Periodic Reporting for period 1 - ICONIC (Smart, Aware, Integrated Wind Farm Control Interacting with Digital Twins (ICONIC))
Reporting period: 2023-12-01 to 2025-05-31
Wind energy is central to Europe’s strategy for achieving climate neutrality, energy independence and global leadership in clean technologies. Offshore wind is expected to grow rapidly in the coming decades, with turbines reaching 20 MW capacity and wind farms expanding into large, dense arrays. These assets will be closely connected to digital, inverter-based power grids and expected to provide not only electricity but also flexibility and grid support services.
This transformation raises critical challenges. Wake effects between turbines reduce energy yield and increase fatigue loads. Current control systems are mostly rule-based, designed for individual turbines rather than entire farms, and cannot adapt dynamically to turbulence, grid requests or component degradation. Monitoring remains reactive rather than predictive, limiting the ability to move towards condition-based maintenance. Finally, data infrastructures are fragmented and proprietary, hindering the interoperability and scaling of digital twin solutions.
ICONIC (Smart, Aware, Integrated Wind Farm Control Interacting with Digital Twins) addresses these challenges by creating a new generation of wind farm technology that is intelligent, adaptive and cybersecure. The project integrates modelling, control and monitoring into a unified framework, combining physics-informed artificial intelligence, digital twins at multiple scales (farm, turbine and component) and advanced predictive control.
ICONIC’s key objectives are to:
1. Develop digital twins for flow reconstruction, load estimation and degradation prediction;
2. Advance turbine-level control through hybrid and stochastic model predictive control;
3. Optimize farm-wide performance with cooperative, AI-based control strategies;
4. Enable predictive maintenance with Remaining Useful Life (RUL) estimation;
5. Validate models and controllers through wind tunnel tests and offshore operational data;
6. Deliver an open, cybersecure software toolchain compatible with industrial SCADA systems.
Expected impacts include reducing operating costs and the levelized cost of energy (LCOE) by 5–8%, extending component lifetimes through condition-based maintenance, improving grid integration, and fostering innovation and transparency through open-source digital tools.
This transformation raises critical challenges. Wake effects between turbines reduce energy yield and increase fatigue loads. Current control systems are mostly rule-based, designed for individual turbines rather than entire farms, and cannot adapt dynamically to turbulence, grid requests or component degradation. Monitoring remains reactive rather than predictive, limiting the ability to move towards condition-based maintenance. Finally, data infrastructures are fragmented and proprietary, hindering the interoperability and scaling of digital twin solutions.
ICONIC (Smart, Aware, Integrated Wind Farm Control Interacting with Digital Twins) addresses these challenges by creating a new generation of wind farm technology that is intelligent, adaptive and cybersecure. The project integrates modelling, control and monitoring into a unified framework, combining physics-informed artificial intelligence, digital twins at multiple scales (farm, turbine and component) and advanced predictive control.
ICONIC’s key objectives are to:
1. Develop digital twins for flow reconstruction, load estimation and degradation prediction;
2. Advance turbine-level control through hybrid and stochastic model predictive control;
3. Optimize farm-wide performance with cooperative, AI-based control strategies;
4. Enable predictive maintenance with Remaining Useful Life (RUL) estimation;
5. Validate models and controllers through wind tunnel tests and offshore operational data;
6. Deliver an open, cybersecure software toolchain compatible with industrial SCADA systems.
Expected impacts include reducing operating costs and the levelized cost of energy (LCOE) by 5–8%, extending component lifetimes through condition-based maintenance, improving grid integration, and fostering innovation and transparency through open-source digital tools.
Farm-level modelling. ICONIC developed DeepWake, a hybrid 3D wake model combining analytical equations with deep learning trained on high-fidelity simulations. It delivers fast, accurate predictions suitable for real-time control and optimization. A complementary approach using Physics-Informed Neural Networks (PINNs) reconstructs flow fields from sparse sensor data, embedding physical laws directly into neural networks.
Farm-level control. The consortium introduced MAOBI (Multi-Agent Offline Behaviour Imitation), an AI framework where each turbine acts as a cooperative agent. Training is performed offline using historical or simulated data, avoiding risks to operational turbines. Tested on both onshore and offshore layouts, MAOBI achieved up to 7% energy gains compared to baseline controllers while also reducing wake losses and balancing structural loads.
Turbine-level control. ICONIC created hybrid models that fuse physics-based equations with machine learning to estimate hard-to-measure quantities such as effective wind speed and structural loads. These models are interpretable, real-time capable, and integrate into predictive controllers. Two new controllers were developed, which include: Stochastic MPC (sMPC): adapts blade pitching under uncertainty, reducing actuator fatigue while maintaining safety; and tube MPC: a robust yet efficient design that contains disturbance effects within a bounded range.
Both outperformed conventional baseline controllers in turbulent inflow simulations.
Component twins and sensing. Advanced models of gearboxes, bearings and generators were developed. The fibre-optic strain sensors demonstrated torque measurement accuracy below 1%. Bearing anomaly detection reached 84% accuracy on real-world datasets, with potential to exceed 95% as models further mature in near future. These advances enable condition-based maintenance and more accurate lifetime predictions.
Experimental validation. A major wind tunnel campaign produced high-resolution flow data measurements, enabling validation of ICONIC’s wake models and controllers. The Rolling Contact Fatigue rig with higher speed and early failure detection is upgrade, generating valuable degradation data for component twins. Preparations for offshore validation at C-Power’s wind farm are underway, including integration of ICONIC algorithms into turbine control hardware.
Cybersecurity and toolchain. ICONIC prototyped privacy-preserving optimization using Secure Multi-Party Computation and anomaly filters against cyber-attacks. Work began on the open, modular software toolchain that will integrate models, controllers and data streams across multiple scales.
Farm-level control. The consortium introduced MAOBI (Multi-Agent Offline Behaviour Imitation), an AI framework where each turbine acts as a cooperative agent. Training is performed offline using historical or simulated data, avoiding risks to operational turbines. Tested on both onshore and offshore layouts, MAOBI achieved up to 7% energy gains compared to baseline controllers while also reducing wake losses and balancing structural loads.
Turbine-level control. ICONIC created hybrid models that fuse physics-based equations with machine learning to estimate hard-to-measure quantities such as effective wind speed and structural loads. These models are interpretable, real-time capable, and integrate into predictive controllers. Two new controllers were developed, which include: Stochastic MPC (sMPC): adapts blade pitching under uncertainty, reducing actuator fatigue while maintaining safety; and tube MPC: a robust yet efficient design that contains disturbance effects within a bounded range.
Both outperformed conventional baseline controllers in turbulent inflow simulations.
Component twins and sensing. Advanced models of gearboxes, bearings and generators were developed. The fibre-optic strain sensors demonstrated torque measurement accuracy below 1%. Bearing anomaly detection reached 84% accuracy on real-world datasets, with potential to exceed 95% as models further mature in near future. These advances enable condition-based maintenance and more accurate lifetime predictions.
Experimental validation. A major wind tunnel campaign produced high-resolution flow data measurements, enabling validation of ICONIC’s wake models and controllers. The Rolling Contact Fatigue rig with higher speed and early failure detection is upgrade, generating valuable degradation data for component twins. Preparations for offshore validation at C-Power’s wind farm are underway, including integration of ICONIC algorithms into turbine control hardware.
Cybersecurity and toolchain. ICONIC prototyped privacy-preserving optimization using Secure Multi-Party Computation and anomaly filters against cyber-attacks. Work began on the open, modular software toolchain that will integrate models, controllers and data streams across multiple scales.
ICONIC’s achievements already extend well beyond current practice by:
• Delivering DeepWake, a real-time hybrid wake model for adaptive control and layout optimization;
• Introducing MAOBI, an offline AI controller for safe, cooperative wake steering and energy maximisation;
• Creating hybrid turbine models and predictive controllers that combine interpretability, adaptability and load reduction;
• Building component-level digital twins and novel sensors validated with laboratory rigs and offshore data, enabling predictive RUL monitoring;
• Establishing experimental validation platforms that bridge simulation, laboratory testing and field data;
• Progressing toward an open, cybersecure software stack that ensures industrial compatibility and trust.
Together, these innovations form a coherent ecosystem where modelling, control and monitoring reinforce each other, paving the way for smarter, more sustainable offshore wind farms.
• Delivering DeepWake, a real-time hybrid wake model for adaptive control and layout optimization;
• Introducing MAOBI, an offline AI controller for safe, cooperative wake steering and energy maximisation;
• Creating hybrid turbine models and predictive controllers that combine interpretability, adaptability and load reduction;
• Building component-level digital twins and novel sensors validated with laboratory rigs and offshore data, enabling predictive RUL monitoring;
• Establishing experimental validation platforms that bridge simulation, laboratory testing and field data;
• Progressing toward an open, cybersecure software stack that ensures industrial compatibility and trust.
Together, these innovations form a coherent ecosystem where modelling, control and monitoring reinforce each other, paving the way for smarter, more sustainable offshore wind farms.