An Artificial Neural Network Model for Predicting the Maximum Allowable Heat Release of Concrete during the Construction of Massive Monolithic Foundation Slabs

Early-age cracking in mass concrete structures, driven by thermal stresses from cement hydration, remains a critical durability concern. One effective way to reduce the risk of early cracking is to select a concrete mixture formulation that reduces heat emission. This study considers, for the first time, the inverse problem of preventing early crack formation: determining the maximum allowable concrete heat release (Qmax) for the given geometric characteristics of the structure and concreting parameters using machine learning methods.

A dataset of 39,200 numerical experiments was collected via thermo-mechanical modeling, considering variables like slab thickness, heat transfer coefficient, concrete grade, ambient temperature, and concreting duration. The target value Q_max was identified using the bisection method, ensuring the tensile stress-to-strength ratio remained below unity. A feedforward Artificial Neural Network (ANN) with two hidden layers was developed and trained on this dataset.

The ANN model achieved exceptional prediction accuracy, with a correlation coefficient of 0.99955 between target and predicted Q_max values. Analysis revealed that the concrete compressive strength grade had a minimal effect on the maximum permissible heat release.

Feature importance analysis showed that the curing rate and slab thickness are the most significant parameters influencing the Q_max value. The negligible impact of compressive strength stems from its competing effects on tensile strength and elastic modulus.

The developed ANN model provides a highly accurate tool for predicting permissible concrete heat release, enabling optimized mix design to mitigate early thermal cracking in massive foundation slabs.