RAC Beam Shear-Capacity Dataset for Study-Independent and Uncertainty-Aware Machine Learning

Abstract

This dataset contains 229 experimental observations of recycled aggregate concrete (RAC) beams used for machine-learning-based prediction of shear capacity. The dataset was independently digitized, standardized, quality-screened, and regrouped at the experimental-study level from the published RAC beam database reported by Yu et al. (2020). The dataset includes geometric, material, reinforcement, loading-related, and recycled coarse aggregate parameters, together with the experimentally measured shear capacity. The variables include beam width and height, effective depth, shear-span-to-depth ratio, recycled aggregate replacement ratio, water-to-cement ratio, maximum aggregate size, concrete compressive strength, longitudinal reinforcement ratio and yield strength, shear-reinforcement contribution, and measured shear capacity. The dataset is intended for research on study-independent validation, uncertainty-aware machine learning, physics-constrained prediction, applicability-domain assessment, and reliability-oriented evaluation of shear capacity in recycled aggregate concrete beams. The final dataset contains 229 observations from 21 experimental study groups. Four observations with censored shear-capacity information and 31 observations involving source-level assumed input quantities were excluded during quality screening. Related research: “Study-Independent Validation and Uncertainty-Aware Machine Learning for Shear Capacity Assessment of Recycled Aggregate Concrete Beams.”