Thermocouple thermometer design

[gaminn]

Hi,
I'm designing thermocouple thermometer. It will consist of several thermocouples which measure temperature on a single metal plate. There is temperature gradient on the metal plate so each thermocouple is on different temperature.

I'm considering making galvanic isolation (on the digital side - after A/D converter) of each thermocouple channel because I'm not sure whether one thermocouple can't influence another thermocouple as all thermocouples are shorted on the metal plate and also all positive legs of each thermocouple are shorted via single BIAS voltage (regarding BIAS voltage I'm thinking about making separate BIAS voltage sources - separate voltage dividers - for each thermocouple if it helps as then thermocouple positive legs won't be shorted together via BIAS voltage).

I would like to avoid galvanic isolation because the device will be battery powered and the galvanic isolation will increase power consumption drastically. But of course, measurement accuracy of the device is the highest priority.

The galvanic isolation is not necessary due to different potential problem - the device is battery powered. It is isolated and completely floating.

Please see picture. I appreciate any comments.

[Kleinstein]

If the thermocouples are connected to the metal, you can not connect the "bias" / DV voltage level of the different thermocouples. This would mix the different TC. It can be acceptable if there are extra series resistors (e.g. 10-100K range) for the DC level of each Thermo couple.

Extra galvanic isolation only makes sense, if the circuits are fully isolated, so separate references and so on. Only the cold temperature measurement  might be common.

Often an external amplifier is better than the input of an ADC. But this depends on the ADC.

With many TC there is also the question if one wants a separate ADC for every channel or can use a single high quality amplifier and ADC and than use a CMOS multiplexer to select one channel at a time. It depends on the required noise and accuracy specs. A single ADC good tends to be more accurate (need only one and can do an extra 0 measurement), but more noisy. It could also make the connection to the outside easier.

[EmmanuelFaure]

Your schematic is fine, I'd just balance the impedance between the "+" and "-" ADC inputs. Mismatched impedance can cause extra offset and/or extra noise with high resolution ADCs.

BTW, why don't you buy this? It's standard and available off the shelf.

[gaminn]

Thanks! I will make separate bias voltage fot the thermocouples.

I'm designing this, because I need long battery life and wireless communication.

[HighVoltage]

May be not what you are looking for, but I came across a nice Cold-Junction Compensated Thermocouple-to-Digital Converter by MAXIM, the MAX31855

[gaminn]

Thanks. I already tried this chip and it is nice. However, with A/D converter I'm able to achieve lower power consumption. With integrated thermocouple converter chip it is also problematic to place the chip on the same temperature as cold junction sees.

[Kleinstein]

If it is for many channels and low power, it might be a good idea to use multiplexing and thus measure one temperature at a time. This could also provide some isolation of the TC from each other, if both sides are switched.

With many channels it makes sense to do the cold side compensation by hand.

[gaminn]

The device has 20 channels and they all have to sample simultaneously in exactly the same time with T = 1 s. I considered multiplexing and then extrapolating measured temperatures to the same time point but using 1 A/D converter per channel has better noise performance in my case (as the A/D converter can average more) and it doesn't add anything to power consumption - A/D converter has very low sleep power consumption.

Good point about the multiplexing is better isolation between channels...

[EmmanuelFaure]

How much accuracy (In °C) is needed by your application?

Could you tell me more about the application? Being myself in the instrumentation field, it's pure curiosity.

[retrolefty]

Unless your operating temps are over 1000F (538C) or so thermocouple measurements are not the best choice for  accuracy. Look into higher temperature rated RTDs or other more accurate temp sensor types.

[EmmanuelFaure]

Retrolefty, you might appreciate this : (This document is not well referenced in google)
http://gecinstruments.com/Ultra%20Accurate%20TC%20Measurements.pdf

[retrolefty]

Retrolefty, you might appreciate this : (This document is not well referenced in google)
http://gecinstruments.com/Ultra%20Accurate%20TC%20Measurements.pdf

 Yes intresting and gives a good recap of all the error sources for a typical thermocouple application. Their solutions sound very expensive also. But to justify my original comment I saw this near the end:

Some customers order a highly accurate and stable thermistor probe or RTD probe
attached to their thermocouple instrument at the time of purchase, for use as a precision
temperature standard and for use in calibrating their thermocouples.

[gaminn]

EmmanuelFaure: We measure 50 - 200 °C, accuracy should be better than 0.5 °C after calibrating individual thermocouples. I (hopefully) designed the device not to introduce any significant error during analog and digital signal processing. The only significant error comes from thermocouple itself.

We use custom designed thermocouples because we need low thermal capacity of the sensor and good thermal contact on the metal plate we measure.

[Kleinstein]

There are still two important informations missing:
1) how many TC
2) how fast does the measurement need to be

Something like an RTD might be more accurate, but slower.

[gaminn]

Sampling period is 1 second, there are 20 thermocouples in the device.

[Kleinstein]

One ADC per input, and directly reading the TC, with just some ESD Protection and RF filtering might be an interesting option.  What type of ADC do you have in mind ?

However one still has to somehow combine the signals together - I2C can have difficulties with many of the same devices on a single bus. SPI wll need quite some IO lines.

Using a MUX for all 20 Channels might need really low noise amplification, but might still be an option: a low noise amplifier (e.g. OPA140 / LT1037) could be in the 5 nV/sqrt(Hz) range, so at a 25 Hz Bandwidth might in theory go down to about 25 nV RMS Noise. With including a zero (or reversed) measurement, one will not need to use an AZ amplifier and can still avoid 1/f noise. It is not simultaneous sampling, but could be twice per second (ADC using 1 PLC), so that not that much interpolation is needed.

Cheap sigma delta ADCs I found (e.g. MCP3421) are more like a 0.5-2 µV RMS noise with direct reading.

[EmmanuelFaure]

EmmanuelFaure: We measure 50 - 200 °C, accuracy should be better than 0.5 °C after calibrating individual thermocouples. I (hopefully) designed the device not to introduce any significant error during analog and digital signal processing. The only significant error comes from thermocouple itself.

0.5°C is typically 20µV at the thermocouple ends. 0.5°C over 150°C span require 300 points. There's absolutely no difficulty achieving this.

We use custom designed thermocouples because we need low thermal capacity of the sensor and good thermal contact on the metal plate we measure.

Welded/Brazed/Glued to the plate?

[gaminn]

Kleinstein: My ADCs have SPI. I use shift register to mux CS signal to them. This saves a lot of pins on my MCU.

EmmanuelFaure: They are ribbon thermocouples of custom design. As the device is portable, they touch the metal surface when attached to it.

[Doctorandus_P]

If I understand you schematic you're shorting the thermocouples to each other via the surface you're trying to measure.

Just wondering,
would it be possible for you to use your metal surface as one side of the thermocouple?
Use a battery spot welder to weld the "other" wire at the right places...
Mechanical stabitity would be an issue though.

// From Jim Williams:
http://cds.linear.com/docs/en/application-note/an28f.pdf