A conservative multi-fidelity surrogate model-based robust optimization method for simulation-based optimization

Multi-fidelity (MF) surrogate model-based robust optimization has been used to deal with engineering design and optimization problems that have uncertainty in their inputs. However, the MF surrogate model constructed by a limited number of samples ineluctable has prediction uncertainty, which often leads to the optimal solutions becoming infeasible. In this paper, a MF surrogate model-assisted semi-nested variable adjustment robust optimization (CMF-SN-VARO) method is proposed to address the impact of prediction uncertainty of the MF surrogate model during the optimization process. A piecewise conservative MF surrogate modeling method is proposed to replace the objective functions and constraints, in which the safety margin is calculated by different error metrics according to their performance in problems with different dimensions. The variable adjustment robust optimization (VARO) framework is adopted to solve the robust optimization problem by adjusting the preexisting design. A switch criterion is utilized to adaptively determine when to use surrogate models or inner optimization problems to evaluate the robustness index of design alternatives. The performance of the proposed method is illustrated with an analytical example, a torque arm design problem, and a micro aerial vehicle fuselage design problem. Results show that the proposed method achieves better optimal design solutions that are both objective robust and feasibility robust.