Efficiency improvement of optimal design for laminated composite structures using Bayesian optimization

Abstract

The present study applies the Bayesian optimization (BO) technique using a regression model obtained by Gaussian process to the open-hole tension (OHT) problem of composite structures. The design variables are lamination parameters that directly define stiffnesses for the total thickness of the composite plate without fiber orientation angles in each layer. To investigate the tradeoff relations in the present multi-objective optimization problem, Pareto solutions were obtained by non-dominated sorting of the samples explored by the BO. As a result of the numerical analysis, enough samples close to the Pareto solutions could be found with a small number of analyses. On the other hand, samples equivalent to single-objective optimization could not be found. This confirms the need to adjust the acquisition function to improve searching ability.