Overview
Inverters
| Inverter | Rating | Data | Model | Valid peers | Ratio | Alarm limit | Health | Band |
|---|
Measured against expected power
Daily ratio, measured over expected
Why this state
Events
Expected power is the peer model (rating-normalised median of the other valid inverters, leave-one-out) when at least 3 valid peers exist, otherwise PVWatts from the site irradiance sensor. The alarm limit is the 1 % conformal quantile of the ratio over the last 90 clean days. The engine sends alerts only and never commands equipment.
Fault Log
Every fault event the engine has recorded, newest first, with its state and close-out. Actions stay on the inverter tiles on Home.
| Start | Inverter | Type | Priority | Status | End | Duration and loss |
|---|
Alarm Performance
How reliable the engine's own fault alerts are, not the plant's uptime. Fault events that start in the range, matched to their close-out.
Close-outs
Per inverter
| Inverter | Days with a channel | Fault events | Not a fault |
|---|
Fault events in the range
| Start | Inverter | Type | Priority | Close-out | Alert latency (min) | To acknowledge (min) |
|---|
Anomaly Log
Every data issue the engine has recorded: feed gaps, inverters not reporting, frozen or out-of-range readings, irradiance sensor drift and timing. Rows affected by a data issue are not scored.
| Start | Device | Issue | Status | End | Duration |
|---|
How it works
A step-by-step simulation of how the engine handles one hour of data. Open the two tables below if you need them, set the choices, then press Simulate fault: the flowchart lights up the path the engine takes. Change the choices to explore the other branches.
1 · What happens: 5 faults, 2 data issues and a normal day
| Choice | Type | When the engine raises it | Where the path ends |
|---|---|---|---|
| Outage | Fault | Output below 5 % of expected for at least 15 valid minutes | High priority · alert · close-out |
| Low output | Fault | Measured ÷ expected below the alarm limit for at least 30 valid minutes | Medium priority · alert · close-out |
| Capped output | Fault | Output flat at a ceiling below 90 % of rating for at least 30 minutes while expected is 1.2 × higher | Medium priority · alert · close-out |
| Partial loss | Fault | Daily ratio below 1 − max(10 %, 3 × the inverter's noise) for at least 3 days in a row | Low priority · reports only, no alert |
| Plant underperformance | Fault, whole plant | Plant ratio against a frozen PVWatts reference below 1 − max(5 %, 3 × its spread) for at least 3 days, most inverters low together | Alert only: a plant event has no tile, so no close-out yet |
| Frozen reading | Data issue | Rows arrive but the value is stuck at one number | Anomaly Log · rows not scored, no fault |
| Not reporting | Data issue | No rows from the inverter while the rest of the feed arrives | Anomaly Log · rows not scored, no fault |
| A normal day | No fault | Measured ÷ expected stays above the alarm limit | Within limits · no event |
2 · The four conditions you can change
| Condition | Options | What it decides | Boxes it switches between |
|---|---|---|---|
| Expected power | As the inverter is now · Peer model · PVWatts | How "expected" is worked out. Peer model (the median of the other inverters) needs at least 3 valid peers; otherwise PVWatts from the irradiance sensor. "As now" uses the inverter's real count; the other two are what-ifs. | Expected power · peer model ↔ PVWatts |
| Time at the site | Now · 10:00 · 22:00 (Singapore) | Day or night at the site. On the live pilot, alerts between 21:00 and 07:00 are held until 07:00. | Alert sent now ↔ Quiet hours, held until 07:00 |
| Acknowledged within 2 h | Yes · No | Whether the technician presses Acknowledge in time. Only a high-priority alert (an outage) becomes due for escalation; nothing is re-sent automatically. | Acknowledged on the tile ↔ Escalation due |
| Site check finds | The fault · Nothing wrong | What the technician records after visiting: Close out (a confirmed fault) or Not a fault (a false alarm). | Confirmed fault ↔ Not a fault, false alarm |
Outcomes (boxes in coloured corner brackets):
Confirmed fault
Within limits
Data issue
Reports only · False alarm
Simulation only, in your browser: nothing is sent to the engine, the database or any equipment.
Pick a fault and press Simulate fault on an inverter.
The steps in full
Readings arrive
Enverse sends one file per hour with minute values for each inverter, the site meter and the irradiance sensor. The engine runs at 20 minutes past each hour and reads the new rows by their content and time, never by file name.
Data checks come first
Feed gaps, inverters that stop reporting, frozen or out-of-range readings, counter resets and irradiance sensor drift or timing become data issues (Anomaly Log). Rows they affect are not scored, so a data problem never turns into a fault.
Expected power
For each inverter and minute, expected power is the peer model: the rating-normalised median of the other valid inverters, leaving the inverter itself out. With fewer than 3 valid peers the engine uses PVWatts from the irradiance sensor instead.
Ratio against the alarm limit
Measured over expected gives a ratio. The alarm limit is the 1 % conformal quantile of that ratio over the last 90 clean days. Staying below it, or producing nothing while the peers produce, opens a fault event: outage, low output, partial loss, capped output or plant underperformance, with a likely cause and a confidence.
Alerts
Each fault event gets a priority (low, medium or high) from the energy lost per day, the confidence and the urgency. Medium and high events send one alert; another follows only if the priority rises, or if the fault returns at least 24 hours after it resolved. Low-priority events appear in reports only. On the live pilot, alerts between 21:00 and 07:00 Singapore time are held until morning, and a daily digest goes out at 21:00. An unacknowledged high-priority alert becomes due for escalation two hours after it was sent; nothing is re-sent automatically.
People close the loop
On Home, the inverter's tile offers Acknowledge, then Close out with a root cause and action, or Not a fault with a reason. Engineers can Shelve an event for 24 hours or 7 days. Every action is recorded in the audit log, and close-outs become the verified labels behind Alarm Performance.
Health, in shadow
Once a week the engine scores each inverter's health and raises warnings in shadow mode: visible to engineers only, never notified. Closing a warning resets that indicator from the following Monday.
Health warnings
Shadow mode: engineer only, never notified. Closing a warning resets that indicator's CUSUM from the following Monday.
Inverter map
Each asset's Enverse ID by date, from the engine's device map. Past days are always scored with the mapping valid at that time.
PdM engine · Inverter map
Mapping
| Engine ID | Field label | Enverse ID | Valid from | Valid to | Note |
|---|
Unassigned channels
| Engine ID | Field label | Enverse ID | Valid from | Valid to | Note | Action |
|---|
Channel requests
| Request | Channel | Asset | Valid from | Requested by | Reason | Status | Action |
|---|
Asset register
| Engine ID | Field label | Rating | Location | Make and model | Serial | Status | Action |
|---|