A rcGAN-based surrogate model for nonlinear seismic response analysis and optimization of steel frames

The combination of the surrogate model and optimization algorithm to solve structural optimization problems is an efficient way to lower computational costs and reduce time consumption. However, the development of surrogate models for structural analysis frequently faces challenges due to limited datasets. To tackle this issue, this paper presents a surrogate model capable of training on limited datasets while simultaneously predicting multiple concerned indicators, and demonstrates its effectiveness in performance assessment and design optimization through two seismic design case studies. Specifically, an improved model architecture based on the conditional Generative Adversarial Network (cGAN) is proposed. The feasibility of this surrogate model for seismic response analysis and optimization is initially demonstrated using an existing planar frame case. Subsequently, to validate the suitability of the surrogate model for Nonlinear Time History Analysis (NTHA) tasks, the proposed approach is applied to optimize a 3D steel frame equipped with nonlinear viscous dampers. Herein, a three-objective optimization problem is formulated, employing the Non-Dominated Sorting Genetic Algorithm (NSGA-II), driven by the trained rcGAN, to identify the Pareto front. The optimum design is subsequently selected from this front utilizing a multi-criteria decision-making technique. The outcomes from three optimization tests indicate that our approach effectively enhances the seismic performance of the frame while achieving substantial economic benefits, ultimately reducing the construction cost of the benchmark structure by up to 31.1 %.