Topology Optimization Benchmark Problems for Assessing the Performance of Optimization Algorithms

摘要

This paper presents a set of benchmark topology optimization problems that are used to evaluate the performance of optimization software. The topology optimization formulations considered in this benchmark set include mass-constrained compliance minimization, stress-constrained mass minimization, mass and stress-constrained compliance minimization, and mass and frequency-constrained compliance minimization. Both structured and unstructured quadrilateral and hexahe-dral meshes are used with conventional and non-conventional topology optimization design domains. In total, the benchmark set contains 108 2D and 72 3D domain and mesh combinations. In this preliminary work, the performance of the optimizers SNOPT, IPOPT and ParOpt are evaluated. Performance profiles are used to assess the performance of an optimizer across the full benchmark set. We find that SNOPT performs the best overall among all optimizers. Furthermore, a modification of the quasi-Newton Hessian update provides a considerable performance benefit in both objective function value and discreteness measure.