Explainable stacked machine learning models for predicting fiber-reinforced polymer (FRP) bars–concrete interfacial bond strength retention in marine environments
Engineering Structures, cilt.365, ss.1-19, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 365
- Basım Tarihi: 2026
- Doi Numarası: 10.1016/j.engstruct.2026.123237
- Dergi Adı: Engineering Structures
- Derginin Tarandığı İndeksler: Academic Search Ultimate (EBSCO), Engineering Source (EBSCO), Scopus, Science Citation Index Expanded (SCI-EXPANDED), Compendex, Geobase, ICONDA Bibliographic, INSPEC, The International Construction Database (ICONDA)
- Sayfa Sayıları: ss.1-19
- İnönü Üniversitesi Adresli: Evet
Özet
The durability of the bond between fiber-reinforced polymer (FRP) bars and concrete is a critical concern for structural applications exposed to aggressive marine environments. This article presents a novel explainable stacked machine learning (ML) framework for predicting the bond strength retention (BST) of FRP bar–concrete systems in marine engineering. An existing and systematically compiled experimental database comprising 601 test results available in the literature was utilized, incorporating geometric, mechanical, material, and environmental parameters. Five advanced tree-based ML algorithms—Random Forest (RF), Gradient Boosting Machine (GBM), Extra Trees Regressor (ETR), Extreme Gradient Boosting (XGB), and Light Gradient Boosting Machine (LGBM)—were employed as base learners, followed by the development of 20 stacked ML configurations to enhance predictive robustness and generalization capability. The outcomes display that stacked ML models consistently outperform base learners, particularly in the testing phase, with the optimal configuration SM-4 (XGB + LGBM) achieving superior accuracy (R = 0.935, R² = 0.845) and reduced error levels. To ensure transparency and physical interpretability, explainable artificial intelligence (XAI) techniques—Shapley Additive exPlanations (SHAP) and Individual Conditional Expectation (ICE)—were applied. These analyses revealed that concrete compressive strength, conditioning duration, and conditioning temperature are the dominant factors governing long-term bond durability, while geometric and material descriptors exert secondary and interaction- dependent effects. Finally, the optimized model was embedded into a GUI-based real-time prediction framework, enabling practical engineering implementation and rapid scenario assessment. Overall, this study provides an accurate, interpretable, and application-oriented ML solution for evaluating FRP–concrete bond durability in marine environments.