A framework based on the twin-delayed deep deterministic policy gradient algorithm is coupled with a validated numerical simulation of a two-bladed VAWT, aimed at maximising power generation.
Abstract
Abstract Content of image described in text. Vertical-axis wind turbines (VAWTs) have emerged as natural complements to their horizontal counterparts but suffer from performance limitations associated with complex aerodynamics, in particular dynamic stall. To address this challenge, we employ deep reinforcement learning to identify optimal blade pitching policies. A framework based on the twin-delayed deep deterministic policy gradient algorithm is coupled with a validated numerical simulation of a two-bladed VAWT, aimed at maximising power generation. The agent receives only onboard measurements of blade pressure and position, rendering the problem partially observable. From these inputs, it learns policies for three distinct flow cases: one uniform flow and two shear flows. The resulting policies outperformed the optimum fixed-pitch setting within 100 turbine revolutions across all cases. Ultimately, the agent found cycle-averaged power coefficients that are 3.1–5.6 times larger than those of a blade fixed at zero degrees, while simultaneously reducing variations in blade loads. Improvements originate equally from both halves of turbine rotation, despite the latter-half’s low absolute power contribution and depleted flow energy. Through flow field analysis using the power partitioning method, the underlying mechanisms responsible for power production are revealed. The presence of a rotational flow around the airfoil leading edge dominates. Its net power contribution is determined by the rotational intensity and the blade’s kinematics. Kinematics also play a central role in mitigating dynamic stall. Flow separation is not intrinsically detrimental, as proper blade positioning can suppress its harmful effects.
This is the first study in which a fully three-dimensional computational fluid dynamics simulation has been employed to train a DRL model for active flow control in wings, suggesting promising directions for extending DRL-based strategies to higher Reynolds numbers and more complex wing configurations where prior physical knowledge may be limited.
R. Montalà, B. Font, Pol Suárez et al.· Journal of Fluid Mechanics· 0 citations
This work investigates Deep Reinforcement Learning (DRL) as a tool for model-free closed-loop active separation control in a fully turbulent wind tunnel flow over a one-sided diffuser. The agent controls an array of magnetic valves (on/off) that eject compressed air into the boundary layer, while the environmental state is reduced to the signal from a single wall-shear-stress sensor placed near the natural transitory detachment point. The control law is learned in real time using Proximal Policy Optimization. Compared to the standard learning design based on the weighted sum of all rewards following an action, we demonstrate that a horizon aligned with the convective time of the flow leads to faster convergence and a more robust control strategy. The resulting control law corresponds to a low-duty-cycle actuation pattern that yields a forward-flow fraction of approximately $53\%$. This compares favorably with conventional and optimized periodic open-loop control ($\sim 40\%$ and $\sim 51\%$, respectively). The findings of this article indicate that, when embedded into an online experiment, DRL represents an efficient tool to identify robust and interpretable active separation control strategies.
Sofia Avdiiv, Andre Weiner, B. Steinfurth· 0 citations
HydroGym is introduced, a solver-independent reinforcement learning platform providing more than 60 validated, openly available flow control environments spanning from canonical laminar flows to complex turbulent flows, with systematic progression in the Reynolds number up to Re = 4 × 105, and Mach number variations in two and three dimensions.
Christian Lagemann, Sajeda Mokbel, Miro Gondrum et al.· Nature· 1 citation
This study aimed to enhance the aerodynamic performance of a Vertical-Axis Wind Turbine (VAWT) airfoil through a multidisciplinary approach that combines Machine Learning (ML), Computational Fluid Dynamics (CFD), and experimental validation. The focus was on enhancing the lift-to-drag coefficients ratio (퐶 푙 /퐶 퐷 ), particularly at higher Angles of Attack (AoA ≥ 20°), a critical operational regime for VAWTs. A baseline airfoil of 12-inch chord length and 10-inch wingspan was analyzed using ANSYS Fluent across a range of AoA (0°–90°) at a constant freestream velocity of 10 m/s (Re ≈ 2.0 × 10 5 ). This served as a performance benchmark. Using ANSYS Design Explorer and Python-coded constraints, ML-based optimization—employing adjoint solvers and stochastic gradient descent—was applied to the baseline geometry at 20° AoA. The resulting AI-optimized airfoil was then evaluated across all AoAs, both with and without specially designed endplates. CFD simulations revealed substantial aerodynamic improvements, with the AI airfoil delivering markedly higher 퐶 푙 /퐶 퐷 across the full AoA range. In the critical 10°– 15° AoA zone, the AI airfoil with endplates achieved more than twice the 퐶 푙 /퐶 퐷 of the baseline without endplates, highlighting improvements in lift generation, drag reduction, and vortex control. Experimental validation using a 3D-printed scale model in an open circuit wind tunnel further substantiated the CFD findings. Across all AoAs, the AI-optimized airfoil—particularly with endplates—demonstrated superior aerodynamic performance, closely aligning with CFD predictions (within 2%–7.8% deviation). Experimental results confirmed a 퐶 푙 /퐶 퐷 gain of over 133% at 0° AoA and consistent improvements of 40–55% between 5° and 15° AoA when using endplates. At higher AoAs (25°–50°), the AI airfoil maintained elevated performance levels, benefitting from delayed stall and improved flow coherence enabled by the endplates. This investigation confirms the dominant role of airfoil geometry and endplate design in aerodynamic optimization. The study highlights the reliability of CFD-ML methods for airfoil design. Future research will implement these AI-optimized geometries in a full VAWT setup, evaluating torque, power coefficients with tip-speed ratio (TSR) through transient CFD and further experimental validation. These results lay a strong foundation for next-generation VAWT development driven by computational intelligence and aerodynamic refinement.
Optimizing wind turbine blades for maximum annual energy production is a complex real-world challenge, marked by high-dimensional search spaces and computationally expensive, non-differentiable aerodynamic simulations that hinder efficient gradient-based optimization. We present a novel, fully differentiable design framework that creates an end-to-end differentiable pipeline from Kulfan design parametrization to power efficiency (CP). Unlike prior "black-box" approaches, our pipeline composes NeuralFoil — a neural surrogate mapping Kulfan parameters to aerodynamic coefficients — with a custom differentiable implementation of Blade Element Momentum Theory (BEMT) with guaranteed convergence. This integration enables the computation of exact gradients of the whole-blade performance via automatic differentiation. To navigate the non-convex aerodynamic landscape, we propose a hybrid memetic strategy: Particle Swarm Optimization (PSO) is employed for global exploration to identify high-quality basins of attraction, followed by a gradient-based Augmented Lagrangian Optimizer (ALO) for precise local refinement. Experimental validation on the NREL 5MW reference turbine demonstrates that this physics-informed hybrid framework achieves superior design performance, reaching the Betz Limit (CP ≈ 59.3%) without losing its physical feasibility.
L. Barbosa, Adriano Maurício de Almeida Côrtes, Laura Silvia Bahiense da Silva Leite· Proceedings of the Genetic a...· 0 citations
PEARL employs an actor-adjoint algorithm that leverages automatic differentiation to compute policy gradients over short horizons and adjoint-based sensitivities of future returns approximated via neural networks, significantly reducing the number of environment interactions, while mitigating long-term gradient instabilities.
Matteo Tomasetto, Nicolò Botteghi, Gabriele Bruni et al.· 0 citations