Device health, alarms, and string performance
SolarSENS gives you three separate signals for investigating equipment behavior — they are not different versions of one health score:
| Signal | Question it answers | Time context |
|---|---|---|
| Alarm and attributed loss | What event occurred, and what energy impact does the model attribute to it? | The event window and its processed analytics |
| Device or plant health | What did the latest status evaluation find? | A near-current snapshot |
| String performance | How do strings compare within the plant? | The selected analysis date and position |
Because each is calculated independently, disagreement between them is normal and is itself diagnostic: a healthy snapshot with an open alarm usually means the snapshot is newer than the fault, or the fault doesn’t affect the evaluated conditions.
Alarm record, alarm analytics, and work order are three things
Section titled “Alarm record, alarm analytics, and work order are three things”An alarm is associated with a plant and an originating device, and carries type, severity, status, start and clear times, acknowledgement, and fault detail. Fault codes and their descriptions are interpreted from the inverter manufacturer’s fault-code documentation, so the available alarm types and detail depend on the brand and how the plant sends data: provider integrations (Gelex, Huawei, Sungrow, SolarEdge) deliver the provider’s alarm records, while logger-connected devices are read through per-manufacturer register mappings. SolarSENS shows it in two ways, and a work order is a third, separate record:
| Question | Use |
|---|---|
| What condition was reported, and what is its current state? | Alarm Triage — the operational list where you acknowledge or clear |
| What patterns or attributed impact appear over time? | Alarm Analytics — processed data with derived duration and attributed loss |
| What work was requested, assigned, or completed? | Work Orders |
Acknowledging or clearing an alarm does not create, update, or close a work order — that is a separate, deliberate action. Where the alarm view offers Add Work Order, use the alarm’s plant, device, type, and time context when describing the work. Analytics values help you compare alarms over a date range; they do not change the source alarm’s state, and they are not a notification or response-time commitment.
Alarm loss: read Energy Loss with Remarks
Section titled “Alarm loss: read Energy Loss with Remarks”Alarm Analytics estimates energy loss for inverter alarms that meet the calculation’s requirements: the alarm window must overlap analytical intervals, and expected production must be computable from irradiance and model inputs. Per interval:
The alarm-level figure is the sum of these interval estimates — a model output, not a meter reading.
Alarms that don’t meet the requirements get no loss figure: Device Offline is excluded from this calculation, non-inverter alarms have no inverter-loss estimate, and any alarm can lack the required production or irradiance inputs. Read Energy Loss with Remarks before treating 0 as no loss.
Device health: a snapshot, not a monitor
Section titled “Device health: a snapshot, not a monitor”A health status describes the conditions found at evaluation time, using the equipment inventory, recent telemetry, data freshness, and alarm context. Between evaluations it can drift from reality.
- Offline — the evaluation did not find the conditions to consider the device online.
- Fault — the snapshot includes qualifying fault or alarm context.
- Stale or missing data — an input or freshness problem. Check data delivery before suspecting hardware.
Check Last Updated before acting on a status: a snapshot older than the event you’re investigating tells you nothing about it.
String performance: comparative, not absolute
Section titled “String performance: comparative, not absolute”String analysis compares electrical observations across the plant’s equipment hierarchy, labeling strings Normal, Underperforming, Unknown, Half, Curtailed, Disconnected, or Offline depending on the analysis type.
The labels rank strings against each other, so only compare within the same plant, reference period, analysis type, and analytical position. Disconnected or Offline classifications can also result from missing observations rather than a measured zero.