The Fault That Is Not There When You Arrive
A subscriber reports drops every evening. You attend, measure, and everything is within specification. You leave. It happens again that night.
Intermittent faults are the most expensive category in fiber operations, not because they are hard to fix but because they are hard to catch. The repair is usually trivial once located.
The good news is that intermittent optical faults have a small number of causes, and each one leaves a signature.
The Three Physical Causes
Temperature. Materials expand and contract. A splice under slight tension, a fiber pressed against a tray edge, or a connector in a marginally mated state will behave differently across a 20 degree swing. Symptoms follow a daily cycle — worst in the coldest hours or the hottest part of the afternoon.
Water. Water in a closure or hand-hole causes loss directly, and much worse when it freezes and expands. Symptoms follow rainfall and season, and often appear days after the rain rather than during it.
Movement. Aerial cable moving in wind, a duct cable shifting as traffic passes, a jumper being knocked in a cabinet. Symptoms are irregular and correlate with activity — wind, traffic, or someone working in a rack.
If you can establish which of the three patterns applies, you have narrowed the search enormously before touching an instrument.
Get the Timing Data First
The essential move is to stop guessing about when it happens.
Ask for exact times. Not "in the evenings" — actual timestamps. Subscriber-reported times are imprecise but even rough ones reveal a daily cycle.
Pull the ONU's own history. Most OLTs record per-ONU optical levels and loss-of-signal events with timestamps. This is the single most useful data source and it is often forgotten. A graph of received power over a week usually answers the question outright.
Correlate with weather. Temperature curve and rainfall for the affected days. This takes minutes and either confirms or eliminates two of the three causes.
Then Measure While It Is Happening
If the pattern is predictable, be there — or leave something there.
Continuous power monitoring. Leave a meter logging on a spare fiber in the same cable, or use the OLT's own per-ONU statistics. A logged level that dips at 4 a.m. and recovers by 8 a.m. localises the cause to temperature immediately.
Compare wavelengths during the event. As covered in the bend radius lesson, macrobend loss is much worse at 1550 nm. If the loss during an event is disproportionately at 1550, it is mechanical — a bend or pressure — not a break or a dirty connector.
Capture an OTDR trace during the event. Hard to arrange, and decisive when you manage it. The event appears at a specific distance and the search is over.
Where They Actually Are
In practice, intermittent faults concentrate in a few places:
Inside closures. Slack under stress, a splice at the edge of its protection sleeve, water ingress. As covered in the enclosure lesson, most closure failures are sealing failures.
At connectors. A marginally mated connector that moves with thermal cycling. Contamination that is borderline.
Where cable enters a building or a hand-hole. Transition points concentrate mechanical stress and water.
Aerial spans near attachment points. Movement is greatest where the cable is constrained.
Notice that all four are accessible without excavation. Intermittent faults are rarely mid-span cable damage — a damaged cable tends to fail and stay failed.
The Trap
The trap is declaring victory. You open a closure, find nothing obviously wrong, reseat everything, reseal it, and the fault does not recur for two weeks.
Did you fix it, or did you disturb it into a temporarily stable state? Often the latter. Which is why:
Leave monitoring in place after the repair. For a full thermal cycle at minimum, and through the next rainfall if water was suspected.
Record what you actually changed. "Reseated tray 3, resealed gland" is a real record. "Repaired" tells the next engineer nothing when it recurs in spring.
Summary
- Intermittent optical faults are almost always temperature, water, or movement, and each has a distinct timing signature.
- Get timestamps before instruments. The OLT's per-ONU power history is the most underused source.
- A 1550 nm loss much worse than 1310 during an event means mechanical stress, not contamination.
- They cluster at closures, connectors, entry points and aerial attachments — all reachable without digging.
- Disturbing a fault can mask it. Monitor through a full thermal cycle before closing the ticket.