Fracture prediction in recycled aggregate concrete using experience-based machine learning with a defective database

The fracture behavior of recycled aggregate concrete (RAC) is highly complex, leading to significant variability in test results and a lack of reliable data, making direct fracture prediction challenging. This study addresses the key scientific problem of how to improve fracture prediction accuracy when working with defective experimental datasets. First, based on experimental analysis and fracture mechanics models, a two-step data processing approach is developed to clean and augment the defective dataset, improving its reliability, richness, and dimensionality. Then, an ensembled learning algorithm is employed to construct a robust predictive model with strong generalization capability (R2 = 0.942). Finally, this study establishes an experience-based artificial intelligence framework for utilizing defective datasets in fracture prediction, providing a novel and practical solution to a long-standing challenge in RAC application.