چكيده به لاتين
In this study, the optimal design of space structures with semi-rigid joints is investigated using a surrogate model based on machine learning. Space structures are a class of structures that exhibit dominant three-dimensional behavior, effectively integrating technical constraints, economic factors, safety principles, and aesthetic considerations. In the typical design process of space structures, joints are ideally assumed to be either rigid or pinned for simplicity. However, the real behavior of joints in space structures is often closer to semi-rigid, and these idealized assumptions neglect the interactions between elements and connections. Designing structures with near-rigid connections leads to increased steel consumption in the connections, resulting in higher overall weight. On the other hand, using pinned joints increases the need for larger member sections, which can also lead to an increase in the structure's weight. Therefore, designing space structures with semi-rigid joints, which account for the realistic behavior of the joints, can contribute to reducing the structure’s weight in the optimal design. Considering that 15% to 45% of the total weight of steel space structures comes from the joints, this study, unlike many optimization problems, includes the weight of the joints in the objective function. The optimal design of space structures, considering the stiffness of the connections, incurs high computational costs when dealing with large-scale models. To manage these computational costs, a surrogate model based on the XGBoost machine learning algorithm is employed in this study. Since the accuracy and performance of surrogate models are highly influenced by the quality of the generated samples, and generating a large number of samples is computationally expensive, a novel active learning approach is adopted in this study for sequential sampling and updating of the surrogate model. The numerical results show that space structures with semi-rigid joints have 4.25% and 14.48% less weight compared to structures with pinned and rigid joints, respectively. Additionally, the surrogate model used in this study can find the optimal solution with fewer analyses compared to the ECBO metaheuristic algorithm