Optimising General Configuration of Wing-Sailed Autonomous Sailing Monohulls Using Bayesian Optimisation and Knowledge Transfer

Wing-sailed autonomous sailing monohulls are commonly designed for specific payloads and target sea areas. Their general configuration—namely, the size matching among sail, hull, and keel—requires a careful trade-off between safety and sailing efficiency. The associated design space is large, and reliable prior performance estimates are difficult to obtain from historical designs; consequently, conventional offline surrogate-based simulation-driven design typically demands extensive computational fluid dynamics (CFD) sampling at high computational cost.

This paper proposes a generalised simulation-driven framework that integrates Bayesian optimisation (BO) with knowledge transfer (KT). BO adaptively concentrates sampling near critical operating conditions, while KT surrogates reuse historical CFD information, enabling high-quality and computationally efficient configuration optimisation even in the absence of an initial design and prior estimates. Under the design constraints of the “Seagull” prototype (for example, a heel angle not exceeding 26°), the optimised configuration improves upwind speed by approximately 56% and downwind speed by approximately 106%; relative to offline methods, simulation cost can be reduced by approximately 40%–60%. The framework provides a transferable methodology for mission-specific autonomous sailboat design and may be extended to more complex systems involving hydrofoil coupling.