Skip to main content
ResearchProjects

iCOWE Project

Combining high-fidelity computational fluid dynamics and artificial intelligence for faster prediction of floating offshore wind turbine wakes.

Floating offshore wind turbines operate under complex environmental conditions that cause platform motions. These motions alter the airflow behind the rotor, influencing wake development, downstream wind conditions, and potentially the performance of neighboring turbines.
The iCOWE research project at the Norwegian University of Life Sciences (NMBU) investigates these aerodynamic interactions by combining high-fidelity computational fluid dynamics (CFD) with advanced machine learning techniques. The research has provided new insights into how platform motions influence wake structures, velocity deficits, and wake recovery. Building on validated CFD simulations, the project also developed motion-conditioned neural network models that predict complex wake dynamics at substantially reduced computational cost.
A key outcome is the development of data-driven surrogate modeling approaches that learn both the evolution of moving wakes and the perturbations introduced by platform motion. These methods demonstrate the potential of integrating physics-based simulations with artificial intelligence to accelerate aerodynamic predictions. The research contributes to computationally efficient tools for floating offshore wind farm design, wake analysis, and future digital-twin applications.

01 Jan 2023 - 31 Dec 2026

Norwegian University of Life Sciences (Internal funding)

About the project

  • Background

    Global wind power is expected to grow three times over the next decade to avoid the worst impact of climate change. The wind power plants are constructed with turbines having large rotor diameters to allow maximum energy yield from the oncoming wind. Special attention is given to reducing the cost of energy and increasing profitability, and this is anticipated to be achieved by developing optimized design, analysis, prediction, and monitoring tools capable of simulating aerodynamic flows accurately in offshore wind farms. To this end, traditional analytical tools employed to model aerodynamic flows in a wind farm are fast but include simplifications that can compromise the overall integrity of solutions. On the other hand, high-fidelity flow models based on computational fluid dynamics (CFD) are more accurate but at the same time require significantly high computational resources to obtain meaningful results.

    The Ph.D. project aims to advance flow simulation for offshore wind turbines by integrating high-fidelity computational fluid dynamics (CFD) models with machine learning (ML) methods. The developed state-of-the-art models are expected to be integrated into future digital twin technology to meet industrial needs for computationally efficient tools and to serve as a basis for condition monitoring, prediction, analysis, and maintenance of offshore wind turbines. Key research areas to be explored in the Ph.D. project will include physics-informed and data-driven surrogate model to simulate and predict the wake flow of a floating offshore wind turbine.

  • Objective

    The iCOWE project aims to revolutionize offshore operational management by developing innovative methods and systems that harness digital technologies. Central to this effort is the creation of a sophisticated toolbox that enables intelligent simulations to predict wake flow around floating wind turbines. By enabling smarter control and predictive capabilities for wind turbine operations, iCOWE aims to lower operational costs, enhance efficiency, and contribute to the sustainable growth of offshore wind energy.

  • Interactive Research Demonstrator – AI-Based Wind Turbine Wake Prediction

    As part of the iCOWE project, an interactive online demonstrator has been developed to showcase how high-fidelity Computational Fluid Dynamics (CFD) simulations can be combined with Artificial Intelligence (AI) to predict the aerodynamic wakes of floating offshore wind turbines.

    The demonstrator illustrates the application of motion-conditioned neural network surrogate models, which learn complex flow behaviour from CFD-generated data. These models aim to reproduce the influence of floating platform motions, particularly pitching and rolling, on downstream wake development without requiring computationally expensive CFD simulations for every prediction.

    By providing an accessible visualization of AI-based wake predictions, the platform bridges the gap between advanced academic research and potential engineering applications. It demonstrates how data-driven methods could contribute to faster aerodynamic assessments, wind farm optimization and future digital-twin technologies.

    Explore the Interactive Demonstrator

    This demonstrator is a research prototype developed to illustrate the potential of AI-assisted wake prediction. It is not intended as a validated industrial design tool.

    FOWT Online Demonstrator
    Photo: Haris Hameed Mian

  • Key research findings

    ✅ Validated High-Fidelity CFD Simulations: Blade-resolved CFD simulations using Improved Delayed Detached Eddy Simulation (IDDES) demonstrated strong agreement with experimental measurements, establishing a reliable foundation for investigating the aerodynamic effects of floating wind turbine motions. Additional URANS–RSM simulations were validated against wind-tunnel measurements to support the development of data-driven wake prediction models.

    🌪️ Enhanced Wake Recovery Under Platform Motion: Roll-induced motion significantly influenced wake mixing and recovery, reducing the velocity deficit by up to 50% at 10 rotor diameters (10D) compared with the stationary turbine. Pitch and roll motions produced distinct wake responses, highlighting the importance of platform motion in offshore wind farm aerodynamics.

    🔁 Different Wake Responses to Pitch and Roll: Pitch motion primarily generated streamwise wake pulsation, whereas roll motion produced pronounced lateral wake meandering. These differences affected the distribution of momentum, wake structure and downstream velocity recovery.

    📈 Influence of Motion Amplitude: Increasing platform motion amplitude promoted the earlier development of coherent wake structures, with the onset advancing by 34% for pitch and 42% for roll in the investigated conditions.

    📊 Wake Structure and Energy Transport: Statistical wake analysis revealed that roll motion broadened the lateral wake distribution, while pitch motion produced a more symmetric wake response. An energy-budget analysis further demonstrated that wake recovery depends strongly on the direction and spatial organization of advective kinetic-energy transport, rather than turbulence enhancement alone.

    🧠 AI-Based Prediction of Moving Turbine Wakes: A motion-conditioned, physics-informed recurrent neural network (MoC-PiRNN) was developed to predict the evolution of aerodynamic flow fields under prescribed platform motions. By combining motion information, temporal learning and physics-based constraints, the model demonstrated the potential for predicting complex moving-boundary flow dynamics without repeatedly running full CFD simulations.

    ⚡ Improved Prediction Through Wake Perturbation Learning: A motion-conditioned neural network was developed to learn wake perturbations relative to a fixed-turbine reference flow. The approach reduced selected wake prediction errors by approximately 49–50% compared with baseline models, while improving the representation of motion-induced wake displacement and velocity deficits.

    🌍 Towards Faster Offshore Wind Engineering Tools: The integration of validated CFD simulations and AI-based surrogate models offers a promising pathway towards computationally efficient wake prediction. These developments could support future floating wind farm design, aerodynamic assessment and digital-twin applications.

  • Funding

    This project has received funding from the Internal PhD Scheme at Norwegian University of Life Sciences.

  • Participants

    Arvind Keprate

    Arvind Keprate

    Professor at OsloMet

Timeline

30 November 2026: Research Article Published: Ocean Engineering Journal - Learning wake perturbations of floating wind turbines using motion-conditioned neural network

A perturbation-learning approach is proposed for floating wind turbine wakes. Platform-motion wake changes are learned relative to a fixed-turbine wake. The model is conditioned on motion type, amplitude, frequency, and phase. Perturbation learning reduces velocity-deficit RMSE by 49.1%. Wake-center error is reduced by 43.6% compared with direct wake prediction.

06 July 2026: Conference Participation: OMAE 2026 - International Conference on Offshore Mechanics and Arctic Engineering- Exploratory Energy-Budget Analysis of a Floating Wind Turbine Wake Under Pitch and Roll Motions

Floating offshore wind turbines (FOWTs) enable access to deep-water wind resources but introduce platform motions that influence wake development and recovery. While previous studies have focused on velocity deficit and turbulence intensity, the redistribution of wake kinetic energy remains less explored. This study analyzes wake energy transport for a validated wind turbine model (MoWiTO 0.6) undergoing prescribed harmonic pitch and roll motions. Using time-resolved planar velocity fields, a pseudo-energy budget framework is developed to decompose the in-plane kinetic energy into advection, diffusion, and dissipation components, and to quantify downstream recovery via directional energy transport across wake crossplanes. The results show that pitch and roll motions lead to fundamentally different wake-recovery mechanisms. Pitch motion primarily modifies stream-wise energy transport while preserving lateral symmetry, leading to relatively uniform recovery. In contrast, roll motion enhances cross-plane energy transport, causing lateral asymmetry, uneven inflow, and non-uniform dissipation. The energy-based diagnostic framework developed in this work offers physical insight into wake-recovery behaviour by examining motion-dependent redistribution of kinetic energy.

23 June 2026: Conference Participation: 40th ECMS 40th ECMS International Conference on Modelling and Simulation: Simulating moving-boundary flows with neural surrogates: latent dynamics vs operator learning

This work compares two representative learning approaches: motion-conditioned latent dynamical models based on recurrent neural networks and motion-conditioned operator-learning approaches based on Fourier Neural Operators enhanced with autoregressive consistency and physics-based temporal regularization (AR-FNO). Using a heaving square cylinder as a canonical test case, we assess short-horizon reconstruction accuracy, error growth during autoregressive rollout, and spatial flow-field behaviour. A floating wind-turbine configuration is included as a supplementary case to assess performance in a more complex flow environment. Both models achieve comparable short-horizon accuracy. During extended autoregressive rollout, the ConvLSTM exhibits progressive phase drift and amplitude distortion, whereas the AR-FNO maintains improved phase consistency and lower cumulative RMSE. In the square-cylinder case, the AR-FNO reduces the long-horizon cumulative error growth by approximately 15-20% relative to the latentdynamical surrogate, thereby demonstrating improved performance.

03 October 2025: Research Article published: Coherent flow structures in the wake of a model floating wind turbine under pitch and roll motions

This research numerically studies the wake dynamics of a model FOWT under platform motion using high-fidelity CFD. It was found that higher oscillation amplitudes shift coherent wake structures closer to the rotor. The pitch motion drives pulsating wakes at low Strouhal numbers, shifting to meandering at high Strouhal numbers, whereas the rolling motion causes vortex merging, thereby increasing lateral spreading and mixing. Also, the optimal Strouhal number range (
  0.3–0.6) enhances wake recovery and aids wind farm design.

25 July 2025: Research Article published: Advances in computational intelligence for floating offshore wind turbines aerodynamics: Current state review and future potential

Our review article has been published in Renewable and Sustainable Energy Reviews (Elsevier). This journal is ranked Level 2 in Norway and is highly recognized in the field of renewable energy research. In this work, we provide a comprehensive overview of computational methods for floating offshore wind turbines (FOWTs), from low-fidelity engineering models to high-fidelity CFD approaches. We also highlight recent advances in data-driven modeling and AI applications for improving FOWT design and performance. The article further discusses the impact of wake aerodynamics, validation efforts, and the future opportunities for integrating AI and physics-based models in wind energy research. This publication is the result of strong collaboration across multiple institutions, and we hope it contributes to advancing sustainable offshore wind energy technologies.

9-12 June 2025: Presented research work at the WAKE conference in Visby Sweden

The iCOWE project was represented at the Wake Conference 2025, held at Uppsala University’s Gotland Campus in Visby, Sweden. Haris Hameed Mian presented the paper “Nonlinear Wake Dynamics of a Model Floating Offshore Wind Turbine Under Pitch and Roll Motions,” which explores the influence of platform-induced motions on wake behavior. The study uses advanced CFD simulations with an Improved Delayed Detached Eddy Simulation (IDDES) approach to analyze how pitch and roll dynamics affect turbulence, wake recovery, and downstream wind conditions. The findings provide valuable insights for optimizing floating wind farm layouts and support the objectives of the iCOWE project. The work is now published in the IOP Journal of Physics: Conference Series and reflects ongoing efforts to advance high-fidelity modeling techniques for offshore wind energy systems.

15-17 Janunary 2025: Presented posters in ERRA DeepWind 2025

Presented two posters at DeepWind 2025, showcasing cutting-edge advancements in wind farm optimization. The first poster focuses on Wind Farm Layout Optimization, where a machine learning-enhanced wake modeling approach, combined with SLSQP optimization, improves energy output at the Horns Rev 1 wind farm. The second poster explores Leveraging Data-Driven Techniques with LiDAR and SCADA Data, applying XGBoost and Bi-LSTM to analyze offshore wind farm performance, offering valuable insights into wind speed predictions and turbine efficiency. These studies highlight the power of data-driven methodologies in enhancing renewable energy solutions.

1st June 2024: Completed three-month research stay at Fraunhofer IWES, Oldenburg, Germany

During his time, the research was conducted on advancing and refining floating offshore wind turbine modeling and simulation using OpenFOAM, contributing to the cutting-edge research in this critical area of renewable energy technology. The financial support for this was provided from the internal funding scheme at the Norwegian University of Life Sciences (project number 1211130114).

Publications

Highlights

H H Mian, RSER Journal (2025)
Wake of Floating Wind Turbine
Rotor_Pitch_motion
Moored FOWT Floater
FOWT Floater Mesh
A neural network-based motion prediction of semi-submersible floating platform for offshore wind
Hydrodynamic Response of Semi-Submersible FOWT Floaters: A Numerical Investigation of Wave and Mooring Parameter Dependencies
Aerodynamic and Structural Assessment of Floating Wind Turbines Under Pitch Motion
Numerical investigation of blade roughness impact on the aerodynamic performance and wake behavior of horizontal axis wind turbine