Temperature calibration practicalities - help!

[DTJ]

I've got a temperature sensor based on a digital temperature sensor IC that I am trying to accuracy test.

It's housed in a 1/2" (12.5mm) stainless tube and potted in polyurethane. The sensor is close to the end of  the tube housing and is specified to be accurate to <0.1°C


I'm trying to check its accuracy over the range of say 0°C to 90°C.

The sensor is placed in an aluminium block along with a calibrated PT-100 reference sensor.
This is immersed in hot water inside an insulated cooler. The top of the cooler is covered to slow cooling.
I have a propeller on the end of a plastic shaft spinning at around 1.5Hz to stir the water.
The set up is shown in the photo below (with the aluminium block shown lifted clear of the water).

I'm still getting a temperature difference between the two sensors of around 0.3°C or 0.4°C. 
The error seems fairly constant - at 80°, 65° and 45° it's around 0.3°C.
I'm still waiting for it to cool to see what it does at ambient but I suspect it will still be off by ~0.3°.


If I get an un-potted sensor and thermally bond it to the reference sensor using some machined brass rod and some silicon heat sink compound they track within about 0.05°. 
This assembly is wrapped in bubble wrap and buried in foam inside a cardboard box and left to stabilise at room temperature.

If I simply put the sensors against one another and again insulate them well I will see an error of say 0.3C or 0.4°C, the thermal bonding seems to work.



So why can't I get the sensor under test to match the reference sensor when immersed in water?
I guess it could be that there really is an error of 0.3° at elevated temperatures but I doubt it as I expect the sensors to comply reasonably well with the data sheet specs.

I can't bond an un-potted sensor IC to the reference sensor and immerse it in water for obvious reasons.



What are people experiences in trying to do this??   

[Ian.M]

Self-heating + poor thermal conductivity of the polyurethane potting compound?

Try a constant temperature bath e.g an melting ice-bath, and also a well insulated water bath at ambient to eliminate errors due to the thermal lag between the isothermal block and the sensor and the decreasing bath temperature.

[Kleinstein]

Both sensors look like they can conduct quit a lot of heat through there connection / tube. So usually the sensor should be emersed quite deep. Just a few inches in the AL block might not be enough.

With a water bath at higher temperature, there can be quite some colder temperature at the top because of evaporation cooling. This is why higher temperature baths prefer oil.

[DTJ]

Self-heating + poor thermal conductivity of the polyurethane potting compound?

Try a constant temperature bath e.g an melting ce-bath, and also a well insulated water bath at ambient to eliminate errors due to the thermal lag between the isothermal block and the sensor and the decreasing bath temperature.

Self heating isn't an issue - it's asleep pretty much all of the time and only wakes for 10's of mS before sleeping again.
The PU potting delays response but I figure as long at the rate of cooling of the whole bath is much much slower than the heating delay through the potting I should get reasonable results. The PU isn't too bad, the step response time constant is around 30 seconds to 63%.


I'm happy the sensor is accurate at room temperature, I just want to confirm it at higher and lower temps.
I'll try an ice bath tomorrow and see how it goes.
Cheers.

[DTJ]

Both sensors look like they can conduct quit a lot of heat through there connection / tube. So usually the sensor should be emersed quite deep. Just a few inches in the AL block might not be enough.

With a water bath at higher temperature, there can be quite some colder temperature at the top because of evaporation cooling. This is why higher temperature baths prefer oil.

When it's under test the aluminium block is about 50mm below the water surface. I handn't considered evaporation cooling - I'd sort of hoped having the water mixed constantly might minimise any thermoclines.

I should try floating an 20mm styrofoam sheet on the water surface.

Oil might be the way to go - very messy!

[DTJ]

OK - progress!

After all this it seems I have a faulty sensor that under-reads ~ 0.3°C.

I tried the ice-point test as per Ian's suggestion and it read 0.3°C low so I put a bare PCB prototype in some heat-shrink and tested that and it read within 0.02° of the reference sensor.

On the plus side I found a great video on creating a good ice-point reference:



and reference sheet:

https://www.measurement.govt.nz/sites/all/files/training-manuals/TG01-July-2009.pdf

I'm not sure why the sensor reads low. All I can think of is that I damaged it during the soldering process.