Ensemble boosting-based soft-computing models for predicting the bond strength between steel and CFRP plate

Fiber reinforced polymers (FRPs) hold great potential for reinforcing and repairing structures due to their high strength-to-weight ratio and excellent resistance to corrosion and environmental degradation. Modeling debonding failures requires an understanding of the behavior of simple FRP-to-steel bonded connections. The use of boosting-based machine learning methods for CFRP-to-steel bonded contacts has received little research attention. The present study examines the bond behavior of CFRP sheets in steel beams using boosting-based ensemble machine learning approaches such as the XGBoost, GBM, CATBoost, LGBM, and ADABoost algorithms. For the machine learning boosting-based model approach, eight total input variables and one output variable were chosen to predict the maximum load (PU) of the bonding behavior between the CFRP and steel. The study uses a database of 317 experimental datasets compiled from previous literature for training and testing the proposed machine learning models. On the other hand, rank analysis was utilized to determine the optimal models. According to the results of rank analysis utilizing many performance criteria, ADABoost overcame the other outcomes, with R² values for training and testing of 1 and 0.99882, respectively. The construction industry benefits directly from the application of established boosting-based machine learning techniques to investigate the bonding behavior of CFRP sheets on steel beams. This methodology enhances the accuracy of the design, reduces costs, and increases the general performance and durability of CFRP-reinforced buildings.