BP-ANN based bond strength prediction for FRP reinforced concrete at high temperature

Abstract

In view of the difficulty in establishing a mathematical model to characterize the interfacial performance of FRP reinforced concrete at high temperature, the back-propagation neural network (BP-ANN) paradigm is used as a computational tool for capturing the bond behavior in this paper. Through taking temperature, FRP type, surface form, FRP bar diameter, anchorage length, failure mode, concrete compressive strength and normalized concrete cover thickness as input variables and the interface bond strength as output objective, the BP-ANN model has been established, well-trained and tested by using 151 sets of pull-out experimental data of FRP reinforced concrete at different temperatures in the literatures. The prediction accuracy of BP-ANN model has been evaluated by comparing the training and testing values with experimental results. The more advantages in universality of BP-ANN model than the currently available mathematical models are demonstrated by comparing their applicable scopes and computational errors.