Framework for AI-Enhanced Offshore Wind Farm Digital Twins

Abstract

This dissertation presents a Semantic Digital Twin Framework (SDTF) for offshore wind farm planning that integrates large language models (LLMs), semantic modeling, and physics-based simulation into a unified, knowledge-driven system. The motivation for this research stems from the increasing complexity of offshore wind energy development, which requires the integration of heterogeneous data sources, regulatory constraints, environmental considerations, and engineering models within a coherent decision-making framework. The proposed SDTF enables the transformation of unstructured domain knowledge—such as regulations, technical documents, and environmental reports—into structured, machine-interpretable representations using LLM-based knowledge extraction. The extracted information is encoded into ontologies and rules within an RDF/OWL-based semantic graph, implemented using the Apache Jena framework. Rule-based reasoning, including geospatial reasoning through GeoSPARQL, allows the system to perform automated compliance checking, infer operational states, and dynamically respond to environmental events. To support engineering analysis, the semantic layer is coupled with simulation and optimization tools, including TopFarm for wind farm layout optimization, PyWake for wake modeling, and OpenFAST/WEIS for aero-servo-elastic turbine simulations. This integration enables a closed-loop workflow in which semantic reasoning informs optimization and simulation, while computed results are continuously fed back into the knowledge graph for validation and further inference. A key contribution of this work is the incorporation of event-driven reasoning and generative modeling for extreme weather scenarios. Synthetic hurricane trajectories are generated using an LSTM-based variational autoencoder, enabling the evaluation of turbine responses under realistic but previously unobserved conditions. Semantic rules are used to trigger adaptive turbine control actions, such as yaw alignment and shutdown, demonstrating the ability of the framework to support dynamic and context-aware decision-making. The framework is validated through a case study of the Maryland offshore wind lease area, incorporating real-world marine spatial planning data, environmental constraints, and turbine specifications based on the IEA 15-MW reference wind turbine. Results demonstrate that the SDTF can effectively integrate diverse knowledge sources, enforce regulatory compliance, and support optimization under both normal and extreme operating conditions. This research advances the state of the art in digital twin technologies by introducing a knowledge-augmented, semantically grounded approach that bridges symbolic reasoning and numerical simulation. The proposed framework provides a scalable and extensible foundation for intelligent infrastructure systems, enabling explainable, adaptive, and data-driven decision-making in offshore wind energy and beyond.