A reliability-based design optimization strategy using quantile surrogates by improved PC-kriging

In recent years, the surrogate-assisted reliability-based design optimization (RBDO) methods have been continuously developed, and numerous advanced optimization strategies have boosted efficiency and accuracy. However, ensuring sufficient accuracy and feasibility at the optimal is still a challenge. In order to achieve a well-balanced between efficiency, accuracy, and optimal feasibility, in this work, a RBDO strategy using quantile surrogates by improved PC-Kriging model is proposed. The novelty of the proposed method lies in the following main aspects: Firstly, an improved learning function has been developed to significantly enhance the convergence efficiency during the construction of the PC-Kriging model. Secondly, in the RBDO analysis process, a novel "MP+EI" combination point addition strategy is adopted to enhance the approximation of the surrogate model to the optimum of the objective function. It can further improve optimization efficiency and accuracy. On the basis of the rough probability constrained surrogate model established by the global enrichment strategy, a local refinement strategy is introduced to guarantee the accuracy of the quantile evaluation of the probability constrained surrogate model for each iteration solution during the optimization process. Finally, the proposed method is validated by three typical RBDO test examples and one engineering application example.