Diagnose in layers, from the switch outward
PoE faults present as device faults — 'the camera is broken' — and are usually power or cable faults wearing a disguise. The discipline that saves hours: prove the switch port, then the cable, then the device, in that order, changing one variable at a time. Keep a known-good short patch lead and a known-good powered device (a small af camera or a PoE tester) in the kit; with those two, every fault below resolves in minutes.
Safety note before the list: standards-based PoE only energises after a handshake, so port testing is safe. The exception is legacy passive-PoE gear, which is live whenever plugged in — one more reason it is segregated and labelled in disciplined estates.
The six faults
1. No power, no link. Port dead or cable open. Swap the device to a known-good port: if it lives, the original port or its configuration is at fault (check PoE is enabled on that port — a managed switch may have it off or scheduled); if it stays dead, test the cable run, then suspect the device.
2. Link but no power. The device negotiates data but draws nothing: the device may be non-PoE (needs a splitter), the port's class may be mis-set, or the budget may be exhausted — a switch at its total budget refuses new devices while existing ones run. Check the budget arithmetic before blaming hardware.
3. Device reboots at night. The seasonal classic: IR LEDs and heaters raise draw past the port class or the remaining budget. Log the device's draw at night, and re-run the budget with night figures. Fix by rebalancing ports, raising the budget (a higher-budget version), or moving the device to an at/bt port.
4. Intermittent drops on long runs. Measure the run honestly — over 100m, or a '100m' run full of couplers and patch panels, is out of envelope. Fixes: long-distance mode, an extender, or fiber. Also check for the CCA (copper-clad aluminium) cable that crept into the install; its resistance breaks both signal and power budgets.
5. One port flapping. Water in an outdoor junction, a corroded contact, a failing device PSU stage. Swap device and cable independently; the fault follows the guilty half.
6. Whole switch cycling. Power supply distress: total load at the budget ceiling, a hot cabinet, or a failing PSU. Split the load across ports/switches, ventilate, and check the budget arithmetic — this fault is arithmetic announcing itself late.
When the switch itself is the tool
Two catalogue features exist precisely because these faults are common. The PoE watchdog (on smart models, and automatic in the AI PoE line) detects a silent powered port and restarts it — clearing fault class 'frozen device' without a visit. Managed switches expose per-port power state, draw and restart to an operator or an NMS, turning most of the six faults into console work. If your maintenance log is dominated by restarts and budget surprises, the troubleshooting is done: the network has told you its next switch tier.
Prevention: the maintenance log is a design document
Troubleshooting ends, properly, in prevention. Keep three records and most of the six faults stop recurring: a power schedule (which port feeds what, at what worst-case draw — the budget arithmetic made permanent), a run-length record for every link over 80 metres, and a change log for anything that alters either. The estates that suffer the same fault quarterly are almost always missing the record that would have made the pattern visible.
Then match hardware to the log you already have. If the log is restart-shaped, watchdog and managed tiers pay for themselves. If it is budget-shaped, your next switch is a bigger-budget version, not a different brand. If it is distance-shaped, the fix is topology — extenders, long-distance models or fiber — not another identical switch installed with fresh optimism. Tools exist on this site for each shape; the log, honestly read, is the specification for choosing among them.
The diagnostic kit: five items, most faults
Professional PoE troubleshooting is mostly preparation. The kit that resolves the six faults fits in a small case: a known-good PoE camera or tester (the negotiation probe), a 2-metre certified patch lead (the reference cable), a cable tester with length measurement (the run-length truth-teller), a clamp meter or inline PoE meter (draw, measured rather than nameplate), and the site file — power schedule, run lengths, configuration backup. With those five, the layered method (port, cable, device) becomes a sequence of substitutions, each under five minutes.
The kit's absence explains the classic multi-visit fault: technician one swaps the camera (fault stays), technician two swaps the switch (fault stays), technician three — with a tester — finds the corroded junction in eleven minutes. The moral is not that the first two were careless; it is that PoE faults live at boundaries (device/cable, cable/switch, budget/class), and boundaries are only visible to instruments and records. Buy the kit before the second visit, not after the fifth.
The method, compressed
Port, cable, device — in that order, one substitution at a time, with a known-good device and a certified patch lead doing the arguing. Read the fault's timing before its symptoms: night faults are power, weather faults are junctions, growth faults are budgets, and distance faults were born at installation and merely waited for a hot day. When the same fault returns a third time, stop fixing the instance and fix the tier — the network is specifying its own upgrade, and the maintenance log is the purchase order it has been writing all along.
One last practical note on escalation: know the two-visit rule. If a fault survives two structured visits — the layered substitutions, the kit, the records — it is no longer a device fault; it is a design fault wearing a device fault's clothes, and the correct escalation is to the design documents: the power schedule, the run lengths, the tier decision. Management hates hearing 'the network is under-specified' from a troubleshooter, but they hate the third, fourth and fifth visit more. Escalate with the evidence the method produced, and the conversation becomes short: here is the arithmetic, here is where reality exceeds it, here is the change that closes the gap.
Keep the log itself with the same seriousness as the kit: date, port, symptom, substitution made, result. Free-text war stories are where patterns go to die; a five-column log makes them visible at a glance — the same port recurring, the same season recurring, the same cabinet recurring. Most chronic PoE faults are chronic because nobody could see the recurrence. The log is the cheapest diagnostic instrument in this entire guide, and the only one that improves every time it is used.