Hi, I am trying to increase the accuracy and precision of my thermistor. My goal is to be able to calibrate to the vernier temperature probe (20k NTC thermistor). Right now I am getting an accuracy of +/-0.5C. My goal is to be dead on. In what ways can I increase my accuracy and precision? Current I am running a voltage gate connected to my 10K NTC thermistor (5 vin). How can I use the full range on the adc-10 on Arduino Uno? My temperature range is 0C to 80C.
kendreaditya:
My goal is to dead on. In what ways can I increase my accuracy and precision?
Perhaps by re-thinking the goal, and then the appropriateness of a thermistor for achieving it.
Nick_Pyner:
Perhaps by re-thinking the goal, and then the appropriateness of a thermistor for achieving it.
I can't change my vernier thermistor.
Really? Then re-think your goal.
I think you might mean "Vernier" a company that makes sensors for organisations with big budgets, in which event you should follow their comprehensive instructions.
If otherwise, I would have thought +/-0.5 pretty good, particularly if degrees F, about which you say nothing. It would also depend somewhat on the required temperature range, about which you also say nothing.
Thermistors are fine, so long as you recognise their shortcomings. I understand throwing money at the problem works to a (ahem!) degree but, in that event, you might consider alternatives that provide a better solution more cheaply and with less effort.
A 10k thermistor with a 10k fixed resistor and a 10-bit A/D has about 10 steps per degree C resolution when close to 25C. Anything further away from 25C results in fewer steps per degree C.
That means you can never make a better than 0.1C display with that setup.
You can ofcourse try to calibrate it to 0.1C yourself by changing values in the code.
What is a "voltage gate".
Thermistor/resistors must be read with a ratiometric A/D for stability.
Post a diagram and picture of the setup.
Leo..
kendreaditya:
Right now I am getting an accuracy of +/-0.5. My goal is to be dead on.
For most applications that is considered "dead on" already.
Are you sure whatever thermometer you use as reference to calibrate your probe against is much better than that? And that your reference thermometer is indeed exposed to the exact same actual temperature as your NTC probe?
Right now I am getting an accuracy of +/-0.5
Fahrenheit or Celsius? Units do matter.
What calibrated thermometer are you using for a reference, and what is its certified accuracy?
Wawa:
A 10k thermistor with a 10k fixed resistor and a 10-bit A/D has about 10 steps per degree C resolution when close to 25C. Anything further away from 25C results in fewer steps per degree C.
That means you can never make a better than 0.1C display with that setup.
You can ofcourse try to calibrate it to 0.1C yourself by changing values in the code.What is a "voltage gate".
Thermistor/resistors must be read with a ratiometric A/D for stability.
Post a diagram and picture of the setup.
Leo..
May I ask how you calculated that 10 step per degree C? Would oversampling help? I am using the Steinhart hart equation to calibrate. I already know the coefficients for the vernier probe, just need to find it for the 10K NTC thermistor. I am currently using the 5V pin as an analog input as the actual vin in the equation (R*(vin/vout))
wvmarle:
For most applications that is considered "dead on" already.Are you sure whatever thermometer you use as reference to calibrate your probe against is much better than that? And that your reference thermometer is indeed exposed to the exact same actual temperature as your NTC probe?
My reference probe has to be the vernier probe despite its comparative low accuracy at +/-0.2C, which is fine as long as the precision is <0.1C. I am sure that both my ref thermistor is exposed to the same temperature as my NTC thermistor.
If you want your required precision at more than a 10 degree span, measure the resistance of your NTC at various temperatures, preferably in 1-2°C steps. Interpolate (linear is now good enough).
Of course this requires you to have a very well controlled and calibrated temperature chamber or oil bath or so. Your final accuracy stand and falls with your calibration accuracy.
wvmarle:
If you want your required precision at more than a 10 degree span, measure the resistance of your NTC at various temperatures, preferably in 1-2°C steps. Interpolate (linear is now good enough).Of course this requires you to have a very well controlled and calibrated temperature chamber or oil bath or so. Your final accuracy stand and falls with your calibration accuracy.
The Steinhart hart equation takes care of that, given 3 points, it makes log curve that is more than my water bath can be. What I need help with now is improving the accuracy of my NTC thermistor's resistance, whether it's improving the reference voltage or/ADC's readings. I think those are the 2 ways that will mainly benefit me, I just don't how to make a more precise ref voltage and increase the precision of the ADC reading.
The ONLY way you can improve the accuracy of your measurement device is to calibrate it using an accurate, certified reference.
Do not confuse precision and accuracy.
jremington:
The ONLY way you can improve the accuracy of your measurement device is to calibrate it using an accurate, certified reference.Do not confuse precision and accuracy.
I understand, my main goal is to have precision, my accuracy does not matter (I was wrong before) as what determines my accuracy is the vernier probe's temperature.
Then read and follow the instructions in reply #9.
kendreaditya:
May I ask how you calculated that 10 step per degree C?
Would oversampling help?
The thermistor/resistor voltage divider outputs a (varying) ratio of it's supply to the A/D.
Calculate that ratio with the help of a "10k thermistor resistance table", for two temps that are say 10C apart.
Then calculate how many A/D values are in between those two temps. That is the step resolution.
I calculated about 10 per degree C for a 10-bit A/D. A 12-bit A/D would ofcourse have four times the steps.
Note that I'm talking about ratio, not voltage.
Ratio stays constant, voltage not.
Leo..
jremington:
Then read and follow the instructions in reply #9.
This is for a competition where I have 20 mins to calibrate, so I won't have time to calibrate for every 2C. I am using the Steinhart hart equation which gives the accuracy I need from only data points, trust me. I only need to know how to increase the precision of the resistance readings of my NTC thermistor, which I don't know how to do.
Why do you have to calibrate on the day of the competition? That doesn't make sense. You of course do that beforehand, arriving with a well calibrated sensor.
If you want a higher resolution of readings, you may try to oversample the ADC (that can gain you a bit or two of resolution), or use an external ADC.
Most of us are quite familiar with the Steinhart-Hart approximation of the thermistor response curve. Trust us.
You are not measuring resistance. To increase the precision of the voltage measurements from the thermistor voltage divider, from which the thermistor resistance is estimated, you could use a ratiometric ADC with more than 10 bits, or use oversampling and averaging to increase the effective resolution of the Arduino 10 bit ADC.
wvmarle:
Why do you have to calibrate on the day of the competition? That doesn't make sense. You of course do that beforehand, arriving with a well calibrated sensor.If you want a higher resolution of readings, you may try to oversample the ADC (that can gain you a bit or two of resolution), or use an external ADC.
Most teams will not know the calibration sensor used to calibrate their devices until they are at a competition, I on the other hand do, and have access to it beforehand. This is what I am doing, the now the curve of the calibration sensor and thus need to match my NTC thermistor curve to it. I tried using a 16-bit external ADC, but I think I did something wrong with the gain and now I can only do 2/3 gain, not to mention external ADC's are not allowed (I know, its very very dumb). I don't know what gain does but using 2/3 gain did not increase my precision to much.
Assuming I figure out a way to get precise readings, is there a way I can use the same coefficient from the Steinhart hart equation for calibration sensor (which is 20K ish) with my NTC 10k thermistor?
I can use the same coefficient from the Steinhart hart equation for calibration sensor (which is 20K ish) with my NTC 10k thermistor?
No.
However, you can get a better set of coefficients for your particular thermistor, if you carefully measure its resistance at three accurately known temperatures, using this on line calculator.