Bond strength prediction of FRP bars to seawater sea sand concrete based on ensemble learning models

Pei-Fu Zhang,Mudassir Iqbal,Daxu Zhang,Xiao-Ling Zhao,Qi Zhao
DOI: https://doi.org/10.1016/j.engstruct.2023.117382
IF: 5.582
2024-03-01
Engineering Structures
Abstract:The bond strength between fiber-reinforced polymer (FRP) bars and seawater sea sand concrete (SWSSC) is a critical factor in structural design and applications. To address the inherent nonlinear problem, where bond strength is closely tied to constitutes of concrete and properties of FRP bar, this paper presented a novel approach to predict the bond strength using ensemble learning models and investigated the effects of input parameters on bond strength. For this purpose, a dataset of tests of pullout failures was determined consisting of 13 input parameters. Based on the determined dataset, two prediction models were developed using ensemble learning algorithms Random Forest (RF) and XGBoost. Simultaneously, a bond strength equation for pullout failures was derived from the ACI 440.1R-15 database since the existing equation in the code was originally developed to address splitting failures. The developed models demonstrated commendable accuracy based on statistical evaluations. Given its superior prediction and comparable generalization performances compared to the XGBoost model, the RF model was chosen for bond strength prediction compared to the developed equation. Comparative results favored the RF model, which significantly outperformed the developed equation in bond strength prediction. Furthermore, the study investigated the effects of input parameters by analyzing SHAP (SHapley Additive exPlanations) values. The analysis unveiled the influential mechanisms of mechanical property and composition of SWSSC, along with the mechanical, geometric, and surface properties of FRP bar on bond strength. In conclusion, the developed models represent an effective approach for predicting the bond strength of pullout failures between FRP bars and SWSSC.
engineering, civil
What problem does this paper attempt to address?
### Problems Addressed by the Paper The paper aims to address the issue of predicting the bond strength between Fiber Reinforced Polymer (FRP) bars and Seawater Sea Sand Concrete (SWSSC). Specifically, the paper focuses on how to predict the bond strength between FRP bars and SWSSC using ensemble learning models to solve the non-linear prediction problem of bond strength in existing methods. ### Background and Motivation 1. **Resource Utilization**: In coastal and marine areas, the consumption of freshwater and river sand is enormous and difficult to obtain. Therefore, using seawater and sea sand for structural construction becomes an economical and low-carbon choice. 2. **Corrosion Issues**: Traditional steel bars are prone to corrosion under the action of chloride ions in seawater and sea sand, affecting the safety and durability of structures. FRP bars, due to their excellent mechanical properties, good designability, and corrosion resistance, are considered promising materials to replace traditional steel bars. 3. **Importance of Bond Strength**: The bond strength between FRP bars and SWSSC is a key factor in the design and application of FRP-reinforced SWSSC structures. Existing experimental studies mainly focus on limited issues and cannot comprehensively explain the influence of different parameters on bond strength. 4. **Limitations of Existing Methods**: In the existing ACI 440.1R-15 specification, the bond strength equation mainly targets splitting failure and is not accurate enough for predicting pull-out failure, with limited factors considered. ### Solution 1. **Dataset Construction**: The paper constructs a dataset containing 13 input parameters, covering the mechanical properties and composition of SWSSC, as well as the mechanical, geometric, and surface characteristics of FRP bars. 2. **Ensemble Learning Models**: Based on the constructed dataset, the paper develops two ensemble learning models, namely Random Forest (RF) and XGBoost. These models can handle complex non-linear relationships and improve the accuracy of bond strength prediction. 3. **Model Evaluation**: The performance of the models is evaluated using statistical evaluation metrics such as the coefficient of determination (R²), root mean square error (RMSE), and mean absolute error (MAE). The results show that the RF model performs excellently in predicting bond strength, outperforming the XGBoost model and the existing ACI equation. 4. **Parameter Impact Analysis**: Through SHAP (SHapley Additive exPlanations) value analysis, the paper reveals the impact mechanism of each input parameter on bond strength. The analysis results show that the mechanical properties and composition of SWSSC, as well as the mechanical, geometric, and surface characteristics of FRP bars, significantly affect bond strength. ### Conclusion The ensemble learning models proposed in the paper provide an effective method for predicting the bond strength between FRP bars and SWSSC, which can serve as an important reference for structural engineering design specifications.