Multi-Fidelity Optimization of High Speed SWATHs Available to Purchase

One of the great challenges of SBD (Simulation Based Design) optimization is balancing the trade-off between the required computational effort and the accuracy of the numerical solver selected to evaluate multimodal objective functions. Two conflicting tasks need to be targeted in an optimization problem: on one hand objective functions must rely on accurate flow solvers able to capture complex multi-phase physics, on the other hand fast computational tools are required to perform a very large amount of numerical simulations. As of now, potential flow solvers have represented the standard, rising concerns about the excessive simplifications at the basis of their formulation. However, for highly multi-dimensional problems, global optimization algorithms based on URANS objective functions predictions are nowadays still computationally too expensive. We address this dualism by presenting a re-formulated multi-fidelity framework that allows nonlinear and discontinuous correlations between shape parameters and quantities of interest. Resistance in calm water is investigated for a SWATH hull in the multi-dimensional design space using a new method to derive a high quality response surface through Bayesean inference and machine learning techniques based on a low number of high-fidelity computations and a larger number of less-expensive low fidelity computations. A first verification and validation study is presented with the goal of comparing and ranking numerical methods against experiments performed on a conventional SWATH geometry. Then the hull geometry of a new family of unconventional SWATH hull forms with twin counter-canted struts is parametrically defined and used in the optimization. Ship resistance in calm water is finally predicted using observations coming from two different fidelity levels. It will be demonstrated that the multi-fidelity optimization framework is successful in finding the true optimum by a limiting number of high-fidelity computations and a larger number of low fidelity computations. Simulation and optimization costs are reduced by orders of magnitude providing accurate certificates of fidelity of the performance of the proposed design.