Uncertainty-aware Bayesian optimisation framework for the design of floating offshore wind turbines

Floating wind platforms are complex systems requiring integrated optimization of interdependent subsystems under highly dynamic conditions. Conventional design approaches often treat the process as a black-box optimization problem, minimizing a cost metric based on simplified assumptions. However, these assumptions evolve during project development, and inaccuracies in early-stage models can significantly affect outcomes. This study introduces an uncertainty-aware design paradigm in which uncertainty is not a byproduct but a central element guiding the optimization process and informing decision-making. The proposed framework embeds Bayesian Optimization within the design workflow, leveraging Gaussian Process regression to model both the objective function and its associated uncertainty. This probabilistic representation enables explicit integration of uncertainty into the optimization strategy, while the Lower Confidence Bound acquisition function balances exploration and exploitation according to user-defined preferences. The approach reduces the number of objective function evaluations and provides valuable insights into cost distributions. Its effectiveness is demonstrated through three case studies of increasing complexity: a multi-modal objective function, the mass optimization of a catenary mooring system, and the same system with penalization terms. Compared to a Genetic Algorithm applied to an equivalent problem, the proposed method achieves convergence with approximately 80 % fewer evaluations.