High Resolution/Precision Thermistor Circuit

Thread Starter

dksoba

Joined Jan 8, 2010
25
Hello,

I am designing a temperature controller for growing human cells for AIDS research. We want to be able to measure the temperature at a certain location in our device (very close to the cells). The temperature will be around 37C, and we want to control it to +/- 0.5C. We have looked into different temp sensors and pretty much have settled on using NTC thermistors, such as the mf51e from Cantherm, or others from Honeywell. Now, to get the highest accuracy, we will calibrate the sensors AT 37C ourselves. My question is, to get high precision, and repeatability, what type of circuit should we use? I made a wheatstone bridge + differential amplifier circuit already in both LTSpice, and on the breadboard with 2% tol resistors. It seems to work okay, but I don't have a reference thermometer yet so I'm not sure how precise my results are, or repeatable. Also, although not in the attached picture, I designed a wheatstone bridge such that the range on the thermistor that correlates to 37+/-5C results in approximately 0-500mv, which is then amplified by an op amp.

Since the 0-500mv corresponds to 0-5V, that's 100mv/V. I'm not sure what my CMRR should be based on that information. In order to increase my CMRR I would need to use higher precision resistors... But I'm wondering if it's just cheaper and more efficient to buy an instrumental amplifier? I would still need a wheatstone bridge though. Does that need to be made with high precision resistors?

Also, the rest of the system consists of heater coils, powered by MOSFETs, and water pump(s), and a PIC18F4550. If I calibrate my temperature sensor to 37C and some other points around that temperature, I can store a calibration curve in the PIC. However, how do I design my electronics so that that calibration curve is accurate when the ambient temperature changes from 20C to 30C? That's about the fluctuation in temperature I'm looking at for room temperature. I know my wheatstone bridge resistors will change values and such due to temperature fluctuations, and the opamp will change a little, etc.

In the end, we'd like our PID to have a feedback from a temp sensor in the range of 37+/-5C, and with a resolution of 0.01C or greater, and an accuracy of 0.1C or greater. I know these are tight tolerances, but is this unreasonable?

Thanks,
Matt
 

Attachments

beenthere

Joined Apr 20, 2004
15,819
Might not life be simpler using an LM35? The response is already linearized, and is 10 mv/degC. You can get a data sheet here - http://www.national.com/mpf/LM/LM35.html#Overview

If you want to stay with the thermistor, why worry about the calibration cure at all? Your only interest is controlling the system to maintain a single voltage level out of the sensor. The departure from absolute linearity over +/- 2 deg is insignificant. If you feel that a bridge is necessary, place all the elements inside the temperature controlled zone and avoid any slightest disturbance from the ambient.

Sensing the temp is not at all challenging. Maintaining the temp can be.
 

retched

Joined Dec 5, 2009
5,208
I second that. The best way would be to keep all sensing in the controlled environment, and transmitting out the results. You can use a very inexpensive uC and a thermistor to read the temperature then use a serial line out of the chamber to report the temperature to whatever device you would like.


As beenthere has said, reading the temp is easy. And very accurate these days with just two parts and a power supply. What are you using as your chamber?
 

ifixit

Joined Nov 20, 2008
652
Some good advice has been posted, but incase that is not possible and you are proceeding with what you have, here are some tips...
  1. Use precision resistor wherever ambient temperature may effect the resistor value, even if you don't need the absolute accuracy, these resistors always have good TC values. Generally, the more precise the resistor, the better the TC will be.
  2. To test the temperature stability of your circuit, replace the thermistor with a resistor that is equilivent to the thermistor at 37.00°C. This data should come with the precision thermistor. Now heat and cool your circuit and note how much the reading changes. Keep the calibration resistor at a fixed temperature.
  3. Repeat test (2) at the ends of your measurement range.
  4. Upgrade any parts that seem to cause too much drift. For ±0.1°C accuracy I would shoot for analog circuit stability in the order of ±0.01°C. This leaves room for other in-accuracies, like the ADC.
  5. Vary the power rails and repeat 2 & 3.
  6. You'll need high resolution, and accurate insturments to do all this.
  7. The mf51e thermistor is ±0.5% device, so that will be your absolute accuracy for now until you can calibrate it to ±0.1°C.
  8. LTspice can be used to test effects of a resistors TC. e.g. CTRL right-click on resistor to chg params, then set Value2 to TC=-0.001 for a 0.1%/°C change with temperature or whatever the manufacturer says it is.
All this is not trivial to design. You'd be better off to buy a box that does all this for you... but where is the fun in that?

Good Luck
Ifixit
 

Thread Starter

dksoba

Joined Jan 8, 2010
25
That was a lot of quick responses!

The "chamber" is a clear plate that's under a microscope. The desire is to grow cells ON a microscope under a controlled temperature, and leave the cells constantly being photographed over a period of several weeks. Because the temperature sensors are located on the cell plate, where real estate is at an extreme premium, they have to be very small, hence the thermistor. Otherwise I'd use an IC and call it a day.

I think I'll use an instrumentation amplifier such as a TI INA122PG4 (http://mouser.com/ProductDetail/Tex...GAEpiMZZMsE1dKaA2ImUCkkzVUbBDY7/i8Y%2bLyLIxc=)

To amplify a wheatstone bridge voltage. The wheatstone bridge will be made out of precision resistors (rated at +/- 10 ppm/C). The entire circuit will be in a controlled temperature biohazard level 3 room, around 25C. I'm not sure how big the temp fluctuations there are, but in the worst case scenario I know it can't be more than +/-5C. That's a 10C swing, in the worst case scenario I'm looking at 100ppm or 0.01%. If I calibrate my thermistor ON that bridge at 25C, then the max temp swing is 5C, which is only 0.005%. My back-of-envelope calculations show that that's about 20 uV error in the worst case scenario, which is small considering I'm amplifying a range of about 500mV. If 1 deg C = 50mV, then this is an error of 0.045 deg (assuming linearity of RT curve over 10deg range), even less if I change my wheatstone bridge to read over a 2deg range, etc.

Of course, probably the biggest source of error, will be using the 10bit ADC on the PIC18f4550. Since temporal resolution is not super important, MANY samples will be averaged over time. We're only trying to control the temperature to within +/-0.5C of 37C, so maybe the accuracy I'm thinking I need is way too high.

Does this seem reasonable?
 

someonesdad

Joined Jul 7, 2009
1,583
Another sensor to consider might be an RTD. They have well-known responses, but of course are more expensive than thermistors.

Instead of building a bunch of stuff, you can buy PID controllers (many use thermocouples to sense the temperature). You can find them for $50 or so on ebay. I haven't used one, so I can't comment on their performance, but it's a lot cheaper than your time. You provide a temperature sensor input and the thing can provide a proportional output to control the heater (they usually come with on/off control also).
 

Thread Starter

dksoba

Joined Jan 8, 2010
25
Another sensor to consider might be an RTD. They have well-known responses, but of course are more expensive than thermistors.

Instead of building a bunch of stuff, you can buy PID controllers (many use thermocouples to sense the temperature). You can find them for $50 or so on ebay. I haven't used one, so I can't comment on their performance, but it's a lot cheaper than your time. You provide a temperature sensor input and the thing can provide a proportional output to control the heater (they usually come with on/off control also).
Please correct me if I'm wrong:

I'm is one of my senior design projects in Mechanical Engineering at UCSD. The whole purpose is to learn AND produce a product for our "sponsor". Our sponsor is a company/organization that provides the budget and design constraints. I looked into RTDs, and I couldn't see any advantages over thermistors. In general, they're bigger, have similar tolerances to standard calibration curves, and are more expensive.

Our PID Controller has to be custom for 3 reasons. 1) If we used an off the shelf component like that, our professor would kill us since control systems is a big area of research at our school and we should be making our own controllers. 2) Our controller will control the temperature of different parts of the cell plate, with multiple inputs and outputs. It's not something we can do off the shelf. 3) The PIC18F will also control other things in the project, and since this will be a commercial product, everything should be integrated.

I'm not trying to roll my own temp controller to save money off commercial PID controller.

I appreciate your response,
Matt
 

beenthere

Joined Apr 20, 2004
15,819
The only comment I might make is coupling "precision" sensing with a 10 bit A to D. You can still sense tighter than 1/2 degC, but 1 part in 1024 is really easy to achieve with simpler methods.
 

Thread Starter

dksoba

Joined Jan 8, 2010
25
The only comment I might make is coupling "precision" sensing with a 10 bit A to D. You can still sense tighter than 1/2 degC, but 1 part in 1024 is really easy to achieve with simpler methods.
Such as? Yea I was also thinking it might be useful to use a 16-bit or 24-bit ADC, but I'm not sure yet if it's necessary. I think if I take many samples and average them out I can also get pretty good accuracy.

Matt
 
Top