Modelling the sensitivity of fresh state performance to the mix design of conventional and self-compacting concretes

The fresh and hardened state performance of conventional concrete and self-compacting concrete (SCC) are affected by multiple factors. The introduction of supplementary cementitious materials (SCMs) or mineral additions, the use of different chemical admixtures, the maximum aggregate size or the grading of aggregates all influence how the mix is designed to the specified performance requirements. In SCC mixes, the fresh state performance is generally the main driver of the mix design, but in all cases, SCC or not, the cohesiveness and workability of a mix can influence its hardened state properties. This study presents the results of the analysis of a large dataset with over 1500 concrete mixes compiled from routine industrial testing, with the objective of developing a tool that predicts the workability based on the mix design. The dataset includes SCC and conventional (i.e., non-SCC) mixes, incorporating different SCMs and water reducers or superplasticisers at different dosages. Machine learning techniques have been employed to develop a classifier which can predict the slump class of a concrete mix. One of the main novelties of this study is that SCC and non-SCC mixes are not analysed separately. Instead, a unified predictive model is developed and discussed. The paper describes how the challenge posed by the diverse range of mix designs has been overcome and in what ways the knowledge obtained from data can be enhanced by following such an integrated approach.