Boosting algorithms for predicting the bond properties of timber and fiber reinforced polymer (FRP) under thermal cycling using single-lap shear tests

Precise modeling of the bond properties between timber and fiber-reinforced polymer (FRP) at varying temperatures is crucial for structural integrity. This study uses boosting algorithms-XGBoost, LightGBM, and CatBoost-to model the impact of thermal cycles on the bond behavior of FRP-timber using single-lap shear tests. A comprehensive dataset of 150 experimental results was compiled and processed to train and test the models. Key parameters were density, stiffness, temperatures, thermal cycling, energy absorption, ultrasonic pulse velocity (UPV) measurements, and fiber types (glass, carbon, and aramid). Genetic algorithms (GA) and a 5-fold cross-validation (CV) technique were employed to fine-tune the hyperparameters of the boosting algorithms. The results demonstrated the superior accuracy of the CatBoost model in predicting the bond characteristics. This comprehensive evaluation highlights the critical factors influencing FRP-timber composite mechanical behavior. Feature contribution analysis revealed that temperature and thermal cycles exert the most significant impact on the bond properties. The types of fibers (glass, carbon, and aramid) showed relatively low importance. This study shows that FRP-timber interface properties inform accurate predictive models and design guidelines for varying thermal cycles.