Efficient simulation-based design optimization for fluid-structure interaction problems affected by uncertainty

A strategy is presented for efficient simulation-based Multidisciplinary Robust Design Optimization (MRDO) of fluid-structure interaction problems affected by uncertainty. The focus is on a racing-sailboat fin, subject to stochastic operating conditions. The elastic deformation of the fin induced by hydrodynamic loads cannot be neglected while evaluating the hydrodynamic performances, thus a fully coupled hydroelastic problem is considered, including fluid mechanics (CFD) and structural analysis (FEM). The multidisciplinary analysis (MDA) identifies the multidisciplinary equilibrium by numerical iterations. The stochastic operating scenario is identified by speed and yaw angle, which are defined by means of a probability density function. The distributions of the relevant output parameters are evaluated using uncertainty quantification methods (UQ), requiring a large number of multidisciplinary analyses (MDA). Solving the MRDO problem represents a challenge from the algorithmic and computational viewpoints, since requires coupling together: (1) a minimization algorithm, (2) a UQ tool and (3) simulation-based MDAs. The objective of the present work is the development and validation of an efficient strategy for MRDO. Specifically, the objective function is the expected value of the fin efficiency - lift to drag ratio (CL/CD) - over the stochastic operating conditions, whereas the design variables pertain to the fin geometry, which is modified using a Free-Form Deformation (FFD) technique; the MDA is solved iteratively with a variable level of coupling between the disciplines involved; the UQ is carried out using the Monte Carlo method, based on metamodels; design optimization is solved using subsequent Design of Experiments (DoE), metamodels, and Particle Swarm Optimizations (PSO). During the optimization process, a DoE refinement is performed on smaller design windows with an increasing density of numerical experiments, level of UQ accuracy and MDA coupling. The method is validated versus a standard MRDO, solved by fully coupled MDAs without metamodels. Effectiveness and efficiency of the method are evaluated in terms of optimal design performances and number of simulations required.