Exploring the compositional effect of eco-friendly ultra-high performance concrete on dynamic strength based on stacking algorithm and explainable artificial intelligence

Abstract

This study proposes a two-layer fusion model (stacking-CARF) to predict the dynamic compressive strength of eco-friendly ultra-high performance concrete (UHPC). Before building the prediction model, a well-balanced UHPC dynamic compression dataset is created using the anomaly detection algorithm. Subsequently, it is experimentally determined that the stacking-CARF model consisting of categorical boosting, random forest and linear regression outperforms other prominent ensemble learning and stacked models, and can be used as a robust strength prediction tool. Moreover, Explainable Artificial Intelligence is utilized to elucidate the intricate relationship between material proportions and dynamic compressive strength from both global and local perspectives, offering insights challenging to quantify using traditional methods. In particular, the interaction analysis affirms the role of reasonable replacement ratios between cement and supplementary cementitious materials in enhancing sustainability and cleaner production practices. Finally, a Python-based graphical user interface is developed to facilitate the implementation of the stacking-CARF model in engineering applications.