Metadatabase Creation, Stochastic Pushover Analysis, and Feature Importance Study of Steel Frame Buildings

Policymakers and property owners have been seeking effective disaster risk management strategies by assessing asset resilience and vulnerability. Typically, nonlinear time history analysis (NLTHA) is used to evaluate the performance of structures for future hazards, but the method is computationally expensive and not suitable for analysing a large number of structures at a city-wide scale. While surrogate models can be an efficient and reliable alternative to NLTHA for evaluating the performance of multiple structures across various hazard scenarios, creating a proper database for surrogate modelling at a city level is challenging. To address this issue, this study proposes a metadatabase for surrogate modelling of steel structures where the framework is hazard-independent and can be utilised worldwide. This metadatabase requires information about the structure’s geometry (such as the number of stories and length of bays) as well as the boundary of the dimensions of beams and columns. In a total of 30,000 steel moment resisting frame buildings were randomly generated which was deemed as sufficient. The creation process checks appropriate criteria for connections and strength of beams and columns to ensure that the final database adheres to contemporary seismic codes. The preliminary design is based on resiliency and practicality criteria and uses pre-designed dimensions for beams and columns. Following this, pushover analysis is conducted for 30,000 buildings and some pushover and modal parameters such as ductility, overstrength, and modal participation ratio are extracted. Lastly, incorporating a random forest machine learning algorithm, sensitivity analyses of the structural parameters are conducted to identify the most crucial factors in designing steel moment resisting frames.