Optimal design of frame structures equipped with viscous dampers using machine learning techniques

It is widely accepted that the use of supplemental damping system is an effective measure to improve seismic performance of buildings. This study aims to develop an automated optimization design method for structures with damping systems to rapidly determine the optimal parameters and placement of viscous dampers in such structures. To achieve this, the proposed method involves the Machine Learning (ML) techniques of Extreme Gradient Boosting (XGBoost) and a Dynamic Programming (DP) algorithm. To generate the dataset for the XGBoost model development, an automated modeling and simulation tool for damping structures was developed. The optimization performance of the framework was validated using three different structural examples (3-story, 6-story, and 8-story). The results demonstrate that the XGBoost is capable of predicting the story drift response of a building structure under any damper arrangement. The built models achieved an average R2 of 0.955 for the testing dataset, indicating a high level of predictive accuracy. Then, with the target of minimizing seismic responses of the structural system, the developed DP-based optimization algorithm effectively identifies the optimal damper arrangement for the considered three structure cases. With varying initial conditions and parameters, the algorithm consistently converges toward the optimal state, demonstrating substantial robustness. The findings validate the efficacy of the proposed methodology in optimizing the seismic design of structures with viscous dampers.