Thermal fluctuations on PCB

[m k]

Quite often I see a quite stable oscillation with a 5-8 seconds period.

Is insulating/delaying only one leg a normal thing?

Then you could at least test how 5 deg tilt is different, if any.

[Kleinstein]

Here are the data from the experiment with tilting the box just a little (~ 5 degree). Depending on the orientation more of the heat from the regulator ( ~ 30-50% of the total heat) reaches the sensor that is nearly in the opposite corner. So after a tilt there is some dirft in the overall temperature. After a chane it takes some time for the oscillation to come up.

[dietert1]

I found this paper: https://ar5iv.labs.arxiv.org/html/0705.1198.
From what i understood the mechanism starts with unstable horizontal layers (inversion = heat on top, cold at the bottom). They break up spontaneously and mix. The flux caused by this triggers the next breakup when it returns to the layer separation after completing its convection path. I think layering will be more with the heater at the top.
In order to confirm that you see a thermal wave emitted by unstable convection you could place a second NTC on the opposite side of your box. It should detect a similar oscillation yet of opposite sign (180° phase shift).

Regards, Dieter

[Kleinstein]

A hot layer on the top would be a stable situation. The problem is getting a hot area on the bottom, especially pockets of hot air traped under horizontal PCB or chips and cooler at the top. For some time this may not be enough to start the convection, but once going it can move for some time.

The other possible problem that I see is that the ideal, steady state convection cells have a given lateral size of some 1.5 (AFAIR) the hight. But the lateral size may not allow the ideal sizes and it may not take much to move the boundery between 2 convection cells. Moving the boundery can effect the local temperature quite a bit.

In my case I don't think it is a single convecion cell and thus simple phase relation. On how it reacts on added shilds / cloths, it is more like an oscillating part near the regulator and than small parts of this still reaching the NTC, more like a wave effecting a 2nd convetion cell more over the main PCB. Shielding more around the NTC had limited effect (similar frequency, less amplitude), changing things around the regulator made a larger difference, especially effecting the frequency.

For me the lesson learned for future designs is to subdivide the case / shilding boxes and if possibly mount strong heat sources directly to the case instead of relying on conveciton. A smaller case probably also helps. In many cases it could be hard to avoid convetion all together, as it is just needed to get the heat away from even a relatively low power PCB. After all a fan may not so bad.

[doktor pyta]

1. Are You sure that diode excitation current is free from these oscillations?
2. Are You sure that diode is not demodulating some nearby WiFi/GSM signal / maybe neighbor  just behind the wall?
3. Air conditioning influence or other temperature regulator ?

[dietert1]

A hot layer on the top would be a stable situation. The problem is getting a hot area on the bottom, ...

The measurement you presented starts without oscillation. This section is labeled "NTC side up". As you wrote the heat source is on the opposite side, this means it is below the NTC. So this is the situation without inversion and with steady convection (no oscillation).
Then you tilt the other way and oscillation happens. This is with the heat source up.

Regards, Dieter

[Kleinstein]

1. Are You sure that diode excitation current is free from these oscillations?
2. Are You sure that diode is not demodulating some nearby WiFi/GSM signal / maybe neighbor  just behind the wall?
3. Air conditioning influence or other temperature regulator ?

The diode current comes with a resistor from the 14 V reference to the ADC. So that should be pressy stable. The NTC also showed the same (or at least similar) frequency.
No air conditioning (it is winter anyway) and the heater system is way slower than 0.2 Hz to make any difference across the room.
The circuit is inside a mainly close metal box - though with an USB cable coming in and the PC using WLAN. Still I don't expect 0.2 Hz modulation from WLAN / GSM - that is usually faster. My initial suspect on low frequency oscillations was the DCDC converter as it uses some frequency modulation/ spread spectrum ( ~50 Hz range) that might hit a beat frequency with mains or the ADC integration time.

The effect reacts quite a bit to changes to the thermal (e.g. tilting). The other indication was the test with more voltage and thus heat at the regulator increasing the amplitude. So I have no doubt it is  thermal.

The NTC side up, it still only a moderate tilt up. The actual regulator part of the heat source was still about the same hight overall. The temperature likely drifted up in that phase because the heat could spread out lateral better. Convection was for sure still active, just not oscillating. Even with the heat source side up, there is still air between the heat source and the top of the cause, so still room for convection, just with a different pattern. The overall temperature drops, as the warm air would not reach the other side that well.
Why it still oscillated with tilt in one direction and not with the other, I don't know - my vague expectation was more of the opposite. The point is not so much how it changed, but that is changed significantly  with only a small tilt.

The convection can be stable, oscillating or with chaotic variations. The transitions are already hard to understand for a simple geomity like in the scientific papers.
The geometry with the PCB is too complicated to fully understand or model it. Fluid dynamics is already tricky with a simple geometry.
Especially in the chaotic case it only takes minute changes get quite different results. My system looks like is somethat oscillating, but not perfectly stable and possibly close to a chaotic range.

[m k]

Late sides of to flat peaks are quite equal.
Same with early side of start and end of NTC down, but up peak of tilt to flat has no obvious oscillation.

Have you tried closed case without screws?