Numerical performance evaluation of debonding strength in fiber reinforced polymer composites using three hybrid intelligent models

Debonding of the fiber-reinforced polymer (FRP) reinforcement is considered as a significant issue in the concrete design because of shear stresses. The main problem is the potential of brittle debonding failures that can highly reduce the effectiveness of strengthening. Shear bond strength and the governing variables have been empirically analyzed several times; however, these experiments cannot provide accurate predictions due to the complexity of debonding process. In this regard, this paper is aimed to investigate the debonding strength of FRP composites using novel models of Extreme Learning Machine (ELM) in co-operation with Teaching-Learning based Optimization (TLBO), Particle Swarm Optimization (PSO) and gray wolf optimizer (GWO). By comparing corresponding values of coefficient of determination (R2) and root mean square (RMSE) in three hybrid models, the best performance in predicting the debonding strength of FRP composites was obtained for ELM-GWO in comparison with ELM-PSO and ELM-TLBO. Considering the best RMSE value as 0, GWO with RMSE = 2.5057 showed the closest value to 0 compared to PSO (2.73) and TLBO (5.58). On the other hand, since the best value of R2 is closest to 1, GWO with R2 = 0.9504 indicated a better performance compared to PSO (0.9431) and TLBO (0.7554).