temperature sensing strategy... what's more clever ?

[JeanF]

Hello,

I would like to ask you guys about one thing... I've learned all day long about sensors, resolution, calibration, precision, etc. so my brain is a bit foggy now.

Let's suppose I have a PT100 temperature probe, for its great precision (when compared to K thermocouples) and extended usable range( when compared to digital i2c temp. sensors)
Now I have to reach a microcontroller, which is several meters away so naturally I had the idea to use a PT100/4-20mA transmitter, which seems to be pretty much the industry standard, at least in France.

So I saw that there are two main solutions : measuring voltage across a precision 100R or 250R resistor directly with an ADC, or using a small converter based on Op-Amps, such as this, or even this single chip receiver before the ADC.

(btw if you know another great solution, or another chip like the rcv420, please let me know :-) )

what would be the most precise architecture between the two (or more) ? Or the best in regard as other criteria of your choice ? The rcv420 claims 0.1% accuracy, is it better than using a 0.1% resistor directly ? Maybe the op-amp based bridge is better when it comes to input protection ?

So I'd like to have your opinion !

Thanks
JF

[Paul Moir]

If it's "several meters" and not "many, many meters", you can just use RTD extension wire.  Use 3 or 4 wire sensing depending on what kind of accuracy you need to preserve.

Going to a current loop seems like a complication to me.
 

[ftransform]

why not just use a current source and a kelvin connection?
pt 1000 rtd's are nice

[nctnico]

A current source can be inaccurate and noisy. I'd use a precision voltage reference and a precision resistor to feed some current through a PT1000 element. To cancel the wiring resistance it will take 4 wires (Kelvin connection). There are many high precision ADCs with differential inputs and several have the ability to multiply the incoming signal.

[ftransform]

 

whatever do you mean?

[Paul Moir]

RTDs are often connected with three wires in industrial settings.  The way it works is that you *assume* all three wires have equal resistance, so one gets doubled up for carrying current and measurement.  That way you save the cost of one wire, which can be a good compromise as long as the assumption tolerance is well below your sensor tolerance.
Example:
http://www.pyromation.com/Downloads/Doc/Training_RTD_Theory.pdf

[JeanF]

Thanks everyone for these informations, and thanks Paul for this good document ! I'll definitely take a look at this idea of reading directly the voltage across the RTD, which seems very attractive indeed, as it saves the cost of one current-loop transmitter per probe, and gets rid of the transmitter accuracy issue.

In these cases (3 or 4 wires), how do you produce the reference current ? Is this current also monitored ? (If it is, how is it done ? A precision shunt resistor + another channel of a differential ADC ?) 

this is going to be used on various spots of a Rankine cycle test bench in my school, and there will be many more than one RTD, so it would be great if adding new channels on the go was easy (hence my question about the precise reference current generation)

Also, there will be 3-phase motors and frequency controllers not so far away, so if you know the best practice to avoid parasites, please tell me 

[ftransform]

I really don't understand why you would need to save on several feet of 28 gauge wire industrially...
Do they have some sort of environments which have 5000 rtd's?

[nctnico]

Thanks everyone for these informations, and thanks Paul for this good document ! I'll definitely take a look at this idea of reading directly the voltage across the RTD, which seems very attractive indeed, as it saves the cost of one current-loop transmitter per probe, and gets rid of the transmitter accuracy issue.

What kind of accuracy do you need? That is the first question you need to answer. Feeding a PT1000 from a 2.2k resistor from 3.3V will work just fine for centrigrade resolution and a few meters of wire.

[SeanB]

I really don't understand why you would need to save on several feet of 28 gauge wire industrially...
Do they have some sort of environments which have 5000 rtd's?

Think of a refinery fractioning column, 50m high. There you will find a temperature sensor every few centimetres so that they can control the separation of the various fractions by controlling the pressure and flow so that the condensing vapours fall into the right trays. Save 1 wire times the number of sensors adds up.

Normally you use a hybrid pretrimmed to give the accuracy desired, and it has all the active circuitry, the 3 opamps, precision resistors and reference voltage source all on the same substrate so that it is a precalibrated plugin that needs no on the line calibration when installing or replacing.

[cybergibbons]

Surely the wire cost is insignificant compared to the pocket, sensor, and other infrastructure.

[ftransform]

Would there ever be a benefit to using a discrete solution or are those all-in-one substrate packages the best you can get?

[JeanF]

As these sensors would be used to precisely determine thermal efficiencies on a test bench in thermodynamics, I guess 0.1°C resolution would be a bare minimum, but I don't know what's the relative extra cost of getting more digits with a known precision...

As for precision, I don't really know what can be reasonably expected... I'm new to this instrumentation world, and as we are equipping a test bench, there are not many examples easily available to get inspiration... the other laboratories keep their little secrets well too. Sadly, it looks like the decision makers (ie the engineering school) don't care either... All they want is to rely entirely on a supplier and recklessy accept poorly drawn quotes. So with 2 teammates we want to find alternative solutions

Sean, do you have an example of these readily built hybrid modules ?

[poorchava]

LMP90098.

Datasheet also contains some really good application examples. Including RTD


http://www.ti.com/product/lmp90098&lpos=Middle_Container&lid=Alternative_Devices

[SeanB]

Thanks Poorchava, the right sort of interface. Consider the wiring will typically be a multicore cable costing $$ per metre, as it has to have individually shielded triplets of cable. It has to survive from -55c to +500C along with being exposed to ultra rough outdoors and extreme swings, all while adding no noise to a millivolt signal, along with having high current cabling right next to it in a cable bundle. Saving 25% will definitely be worth it, as the hybrids will work with a 3 wire or 4 wire connection.

[ftransform]

Wouldent you fix a standard cable to it once its been out of the hot zone?

also, will those modules beat out a discrete solution (differential amplifier, 2 op-amp current soruce, adc)


I'm just wondering because I am curious why companies like TI would include appnotes for making electronics with discrete opamps... the amplifiers in the app-notes I saw with differential amplifier, ADC, voltage source, filters, reference would probably go for 40$+.
Are they just shamelessly trying to rip us off considering they sell single chip solutions which do it for 3-5$?

[JeanF]

thank you very much to everyone who replied ! These chips have some really interesting datasheets indeed. I definitely need to take a further look inside...

If you know a good book or pdf more or less related to the topic, don't hesitate to suggest, as I'm always looking for useful bits of information and the web is, well, very wide...

JF