed.

Part 2 and 3 will be interesting.

Quote
Dyze_Design

Quote
dc42
A nice article, but inaccurate in places:

Many thanks!

You've made great points and I thank you for sharing this information with us. I will take the time to improve this blog based on your comments. Below are my explanations regarding your remarks.

Quote
dc42
- The electronics table is not right, because Duet electronics can use RTDs and thermocouples as well as thermistors. The new Duet WiFi has purpose-made 2-channel daughter boards to support a total of 4 RTDs and/or thermocouples. See [duet3d.com] and [duet3d.com] for details.

I wanted to compare boards as they are, without add-on electronics. As cost is a factor for this blog post and in the following, I want to compare the boards alone.

Fair enough; but when electronics specifically provides daughter boards for supporting other types of temperature sensor, shouldn't they be considered part of the board, perhaps with a note that it is an additional option?

Quote
Dyze_Design

Quote
dc42
- You state the resolution for a thermocouple as 0.5C, and for an RTD as 1.2C. But the resolution depends entirely on what electronics you use to read it. I suspect you are assuming that an amplifier board is used and then the ADC in the 3D printer electronics is used to convert the amplifier output. A better solution is to use a purpose-made chip to read the thermocouple or RTD and send a digital readout to the main electronics. This is what the Duet and Duet WiFi do. The resolution obtained is 0.25C for a thermocouple, and better than 0.05C for a RTD. The quoted error for the conversion chips over the whole temperature range is 2C for the thermocouple up to 700C, and 0.5C for the RTD.

As very few boards include a specialized IC (covered in my part 2 ) such as MAX31855, AD595, MAX6675 for thermocouple and MAX31865 for RTD, I wanted to compare based on a common microcontroller ADC resolution basis. By having a better ADC, as the DUET and most 32 bits boards have, each sensor benefits, rather than comparing the specialized IC with 14 bits or 15 bits integrated ADC.

Your article implies a much worse resolution for thermocouples and RTDs than the recommended solution for some electronics provides. This is misleading. A novice user reading your post might well conclude that an RTD is no good because it can't provide better than 2.5C resolution (which I see you have now changed to 1.2C).

You quote the resolution for a thermistor as "up to 0.16C". But you should point out that this varies a lot over the temperature range. For example, using a 10 bit ADC (AVR processor) and 4.7K series resistor, the popular Semitec GT104 provides 0.5C resolution at 20C, 0.12C resolution at 108C, and 3.7C at 300C. The resolution at high temperatures can be improved in principle by reducing the series resistor, but then the ADC offset tends to introduce large errors at low temperatures, and users don't like seeing room-temperature readings that are so obviously a long way out. 32-bit electronics usually provides a 12-bit ADC and so better resolution by a factor of 4.

---

Large delta printer [miscsolutions.wordpress.com], E3D tool changer, Robotdigg SCARA printer, Crane Quad and Ormerod

Disclosure: I design Duet electronics and work on RepRapFirmware, [duet3d.com].

Quote
aussiephil

Good start but personally I'd like to see the table include the 32 bit boards separately, as it stands it tells the Newbie that a Duet board falls into the Ramps 1.4 basket due to the way the pictures sit above the table....... not even close

DC42 is right in that calibrated accuracy is important but in my humble opinion more for that ability to provide accurate temps to the rest of the community.

I quoted the passage above to comment on the filament fine tuning. I agree that jumping from PLA to ABS to PET requires you to learn what temperatures to use I find that the fine tuning is more about tuning the temperature to the print speed whereas the general "print abs at 230" type guides get you in the ballpark at sub 40mm/s speeds.

I changed the mk10 flashforge setup to all metal at the beginning of the year and upped the ABS temps from 236 to 242 but unless i'm planning a slow print that 242 is used across all the brands/colours of ABS I have as I print nearly everything at 55/60 mm/s and this works at that speed.

I have PLA to be a little bit more sensitive through the dyzdesign hotend on the Delta but have settled on 210c as a normal temp for 90mm/s prints, again fine tuned to the print speed.

Part 2 and 3 will be interesting.

Just updated the blog!

• 32 bits boards separated
• accuracy added
• mentionned about Duet board add-on and general add-on for other boards
• note regarding the thermocouple housing
• note regarding 10 bits ADC as a standard and 12 bits would improve

Quote
dc42
which I see you have now changed to 1.2C

This is the first modification, the RTD was at 1.2°C resolution since the beginning, you may check your first comment!

Thanks for your help guys!

---

DYZE DESIGN
Hotends, Extruders, Liquid Cooling and Accessories.

Thanks, that's looking better. I guess I must have mis-remembered the RTD resolution.

How you arrive at 0.1C for the thermistor accuracy without calibration, even ignoring the fact that the resolution is worse than this towards the ends of the temperature range?

---

Large delta printer [miscsolutions.wordpress.com], E3D tool changer, Robotdigg SCARA printer, Crane Quad and Ormerod

Disclosure: I design Duet electronics and work on RepRapFirmware, [duet3d.com].

Hi dc42!

I'm glad you like it.

I found it on a spec sheet: [www.picotech.com]
The link is added to the source.

I must say, I'm on your side regarding the accuracy, as the thermistor will drift more than the RTD with time, and 0.1°C is pretty low overall. The more I try to find data regarding sensors, the more I find different values and misleading information! I even found a datasheet that would state a thermostat having "high temperature range" from 80°C - 150°C compared to RTD, thermistors and thermocouple.

---

DYZE DESIGN
Hotends, Extruders, Liquid Cooling and Accessories.

This panasonic [www.mouser.com] datasheet is quite interesting even if not a high temp version, of the 4300 thermistors in Mousers catalogue only 23 of them are 100k/300c versions and the cheaper ones are 2% or 5% on the resistance tolerance value.
Further datasheet diving [www.mouser.com] and [www.mouser.com] shows that for glass encapsulated versions the still air time constants are measured in many of seconds, up to 15 seconds in at least one spec and they use liquid immersion to get short time constants.
In the better data sheets they also list the beta tolerance.
So whilst the repeat-ability of a reading for any one thermistor may be good, the variance across units/brands may require more individual tuning. I just bought two B57560G104F units sold as 3D printer specific that are 1% but haven't wired them to anything yet

Interesting line from the picotech link "Thermistors, because of their high sensitivity, are ideal for detecting small changes in temperature — especially when it is the change and not the absolute value that is important."

I would point out that it is not the NTC thermistor itself that varies in accuracy with temperature as DC42 says but rather the electronics ability and bit depth of the ADC to do the resistance reading/comparison.

Quote
Dyze_Design
Hi dc42!

I'm glad you like it.

I found it on a spec sheet: [www.picotech.com]
The link is added to the source.

I must say, I'm on your side regarding the accuracy, as the thermistor will drift more than the RTD with time, and 0.1°C is pretty low overall. The more I try to find data regarding sensors, the more I find different values and misleading information! I even found a datasheet that would state a thermostat having "high temperature range" from 80°C - 150°C compared to RTD, thermistors and thermocouple.

Here is why I don't believe that 0.1C accuracy for an un-calibrated thermistor. I attach a spreadsheet that simulates a thermistor using the B value equation. This equation is of course not very accurate, but it is good enough to illustrate the effects of the thermistor tolerance.

The input parameters in the top left are the thermistor R25 and B, and the tolerances on those values.

In the main section, you can put whatever temperatures you like in column A. Column B shows the nominal resistance at that temperature. Column D is a check on the formula I am using to calculate the temperature. Column H gives the actual thermistor resistance when the R25 has the minimum value implied by the tolerance and B has the maximum value. I did it this way round because above 25C these errors add, so this gives the minimum resistance at temperatures of interest. Column I gives the actual resistance assuming that R25 is at the upper end of its range and B is at the lower end; so above 25C this is the maximum resistance. Columns J and K give the calculated temperatures for those resistances, assuming that the firmware uses the nominal R25 and B values.

I put in a tolerance of 2% for R25, and 3% for B, which are the values for the populate Semitec GT104 thermistor.

You can see that at 25C where only the R25 tolerance has an effect, the error is already +/-0.4C. At 240C the error introduced by the tolerance on R25 and B is +/-12C.

Closer tolerance thermistors are available, but cost more and I don't think they are commonly used in 3D printers. If you do want better accuracy with a thermistor, at 3D printing temperatures the tolerance on the B value is more important than the tolerance on the R25 value.

I conclude that if you use a common thermistor to measure hot end temperature, you can't rely on the temperature reading being accurate. This doesn't necessarily matter to a typical user with just one 3D printer, because finding the right temperature for printing with a particular filament is largely a matter of experimentation anyway.

Edited 5 time(s). Last edit at 07/08/2016 04:09AM by dc42.

---

Large delta printer [miscsolutions.wordpress.com], E3D tool changer, Robotdigg SCARA printer, Crane Quad and Ormerod

Disclosure: I design Duet electronics and work on RepRapFirmware, [duet3d.com].

Thanks for both of your comments!

aussiephil, as you can see, the bigger the thermistor is, the faster the response time. The 0.3mm thermistor has a response time of 0.8 seconds while the 0.58 has 1.7 seconds.Same thing applies to thermocouple and RTD inside their probe. The larger the probe, the longer the thermal constant. A 6.35mm housing can take more than 30 seconds to reach 62% of the temperature.
I like the quote, it points out very well the good side and the weakness of a thermistor.

dc42, many thanks for taking time with this explanation. Very interesting, and the calcs sheet is great too. I really liked it.

Our current thermistor is rated at R200 by the manufacturer and we also have the resistance table for very accurate thermistor table inside the firmware, without having an additional beta consideration.

---

DYZE DESIGN
Hotends, Extruders, Liquid Cooling and Accessories.

What is the percentage tolerance on the resistance of your thermistor, both at 25C and at around 250C?

---

Large delta printer [miscsolutions.wordpress.com], E3D tool changer, Robotdigg SCARA printer, Crane Quad and Ormerod

Disclosure: I design Duet electronics and work on RepRapFirmware, [duet3d.com].

Quote
Dyze_Design
Thanks for both of your comments!

aussiephil, as you can see, the bigger the thermistor is, the faster the response time. The 0.3mm thermistor has a response time of 0.8 seconds while the 0.58 has 1.7 seconds.Same thing applies to thermocouple and RTD inside their probe. The larger the probe, the longer the thermal constant. A 6.35mm housing can take more than 30 seconds to reach 62% of the temperature.

Umm, the example you show says the smaller thermistor has a faster response time, 0.3<0.58 smaller and 0.8<1.7 (faster), which would make a lot of sense as there's less mass to change the temp of. Either change it to be "smaller thermistor has faster response, or larger thermistor has slower response.