Data-Driven Optimisation of Cement-Based Mixes with Chemical Admixtures and Supplementary Cementitious Materials

Concrete is no longer a construction material but an ever-expanding range of cement-based products which can be tailored to specific performance requirements through the combination of chemical admixtures and supplementary cementitious materials (SCMs). During the development of such admixtures and additions, the industrial partner in this project has generated an enormous wealth of data through the proportioning, production and testing of trial mixes where they are used at different dosages and in various combinations. This project aims at utilising the methodological framework and techniques of data mining and multivariate statistics to combine and integrate all this information in a consolidated database and to derive semi-empirical models that can be used to optimise the proportioning of mixes from the point of view of their engineering properties.
The focus of the project will be on the engineering properties, that is, fresh and hardened state properties that are relevant to engineers, designers, manufacturers and contractors in terms of product specification and quality control.
Experimental work is planned to generate validation data, as well as to link microstructure and composition to performance parameters. Here, performance will be understood from a basis of fundamental cement science, giving a physical aspect to the models.
The project will generate a holistic understanding of the interactions between different SCMs and chemical admixtures at different dosages and provide quantitative tools for the prediction of their combined effects on the fresh and hardened properties of concrete and related cement-based materials (grouts, mortars). This is a completely novel approach which will address a significant research gap.
To date, an accurate prediction of such properties is almost never possible without significant efforts, involving parametric adjustment of mix designs in trial batches, i.e. a process of trial and error. Precise performance predictions are complicated by variations in material composition and mix designs, severely limiting the validity of such predictions. By bringing the number of data from a scale of a few dozens to a scale of thousands of cases, the predictive power of the data-based models will improve, leading to more robust conclusions and wider applicability.
It is anticipated that this will lead to semi-automated tools for the proportioning of cement-based products, which will help the construction industry to make a more efficient use of resources. This can be linked to decarbonisation of construction.