A variable-accuracy metamodel-based architecture for global MDO under uncertainty

作者：Leotardi Cecilia; Serani Andrea; Iemma Umberto; Campana Emilio F; Diez Matteo

来源：Structural and Multidisciplinary Optimization, 2016, 54(3): 573-593.

DOI：10.1007/s00158-016-1423-4

• Simulation-based design
• Multidisciplinary robust design optimization (MRDO)
• Uncertainty quantification (UQ)
• Metamodel-based optimization
• Global optimization

A method for simulation-based multidisciplinary robust design optimization (MRDO) of problems affected by uncertainty is presented. The challenging aspects of simulation-based MRDO are both algorithmic and computational, since the solution of a MRDO problem typically requires simulation-based multidisciplinary analyses (MDA), uncertainty quantification (UQ) and optimization. Herein, the identification of the optimal design is achieved by a variable-accuracy, metamodel-based optimization, following a multidisciplinary feasible (MDF) architecture. The approach encompasses a variable (i) density of the design of experiments for the metamodel training, (ii) sample size for the UQ analysis by quasi Monte Carlo simulation and (iii) tolerance for the multidisciplinary consistency in MDA. The focus is on two-way steady fluid-structure interaction problem, assessed by partitioned solvers for the hydrodynamic and the structural analysis. Two analytical test problems are shown, along with the design of a racing-sailboat keel fin subject to the stochastic variation of the yaw angle. The method is validated versus a standard MDF approach to MRDO, taken as a benchmark and solved by fully coupled MDA, fully converged UQ, without metamodels. The method is evaluated in terms of optimal design performances and number of simulations required to achieve the optimal solution. For the current application, the optimal configuration shows performances very close to the benchmark solution. The convergence analysis to the optimum shows a promising reduction of the computational cost.

• 出版日期2016-9