Frost durability of cementitious materials: What's next?

Frost durability, a critical parameter for concrete, especially in harsh exposure regions, has been extensively researched, with almost four thousand papers published since the 1970s. However, a systematic mapping of this research is yet to be explored. This paper presents a novel approach based on Natural Language Processing (NLP) and machine learning to semi-automatically analyze the existing literature on frost durability of cementitious materials. The aim is to identify research gaps and provide insights for future work, offering a comprehensive understanding of the freeze and thaw (FT) research area. Data sets containing academic abstracts on FT tests have been created, and the identified articles are topically structured using a latent Dirichlet allocation (LDA) topic modeling approach. The publication volume associated with each topic over time has been quantified, providing an overview of the research landscape. The results show that NLP and t-SNE effectively review large volumes of technical text data, identifying 12 dominant themes in FT research, such as mechanical properties and material composition. Over recent decades, there has been a shift from focusing on structural performance to emerging topics like cracking and Supplementary Cementitious Materials (SCMs). Additionally, t-SNE and K-means clustering revealed four main clusters, suggesting future research should focus on the FT durability of ecofriendly materials, accelerated testing, and enhanced FT durability materials. These findings not only facilitate the identification of gaps and opportunities for future work but also have practical implications for developing more durable and sustainable concrete.