Machine Learning‐Based Prediction of Drift Limit States of Concrete‐Filled Steel Tube Columns

This study discusses a machine learning (ML)‐based framework for estimating the drift limit states of concrete‐filled steel tube (CFT) columns. The calibration process used a comprehensive database assembled from numerous experiments reported the in literature. An ML regression model was trained based on the calibration data to predict numerical model parameters. Subsequently, 10,000 CFT column design parameters were randomly sampled, and a parametric study was conducted to investigate the influence of the design parameters on the drift limit states using an ML model and the simulation results obtained from the developed numerical model. The results showed that the drift limit states associated with slight, moderate, and extensive damage were mainly affected by the parameters related to the strength and stiffness of the CFT columns (depth‐to‐thickness ratio, steel tube yield strength‐to‐concrete compressive strength ratio, and steel tube stiffness ratio). The drift limit state associated with complete damage was primarily influenced by the axial load ratio. Additionally, it was found that CFT columns are more likely to experience moderate damage owing to the local buckling of the steel tube rather than from concrete crushing. Finally, fragility curves for the CFT frame systems were developed using the nonlinear time‐history analysis results in conjunction with the developed ML‐based drift limit state prediction model.