Databases of Cementitious Materials Including Clinker and Their Applications

In light of the policy of ‘carbon peaking and carbon neutrality' and the advent of artificial intelligence, there is an urgent need for research and development in the field of cementitious materials to advance low-carbon alternatives. Given the lengthy research periods associated with traditional experimental methods, digital research and development has recently emerged as a dominant trend in the field. Currently, digital models are being developed for purposes such as performance prediction, composition optimization, and low-carbon innovation, with a primary focus on algorithm optimization. However, the success of these models relies heavily on the availability of accurate, high-quality databases, which serve as the foundation for model implementation. Utilizing such databases can significantly simplify model construction and reduce the need for extensive optimization procedures. Consequently, building comprehensive databases for cementitious materials research and development has become a key objective in this field.

This paper reviews the current status of databases for cementitious materials, including those related to clinker, cement, concrete, mineral admixtures, and construction mortar. The types of data, application methods, and scope of application encompassed by these databases are summarized. Databases for clinker were among the earliest to be developed, primarily cataloging diverse mineral properties, including hydration products. They include crystallographic data, thermodynamic data, and force field data, with some databases integrating models for analyzing cement hydration. Cement databases, while recording thermodynamic properties, cement types, and characteristics, have largely been focused on production control in cement plants. Moreover, the construction of databases for concrete and related materials has often been driven by specific research objectives, such as studying chemical composition, strength, and durability. Some researchers have also conducted property predictions based on their collected datasets.

Despite the existence of numerous databases on cementitious materials, a significant volume of relevant data within scientific literature remains underutilized. The advancement of artificial intelligence in natural language processing has enabled the adaptation of data extraction algorithms across various domains, including metals, medicine, and biology. Named entity recognition and textual relationship extraction—two critical components of literature data mining—can be implemented through AI algorithms such as ChemDataExtractor and BERT. ChemDataExtractor and related algorithms have demonstrated accurate chemical data extraction from compounds and semi-supervised relationship extraction. Similarly, BERT, like ChatGPT, is a state-of-the-art language model developed using the Transformer architecture and has been successfully applied in automated text data extraction. However, the complex mineralogical and chemical composition, multi-scale particle characteristics, and hydration processes of cementitious materials pose challenges to the direct application of these algorithms.

The construction of comprehensive databases represents the cornerstone of the digital transformation of low-carbon cementitious material research and development. Although existing databases contain extensive datasets on cementitious materials, several challenges persist, such as non-standardized database structures, insufficiently considered data categories, and incomplete material coverage. To address these issues, future database development should prioritize unifying data formats and linking upstream and downstream processes to create a cohesive and interconnected database. Such a unified database would enable the establishment of a fully connected data chain for cementitious materials, enhancing the accuracy of predictions, supporting reverse design processes, and saving time on data cleaning.

Particular emphasis should also be placed on refining databases for related materials, such as mineral admixtures. These refined databases would provide critical data for improving the durability of cementitious materials and reducing their carbon emissions. Furthermore, combining artificial intelligence algorithms, such as ChemDataExtractor and BERT, with domain expert knowledge holds significant potential for advancing literature mining techniques tailored to cementitious materials. This process could begin by focusing on individual performance attributes and gradually expand to encompass the comprehensive extraction of data for all cementitious materials.

The future of cementitious materials database development is promising, with the potential to drive innovations in low-carbon materials research and development, ultimately contributing to achieving global carbon neutrality goals.