NGBoost-based prediction of carbonation in Alkali-activated slag concrete enhanced by metaheuristic algorithms

Concrete carbonation represents a critical deterioration mechanism that significantly compromises the long-term structural integrity of reinforced concrete systems. Traditional methods for assessing carbonation depth (CD) in Alkali-activated slag concrete (AASC) are labor-intensive, costly, and dema...

Celý popis

Uložené v:
Podrobná bibliografia
Vydané v:Construction & building materials Ročník 500; s. 144257
Hlavní autori: Chen, Yun, Fu, Qianwang, Zheng, Wenbo
Médium: Journal Article
Jazyk:English
Vydavateľské údaje: Elsevier Ltd 21.11.2025
Predmet:
Alkali-activated slag concrete
Beluga Whale Optimization algorithm
Carbonation depth
Crayfish Optimization Algorithm
Natural gradient boosting
Remora Optimization Algorithm
Beluga Whale Optimization algorithm
Natural gradient boosting
Alkali-activated slag concrete
Remora Optimization Algorithm
Carbonation depth
Crayfish Optimization Algorithm
ISSN:0950-0618
On-line prístup:Získať plný text
Tagy: Pridať tag
Žiadne tagy, Buďte prvý, kto otaguje tento záznam!
Popis
Shrnutí:Concrete carbonation represents a critical deterioration mechanism that significantly compromises the long-term structural integrity of reinforced concrete systems. Traditional methods for assessing carbonation depth (CD) in Alkali-activated slag concrete (AASC) are labor-intensive, costly, and demand specialized expertise. To address the current gap in applying machine learning (ML) techniques for accurate CD prediction in AASC, this study developed ML frameworks for carbonation forecasting by integrating natural gradient boosting (NGBoost) with three bio-inspired optimization techniques: beluga whale optimization (BWO), remora optimization (ROA), and crayfish optimization (COA). The models were trained on a dataset of 136 data points with 9 input variables. The study employed a comprehensive evaluation framework incorporating four key metrics (Coefficient of determination (R2), Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE)) and Taylor diagram analysis to assess comparative model performance. The results showed that the COA-NGBoost algorithm achieved superior performance with R2 of 0.946, MAPE of 0.145, MAE of 1.451 mm, and RMSE of 2.099 mm. Shapley additive explanations (SHAP) analysis identified environmental conditions, specifically CO2 concentration and exposure time, as key factors influencing CD. Furthermore, an interactive software tool incorporating the COA-NGBoost framework was developed to facilitate probabilistic CD prediction and optimize AASC mixture proportioning. This investigation demonstrates the successful implementation of the proposed framework in parametric analysis of CD factors within AASC systems, providing valuable insights for durability-based design strategies. [Display omitted] •A novel natural gradient boosting (NGBoost) model accurately predicts carbonation depth (CD) in alkali-activated slag concrete (AASC).•Hyperparameter optimization using three novel bio-inspired algorithms, significantly improves NGBoost model performance.•The optimized COA-NGBoost model demonstrates exceptional accuracy (R2 = 0.946) and low error metrics.•SHAP analysis reveals CO2 concentration and exposure time as primary drivers of AASC carbonation.•A GUI enables easy prediction of AASC-CD progression, facilitating practical implementation in infrastructure management.
ISSN:0950-0618
DOI:10.1016/j.conbuildmat.2025.144257