Machine learning-based seismic fragility curves of regular infilled RC frames

Infilled reinforced concrete (RC) frame building is prone to collapse during earthquakes, highlighting the importance for seismic vulnerability assessment of infilled RC frame. Incremental Dynamic Analysis (IDA) is widely used for vulnerability assessment, which required to conduct nonlinear time history (NTH) analysis, leading to high computational costs and significant time consumption for extensive analyses. In this study, machine learning (ML) models which could accurately predict structural response of infilled RC frames are firstly developed, and are used to replace the NTH analysis and quickly obtain the structural dynamic response. Thereafter, ML methods are employed instead of the IDA process to directly predict the dynamic response of structures and subsequently establish fragility curves. This approach significantly reduces computational and time costs, better balances accuracy and efficiency. This paper firstly develops a seismic damage database for infilled RC frames utilizing the OpenSees finite element model. Two methods, IDA and ML, are employed to establish fragility curves for infilled RC frames. The results indicate that eight ML models can predict the structural response excitated by scaled ground motion, with the Artificial Neural Network (ANN) model demonstrating the highest prediction accuracy. Compared to other ML models, the fragility curves predicted by ANN model are generally more consistent with those derived from the IDA method. Eventually, ML method is applied to actual damaged buildings in Taiwan earthquake, indicating that the method of ML-based seismic fragility curves could accurately predict the damage state of infilled RC frames. © 2024 Elsevier Ltd