Scholarlink

<p>The interfacial bond strength between fiber-reinforced polymer (FRP) and concrete is critical for structural design, especially in retrofit applications. This study presents a novel hybrid machine learning (ML) framework for predicting the interfacial bond strength between FRP and concrete, which is critical for structural design in retrofit applications. The framework integrates data expansion (TabPFN), physical constraint learning (VW-PINNs), adaptive reinforcement learning (RL-CMOO) and adaptive evolutionary optimization (AEO). Eight different machine learning models were used in the study for comparison, specifically Artificial Neural Networks (ANN), Support Vector Regression (SVR), Decision Trees (DT), Bagging, Random Forests (RF), AdaBoost, Gradient Boosting (GB), and XGBoost. These models were trained and tested using a sample dataset of 855 samples from existing literature with 80% of the data used for training and 20% for testing. The results showed that the hybrid models significantly outperformed the traditional machine learning models in terms of accuracy, obtaining an R² of 0.9115, an RMSE of 2.6463 kN, and an MAE of 1.7977 kN. Additionally, the newly developed models were compared to six existing empirical models used in the design specification, and the results showed that the hybrid framework outperformed these models in terms of accuracy, with an average R² of 0.9115, an RMSE of 2.6463 kN, and an MAE of 1.7977 kN, when compared to the best performing empirical models with an average improvement of 15% in R². Finally, analysis by the LIME local and SHAP global interpretation techniques showed that FRP material and geometric properties (e.g., FRP width, thickness, and modulus of elasticity) had a greater impact on bond strength predictions than concrete properties.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Bond strength prediction of FRP-concrete interface using a hybrid physics-constrained and AI-Driven framework

  • 期刊论文

  • 发表日期：

    2026年1月14日

查看全文
  • Wenhao Ren,
  • A. Siha,
  • Xue Li,
  • Changdong Zhou

摘要

The interfacial bond strength between fiber-reinforced polymer (FRP) and concrete is critical for structural design, especially in retrofit applications. This study presents a novel hybrid machine learning (ML) framework for predicting the interfacial bond strength between FRP and concrete, which is critical for structural design in retrofit applications. The framework integrates data expansion (TabPFN), physical constraint learning (VW-PINNs), adaptive reinforcement learning (RL-CMOO) and adaptive evolutionary optimization (AEO). Eight different machine learning models were used in the study for comparison, specifically Artificial Neural Networks (ANN), Support Vector Regression (SVR), Decision Trees (DT), Bagging, Random Forests (RF), AdaBoost, Gradient Boosting (GB), and XGBoost. These models were trained and tested using a sample dataset of 855 samples from existing literature with 80% of the data used for training and 20% for testing. The results showed that the hybrid models significantly outperformed the traditional machine learning models in terms of accuracy, obtaining an R² of 0.9115, an RMSE of 2.6463 kN, and an MAE of 1.7977 kN. Additionally, the newly developed models were compared to six existing empirical models used in the design specification, and the results showed that the hybrid framework outperformed these models in terms of accuracy, with an average R² of 0.9115, an RMSE of 2.6463 kN, and an MAE of 1.7977 kN, when compared to the best performing empirical models with an average improvement of 15% in R². Finally, analysis by the LIME local and SHAP global interpretation techniques showed that FRP material and geometric properties (e.g., FRP width, thickness, and modulus of elasticity) had a greater impact on bond strength predictions than concrete properties.