Machine learning models for predicting the resistance of axially loaded slender steel columns at elevated temperatures

Abstract

Due to the interaction between local and global instability modes at elevated temperatures, predicting the capacity of thin-walled columns in the fire situation is a complex endeavor. This work investigates the application of machine learning techniques to assess the resistance of slender steel columns with I-shaped cross-sections at elevated temperatures. First, a validated finite element model is used to evaluate the columns’ response and generate a large dataset for a range of cross-sections, slenderness, and temperature. The dataset is then used for training and testing machine learning models based on support vector regression, artificial neural network, and polynomial regression. The machine learning models outperform the current analytical design method for the training and testing datasets, demonstrating the potential usefulness of data-based methods for structural fire design. The limits of the approach are explored by applying the trained models for predicting experiments with features outside those of the dataset. The results suggest that machine learning techniques can be used to derive efficient surrogate models for the capacity prediction of slender steel members in fire within the boundaries of the training dataset.