Data-driven model for predicting intermediate crack induced debonding of FRP-strengthened RC beams in flexure

There are a kinds of failure modes in flexurally strengthened RC beams using externally bonding FRP sheets. To evaluate the flexural strength of FRP-strengthened RC beam, the failure mode of the strengthened beam must be determined first. Intermediate crack (IC) induced debonding is the main failure mode in flexurally FRPstrengthened RC beams. In this study, a Back-Propagation Neural Network (BPNN) model is proposed to predict the intermediate crack (IC) induced debonding of FRP-strengthened RC beams. A database of 101 beams from different researchers are used to train and test the BPNN model. This model established in this study has eight inputs, including concrete strength, shear span ratio, tensile reinforcement ratio, the yield strength of reinforcement, stirrup reinforcement ratio, FRP stiffness, the ratio of sheet width to beam width, and the ratio of anchorage length to shear span, the output is FRP strain at the initial debonding, that is, the ultimate debonding strain of FRP reinforcements. The importance and sensitivity of the parameters are studied by the BPNN model and collected data. Based on the parametric study and Levenberg-Marquardt algorithm for nonlinear fitting, a design model is established for the prevention of IC debonding failure. The BPNN model and design model established in this study can accurately predicted the IC debonding of FRP-strengthened RC beams.