Experiment 05 · Predictive maintenance
102 Wind Turbines, One Screen: Where Energy Is Lost and What Breaks Next
A gearbox bearing in Nordfriesland is getting close to failure, a turbine in Mecklenburg never goes above 3.0 MW, and the blades of another sit at the wrong angle. Pick a turbine on the map and read its power curve and its vibration side by side: two lines on each side of the curve show when the pattern is no longer normal. Then plan the repair: which depot has the part, which crew is free, how long the drive takes, what it costs, and what doing nothing would cost. Weather, wind directions, clouds, rain, fog and ice come from the live DWD forecast.
What you can explore
One circle, one century, every country
Power curve with alarm lines
30 days of 10-minute values against the expected curve, with a normal range and an alarm limit on each side. Zoom with two sliders per axis; two turbines show a fault you can see and fix.
From vibration to repair plan
Spectra and trends show which part vibrates and the 3D turbine highlights it. The plan picks the crew, the depot with the spare part, the tools, the weather window and the cost.
Weather and routes on the map
Wind arrows and flow, clouds, rain, fog and ice across Germany, each switchable. Road routes from every depot with the drive there and back, and a tour you can change and optimise.
- turbines in 10 wind farms
- 102
- failure risk and weather ahead
- 72 h
- live forecast points
- 33
- depots and service crews
- 5 + 8
The AI use case
How we built it with AI, step by step
From physics and published prices to a working planning page: every model is small, readable and tested in a headless browser.
Physics first
Power curve with air-density correction (IEC 61400-12-1), Jensen wakes, storm cut-out with restart hysteresis, and icing from temperature, humidity and fog.
Condition monitoring
Fault frequencies from the drivetrain geometry, VDI 3834 vibration zones, power-curve alarm lines, and 400 simulated futures per fault for the chance of failure.
Real costs and logistics
Technician rates, cranes, gearbox and bearing prices from 2024–2026 sources; crews, depots, stock, tools, weather windows and road routes on the motorway network.
A packaged site
Texts and data as JSON, modules bundled with esbuild, a 3D view that loads only when needed, 36 automated click tests, delivered as a WordPress files-only plug-in.
Sources and licences
Weather: DWD MOSMIX forecast via Bright Sky (Deutscher Wetterdienst, open data). Map: Natural Earth (public domain).
Methods: IEC 61400-12-1 (air-density correction), VDI 3834 Part 1 (vibration zones), NREL drivetrain condition monitoring, § 51 EEG (negative prices).
Costs: Deutsche WindGuard 2024, Nordex dismantling cost sheets (crane day rate and mobilisation), EPRI main bearing 2024, Wind Systems bearing cost 2024, Fraunhofer ISE 2024 (full-load hours), netztransparenz.de (market value of onshore wind 2025).
The fleet, its faults, crews, depots and stock are fictional; prices without a published figure are labelled proxies in the app. Libraries: three.js (MIT), topojson-client (ISC).
Your Next Step
Your own SCADA, condition-monitoring and service data could answer the same questions. We build data platforms, analytics and AI that your team can run.