Machine Learning to Derive Unified Material Models for Steel Under Fire Conditions

Abstract

Once exposed to elevated temperatures, structural steel experiences physio-chemical reactions that mirrors material degradations. Such degradations are represented through temperature-dependent material models. Models, as valuable tool in fire analysis and design, can showcase how steel properties vary with rise in temperature. Since we continue to lack a standardized testing procedure, the bulk of the available models does not seem to converge onto a particular trend, but rather a closer look into these models shows a wide scatter. As a result, fire engineers are often faced with a few questions, for instance, which temperature-dependent material model to use in a fire design? How to verify or equate structural fire designs carried out by different engineers/firms from around the globe that adopt different material models? In order to encourage a more standardized fire assessment of steel structures, this chapter leverages machine learning to develop a unified temperature-dependent material model that can be used to evaluate fire response of steel structures. Predictions obtained using the proposed ML-based material model showed relative superior to those compared against those using commonly used material models. Findings from this chapter can be extended into other construction materials (i.e., concrete, timber, etc.) and can help facilitate harmonization efforts within the fire engineering community.