Integrating machine learning and CFD for enhanced trailing edge serration design on a NACA 0012 wind turbine blade

Marwa Khaleel Rashid, Khuder N. Abed, Mohammed Alharbi, Mohammed Waleed Muayad, Tarik Alkharusi, Arman Ameen,
Integrating machine learning and CFD for enhanced trailing edge serration design on a NACA 0012 wind turbine blade,
International Journal of Thermofluids,
Volume 30,
2025,
101446,
ISSN 2666-2027,
https://doi.org/10.1016/j.ijft.2025.101446.
(https://www.sciencedirect.com/science/article/pii/S2666202725003921)
Abstract: This study combines machine learning (ML) techniques with computational fluid dynamics (CFD) simulations using ANSYS Fluent to investigate the impact of different trailing-edge serration designs on a NACA 0012 airfoil, a commonly used design in wind turbine blades. Building on a fundamental CFD analysis, ML-driven data augmentation—including synthetic data creation, geometric transformations, and noise sensitivity analyses—is employed to enhance and accelerate the design process. The CFD simulations utilize the k-omega SST turbulence model to calculate the lift coefficient (CL) and drag coefficient (CD) over a range of angles of attack (0°–20°). The ML framework evaluates the model's robustness and prediction accuracy under various noise levels and augmented training datasets. Results indicate that the rounded serration design achieves the best lift-to-drag ratio (CL/CD), with approximately a 15 % improvement over the baseline at an angle of attack (α) of 12° In contrast, sharp-edged serrations produce more lift at middle angles but generate increased drag at higher angles. Using hyperparameter-tuned ML models—such as Ridge regression, Random Forest, and a feedforward neural network—improves predictive accuracy and facilitates exploration of the complex design space. This combination of CFD and ML provides a robust method for optimizing wind turbine blade performance, striking a balance between aerodynamic efficiency and computational cost. Unlike previous studies that relied solely on CFD or experiments, this research integrates machine learning with CFD. This dual approach aims not only to analyze the serration geometry but also to optimize it effectively through alternative models and data augmentation. By merging fluid dynamics and machine learning, this approach transforms the design process from trial and error to a data-driven, predictive methodology, marking a significant advance in the aerodynamic design of wind turbine blades.
Keywords: Aerodynamics; Trailing edge serrations; NACA 0012 airfoil; Machine learning; CFD; Wind turbine blade