Authors: Hong-Hui Qi , Guo-Qiang Li , Shaojun Zhu
Abstract: The axial force plays a critical role in assessing the functional integrity of columns within a building in fire. However, it cannot be measured directly and is influenced by factors such as temperature, load ratio, and axial restraint. This study proposes a real-time methodology to predict the axial force of restrained concrete-filled steel tubular (CFST) columns exposed to real fires, utilizing modular artificial intelligence. A module is developed that combines a convolutional neural network (CNN) and long short-term memory (LSTM) networks to predict the temperature field of CFST columns caused by fire in real time. This module estimates the current temperature field using past data and the current surface temperature, which is continuously monitored with inherent noise. It effectively mitigates noise interference, achieving an R2 of 0.97 on the test dataset, which ensures accurate estimations. Additionally, a separate LSTM module with a skip connection is employed to predict the axial force ratio, integrating temperature predictions and real-time measurements of axial deformation. Finally, the accuracy of this modular model demonstrates better performance in predicting real-time axial force compared to the conventional integrated deep learning model, achieving an R2 of 0.99. The proposed approach enables accurate prediction of axial force in restrained CFST columns across various fire scenarios and structural conditions, aiming at increasing the scientificity of fire rescue decisions.