Multi-Fidelity Adaptive Metamodel for Ship Hull Performance via CFD

The simulation-based design (SBD) process of complex engineering systems (such as ground, aerial, and marine vehicles) requires computationally expensive physic-based solvers, in order to achieve accurate solutions. Often, structural and/or computational fluid-dynamic (CFD) solvers are used in order to assess the design performance, e.g. the hydrodynamic performance of a ship hull with the resulting resistance force. The SBD process can integrate optimization algorithms in order to perform a fully-automated de- sign optimization. In this case, a large number of computer simulations is required to converge to the op- timal solution, and the computational cost of the process is usually very high. Furthermore, when dealing with real-life applications, uncertainties (stemming from environmental and operating conditions) must be taken into account in the design process, including uncertainty quantification (UQ) procedures and requiring very large computational resources. Metamodels are used to reduce the computational cost of the SBD and have been successfully applied in diverse engineering fields. Among others, accuracy and efficiency of radial basis functions (RBF) have been demonstrated for several engineering applications. In order to combine the accuracy of high-fidelity solvers with the computational cost of low-fidelity solvers, several multi-fidelity approximation methods have been developed. Combining metamodelling methods with multi-fidelity approximations potentially leads to a further reduction of the computational cost. The objective of the present work is to apply and assess the effectiveness and efficiency of a multi- fidelity global metamodel (Pellegrini et al., 2016a, Pellegrini et al., 2016b) able to manage high- and low-fidelity solvers through a multi-fidelity adaptive sampling procedure. The metamodel is used here to evaluate the ship hull performance versus the operating conditions.