Developing Empirical and Graphical Indices for Fire-Induced Spalling of Concrete

Abstract

Abstract Fire-induced spalling remains an ongoing concern for concrete structures exposed to fire conditions as existing predictive methods struggle to achieve reliable predictions. To bridge this knowledge gap, this study proposes a methodology to develop empirically driven indices to predict the spalling phenomenon. The methodology comprises three approaches: systematic data transformations, logistic regression, and a data-informed boundary decision based on machine learning (ML). The obtained indices, evaluated over a comprehensive dataset of over 1,000 fire tests and 200,000 cases, exhibited an accuracy ranging from 71–82%. The proposed indices were then refined per concrete grade and achieved improved performance. These findings highlight the potential for developing empirical indices to arrive at simple and easy to use tools to predict fire-induced spalling of concrete. Such indices could set the stage for new predictive methods that go beyond simple statistical and ML inference approaches in the near future.