A PARALLEL NON-PROBABILISTICALLY-DOMINATED SORTING GENETIC ALGORITHM FOR ROBUST OPTIMIZATION

Successful engineering design generally requires the resolution of various conflicting design objectives. In order to identify an optimal design, conflicting objectives are typically considered simultaneously. One of the most powerful tools for resolving such objectives, in a computational setting, is multi-objective optimization. By definition, Pareto solutions are considered optimal because there are no other designs that are superior in all objectives. In a deterministic optimization the effect of uncertainty is not considered when drawing conclusions, hence we want to extend the concept of Pareto dominance in a probabilistic sense. In each evolution era, the different chromosomes can be evaluated with respect to the objective functions in parallel. For each one of the chromosome the Simplex Stochastic Collocation is used to handle uncertainty. The method is based on a serial sequential-solution refinement of the probability space until a defined threshold: different simplexes converge at different times. A nested loop of parallel-serial evaluations is handled by the R-Opt to achieve the best computational efficiency.