Development of an early stage hydrofoil design tool including unsteady hydrodynamics

Conceptual design of hydrofoil craft and in particular hydrofoils is challenging since the hydrodynamics is governed by multiple complex physical phenomena such as the effect of the free surface, turbulence and unsteadiness due to waves and/or operations. High-fidelity simulations capturing these physical phenomena are computationally too demanding and costly to be used directly in conceptual hydrofoil design. An early stage hydrofoil design tool is developed by training a recurrent neural network on the unsteady lift and drag due to forced pitch motions and due to forced heave motions of a geometrically parameterized hydrofoil. By combining static and dynamic features, the machine learning model learns the unsteady response of the hydrofoil for an arbitrary set of parameters describing the geometry. The mean relative test error with respect to high-fidelity RANS results in lift and drag due to forced pitching is 13.16 % and 9.67 % respectively. For forced heave, we find 1.78 % and 5.78 % respectively. Evaluation of the machine learning models is very cheap. This makes it possible to solve multi-objective optimization problems in a limited amount of time by using the machine learning model as a surrogate in the optimization process. A web application is developed for this purpose, resulting in an useful tool for conceptual hydrofoil design. This development is followed by a verification and validation case and a case study to assess the value of the tool in practice. The tool can quickly propose new and interesting hydrofoil designs with favorable performances for given operational conditions. This motivates to explore new machine learning strategies and enhance tool development with the purpose to improve conceptual hydrofoil design.