Local Latin hypercube refinement for multi-objective design uncertainty optimization

Optimizing the reliability and robustness of a design is important but often unaffordable due to high sample requirements. Surrogate models based on statistical machine learning methods are used increase efficiency. However, for higher dimensional or multi-modal systems, surrogate may also require large amount samples achieve good results. We propose sequential sampling strategy solution multi-objective robust optimization problems. Proposed local Latin hypercube refinement (LoLHR) model-agnostic can be combined with any model because there no free lunch possibly budget one. The proposed method compared stationary as well other strategies from literature. Gaussian process support vector regression both models. Empirical evidence presented, showing that LoLHR achieves average better results tested examples.