Reliability assessment of IC debonding in FRP strengthened RC beams using neural network

Abstract

IC debonding is a common failure mode in fiber-reinforced polymer (FRP) strengthened concrete beams. Guidelines proposed several models to predict the ultimate strain in the FRP when IC debonding occurs. The reliability of these models has great importance in the design context. Most studies used assumptions that decrease the accuracy of the reliability analysis. In this study, by using the finite element method and neural network, the use of simplifying assumptions is reduced, and the accuracy of analysis is increased. To this end, the bond behavior between the FRP sheet and concrete is accurately modeled by finite element modeling. The neural network is used to find the relationship between the inputs affecting the debonding behavior and the outputs obtained from the finite element analysis, providing an implicit relationship for the reliability analysis. The results indicate the high accuracy of the neural network (NNET) method. The NNET method's error in predicting the beam's ultimate load for nearly 50% of the samples was in the range of 0–5%. The reliability index obtained from the NNET method for all models was smaller than the normal method. The NNET method suggests that the effect of large live loads on reducing the reliability index is much more than the value predicted by the normal method. To achieve the reliability index in the range of 3.3–3.5, the resistance reduction factor of 0.8 can be suggested for the ACI IC debonding model.