Choosing right RTD for a narrow 0-50 °C range: Pt100 vs Pt1000

[e-pirate]

Hi folks!
I'm thinking about design of a narrow range thermometer. Target temperature is about 20-25 °C, but lets accept it to be 0-50 °C. It should be at least 0.1 °C accurate. Fast respond is not a requirement (the thing will be submersed in a water flow with extremely slow changing rates: a couple of degrees/hour in worst case). No ambient temperature change (room air conditioned environment with same temperature range). The cable length is under 2m. Thermal mass is not an issue. After reading this https://www.acromag.com/page/white-paper-comparison-thermocouple-and-rtd-temperature-sensors papers, I came to a conclusion, that RTD is the best choice for my task. I'm going to buy cl. A RTD, but since they come in 2 different "flavors" (well, actually 3, but Pt500 are very rare) , I need some advice here. In my "feeling", for my application its a matter of OPAmp gain, but on the first glance, I prefer Pt1000.
Opinions with arguments, please.

[mycroft]

Take a look at https://edwardmallon.wordpress.com/2016/06/09/better-thermistor-readings-with-an-arduino-series-resistors-aref/. Lots of interesting information.

[jbb]

A comment from Yokogawa (they do good stuff): http://www.yokogawa.com/us/support/knowledgebase/what-is-the-difference-between-pt100-and-pt1000-temperature-sensor-and-which-one-is-better.htm

Pt100:

• Typically 0.391 Ohms / K
• + perhaps more common when it comes to buying sensors
• + has very slightly lower electrical noise
• - more sensitive to wiring resistance (most use 4 wire Kelvin sensing)
• - requires more drive current - consumes more power
• - may experience more heating due to its internal power consumption

Pt1000:

• Typically 3.91 Ohms / K
• - might be a narrower range of products
• - has very slightly higher electrical noise
• + less sensitive to wiring resistance (you may not need 4 wire Kelvin sensing)
• + requires more drive current - consumes less power
• + experiences less heating due to its internal power consumption

Assuming you're using a Pt1000, your effective range will be 1000 to 1194 Ohms for 0 - 50 deg C.

I do have one sad thought: if you really need 0.1 deg C absolute accuracy, you will have to pay someone to calibrate your equipment once it's built.  This could be more expensive than just purchasing a pre-built piece of (quality!) equipment.

Here's some hints for the driving / measuring circuit:

• The Atmel AVR ADCs probably aren't good enough
• The PT1000 can be placed in series with a precision resistor, and their ratio measured.  Connect the mid point of the divider to the voltage reference rail (which must be clean and stable but not very accurate)
• Remember to add some ESD protection (e.g. a TVS diode) and an EMI filter to the sensor inputs
• Calculate your cabling resistance: it must be lower than the measurement error you want to achieve.  Remember that the electricity needs to go out to the sensor and and back (i.e. double the length)
• If you use a high precision ADC (e.g. 16b or more) then you don't have to mess about with an op amp.
• Remember to calculate how much power will be dissipated by the Pt100/Pt1000; this power will actually warm the resistor up a bit and reduce measurement accuracy.

I did a quick little sketch design which might be a good starting point.

(Edit) It shows a 16b SINGLE-ENDED ADC yielding 57.2 counts / deg C.  For comparison, a 10b AVR ADC would give 1/64th the resolution, i.e. a bit less than 1 count / deg C.

[Kleinstein]

Pt100 tends to need 4 wire connection, while PT1000 might still get away with 2 wires.

With the expensive wire based resistors PT100 tends to be slightly more accurate. With thin film sensors the problem is more with the contacts and thus with slight favor for PT1000. The other advantage of PT1000 is less sensitivity to thermal EMF - the advantage of less aging at high temperatures is with PT100, but not relevant here. Also cable insulation is easier with PT100 - but this is more like a high temperature problem too.

Self heating and noise are the same, if the same power is used - these two are points one can choose the balance.

Both types work, I would slightly prefer PT1000. However there might be better offers on precision PT100s. I would look for ready made probes, not just a bare sensor.

[doktor pyta]

I don't know Your application but You can also consider using 10k NTC thermistor which will give You high sensitivity and no problems with wire resistance. If not go for Pt-1000.

[jbb]

Self heating and noise are the same, if the same power is used - these two are points one can choose the balance.

That's a great point.

If you were looking for the best possible accuracy, possibilities include:

• a Pt100 with 4 wire Kelvin sensing
• use a high quality ADC (e.g. 24b with low INL) with plenty of input protection and EMI filtering
• a fairly strong drive signal to give good output amplitude & low noise
• pulse the system to keep average power (and sensor self heating) low
• use +ve and -ve excitation (i.e. H bridge drive circuit) so that alternate readings are +ve and -ve voltage.  Then de-invert -ve pulses in software and average to remove the effect of parasitic thermocouples

If you want to play this game, the LTC2983 www.linear.com/LTC2983 offers some serious performance with all the analogue stuff integrated.  You can drive 4x 4-wre Pt100s with one, see Page 40 of the data sheet.  While they are expensive, you could ask yourself "how much is my time worth?"

[e-pirate]

The design of the front end is up to my personal chose. It is "one of a kind" instrument, so the costs don't matter at all. For calibration I can use melting ice, which is 0.00 C and something else. Besides, I have an old Soviet mercury laboratory thermometer 0-50 C with 0.1 C step with it's original "passport" with calibration temperatures each 5 C (the thing is 60 cm long and in a perfect condition). And I will be most likely able to sneak to laboratory to calibrate it against something really-really accurate.

[awallin]

Self heating and noise are the same, if the same power is used - these two are points one can choose the balance.

Both types work, I would slightly prefer PT1000. However there might be better offers on precision PT100s. I would look for ready made probes, not just a bare sensor.

some notes on pt100 vs NTC thermistor from a few years back:
http://www.anderswallin.net/2013/03/noise-equivalent-temperature-in-an-rtdthermistor/
although the thermal noise (1 Hz bandwidth) of the sensor itself is equivalent to some uK, in practice it is quite hard to get below mK noise-levels - if anyone knows good solutions let us know!

For the calibration I've heard pt100s are quite sensitive to mechanical stress and shocks. So once you calibrate them you should treat them carefully and not drop them on the floor or bang them around. Mechanical changes in how the pt-wire sits in the sensor affect calibration.

[ebclr]

If 42 C still good for you make a try

http://www.ti.com/product/lmt70/datasheet

[ahbushnell]

Try Omega's web site.

http://www.omega.com/?gclid=Cj0KEQjwqtjGBRD8yfi9h42H9YUBEiQAmki5OvYix4Iy_bRhPk60qBoLBzyaEkSPlxsbboifd8wEkFIaAmxL8P8HAQ

That's where I always buy RTD sensors.  Look at the application notes section.  Lots of info. 

Accurate measurments on the .1 C level can be tricky.  We had a slow flowing system that I was measuring temperature change to check energy balance.  I had a RTD with about a 2" length tube holding the probe in the liquid.  I was making energy.  I considered cold fusion but thought it could be heat draining down the tube.  We changed to 6" and conserved energy.    The 2" tube was cooling my probe. 

Good luck

Andy