Framework for Single and Multi-Hazard Surrogate Models for Steel Moment Frames

This article introduces a surrogate modelling framework based on machine learning (ML) to forecast the response of steel special moment resisting frames (SMRFs) under single-hazard and multi-hazard conditions. While traditional methods for assessing structural performance, such as nonlinear time history analysis (NLTHA) and pushover analysis, are precise, they are also computationally demanding. To address this issue, a comprehensive dataset of 30,000 SMRFs is created, encompassing various material and geometrical uncertainties. The development of the surrogate model takes into consideration two common natural hazards: floods and earthquakes. In the case of single-hazard scenarios involving seismic loads, the study assessed multiple machine learning algorithms, ultimately finding that CatBoost demonstrated the highest predictive accuracy for key engineering demand parameters (EDPs), including inter-story drift ratios and floor accelerations. The research emphasises the significant impact of pushover parameters on NLTHA predictions and underscores the importance of integrating these parameters into seismic surrogate models. Graphical user interfaces (GUIs) have been designed for single-hazard scenarios, allowing engineers to quickly and accurately predict seismic responses, thereby reducing computational efforts. This framework is intended to be applied to multi-hazard scenarios considering flash flooding and seismic excitation, providing a practical and efficient alternative for structural performance assessment, ultimately contributing to improved multi-hazard structural safety and resilience.