Skip to content

Frequently asked questions

Last updated

What analyses does Tipspeed perform?

For each turbine and each 10-minute period, Tipspeed compares the measured production with the expected production of the digital twin. The analyses built on this comparison are:

These analyses are delivered as a one-time assessment, typically over two years of data, or as monitoring, with results updated daily. See Services.

What is Tipspeed's track record?

As of October 2026, Tipspeed has worked on 38 wind farms for 25 customers, totalling 6.5 GW in eleven countries on three continents:

  • terrain from flat to highly complex, onshore and offshore;
  • turbines from Nordex, Vestas, Enercon, Siemens Gamesa, GE, Senvion (Repower), Ming Yang;
  • more than 25 turbine models, from 2.0 MW to 8.0 MW;
  • turbines from newly commissioned to more than 20 years in operation.

What data is needed, and how long does an analysis take?

The analysis needs the layout and turbine specifications, 10-minute SCADA data with operating status, curtailment plans, and any pre- or post-construction wind measurements or reports. Data can be transferred through the API, shared folders or files in any format; a checklist is provided at kick-off. First results are delivered six weeks after all input data has been received; with monitoring, they are then updated every day.

How does Tipspeed distinguish wind farm underperformance from low wind?

Tipspeed computes the production expected with the wind conditions of the same period and compares it with the measured production. The energy waterfall then separates the effect of a less windy period, the windiness loss, from those of wakes, availability and turbine performance. It can start from the P50 of the energy yield assessment. See Performance assessment and Reading the energy waterfall.

Is a turbine underperforming, or only waked by its neighbours?

The digital twin computes the expected production of each turbine with its own wind and wake conditions. A turbine that produces less than its neighbours but as much as expected is waked, not underperforming: its performance ratio stays close to 100%. See Performance ratio.

What is the performance ratio of a wind farm?

The performance ratio is the production at 100% availability divided by the production expected after wake losses over the same period, in percent. It is insensitive to wind level, wake losses and downtime. It falls below 100% when running turbines produce less than expected, for example because of yaw misalignment. See Performance ratio.

What data does the digital twin use?

The digital twin uses SCADA data, the site layout, turbine specifications and reference power curves, curtailment plans, wind measurements, satellite observations and reanalysis data such as ERA5, the global reanalysis of ECMWF. Additional site data raises the calibration level, from a pre-construction model (level 0) to the assimilation of operational wind measurements (level 4). See Digital twin.

How is the wind at each turbine modelled?

The wind is computed by an atmospheric large-eddy simulation of the site, at 100 m horizontal and 25 m vertical resolution. Its boundary conditions are provided by a mesoscale model covering 250 km × 250 km, initialised from ERA5 reanalysis. A surrogate model then interpolates the results in space and time across the site. See Wind flow.

Which wake models does Tipspeed use?

Wake and blockage losses are computed with PyWake on 10-minute time series. The wake model is selected per site among TurbOPark Gaussian, Park2 and Zong Gaussian, then tuned against operational data. Blockage is computed with the scaled Rathmann model, and wake-induced turbulence with the Crespo–Hernández model. See Wakes and blockage.

Was production reduced by a neighbouring wind farm?

Wake losses include the external wakes of neighbouring wind farms. When no up-to-date layout is available, the neighbouring turbines are identified on satellite images. See Wakes and blockage.

Was the pre-construction wake loss estimate correct?

Operational wake losses, computed with the wake model tuned on SCADA data, can be compared with the assumptions of the energy yield assessment. See Wakes and blockage.

Is a lidar required to measure yaw misalignment?

No. Tipspeed estimates yaw misalignment on every turbine without additional sensors, by combining high-resolution satellite imagery with a numerical weather model. A nacelle lidar remains the reference for a contractual test, and the two can be combined. See Yaw misalignment estimation.

How accurate are the wind speed and yaw misalignment estimates?

Using pre-construction data only, the wind speed at each turbine has an absolute bias of 0.11 m/s against 18 masts, lidars and spinner anemometers on six wind farms. Yaw misalignment has a mean absolute error of 2.25° against nacelle-mounted lidar, over 23 tests on seven sites across two continents. Both tests were blind: the reference measurements were disclosed only after Tipspeed had delivered its estimates. See Validation.

How does Tipspeed compare with a nacelle lidar campaign?

A nacelle lidar measures one turbine with reference accuracy, for the duration of the campaign. Tipspeed covers every turbine, over the whole history of the farm, without additional hardware. A lidar campaign on a few turbines can therefore be extended to the whole farm and over time. See Yaw misalignment estimation.

What is the difference between yaw misalignment and optimal yaw offset?

Yaw misalignment is the orientation of the rotor relative to the wind direction. The optimal yaw offset is the misalignment at which the power output of the turbine is highest. It is the target value of a yaw recalibration, rather than zero. See Optimal yaw offset.

How is lost production calculated?

Lost production is calculated with the production-based method of IEC 61400-26-1. For each 10-minute step, the loss is the expected power minus the power actually produced, with a lower bound of zero. The loss is attributed to the operating status of the turbine: stopped, partial performance, out of environmental specification or missing data. See Lost production.

How is missing SCADA data handled?

Tipspeed estimates expected power for every turbine and every 10-minute step, including steps in which the turbine was stopped, derated or had missing data. The estimate is produced by one gradient-boosted model per turbine, with the neighbouring turbines, the digital twin and the power curve of the turbine as inputs. During training, inputs are masked, so the estimate degrades gradually as data becomes unavailable. This estimate serves the availability losses only; underperformance is measured against the digital twin. See Expected power.

How is the gain of a turbine upgrade measured?

Tipspeed compares a test group with a control group over the same months, using only 10-minute records at full performance. Each group is first compared with its own expected production, so that wind conditions affect both groups equally and cancel out in the difference. The result is the difference between the two groups, in percentage points. See Upgrade analysis.

Does Tipspeed provide an API?

Yes. Tipspeed provides a read-only HTTP API over the same analytics engine as the application. The API is in closed beta, with access granted per user. The limits are 60 requests per minute and 1000 requests per day per key. See HTTP API.