Design automation of sustainable self-compacting concrete containing fly ash via data driven performance prediction

Abstract

Self-compacting concrete (SCC) is a highly flowable and segregation-resistant material, effectively facilitating proper filling and ensuring exceptional structural performance in confined spaces. Incorporating fly ash as a supplementary cementitious material in SCC mixtures yields numerous benefits, including enhanced cost-effectiveness in construction and the advancement of environmental sustainability. Nevertheless, the addition of fly ash in SCC poses significant challenges in modelling and predicting the properties of SCC due to lack of understanding of its influence on material rheology and bonding. It is therefore desirable to develop more appropriate machine learning approach to compliment the large scale and costly laboratory-based experiments. This paper presents four well trained supervised machine learning models for the prediction of fresh and hardened properties of SCC containing fly ash: support vector machine (SVM), decision tree, random forest, and artificial neural network (ANN). Training datasets gathered from publicly available existing relevant literature, were analysed and processed prior to shape the required machine learning models. Optimization strategies of hyperparameters were also implemented for each model. To evaluate the performance of these machine learning models and to compare their accuracy, regression error characteristic curves and Taylor diagrams were utilized. The findings reveal that all models demonstrate promising results, with the random forest model outperforming the others in predicting SCC properties with higher accuracy. This underscores the potential of random forest algorithms in accurately modelling and predicting the properties of fly ash-infused SCC. Finally, a data driven implementation framework has been developed, thereby offering robust and logical strategy for experimental designs and guidance for developing sustainable SCC.