Machine learning-enabled prediction of wind turbine energy yield losses due to general blade leading edge erosion

Blade leading edge erosion is acknowledged to significantly reduce the energy yield of wind turbines. The problem is particularly severe for offshore installations, due to faster erosion progression boosted by harsh environmental conditions. This study presents and demonstrates an experimentally validated simulation-based technology for rapidly and accurately estimating wind turbine energy yield losses due to general leading edge erosion. The technology combines the predictive accuracy of two- and three-dimensional Navier–Stokes computational fluid dynamics with the runtime reductions enabled by artificial neural networks and wind turbine engineering codes using the blade element momentum theory. The main demonstration is based on the assessment of the annual energy yield of the National Renewable Energy Laboratory 5 MW reference turbine affected by leading edge erosion damage of increasing severity, considering damages based on available laser scans and previous leading edge erosion analysis. Results also include sensitivity studies of the energy loss to the wind characteristics of the installation site. It is found that the annual energy loss varies between about 0.3 and 4%, depending on the damage severity and the site wind characteristics. The study also illustrates the necessity of resolving the geometry of eroded leading edges rather than accounting for the effects of erosion with surrogate models, since, after an initial increase of distributed surface roughness, erosion yields leading edge geometry alterations causing aerodynamic losses exceeding those due to the loss of boundary layer laminarity consequent to roughness-induced transition. The presented technology enables estimating in a few minutes the amount of energy lost to erosion for many-turbine wind farms, and offers a key tool for predictive maintenance.