Prediction and comparison of ice accretion on wind turbine blades between central and northern Europe by neural network and FEM

Wind energy in inland regions is limited mainly due to low mean annual wind speed and due to turbulence. However, advancements in wind turbine technology made it possible to utilize wind at higher altitude which have a stable and faster wind flow but introduce new issues such as ice accretion. This study analyses ice accretion in January in two European locations, which are Mosonmagyaróvár, Hungary (Central Europe), and Piteå, Sweden (Northern Europe), on a NACA 64–318 airfoil. ANSYS FENSAP-ICE was used to simulate ice accretion for a range of different parameters and conditions such as LWC, temperature, time, cracks and delamination, and snowflakes. The results of the simulation show that cracks and delamination have negligible influence (<3 %) when liquid water content (LWC) is 1 g/m³ but have higher influence over the ice mass when LWC is lowered for example its influence reaches 30 % at LWC of 0.3 g/m³. Furthermore, different Neural network models were trained on MATLAB and IBM SPSS 20 which were able to accurately predict the mass of ice using different regression evaluation metrics such as R², MAE, and RMSE where MATLAB's Bayesian Regularization (BR) model performance was better than the others.