ISSN 2096-4498

   CN 44-1745/U

二维码

Tunnel Construction ›› 2026, Vol. 46 ›› Issue (10): 2067-2079.DOI: 10.3973/j.issn.2096-4498.2026.10.001

Previous Articles     Next Articles

Interpretable Machine Learning and Prediction of Bonding Durability Performance of Fiber-Reinforced Polymer Reinforcement in Marine Environment

LI Jiuyang1, CHEN Li2, WU Jishu2, LUO Chengsheng1   

  1. (1. College of Civil Engineering, Changchun Institute of Technology, Changchun 130012, Jilin, China; 2. College of Civil Engineering, Tongji University, Shanghai 200092, China)
  • Online:2026-10-20 Published:2026-10-20

Abstract: To explore the evolution patterns and key controlling factors of the interface bonding performance of fiber-reinforced polymer (FRP)-reinforced concrete in marine environments, nine input variables—FRP bar type, FRP surface profile, exposure duration, temperature, FRP elastic modulus, concrete compressive strength, bonding length, bar diameter, and concrete cover thickness—were selected from 318 experimental bonding strength samples in the existing literature, with bonding strength as the sole output variable. These variables were used to optimize hyperparameters and compare the performance of six ensemble learning models using the tree-structured Parzen estimator and Bayesian optimization. Four interpretability analysis methods—Shapley additive explanations analysis, individual conditional expectation, partial dependence plots, and accumulated local effects—were employed to investigate the influence of each input variable on bonding strength. A graphical user interface was then developed to enable interactive visual design and prediction of bonding strength. The results show that (1) the CatBoost model achieves the highest prediction accuracy, with an R2 of 0.889, a root mean squared error of 1.722, and a mean absolute error  of 1.088; and (2) the key factors influencing bonding strength are elastic modulus, compressive strength, FRP surface profile, FRP bar type, and bonding length.

Key words: marine environment, fiber-reinforced polymer reinforcement, durability performance, bonding strength, machine learning, Shapley additive explanations analysis