Surrogate-Based Optimization of Steel Girders for Cost and Dynamic Performance

Resumo

Multi-objective optimization and surrogate modeling have become essential tools in pursuing sustainable and efficient structural design. This study investigates the optimization of steel girders of pedestrian bridges, focusing on two key objectives: cost and natural frequency. Natural frequency is a critical parameter for the dynamic performance of the structure, directly influencing pedestrian comfort through its role in calculating structural accelerations. To address the high computational cost associated with structural analysis, the Kriging surrogate model is employed, offering accurate predictions while significantly reducing runtime. The optimization framework integrates the Kriging model with the Multi-Objective Harmony Search (MOHS) algorithm and high-fidelity finite element method (FEM) simulations in a Python-based environment. Results demonstrate the effectiveness of the surrogate model, achieving a predictive relevance of 97% with minimal error, while enabling a 98.79% reduction in optimization runtime compared to direct FEM simulations. The multi-objective optimization produced a Pareto front of non-dominated solutions, revealing valuable trade-offs between cost and natural frequency. For instance, the lowest-cost solution achieved a natural frequency of 13.38 Hz, while a 3.87% increase in cost resulted in a frequency of 19 Hz, substantially improving dynamic performance and pedestrian comfort. These results underscore the potential of employing surrogate models to enable fast and efficient multi-objective optimization of structural systems. The optimization framework was also validated by comparing its results to existing literature, where the developed method provided competitive solutions. The lowest-cost solution identified is comparable to the reference benchmark, despite differences in constraints and assumptions. These findings highlight the effectiveness of combining surrogate models and multi-objective optimization for complex structural problems, demonstrating their potential to promote the adoption of sustainable and efficient design practices. By reducing computational costs and offering actionable insights, the proposed methodology opens new opportunities for large-scale applications in real-world engineering projects.