thermistors in parallel

lørdag den 6. maj 2023 kl. 04.10.39 UTC+2 skrev John Larkin:

if you have enough pins on an FPGA, make 8 ADCs?

or use a cheap 8 channels SPI ADC

Why bother with an ADC? If you know the threshold for speedup, just a comparator is sufficient; dual transistor per thermistor, and a single DAC output can do the adjustment, while collectors-connected-together is your OR-invert function.

Extra functionality comes if you dither the threshold for the various sections, or separate the multiple logic inputs instead of just taking the OR combination.

But piece-wise linear.

The hottest thermistor will have more influence than the cooler parts - if they are interchangeable thermistors - but "dominate" isn't really the right word.

Better.

IMO A simple NO temp switch, one on each module, wired in parallel would do that.

Shame that gives nothing for JL to boast or bullshit about.

.... Phil

The FPGA LVDS inputs make decent RRI comparators, at the cost of two pins each.

It's nice to be able to read real temperatures, but the the fan control algorithm is just a bang-bang thing and the only thing it needs to know is one bit, is this board too hot? So 8 thermistors into LVDS or even single-ended FPGA gpio's would actually work. Or 8 comparators.

Paralleling thermistors does an interesting nonlinear (aka complex) weighted average of my 4 or maybe 8 temperatures.

is pretty radical.

We have an 8-channel BIST mux and ADC on every board in this family; they share a common "template" starting-point schematic and layout. I guess I could add another analog mux ahead of one of the unused BIST inputs and have the FPGA guy work that into his scan logic. The FPGA does the 8-channel analog scan and stashes the results in registers; he'd just have to add 8 more sub-scan thermistor registers. Share the pullup resistor. Easy for me to say.

It is appealing to know all 8 temperatures. Each is a cluster of wirewound resistors on a section of a PCB. Without enough air flow, they will eventually toast the FR4.

The board already has a multiplexed BIST ADC with three available inputs.

Yes, but it would be nice if, in development, we could know the actual numbers, if that was basically free.

Eight separate measurements is appealing. Paralleling thermistors is interesting but maybe too tricky. The idea might be interesting in some other situation.

The wirewpund resistors can run at 250c. The hazard is melting solder joints and, longterm, toasting the FR4. I'm sure you have seen sections of PCBs under resistors that have turned brown and got crispy. That's what I need to avoid, and there is no precise criterion for the time-temperature profile that's safe.

These will be clusters of the stand-up rectangular ceramic 5-watt wirewound resistors. I want to run them at 7 watts, which should be fine with lots of air flow.

The goal of the fan control is to limit the temp of the hottest part in the box, at minimum fan noise when nothing is too hot. What's wrong with that?

Absolutely. Sometimes it takes a lot of engineering to make things simple.

I'd love to have a surface-mount 0603-size temperature switch. I could use it in lots of places. A thermistor comes close.

I find it helpful to discuss designs; I get ideas and it helps me think. And this discussion group is sci.electronics.design.

Try designing electronics. It's fun.

For a pretty simple minded idea of "working".

It even makes sense. A single over hot resistor can cook it's segment of FR4 and that segment stays cooked forever.

Fairly obviously.

Forced Air Cooling power is a trade-off with dust collection and nuisance noise. Proportional Control is ideal but accurate hotspot sensing is required. Active cooling is critical to air velocity over the surface and not volumetric air flow. Thus turbulent air flow enhances cooling rate more than laminar flow well above the hotspot.

Thus to solved this design problem, you need a thermal sensor for each hotspot, which could be an LM35 or a diode or a thermistor epoxied to each hotpot and routed to a shared single point with an LDO controlled fan speed by shunting the ADJ voltage-controlled. Diode OR linear control is one method or current pumped by thermal sensor. Fans are Vdc speed control using Hall sensor commutation so PWM causes problems but LDO control works fine.

I discovered using smoke & thermocouple tests how using air velocity design optimized cooling. So I made an insulating material plenum over the PCB to reduce the aperture to increases the air velocity for all flow and create eddy current effects over the hotspots to significantly lower case temperature. I used this on a 180W 1U high 19" rack design I did for Lucent/Avaya.

Tony Stewart EE since 1975.

But there is an obligation to do at least some homework before you expose your rather low level thinking here.

Not knowing about the Steinhart-Hart relationship suggests that you thinking isn't up to much.

Getting that kind of idea might be helpful.

John Larkin's favourite put-down. The joke is that he seems to evolve his electronics, rather than design his circuits.

Use 1N4148 (or equivalent) diodes in parallel and a constant feed of

Our boxes are mostly used in aerospace test, in clean control rooms.

A bunch of our gear runs in the world's biggest clean room. We don't use air filters and RMAs are not dusty. We're lucky in that respect.

My intent is precisely that, namely to sense the temperature of eight hot spots on this particular board and adjust the enclosure fans to limit the worst temperature. The thermistors will be 0603 surface-mount parts.

There was some debate about the fan control algorithm, given the variety of plugin boards. Some of my guys wanted to do classic PID control which, predictably, turned out to be a nightmare. We're back to the simple up-down counter approach.

The two 48-volt fans each have a PWM input pin for speed control. At

We will have two fans on the front panel. One problem is that that [1] makes two fire hoses of air onto some boards and leaves some with no air. So we'll have a flat perforated baffle to redistribute the air to the eight modules.

lørdag den 6. maj 2023 kl. 15.55.16 UTC+2 skrev John Larkin:

a ninth output to make a ramp or use IOs and capacitors, measure the recharge time that's basically how the original PC joystick port measured ~100k potmeters (with a 558)

That would behave similar to the parallel thermistors. The diode drops would be about -2 mV per degree c and the current per diode would be about 60 mV per decade. The result is a weighted temperature average.

Thanks for the suggestion. We have lots of single and dual surface-mount diodes. Dual would double the signal.

That would work. I designed some electric meters that used single-slope a/d conversion. Single slope has perfect differential linearity so is ideal for dithering and signal averaging, which an electric meter does a lot of. The nuclear spectroscopy folks used to do that, maybe still do.

But I do have a couple of spare ADC channels.

lørdag den 6. maj 2023 kl. 17.56.04 UTC+2 skrev John Larkin:

maybe scan by grounding each diode in turn with an FPGA output?

Version 4 SHEET 1 920 820 WIRE 192 16 32 16 WIRE 464 16 272 16 WIRE 640 16 464 16 WIRE 848 16 640 16 WIRE 32 32 32 16 WIRE 272 48 272 16 WIRE 464 48 464 16 WIRE 640 48 640 16 WIRE 272 192 272 112 WIRE 464 192 464 112 WIRE 640 192 640 112 WIRE 272 304 272 272 WIRE 464 304 464 272 WIRE 464 304 272 304 WIRE 640 304 640 272 WIRE 640 304 464 304 FLAG 32 112 0 FLAG 272 304 0 FLAG 848 16 Vout SYMBOL diode 256 48 R0 WINDOW 123 22 83 Left 2 SYMATTR InstName D1 SYMATTR Value 1N4148 SYMATTR Value2 temp={T} SYMBOL res 176 32 R270 WINDOW 0 32 56 VTop 2 WINDOW 3 0 56 VBottom 2 SYMATTR InstName R1 SYMATTR Value 50k SYMBOL voltage 32 16 R0 WINDOW 123 0 0 Left 0 WINDOW 39 0 0 Left 0 SYMATTR InstName V1 SYMATTR Value 3 SYMBOL diode 448 48 R0 SYMATTR InstName D3 SYMATTR Value 1N4148 SYMBOL voltage 272 176 R0 WINDOW 3 -5 161 VRight 2 WINDOW 123 0 0 Left 0 WINDOW 39 0 0 Left 0 SYMATTR InstName V2 SYMATTR Value PULSE(3 0 0 1u 1u 1m 3m) SYMBOL diode 624 48 R0 SYMATTR InstName D2 SYMATTR Value 1N4148 SYMBOL voltage 464 176 R0 WINDOW 3 -5 161 VRight 2 WINDOW 123 0 0 Left 0 WINDOW 39 0 0 Left 0 SYMATTR InstName V3 SYMATTR Value PULSE(3 0 1m 1u 1u 1m 3m) SYMBOL voltage 640 176 R0 WINDOW 3 -5 161 VRight 2 WINDOW 123 0 0 Left 0 WINDOW 39 0 0 Left 0 SYMATTR InstName V4 SYMATTR Value PULSE(3 0 2m 1u 1u 1m 3m) TEXT -2 296 Left 2 !.tran 0 10m 1u TEXT 208 -200 Left 2 !.step param T 25 100 5

The old-school approach was standoff posts to mount hot parts; if you want it to be printed-wiring, though, and are getting to softening of the solder joints, maybe perforating the printed wiring board, and having airflow THROUGH the board rather than parallel to it, is the answer.

Cool air entering from the wiring side, hot parts on the component side, will let the solder joints chill while the resistors do the work.

Lead-free solder will help as well. John

We have a pick-and-place machine for surface-mount parts and a selective solder machine for thru-hole; zero hand wiring. We can have surface mounts on the bottom too, if they are not too close to leaded parts.

if you want

Only parallel in our case. Cutting holes in the board reduces heat spreading in the copper planes anyhow.

The resistors do have a sort of tunnel on the bottom that might help cool the pads a bit. This style:

I'll have air flow, up to 200 LFPM, on both sides of the board. Big copper pours on the bottom side will help cool the resistor leads.