Intelligent prediction method for the collapse time of steel frame structures under fire.

To accurately and rapidly predict the collapse time of steel frame structures subjected to fire conditions, a three-story steel frame structure with floor slabs was chosen as the research subject. A thermo-mechanical coupled analysis model, based on beam and shell elements, was developed using the finite element software ABAQUS. A realistic fire scenario was created in the FDS software, and the fire dynamic simulation was performed to obtain the temperature curve throughout the entire duration of a single-room fire. The fire responses of the steel frame structure including failure modes, displacement-time curves and deformation rate-time curves of the beam and column components were investigated by the numerical simulation. Subsequently, the key parameters affecting the failure of steel frame structures under fire conditions were determined, and a collapse criterion was proposed. Through extensive parameter analysis, a total of 3888 models were established, and a dataset of collapse times for steel frame structures under fire conditions was obtained. Machine learning algorithms, including Random Forest (RF), LightGBM, Convolutional Neural Networks (CNN), and Long Short-Term Memory (LSTM) networks, were utilized to establish nonlinear mapping relationships between design parameters and collapse times. The results indicate that the Random Forest algorithm shows superior accuracy and robustness in predicting the collapse time of steel frame structures under fire conditions. • An innovative thermo-mechanical coupling modeling method for steel frame structures was proposed. • The fire response of steel frame structures under single-room fire conditions was analyzed by numerical simulation. • A collapse criterion for steel frame structures in fire conditions was proposed. • The collapse time of steel frame structures under fire conditions was predicted using machine learning algorithms. [ABSTRACT FROM AUTHOR]