Prediction of mechanical strength through machine learning models for Alkali-Activated concretes

Abstract

Alkali-activated concrete has become an alternative construction material to
overcome the environmental impacts of traditional Portland cement concrete,
with its better performance, environmental friendliness, mechanical properties,
and durability. In recent years, advanced machine learning techniques have
been used to predict the mechanical properties of alkali-activated concrete;
however, prediction accuracy still needs to be enhanced. In this innovative study,
particle packing theory was used in mechanical properties prediction through
machine learning techniques, where packing density was introduced as a new
input variable in machine learning models. The dataset used in machine
learning models training was collected from an experiment study, which
involved 99 data points of compressive strength and 33 data points of flexural
strength test. This study presents the possibility of predicting the compressive
strength and flexural strength of alkali-activated concrete through mix
proportions using four machine learning algorithms: Random Forest, Extreme
Gradient Boosting, Support Vector Machine, and K-Nearest Neighbors. The
results showed that all machine learning models performed with reasonable
accuracy in the training set and testing set. Within four models, Extreme
Gradient Boosting showed the best performance. According to the influence
analysis for input variables, packing density showed an intermediate effect of
both compressive strength and flexural strength, proving that packing density
can be an influential factor in the mechanical properties prediction of alkali-activated concrete. This innovative research is able to reduce the materials,
time, and costs in experiments, and it will also be beneficial in alkali-activated
concrete mix design in the concrete industry.

Date of Award	15 Mar 2025
Original language	English
Awarding Institution	University of Nottingham
Supervisor	Weizhuo Shi (Supervisor) & Bo Li (Supervisor)