Dottorato in ENERGIA E AMBIENTE

Titolo della tesi: High-Performance Computing Tools for Simulating Offshore Wind Turbines in Deep-Water Environments

The work presented in this thesis is motivated by a practical, engineering-driven question: how do atmospheric boundary-layer conditions and sea waves modify near-surface shear, wake development, and wind-turbine performance in Mediterranean scenarios? This question naturally arises in the context of accelerating interest in offshore wind as one of the most promising renewable energy technologies, especially for deep-water installations in the Mediterranean Sea, where floating platforms and Mediterranean metocean conditions reshape the operational environment compared with Northern European sites. To answer this question, from both physical and engineering perspectives, tools that can resolve the joint wave–wind–wake–turbine dynamics are required. This work adopts a high-fidelity prediction-science perspective, where offshore wind-energy predictions sit at the intersection of physics-based simulation, high-performance computing, and data-driven analysis and machine-learning modeling. Two complementary aims follow from this perspective. First, to develop and validate high-fidelity computational fluid dynamics (CFD) frameworks that incorporate realistic atmospheric boundary-layer (ABL) and sea-state effects while capturing rotor-wake dynamics. Second, to extract coherent physical structures from the resulting high-fidelity datasets by means of a lightweight, unsupervised pipeline designed to operate at runtime. To address the primary physics aim, two different simulation frameworks were developed, both built around actuator-line modeling but implemented within two different numerical solvers: i) UTD-WF, a high-order finite-difference large-eddy simulation (LES) code developed at the University of Texas at Dallas, and ii) OpenFOAM, an open-source finite-volume code. Across the two frameworks, the central question is when ocean waves inject enough energy into the surface layer to alter rotor-scale metrics, and when their imprint remains confined to the surface layer. For the UTD-WF simulation campaign, a case study located in the Sicilian Strait was investigated, where the metocean climate is swell-dominated, with a representative wave age of 1.17 and prevalent wind–wave alignment. By contrast, the OpenFOAM framework was validated through an improved-delayed detached-eddy-simulation (IDDES) campaign at Tavolara Island (Sardinia), focusing on frequent, moderate sea states. Moreover, the thesis introduces a data-driven methodology for the automatic identification and segmentation of the wake regions of a horizontal-axis wind turbine. The approach relies on fluid-dynamic features of the wind-turbine flow field and a k-means clustering algorithm trained on instantaneous two-dimensional high-fidelity CFD data, then temporally propagated onto new time steps. The result is a data-driven model that is both mesh- and solver-agnostic, able to identify physically interpretable wake regions in a pipeline lightweight enough to be deployed at runtime during CFD simulations.