Treat a ground fault, an arc fault, an insulation or isolation alarm, and a shutdown that keeps returning as stops. Treat one low-production hour under cloud as weather. The monitoring screen is useful when it separates those. A portal that only shows a pretty daily total will hide the alarm that mattered and highlight the cloud that did not.
| What the screen says | First reading |
|---|---|
| Ground fault | Stop. Photograph the code. Do not cycle it. A qualified person looks for a path the design did not want. |
| Arc fault | Stop. The DC conductors need a qualified inspection. A trip that seemed to clear is still an event. |
| Insulation or isolation | Stop. Makers use different words for leakage toward ground. Write the word on the screen and treat it as a ground-side alarm. |
| Shutdown that returns | Stop calling it a glitch. Write the times and the code. The cause is still present. |
| One low hour while cloud is on the array | Compare with the sky. A single dull hour that tracks cloud is weather. |
A cloud hour and a real stop
Irradiance moves with cloud. A curve that dips and recovers with the sky is the array following the weather. Solar inverter faults and the weather sorts heat, cloud, and genuine faults so a hot afternoon is not filed as a dead inverter. Use that page for the production shape. Use this page for the alarm names that override the shape. A ground-fault code during a sunny hour is not a cloud. An arc-fault code at noon is a fault code during good sun.
The owner-side meaning of a ground-fault light is in what a PV ground fault means. Arc-fault protection, and the reason a nuisance trip still gets a qualified look, is in what PV arc-fault protection does. Insulation and isolation alarms sit in that ground-leakage family. Exact sensor behavior is the inverter manual. Your note should quote the screen, not a synonym you prefer.
What the log is for
A monitoring system should show production and should keep the alarms. What that system is, and what a useful one makes visible, is in what a solar monitoring system is. Export or photograph the alarm with the timestamp. A later warranty conversation, and the technician's first hour, both go faster with the list than with a memory that it beeps sometimes.
A repeated shutdown is a pattern even when each event looks small. Record the clock time, whether the grid was up, and the code. Resetting between events without the list throws away the pattern. If production is low on a clear day and no alarm is present, look at shade, soiling, a disconnected string, and clipping before you invent a failure. Those checks are on the weather and shade pages. An alarm from the table above skips that queue and goes to a qualified person.
Keep the alarm list next to the commissioning folder. A code that appeared on day three and a code that appeared after a storm are different histories, and the monitoring export is how you show the difference. What to record when solar is commissioned is the starting file. Inverter maintenance is the later visit. Neither visit should begin with a wiped fault log.
Ignore a single percentage the portal prints for the day until you know what it is comparing. A day-ahead guess, a clear-sky model, and yesterday are three different baselines. One low hour against a clear-sky model, while cloud sat on the roof, is the weather row in the table. The same hour with an insulation alarm is the stop row. Read the alarm first, then the curve.
