Advanced machine learning techniques for predicting compressive strength and ultrasonic pulse velocity of concrete incorporating industrial by-products

The incorporation of fly ash, ground granulated blast furnace slag (GGBFS), and other pozzolanic materials into cement formulations presents a sustainable solution for utilizing industrial waste. This study introduces a novel, dual-stage machine learning framework that uniquely integrates both physi...

Full description

Saved in:
Bibliographic Details
Main Authors: Ehsan Mohsennia, Alireza Javid, Vahab Toufigh
Format: Article
Language:English
Published: Elsevier 2025-07-01
Series:Case Studies in Construction Materials
Subjects:
Online Access:http://www.sciencedirect.com/science/article/pii/S2214509525005996
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The incorporation of fly ash, ground granulated blast furnace slag (GGBFS), and other pozzolanic materials into cement formulations presents a sustainable solution for utilizing industrial waste. This study introduces a novel, dual-stage machine learning framework that uniquely integrates both physical mix parameters and detailed chemical compositions of cement and multiple industrial by-products (IBPs)—including fly ash, GGBFS, silica fume, and metakaolin—into a unified predictive model. This approach significantly advances current methodologies, which often rely on single IBPs or exclude oxide-level descriptors. The framework first predicts ultrasonic pulse velocity (UPV), then uses the predicted UPV as an input to enhance compressive strength (CS) prediction—an underexplored configuration in previous studies. This novel approach is aligned with environmental management principles, aiming to maximize the effective use of diverse IBPs, reduce waste, and enhance resource efficiency. A robust dataset, comprising 162 structured IBP concrete samples and 524 data points from existing literature, enabled rigorous training and validation of sophisticated ML models. Among the models tested, the CatBoost (CB) algorithm, optimized with the Whale Optimization Algorithm (WOA), exhibited outstanding predictive performance. The CB-WOA model achieved coefficients of determination (R²) values of 0.96 and 0.94 for UPV during training and testing, and 0.99 and 0.98 for CS, respectively. These results underscore the potential of UPV as a non-destructive metric for assessing concrete structural integrity and predicting material strength. A comprehensive sensitivity analysis further highlighted the importance of targeted feature selection, enhancing prediction accuracy and contributing directly to sustainability goals by optimizing material usage and improving construction practices. To facilitate the practical application of these findings, a user-friendly web application has been developed (https://ibp-concrete-upv-and-cs-prediction-mohsennia-javid-toufigh.streamlit.app/).
ISSN:2214-5095