Bergen Offshore Wind Centre (BOW)

Wake detection using image processing techniques

Maria Krutova and Mostafa Bakhoday-Paskyabi demonstrate the use of image processing techniques for the wake detection and possible applications for the wake analysis.

Pictures of wake data-set
An example of wake detection from lidar measurements taken during the Offshore Boundary-Layer Experiment at FINO1 (OBLEX-F1) campaign. The image shows the original data and various way to process the detected wake: detect the wake centerline and direction, separate near and far wake, process different parts of the wake individually.
Maria Krutova

Main content

The wake is a structure characterized by the decreased wind speed and increased turbulence intensity, which forms behind a wind turbine due to the extraction of kinetic energy. Eventually, the wake speed recovers to the free-flow speed on the distances of 10-20 rotor diameters, depending on the atmospheric conditions. The distance between wind turbines in a wind farm is usually shorter than the mentioned recovery distance. Therefore, the downstream turbines are often subjected to the wakes of upstream turbines. The generated wind power depends on the cube of the wind speed and thus decrease gradually if the wake speed falls below the rated speed. The increased turbulence intensity affects the fatigue loads, which decrease the lifetime of a wind turbine. Moreover, the instantaneous wake oscillates over time. The oscillations become particularly strong in the far-wake region – where downstream turbines are usually located.

Identifying a wake shape is an essential step towards studying the evolution of the instantaneous wake and detecting the regions with the highest wake deficit. Several wake detection methods already exist with their advantages and limitations. We propose an automatic thresholding method to separate the wake shape from the free flow. The input data can be provided both as wind speed values and as a grayscale image of a 2D cross-section. Although the supplementary data, such as free-flow wind speed and wind direction, improve the detection accuracy, they are not mandatory for the method. The low requirements to the input data make the method versatile and allow the use of diverse data sets: lidar and satellite measurements, particle velocimetry, or LES output. With the automatic thresholding, we can also define criteria for the algorithm to distinguish between far and near wakes.