Precision Thermistor Temperature Sensing

Thread Starter

picotrain

Joined Apr 12, 2013
32
Hi all,
I am working on a project that requires temp sensing from room temp (~70F) to 212F, with my range of interest from 195F to 208F with a resolution of 0.1F or better.
While there are dedicated ICs such as the AD7711 that do the job, they tend to be expensive, and I am hoping to learn something while I work on this project.

The research I have done seems to suggest :
1. Place the thermistor in a voltage divider, where the reference resistor is the same value as the thermistor in the middle of my range of interest, in order to maximise sensitivity.
2. The output of the voltage divider is then passed through a Instrumentation Amp with the gain set to fill as much of the ADCs input window as possible.
3. In code use the Steinhart-Hart equation to convert the ADC output to a temperature.

Does this sound right? Or have I missed something?
 

Thread Starter

picotrain

Joined Apr 12, 2013
32
I should add the following

I have selected a thermistor as my sensor due to its accuracy (I have identified USP11492 with accuracy of 0.2C), fast response time & relative cheapness compared to RTDs and Thermocouples.
My application requires an immersion probe, coupled with fast response time, which rules out LM35 and most other semiconductor sensors.

I plan to obtain a reference thermometer to calibrate my system.
 
Take a look at the PG309 from TI. I'd like to do a project with it to some day. The curves are VERY non-linear.

Look-up tables and interpolation are also possible.

Mike also nailed it.
 

Thread Starter

picotrain

Joined Apr 12, 2013
32
Thats a 1% tolerance on a 100 ohm device. The 4 wire RTDs with tighter tolerances that I have looked at are closer to $40.

Also my research seems to indicate that RTDs are far more susceptible to noise and are far less sensitive than say a 10K thermistor.

Please correct me if I am wrong.
 

ebeowulf17

Joined Aug 12, 2014
3,282
+1 for platinum RTDs. We use 2000ohm RTDs from Tempco with great success. I'm not involved in purchasing, but I'm almost positive they cost less than your thermistor. Using 2000ohm units instead of the more common 100ohm units makes the system far less susceptible to errors resulting from extension wire resistance variations. We simply run the wires in a twisted pair, reasonably separated from high power wires and have no noise problems.

Regardless of sensor choice, you may want to consider a wheatstone bridge as the front end, as it simplifies the scaling/level shifting greatly. [EDIT: I was only referring to thermistors and RTDs here. You wouldn't want a wheatstone bridge on a thermocouple.] I'd also recommend lookup tables with interpolation over formulas. I have less experience with thermistors, but with RTDs and thermocouples, a lookup table provides much better accuracy than all but the most complicated formulas. [EDIT: My memory seems to have failed me here. I've done a bit more reading and it looks like RTDs are remarkably linear and their non-linearities are pretty easy to account for with simple formulas. There's still no reason not to use lookup tables with them, but it may not be necessary.]
 
Last edited:
We kept to thermocouples most of the time. R ,S, T, J. C, K. All over the map. The RTD was annoying, but worked extremely well. I used the 3 wire method. The RTD got it because of the better accuracy and resolution needed and it was electrically insulated. You have to watch the meters for thermocouples. The isothermal junction wasn't for one instrument. For the RTD, lead length and wire gauge has to be the sane,

Thermistors have their purpose. Fast response over a limited range and cheap. These http://www.omega.com/temperature/tsc.html guys were always a good source of information.
 

Thread Starter

picotrain

Joined Apr 12, 2013
32
Using 2000ohm units instead of the more common 100ohm units makes the system far less susceptible to errors resulting from extension wire resistance variations. We simply run the wires in a twisted pair, reasonably separated from high power wires and have no noise problems.

Regardless of sensor choice, you may want to consider a wheatstone bridge as the front end, as it simplifies the scaling/level shifting greatly. I'd also recommend lookup tables with interpolation over formulas. I have less experience with thermistors, but with RTDs and thermocouples, a lookup table provides much better accuracy than all but the most complicated formulas.
Thanks for the tips, much appreciated :)

You have to watch the meters for thermocouples.
I dont understand what you mean?

For the RTD, lead length and wire gauge has to be the sane,
Do you mean the length of the current supply wires and the sensing wires have to be the same? Or something else?
 
I dont understand what you mean?
Some meters/controllers are of a very poor design. Isothermal half-junctions with temperature is expensive to make.

RTD's have 3 and 4 lead sensing, For three lead to work, all of the wires used should be the same, including the length,
 

wayneh

Joined Sep 9, 2010
17,153
Another vote for platinum RTD. That's the state-of-the-art solution any professional would turn to.

On the other hand, you said you need a resolution of 0.1°F but that doesn't necessarily mean you need that level of accuracy. I think most of us were assuming you did, but it would be good to confirm; what accuracy do you require? You're definitely getting into the range of traceable calibration standards.
 

Thread Starter

picotrain

Joined Apr 12, 2013
32
I feared that may have been the case.

I'm wondering now, just to keep things simple, whether I should use a LM35 (+/- 0.75C accuracy in my range of interest)
By oversampling with a 10 bit ADC I should be able to get the precision I want.
I just won't have the degree of accuracy I would like.
 

ebeowulf17

Joined Aug 12, 2014
3,282
My application requires an immersion probe, coupled with fast response time, which rules out LM35 and most other semiconductor sensors.
How would you plan to use the LM35 in an immersion application?

I plan to obtain a reference thermometer to calibrate my system.
If you were already planning on calibrating to a reference thermometer, why would the recent accuracy vs. precision questions change your plans? You need a sensor/circuit combination that provides linear response, precision, fast enough response times, and immersion capabilities. If the initial accuracy of your components isn't quite right, you adjust for that when calibrating to your reference, right?
 

ebeowulf17

Joined Aug 12, 2014
3,282
That's cool. Didn't know probes like that existed.

Still, you mentioned response time early on... what do you really need in terms of response time? A 6mm probe is going to respond a lot slower than the 1/8" or even 1/16" probes that are common for RTDs and thermocouples (maybe thermistors, too. I've just never checked.)
 

Thread Starter

picotrain

Joined Apr 12, 2013
32
That's cool. Didn't know probes like that existed.

Still, you mentioned response time early on... what do you really need in terms of response time? A 6mm probe is going to respond a lot slower than the 1/8" or even 1/16" probes that are common for RTDs and thermocouples (maybe thermistors, too. I've just never checked.)
My application is pressurized boilers, such as those used in small "prosumer" espresso machines (typically <600mL), so the response time required would have to be pretty fast. While I agree that the 6mm probe would be far too slow, it gives me some hope that I will be able to find a IC based temperature sensor (which addresses most of my off the shelf accuracy needs without calibration) that can be potted into a thinner probe.

I was originally looking at thermistors for their fast response time, however if I develop a precision temperature controller based on thermistors, I would need to calibrate every unit I build.
 
Top