Why it is the most awkward kind
A permanent fault can be measured; an intermittent one tends to be absent just when you check. The equipment passes the test on the bench and fails again in operation. The difficulty is not technical but one of observation: the defect has to show itself while it is being measured.
Frequent causes
- Contacts and connections: loose terminals, cracked solder joints, connectors with oxidation or variable resistance.
- Thermal effects: a component that fails only when heating up or cooling down, because of differential expansion.
- Vibration or movement: a cable or trace that breaks contact with mechanical movement.
- Moisture or contamination: leakage paths that appear with environmental conditions.
- Noise or interference: external electrical events that coincide with the fault.
- Software or timing: race conditions that depend on the state of the system.
Strategies for capturing it
- Record the moment: note what the equipment was doing, the temperature and the conditions when it failed.
- Provoke it: apply localized heat or cold, gently flex the board or the cables, tap carefully.
- Instrument with a trigger: use an oscilloscope or recorder with event triggering to capture the signal just as it is lost.
- Measure continuously: a long recording of voltage or current reveals brief dips that a multimeter would never see.
- Change one variable at a time, to narrow down the suspect area.
When it is considered resolved
Because the fault does not always occur, a single successful test proves nothing. It is considered corrected when a plausible mechanism has been identified, the fault has been reproduced in a controlled way, and it has stopped occurring after the corrective action under conditions that used to provoke it.