A novel design framework: shape optimization of a Moth class hull with high-fidelity RSM technique trained on CFD simulations

This work is aimed at providing a new design framework for hull shape optimization with a high fidelity surrogate model trained on CFD results. The geometry of the hull is initially generated with 11 parameters in Grasshopper (a Rhinoceros 3D plug-in). All the hydrostatics coefficients relative to the hull are exported as an output in the modeFRONTIER project and the geometry is simulated with StarCCM+ in order to obtain the drag of the hull in addition to the previous coefficients. An initial DOE is created with the ULH algorithm and subsequently enriched with an Adaptive Space Filling method, in order to improve the predicting capabilities (evaluating the RSquared value) of the Neural Network generated RSM. Subsequently we use the RSM to optimize the hull shape with an NSGA-II genetic algorithm, replacing the CFD simulations in StarCCM+ to reduce the computational cost of the whole process.