Machine learning methods for predicting the durability of concrete materials: A review

Evaluating the durability of concrete materials through experimental research is a lengthy, costly and inefficient process. Traditional empirical formulae offer limited accuracy in predicting durability and fail to guide concrete proportioning based on performance. Consequently, it is crucial to develop new, efficient tools for material quality control and performance prediction. This can be achieved by elucidating the process involved in machine learning (ML) models, delineating the fundamental operating principles and benefits of prevalent algorithms, and critically reviewing ML-based durability index prediction algorithms and their practical applications and future directions. In this study, CiteSpace software was used to assess the current state of ML research in the prediction of concrete durability. A comprehensive analysis of the number of publications, research focal points and emerging trends was conducted. This not only furnishes references for future research but also aims to facilitate more effective utilisation of ML technology, thereby fostering the development of innovative construction materials and advancing the goal of environmental sustainability.