Concrete Spalling Data
Description
This dataset compiles comprehensive experimental results on the thermal-induced spalling behavior of concrete from 55 published studies covering a broad range of material compositions, specimen geometries, and fire exposure conditions. It integrates 1,430 samples featuring multi-scale descriptors at the material, geometric, and thermal levels — including parameters such as binder ratios (W/B, FA/B, CA/B), fiber geometry, density, moisture content, specimen dimensions, heating rate, and peak temperature. The dataset is primarily used to support the development and validation of a hierarchical, data-driven framework for predicting and interpreting post-fire concrete spalling. It enables the study of both binary classification (spalling occurrence and explosivity) and regression analysis (residual mass loss and spalling depth) through explainable machine learning models such as XGBoost and SHAP. All variables are standardized in consistent physical units and organized according to three feature groups — Material, Physical & Geometric, and Thermal. This dataset provides a unified benchmark for future research on data-driven fire performance assessment and mechanistic understanding of spalling phenomena in cementitious materials.
Files
Steps to reproduce
The dataset file Spalling data.csv contains all raw and preprocessed variables used in the analysis. Each row represents an experimental specimen from published studies, and each column corresponds to a standardized input or output variable (see Appendix A of the associated paper). Categorical descriptors such as aggregate type and fiber category were encoded using one-hot encoding. Data analysis and modeling were conducted in Python 3.10 using XGBoost, Scikit-learn, and SHAP libraries. Reproduction of results follows the workflow described in “Fire-Induced Concrete Spalling: A Data-Driven Framework for Quantitative Prediction and Interpretation.”
Institutions
- University of Auckland