The use of alternative materials and recycling in construction has gained popularity in recent years as part of the industrys commitment to sustainability. One such material, recycled aggregates, has been extensively studied over the past two decades for its potential to replace natural aggregates in cement-based composites. However, the unique properties of recycled aggregates make traditional concrete mix design methods ineffective in determining their target compressive strength. To address this challenge, four machine learning models based on ensemble learning algorithms, including CatBoost regressor (CatBoost), light gradient-boosting machine regressor (LGBM), random forest regressor (RFR), and extreme gradient-boosting regressor (XGBoost), were employed to predict the compressive strength of recycled aggregate concrete. Results demonstrate that the proposed models are highly accurate and generalizable, with high coefficients of determination and low error predictions. The CatBoost model performed the best, exhibiting an R² of 0.938 and low mean absolute error and root mean squared error values of 2.639 and 3.885, respectively, in the blind evaluation process. Although the random forest regression algorithm performed the least well among the four models, it still outperformed conventional machine learning algorithms such as support vector machines and artificial neural networks. The findings in this study suggested that the CatBoost model is the optimal choice for predicting concretes compressive strength due to its high accuracy and low prediction error.

Corrosion in reinforced concrete (RC) structures has led to the increased adoption of non-corrosive materials, such as carbon fiber-reinforced polymers (CFRPs), as replacements for traditional steel rebar. However, ensuring the long-term reliability of CFRP grids under sustained stress is critical for achieving safe and effective designs. This study investigates the long-term tensile creep rupture behavior of CFRP grids to establish a design threshold for their tensile strength under sustained loading conditions in demanding structural applications. A comprehensive laboratory experiment was conducted over 10,000 h, during which CFRP grid specimens were subjected to constant stress levels ranging from 92% to 98% of their ultimate tensile strength. The results confirm the excellent creep rupture resistance of CFRP grids. Specimens subjected to a sustained stress ratio of 92% of their ultimate tensile strength remained intact throughout the testing period, with minimal creep strain ranging from approximately 1% to 4% of the initial strain. The mean extrapolated creep rupture factors were found to be 92.1% and 91.7% of their ultimate tensile strength for service lives of 50 and 114 years, respectively. Based on the results of this study, a tensile stress limit of 48% of the ultimate tensile strength is recommended for CFRP grids to ensure long-term creep rupture resistance over a 100-year service life.

Shear failure in reinforced concrete (RC) beams is abrupt and brittle, occurs without warning, and leaves no opportunity for internal stress redistribution. Despite the critical need for accurate shear strength assessment, existing methods vary widely across regions, leading to inconsistencies in practice. This study presents a unified shear strength equation for non-prestressed rectangular RC beams without stirrups, developed for simplicity and broad applicability. The model requires only basic geometric and material properties and applies to both shear-slender and non-shear-slender beams. It was formulated using a data-driven approach that combines an extensive experimental database collected up to 2007 with advanced computational techniques, including Artificial Neural Networks, Generative Adversarial Networks, and Bayesian optimization. The proposed equation was evaluated against established shear provisions, such as ACI 318-25 and CSA A23.3-24, and benchmarked with an experimental database. The results show that the model improves prediction accuracy, reduces uncertainty, and provides a more consistent method for shear strength assessment. The robustness of the equation was further confirmed through additional experimental database gathered after 2007, demonstrating strong agreement with test results and lower prediction uncertainty than current code provisions. These findings support the potential adoption of the proposed equation in engineering practice.

The healing products reported for sodium silicate, a leading agent for self-healing concrete, are inconsistent across studies. The reactions involved are individually known, but which of them controls the healing product cannot be isolated experimentally, because agent composition, reacted dose, pore-solution chemistry, and matrix dissolution change together. This study separates these effects by thermodynamic equilibrium modelling in PHREEQC with CEMDATA18 and the CNASH_ss solid-solution model, confirmed against three independent experimental datasets with no parameters adjusted, and calibrated against a reported healing experiment through a reaction-extent parameter. The healing reaction operates in two regimes separated by the exhaustion of substrate portlandite. While portlandite survives, the agent controls the healing gel composition, and raising the silicate modulus raises both the gel Ca/Si and net healing volume. Once portlandite is exhausted, the substrate sets the available calcium, raising the modulus lowers the gel Ca/Si, and the net healing volume becomes non-monotonic. The influence of substrate aluminate is regime-dependent: negligible while portlandite survives, because portlandite fixes the gel composition, but strong once portlandite is exhausted. The regime boundary is forced by the calcium mass balance of the substrate; it is preserved in 22 to 24 of 24 conditions under perturbation of every uncertain input and in 23 of 24 with a structurally dissimilar solid-solution model, whereas absolute gel compositions are model-dependent. Because the outcome is co-determined by agent and substrate, the regime structure offers an explanation for the inconsistent products, and an agent optimised for one substrate cannot be assumed to transfer to another.