An approach for developing probabilistic models for temperature-dependent properties of construction materials from fire tests and small data

Probabilistic approaches provide a more realistic look into assessing structures under fire conditions and overcome some limitations observed in the more traditional (deterministic) approaches. These approaches have also been introduced to the fire engineering domain, for example, fire probabilistic risk analysis and probabilistic structural fire engineering. In order to perform probabilistic-based analysis, temperature-dependent probabilistic models for material properties are needed. This paper presents a methodology to develop temperature-dependent probabilistic models for the thermal and mechanical properties for commonly used construction materials, including normal-strength, high-strength, and high-performance concrete and mild, high-strength, and cold-formed steels. The presented approach analyzes a comprehensive list of surveyed experimental data at different temperature groups, tests the goodness of fit for a number of distributions, and derives a continuous function to quantify temperature-dependent parameters of the distribution. In addition, the newly derived models are also compared against those adopted by fire codes, and standards and others derived using machine learning. The newly developed models will complement existing efforts to facilitate probabilistic performance-based structural fire engineering