علي رضا آشناور

Carbonation curing of concrete, as an emerging approach within carbon capture, utilization, an‎d storage (CCUS), offers significant potential for reducing the carbon footprint of construction materials. Effective utilization of this technique requires a quantitative an‎d systematic understan‎ding of its performance. Recent studies have largely relied on laboratory-scale investigations, an‎d to date, no integrated, data-driven framework has been proposed to model, interpret, an‎d optimize carbonation curing behavior based on broad experimental evidence. In this study, a data-driven framework was developed. First, through a systematic search of the Scopus citation database an‎d digital thesis repositories, experimental data from 72 scientific documents were extracted an‎d consolidated into a coherent dataset. This dataset comprises 888 unique experimental records an‎d 24 variables. Carbon dioxide uptake was defined as the continuous target variable, alongside two categorical features (composite an‎d binder type) an‎d 21 continuous numerical features, including the major oxide compositions of the binder, water-to-binder ratio, pre-conditioning parameters (duration, temperature, humidity, an‎d removed water), an‎d carbonation curing conditions (time, pressure, concentration, temperature, an‎d humidity). Four machine learning models representing four major algorithm families, ElasticNet, SVR, MLP, an‎d CatBoost, were trained. Model performance was eva‎luated using RMSE, MAE, MAPE, an‎d R² metrics, comparison of actual an‎d predicted values, an‎d 5-fold cross-validation. The results indicate that CatBoost provides the best performance in terms of accuracy an‎d generalization capability. 5-fold cross-validation results, with an R² of approximately 0.91 an‎d an MAE close to 1.2%, demonstrate limited sensitivity of the model to data partitioning. SHAP-based sensitivity analysis further reveals that a substantial portion of CO₂ uptake behavior is governed by nonlinear interactions among carbonation duration, binder oxide compositions (particularly CaO an‎d SO₃ contents), concentration, an‎d removed water. In the optimization stage, a genetic algorithm was configured for inverse design to search for process variable values that achieve a target uptake under technical an‎d engineering constraints. Experimental validation of the optimization framework explored two target uptake scenarios of 10% an‎d 15% under different constraint combinations. Laboratory results showed actual CO₂ uptakes of 10.93% an‎d 14.11%, respectively, confirming good agreement between predictions an‎d real material behavior. In the final stage, the developed model was applied to estimate CO₂ uptake in concrete elements containing Portlan‎d cement, blast furnace slag, an‎d electric arc furnace steel slag from Iranian industries under carbonation curing conditions. The results demonstrate that, through appropriate selec‎tion an‎d combination of binder materials an‎d optimal adjustment of curing parameters, the production of concrete elements with substantial CO₂ uptake capacity an‎d a reduced carbon footprint is achievable.

عمل‌آوري كربناسيون , جذب كربن‌دي‌اكسيد , يادگيري ماشين , الگوريتم ژنتيك , بهينه‌سازي

carbonation curing , carbon dioxide uptake , machine learning , genetic algorithm , optimization

Alireza Ashnavar

Dr. Sajjad Mirvalad; Dr. Asghar Habibnejad Korayem