Heater wrote: ↑
Wed Jun 27, 2018 6:44 am
They are in NEMA enclosures in the baking California sunshine.
A few year back I had one outdoor PI system fail. Turned out not to be a PI problem but the SD card went on fire. Almost literally, the SD got hot and smoked when I got it back on the bench and powered it up.
In general, the Pi2/3 should avoid plastic enclosures in industrial settings. Or you too will discover the EMI/EMC profile from the Ethernet chip-set will just barely pass FCC+CE, and fail most other ISO testing. The new pi3 does use a metal can on the SoC, but the other chips should at minimum still use a heat-sink glued on with RTV silicone (don't trust the 3M thermal tape at all temperatures).
Keep in mind our work was mostly for equipment in the Northern oceans, and needed to pass sub -17'C testing in salt-water environments. Almost all plastics (including epoxy) allow some moisture to diffuse into the enclosure over time, and this is is why the equipment we make has large bags of Silica-gel glued into the bottom of the sealed enclosures.
NEMA level certification also doesn't really cover "slow" water-ingress along wiring insulation, and this wire-rot oxidation can even occur during storage. Those expensive weather-proof connectors do offer some protection (albeit a physically brittle one), but a flexible adhesive from Loctite is still recommended in addition to the rubber seals (or at least some Vaseline). Anti-Corrosive Zinc-Paste on the connectors is also a good idea in most cases, as even slightly corroded battery terminals can get very hot.
The sun will bleach and rot most exposed labels/glues/paints within 6 months, so your mounting hardware should take this into account. Most plastics already become very brittle at low temperatures, and not all formulations include UV resistance.
The SDCARD fire could be related to a few issues, but in general... they only fail-closed like that when the supply voltages are exceeded. I would suggest scoping the Pi rail supply under operational loads. Personally, I typically supply regulator power directly into the Pi 5v pins, and use an over sized 2A 3.3v linear LDO regulator for additional sensor chips etc....
Story time, we had a TI chip based power-supply run with bad ceramic capacitors (often an expensive choice normally avoided for filtering switching transients) that would cause ringing well over ripple spec. This ring would take out the rest of the board, and was considered odd as the inductor was purposely over-sized to reduce operating temperatures. Even though the TI chip had respectable over-current protection above the hardware load max... the de-rated high-temperature MLCC could still failed closed after a few months. We even measured a surprising 8A fault through one cap's EIA3216 package. What saved the board, was the fact that we have a design rule where the first part on every module is always a real fuse calibrated for the maximum expected load capacity (we will usually also have a lower valued external in-line fuse for users). Note that ceramic fuses that fail open are usually more reliable than the poly option, but they can also corrode open in high humidity.