Hardware Reliability: Storage & Power

The two things that kill fielded Pi devices are worn-out SD cards and bad power. Both are cheap to get right up front and expensive to discover mid-mission. This guide covers what to buy and how to operate so your AirCast device outlives the airframe.

How an SD card tells you it's dying

Consumer SD cards in a streaming device typically last 6–18 months. Failure escalates in stages:

  1. Intermittent slowness — the dashboard takes longer to load, boots get slower. The card's controller is remapping failing cells.
  2. Random write failures — settings don't persist across reboots, logs show I/O errors, the filesystem occasionally remounts read-only.
  3. Boot failure — solid red LED with no green ACT activity on power-up: the boot files landed on a dead cell. Re-flashing the same card may work once by landing on healthier cells, but the card is untrustworthy from here on.

Storage choice

Endurance is rated in TBW — terabytes written before the card is expected to fail. The classes differ by an order of magnitude:

ClassEndurance (32 GB)Use case
Consumer SD (avoid in the field)3–10 TBWBench and development only
High Endurance SD17–30 TBWField deployment — the default choice
Industrial SD30+ TBW (MLC/pSLC)Recording-heavy or always-on devices
NVMe (Pi 5 + M.2 HAT)600+ TBWLong-life operations; ~20× the endurance of an industrial SD
  • Bigger cards last longer — a 64 GB card has more wear-leveling headroom than a 16 GB one — but endurance class trumps capacity class.
  • Recording multiplies wear. 1080p H.264 at 4 Mbps writes roughly 30 MB per minute — about 1.8 GB per hour of recording. If you record on the device, buy industrial or record to a USB stick instead of the boot card.

Power: the failure that impersonates SD wear

Undervoltage is the second-biggest cause of SD corruption after wear — a brownout to 4.5 V makes SD writes fail mid-stripe in a way that looks identical to a worn-out card. If cards keep "dying" on one airframe, suspect the power rail before the card.

  • On the bench: use the official Raspberry Pi PSU — 5 V / 3 A for the Pi 4. Phone chargers work most of the time and fail at exactly the wrong moment.
  • On the airframe: use a dedicated 5 V / 3 A+ BEC for the Pi — not a hobby BEC tapped off the same rail as the ESCs. ESC switching noise on a shared rail causes brownouts that are indistinguishable from a bad PSU.
  • Wiring matters: undersized power leads drop voltage under load. Short, thick wires from the BEC to the Pi's 5 V input.

To check whether the Pi has seen undervoltage since boot, run this over SSH:

vcgencmd get_throttled

throttled=0x0 is healthy. Any other value means undervoltage or thermal throttling occurred — fix the power (or cooling) before the next flight.

Power-loss practice

Yanking power mid-write is the top cause of boot-partition corruption. The operational rules:

  • Shut down cleanly when you can: sudo shutdown -h now over SSH, wait for the green ACT LED to stop blinking entirely, then remove power.
  • When you can't control the power sequence (battery swaps, field ops), a small UPS HAT (~$25) detects supply loss and gives the Pi ~10 seconds on its own cell to shut down cleanly.

Pre-deployment checklist

  • High Endurance or Industrial SD card (32 GB+), not a leftover consumer card
  • Dedicated 5 V / 3 A+ BEC on its own rail, short thick leads — vcgencmd get_throttled reads 0x0 after a test run
  • Spare flashed SD card in the field kit — a re-flash in the field is minutes, ordering a card is days