A novel time-dependent system constraint boundary sampling technique for solving time-dependent reliability-based design optimization problems

摘要

Solving the time-dependent reliability-based design optimization (TRBDO) problems is quite cost-consuming in realistic engineering application. Surrogate model-based TRBDO methods can provide a good tradeoff between accuracy and efficiency of the solution. However, the existing surrogate model-based TRBDO methods pay efforts on enhancing surrogate models of constraint functions in both feasible region and infeasible region. While only the estimation accuracy of boundaries between feasible region and infeasible one has large effects on the calculation accuracy of design parameter, and the estimation accuracy of limit state surfaces of constraint functions in infeasible region has little effects on the calculation accuracy of design parameter. Therefore, it is no major demand to deliberately enhance the accuracy of surrogate models of constraint functions in the infeasible region, and the existing surrogate model-based TRBDO methods introduce extra computational cost. In this paper, a new surrogate model-based method called time-dependent system constraint boundary sampling (TSCBS) is proposed to overcome the limitation of existing surrogate model-based TRBDO methods. By employing the TSCBS, samples around the boundaries between feasible region and infeasible one are selected, and samples in the infeasible region far away from the boundaries are avoided as much as possible, to improve the surrogate models of constraint functions. Furthermore, in order to obtain the feasible region of the TRBDO, the safe region of each probabilistic constraint is to be sure first with an established adaptive coefficient. The adaptive coefficient can provide a good tradeoff between the input dimension and iteration. Numerical and engineering examples show that the proposed TSCBS generally performs higher computational efficiency than existing surrogate model-based TRBDO methods. (c) 2020 Elsevier B.V. All rights reserved.