Precision Thermistors

Thread Starter

joeyd999

Joined Jun 6, 2011
5,237
A couple of days back, there was this thread going on:

http://forum.allaboutcircuits.com/showthread.php?t=63934

One of the reasons that I so heavily participated was that I actually in the process of developing a high-precision temperature controller for the past two and 1/2 weeks, and the material was fresh in my mind.

The controller supplies up to 20 watts to a small volume of air (about 1 1/2 gallons or so). I implemented a PID controller on a PIC 18F processor using a US Sensor PS502J2 5K@25C thermistor with 0.1C tolerance as the temperature sensor. The only other precision component in use is a 0.1% fixed resistor in series to form a voltage divider, the output of which goes to the PIC's A/D.

No other precision components, voltage or current sources were used.

Unfortunately, I don't have possession of any high accuracy and high precision thermometry equipment, at least not as good as the thermistor is supposed to be. So I decided to test at 37C (body temp), and went to Walgreens and got a fever thermometer. These thermometers supposedly are accurate to 1/10C, and have really good resolution (about 1C/cm), albeit with a limited range.

The volume took about 3 minutes to stabilize to 37C. I then stuck the thermometer in. It took about another 5 minutes for the thermometer to stabilize.

I expected at most an 0.1C error due to the tolerance of the thermistor, a 0.1C error due to the 0.1% precision resistor, and 0.05C error due to self-heating, for a total error budget of at most 0.25C without any calibration whatsoever.

Here is the result of the test:



Sorry, I cannot post additional details or photos of the circuit or setup. I just wanted to illustrate what's possible with very few precision parts.
 

Attachments

tracecom

Joined Apr 16, 2010
3,944
I implemented a PID controller on a PIC 18F processor using a US Sensor PS502J2 5K@25C thermistor with 0.1C tolerance as the temperature sensor. The only other precision component in use is a 0.1% fixed resistor in series to form a voltage divider, the output of which goes to the PIC's A/D.

No other precision components, voltage or current sources were used.
I, too, have used thermistors as the sensing element for a thermostat, and I agree that the results can be quite good. I have used them with op-amp comparator circuits and with microcontrollers.

However, I don't understand the value of high precision components unless the design does not provide for temperature adjustment. I can understand the need for components with predictable rates of change, but doesn't the inclusion of an adjustment feature preclude the need for .1% resistors?
 

crutschow

Joined Mar 14, 2008
34,285
I, too, have used thermistors as the sensing element for a thermostat, and I agree that the results can be quite good. I have used them with op-amp comparator circuits and with microcontrollers.

However, I don't understand the value of high precision components unless the design does not provide for temperature adjustment. I can understand the need for components with predictable rates of change, but doesn't the inclusion of an adjustment feature preclude the need for .1% resistors?
It depends upon how you do the adjustment.

If the you adjust the set point of the thermostat by changing a variable in the microprocessor to accurately detect a certain output voltage (or voltage ratio) of the thermistor circuit, then the accuracy of the circuit is preserved, since the rate of change of the thermistor circuit voltage with temperature is accurately predicable.

If you adjust the set point by adjusting the resistance or voltage of the thermistor circuit, then the accuracy is generally not preserved.
 
Last edited:

Thread Starter

joeyd999

Joined Jun 6, 2011
5,237
However, I don't understand the value of high precision components unless the design does not provide for temperature adjustment. I can understand the need for components with predictable rates of change, but doesn't the inclusion of an adjustment feature preclude the need for .1% resistors?
The value is that you know the ratiometric output of the voltage divider for any given temperature to very high precision and accuracy, without any calibration requirement.

The moment you tack a non-precision component into the network, the accuracy gained by using (expensive) precision components vaporizes.

If you wish to use the precision network as an adjustable precision thermostat, while retaining accuracy, this is easy. One input to a comparator is tied to the fixed thermistor voltage divider. The other input is tied to another divider with a set-point potentiometer. Both dividers are tied to the same V+ and ground (it doesn't matter what V+ is, except that higher V+ will cause more internal heating of the thermistor and lower V+ will cause more noise).

Then, you mark the temperature points on the pot based on the ratiometric output derived from the thermistor mfgr's spec sheet. The comparator's output will switch at the set-point.

You will want to add hysteresis to the comparator circuit so as to avoid *lots* of noise while the temp is changing slowly near the set-point.
 

Thread Starter

joeyd999

Joined Jun 6, 2011
5,237
Obviously forced...but you knew that. The air is forced over a heat sink upon which the heater is mounted.
 

tracecom

Joined Apr 16, 2010
3,944
I have found that increasing thermal mass of the heating element makes air temperature regulation more difficult. Are you varying the power to the heating element or just switching it on and off?
 

Thread Starter

joeyd999

Joined Jun 6, 2011
5,237
I have found that increasing thermal mass of the heating element makes air temperature regulation more difficult. Are you varying the power to the heating element or just switching it on and off?
Not more difficult, just slower. You need lotsa 'D' in the PID to prevent overshoot. And just enough 'I' to ensure zero error after a good period of time. (Plus a teensy amount of code to prevent integrator wind-up).

I tune by hand. I don't do any rigorous analysis to optimize the loop.

In this application, the PID drives a *very* slow PWM with a 1 second period (faster would only make more electrical noise for no additional benefit) and 10 bit resolution.
 
Top