Wind-farm operations

Know which alarms actually matter.

Turbines cry wolf constantly. WindOps separates real faults from nuisance chatter, guarantees safety-critical codes are never silenced, and sends crews where they're needed — on a two-hour reporting loop.

Launch the live demo See how it works
6
Critical
1
High
3
Elevated
0
Routine
What it does

Four jobs, one operations view

Every wind site drowns in SCADA alarms, most of them noise. WindOps turns that stream into a short, ranked list of what a crew should actually do today.

🛡

Safety interlock

Critical codes — high vibration, gearbox and generator over-temp — can never be suppressed by the model. The ML is a priority engine, not a silencer, so a real fault is never hidden.

🔊

False-alarm triage

A gradient-boosted classifier scores each alarm as a real fault or nuisance chatter, weighting a missed fault far more heavily than a wasted truck roll.

🗺

Crew routing

An optimizer plans the shortest service route over the site road network, prioritizing the turbines that need attention first.

📄

2-hour reporting

A written PDF goes to management every two hours, with per-turbine Slack alerts to the units that need a crew this shift.

The triage ladder

Every turbine lands in one of four tiers

The ordering is the product. A crew reads top to bottom and knows exactly what today looks like.

Critical
Safety interlock tripped or catastrophic risk — act immediately, regardless of model output.
High
Confirmed fault or advanced blade wear — dispatch a crew this shift.
Elevated
Worth watching — schedule within the week before it escalates.
Routine
Nominal — monitor, no action needed.
Under the hood

From raw SCADA to a crew's to-do list

Five stages, fully reproducible. The demo runs the whole chain on synthetic data you can regenerate live.

1

Ingest

SCADA, alarms, blade wear

2

Classify

Real fault vs nuisance

3

Interlock

Critical codes escalate

4

Route

Optimize crew path

5

Report

PDF + Slack alerts

Honest note: the demo runs on synthetic data, so it proves the pipeline works end to end — not real-turbine accuracy. Model performance is reported at its true, leakage-corrected numbers (~0.77 AUC, ~89% recall), which is exactly why the safety interlock is load-bearing rather than decorative.

Try the live fleet view

Generate a synthetic fleet, watch the triage tiers update, download the management PDF, and send a Slack alert — all in the browser.

Launch the demo