Bob Brown . writes:

On 25 Feb 2007 04:07:26 -0800, "David Ames"
wrote:

I remember several years ago, on one of the local weather broadcasts,
a reporting station was out of service because of a defective
thermocouple. I seem to recall, also, mutterings as to whether a
thermocouple could be correctly calibrarted.

Thermocouples use the principle that a circuit with two dissimilar
metals will create a voltage that is proportional to the difference in
temperature between the two junctions. There are many standard
combinations of metals; one common one is copper-constantan, which
produces a voltage of about 40 micro-volts per Celcius degree of
difference.

To clarify: by taking one copper wire and one constantan wire and
joining them at both ends (this creates two junctions where copper and
constantan meet), you create a circuit. If you put the two junctions/ends
at different temperatures, a voltage is created. The exact voltage
difference is temperature-dependent itself, but the properties of standard
thermocouple materials are well known, and manufacturers will guarantee
how closely their wire will meet the industry standards.

Thermocouple are *great* for measuring temperature *differences*, but
knowing the difference between the two junctions tells you nothing about
what their actual temperatures are. To use a thermocouple to get an
accurate absolute temperature reading, you need to place one junction (the
"reference" junction) somewhere where you can measure its temperature can
be measured or controlled independently. A common option used as an
example is an ice-water bath (0C), and thermocouple voltage specifications
are often expressed on this basis. Ice-water baths are not particularly
practical for most people, though, so you usually put the reference
junction into somthing else and measure the temperature of that by some
other means.

You need a good volt-meter to read the thermocouple, and a good
independent measurement of reference temperature, but I've had no problems
measuring temperatures to within a few hundredths of a degree with
thermocouples, and better results just cost more money.

Thermocouples do not typically degrade over time, and
interchangeability is not an issue (due to manufacturing standards).
Accuracy is limited by the volt-meter and the reference temperature
measurement.

Thermistors, what about those?

A thermistor is a semi-conductor with a high (and negative) resistance
vs. temperature sensitivity. Again, there are a fairly wide variety of
standard sensitivities available, as well as a range of
"interchangeablities" - which means how closely do any two thermistors in
a batch have the same resistance at the same temperature. +/-20%
thermistors are pennies each, but aren't good for temperature measurement
unless you individually-calibrate them and prepare a
temperature-vs-resistance curve for each. For dollars to tens-of-dollars
each, you can get thermistors that are interchangeable to the equivalent
of 0.1C or 0.2C, which is good enough for many purposes without individual
calibration. With calibration, 0.01C is easy.

Thermistors may be used to determine reference temperatures for
thermocouple measurements.

Thermistors can degrade (and thus shift calibration) over time, so
periodic checks or calibrations are a Good Idea.

They seem accurate enough by using the known properties of
expansion/contraction rates relative to known temps.

Even better are platinum resistance sensors. Very stable, very
accurate. Very common in a lot of the automated meteorological station
sensors.

Mercury seems outdated but then again some people considered walking
outdated. Personally I find walking to be boring.

Mercury-in-glass may still be a common type of reference transfer - I
worked in a lab that had a mercury thermometer that measured from -5 to
0C, in graduations of 0.01C. It also came with a chart of corrections
between what it read and what the true temperature was, so you really
could trust it to within 0.01C. You had to make absolutely sure you
immersed it to the line engraved on it (and no further), though, as
immersing it more or less affected its calibration.