Title:
Can Machine Learning Correlate Accelerated and Natural Carbonation? A Case Study of Low-Carbon Concrete Using Probabilistic Deep Learning
Author(s):
Marani
Publication:
Web Session
Volume:
Issue:
Appears on pages(s):
Keywords:
DOI:
Date:
3/30/2025
Abstract:
Carbonation is a complex physicochemical reaction of carbon dioxide with cement hydration products that can reduce the pH of concrete and thus initiate the corrosion of reinforcement steel. Since natural carbonation takes decades to occur, accelerated carbonation tests have been widely used to study the carbonation performance of different concrete mixtures. Several studies attempted to investigate the relationship between accelerated and natural carbonation rates of various concrete mix designs under specific test conditions. Yet, the correlation between the natural carbonation of concrete exposed to different geographical-specific climatic conditions and the accelerated carbonation of concrete under specific test conditions is still lacking. This study proposes a machine learning approach to estimate and compare the carbonation rate of different concrete mixtures exposed to natural and accelerated conditions. For this purpose, two large datasets of natural and accelerated carbonation of low-carbon concrete mixtures incorporating different types of supplementary cementitious materials (SCMs) were collected from peer-reviewed published research in the open literature. Two probabilistic neural networks (PNN) models were developed to estimate the accelerated and natural carbonation depths of different mixtures. Fick’s law was employed to calculate the carbonation rate to establish the relationship between natural and accelerated carbonation for a wide range of low-carbon concrete mixtures incorporating fly ash, ground granulated blast furnace slag, limestone, and calcined clay. Results indicated that the models could be employed as a reliable framework for relating the natural and accelerated carbonation rates of different concrete mixtures exposed to various environmental exposure conditions with relatively high correlation coefficients.