Variable-Fidelity Optimization of Stiffened Panels

摘要

Genetic algorithm optimization approaches have grown in popularity over the last decade because of their robustness in handling difficult optimization problems and their ability to develop complete Pareto fronts in a single execution of the optimizer. Unfortunately, this robustness comes with a price; genetic algorithms often require numerous objective function evaluations in the process of solving the optimization problem. This makes it difficult to maintain practical turn-around times when coupling genetic algorithms with computationally - expensive high-fidelity objective functions such as non-linear finite element analysis for structural design. As a potential solution to this problem, a variable-fidelity optimization scheme is proposed using a Kriging model constructed on-line to scale the results of rapid but lower-fidelity strength methods to that of a smaller number of high-fidelity finite element simulations. This variable-fidelity approach is demonstrated for a stiffened panel design problem with a modern cluster-computer approach.