Adapt, Adapt, Adapt: Recent Trends in Multi-fidelity Digital Modelling for Marine Engineering

The paper presents some recent trends in multi-fidelity digital modelling for marine engineering appli-cations. Digital modelling is achieved by machine learning methods, namely multi-fidelity surrogate models, trained by computational fluid dynamics (CFD). Adaptative approaches are discussed for ra-dial basis functions and Gaussian process models. Simulation-based design optimisation problems are presented to discuss the use and effects of different adaptivity concepts: (1) adaptive refinement of the computational-domain discretization in CFD; (2) adaptive sampling of the design/operational space; (3) adaptive selection of the fidelity used for the surrogate model training in a multi-fidelity environ-ment; (4) adaptivity of the models to noise. Model adaptation allows for the efficient training of ma-chine learning models, reducing the computational cost associated to building the training sets and improving the overall accuracy of the digital representation.