NTC thermistor grid design

[zoltan]

Hi,

My plan is to install a temperature sensor grid under the (soon to be installed) styrofoam insulation of my house, to monitor the thermal performance. The sensors would be installed between the brick wall and a styrofoam in a 1x1m pattern.
In total 7 rows and 40 columns. I will try and equalize the wire length to cancel out errors due to cable resistance difference, and will use a somewhat large cross section (0.75mm is around 350 miliohm, 2.5mm around 100) wire.
In order to use a single channel ADC I'll feed all the wires trough a reed relay multiplexer.

The resolution I want is 0.1K or better, but I'll be satisfied with less (up to 0.5K) if it's budget breaker. The reading period not less than 1 per minute, probably once per 15 minutes. Ideally, the temperature range will sit between 21-24C, absolute min-max is -15..30C.

As you can see, there is a lot of sensors (280) to measure, measuring time, accuracy and precission are not critical, but price is. I cannot afford 2-3$ per piece, so I bought the cheap ones from ebay (MF58 3950 B 10K 5% NTC). The same goes for the relays, I'll use 47x SIL05-BV07056 in the multiplexer (they are cheap, with around 150 miliohm contact resistance).

The sensors will be mounted on a simple +/- 20x20 mm PCB with a screw hole and wire solder points. All enclosed in some sort of goo. Attached is a sketch of the board.

For readout I'll use a dedicated ADC and a micro, but for now I'll just put my Keithley 2015 to cicle through the sensors (in sync with the multiplexer).

My main concern is the reliability of the used sensors and relays. Never did something like this and as it is a single shot job, after the insulation is glued in there will be no more access except via the dangling wire in the atic. And any mistakes stay buried and unrepairable.

All the suggestions are welcome. Just keep in mind, it is a low budget project, I can't afford to put hundreds or thousands of euros in this.

[Kleinstein]

I don't think one can easily connect NTCs in a matirx like form. There would always be parallel paths for the current. It may work with quite some math to get the back the individual resistors, but this would be quite some math and it would increase the errors, if it works at all.

Instead of using NTCs one could consider using diodes as sensors - these can be used in a matrix. With a large number the leakage currents may still sum up, but chances are they are OK, even with not so ideal 1N4148. Similar to high tolerance NTCs the diodes may need individual calibration, though there is an option to use 2 currents and this way get away without this step.

In both cases the weak point is less the resolution / noise, but more with the calibration and long time stability.

Chances are the switching part could be still accessible and thus repairable if needed. So this part would be less critical.

[Twoflower]

Sending analog signals through long wires is possible but of course they'll have an influence to the measured value as well. And the wires might pick up noise easily I personally would avoid that. If you really want to use NTCs I would digitize the signal close to the temp sensor. But that would be far too expensive.

If you're not stuck to the NTC solution, I would spend some time checking for the Maxim 1-wire temp sensors (for example MAX31820). They reach up to +/-0.5K accuracy I would not expect the NTC solution will be better. Given they're more expensive ($1.54 at mouser), but they still might fit the bill as you can use cheaper wire without concern that this will influence your measurement and much less wiring as they using one wire for all. But keep in mind that this might also be a single point of failure. Using multiple independent strings interleaved to catch you if a full string dies would be best.

Also with decent solder quality and heat-shrinking  you might also skip the PCB. Unless you need that for mounting the sensors (there might be cheaper solutions).

[zoltan]

I don't think one can easily connect NTCs in a matirx like form. There would always be parallel paths for the current.

I was thinking of relay switched matrix. When reading, only one row and one column relay would be active at a time, no parallel connections. But, as Twoflower said, lot of antenna in this solution.

Instead of using NTCs one could consider using diodes as sensors - these can be used in a matrix. With a large number the leakage currents may still sum up, but chances are they are OK, even with not so ideal 1N4148.

This was my first idea, but I was under the impression it would be easier/less prone to noise with high resistance NTC.

In both cases the weak point is less the resolution / noise, but more with the calibration and long time stability.

I don't need very high accuracy, and I can live with less resolution, more important is the stability of the system. It can be 5C off, but must keep it like that for years. I can also make a control array that can be accessible later for the purpose of re-calibration (assuming all NTC's would age at similar rate).

Chances are the switching part could be still accessible and thus repairable if needed. So this part would be less critical.

Yes, I plan to install it in the attic.

[zoltan]

Sending analog signals through long wires is possible but of course they'll have an influence to the measured value as well. And the wires might pick up noise easily I personally would avoid that.

I completely ignored the fact that 20 meter wire will act as a nice antenna. Yeah, that will work until first thunderstorm. I think there is no other option than to bring a shielded cable to each sensor. But that bring another issue, now I need 280 ADC channels, or that many relays.

If you really want to use NTCs I would digitize the signal close to the temp sensor. But that would be far too expensive.

Yes, I already have them. Found a ADC128S102 for less than $4 a piece, I think I'll use them to create a 8 sensor node, then  send the readings via modbus to the master.
And use a J-Y(ST)Y cable for each sensor. We use this cable at work a lot to connect different sensors with PLCs - mostly the passive Ni1000 or active 0-10V. It works fine over 50-60 meters. In this configuration I need max 8 meters.

If you're not stuck to the NTC solution, I would spend some time checking for the Maxim 1-wire temp sensors (for example MAX31820).

Out of stock almost everywhere. Where not, they price is crazy.

But keep in mind that this might also be a single point of failure. Using multiple independent strings interleaved to catch you if a full string dies would be best.

If I use a shielded cable to each sensor this risk is solved. The problem is hovever the reliability of a sensor. Will they survive for decades or I'll lose half of them within months?

Also with decent solder quality and heat-shrinking  you might also skip the PCB. Unless you need that for mounting the sensors (there might be cheaper solutions).

I needed it to "fix" the matrix until the insulation is glued in place. But if I use a single cable for each sensor this is no issue anymore.

I will also sprinkle a few DS18B20 around, I can probably use them to periodically check for drift in NTC's, assuming the DS18B20 will stay stable and alive.

[mikerj]

Just wondering how you can determine the performance of the insulation with only sensors behind it?  Wouldn't you also need a 1x1m grid of sensors in front so you can actually measure temperature delta?  If you are assuming the temperature in front of the insulation is constant across the wall (which seems unlikely), why can't you also assume the temperature is fairly constant behind the insulation and reduce the number of sensors?

[Kleinstein]

I don't think one can easily connect NTCs in a matirx like form. There would always be parallel paths for the current.

I was thinking of relay switched matrix. When reading, only one row and one column relay would be active at a time, no parallel connections. But, as Twoflower said, lot of antenna in this solution.

I think I understand the idea of the grid with relais to switch the raws and columns, a bit like a LED matrix. However there would be still parallel paths for the current, not just the one between the active column and active row. other paths would include 3 resistors in series. It really needed one could suppress the other paths, by grounding the non active colums, but this would complicate the readout as there would be additional resistors to ground at the read put line. This would make is much more sensitive to cable resistance.

The grid solution only works much better with diodes, so the other paths would have at least 1 diode in reverse direction.

Diodes as a sensor are relatively easy on the cables. One can use a moderate current (e.g. some 100 µA) and still have a relatively low differential ouput resistance  ( ~ 26 mV/I   = 260 Ohm) for the sensor signal. It would need some 20 Ohms of cable resistance to get 2 mV and thus 1 K of error.

The glass encapsulated diodes are also extected to be relatively stable, as they are sealed from humidity.  There are also 2 ways to read the diodes: one is the absolute votlage at a fixed current and the 2nd way is the difference with 2 currents of e.g. 20 µA and 100 µA. That voltage change is about proportional to the temperature in Kelvin with a fixed ln(I1/I2)*k/e slope.

The antenna action can however be a problem, especially with a strong AM radio station.
How bad spikes with a thunder storm are, depends on roof above. It would mainly effect the read-out circuit. 

[T3sl4co1l]

Use RTDs.  NTCs just aren't that accurate.  Maybe if you take the time to calibrate them first?  But you say cost is critical, and you'll easily more than double your cost doing that.

Three or four wire connections are preferred of course, and they'll have to be well sealed (IP67 or whatever) to sit in an environment like that.  Which probably comes to something like $20+ per sensor?  Probably an industrial or lab receiver/DAQ unit will do to gather them all together, just a few thousand bucks extra.  Yeah, it can be relay muxed, nothing wrong with that, but you don't save much wiring (length or number of connections) unless the relays are distributed with the sensors, raising costs even further.

It's not going to be a cheap installation, but as a research project / demonstration, you only ever have to do it once, at least.  And you can get some back with whatever papers/grants/services you can generate from it I guess.  This... this IS some kind of demo thing, right?  You don't actually mean to just casually monitor the interior wall temperature of your own house, just because, right?  'Cuz there are MUCH better ways to do that...

Tim