Prediction model for the failure behavior of concrete under impact loading base on back propagation neural network

Predicting the failure behavior of concrete under impact loading is of great significance for the repair of buildings and improving the protection capabilities of modern defense systems. Conventional approaches such as experiments, theoretical analyses and numerical simulation methods have been widely used in the analysis of this problem. However, they are not always accessible in situations where high accuracy, fast computation and simple modeling are required at the same time. In this study, with projectile penetration as impact loading, an artificial neural network (ANN) model for predicting the failure behavior of concrete is proposed based on the back propagation (BP) algorithm. Feature selection is conducted when constructing the model and the direct mapping between penetration features and failure behavior parameters of concrete is established. The model outperforms commonly used empirical formulas in terms of prediction accuracy. Compared with theoretical analysis method, it does not rely on the failure mechanism which is not yet clearly under stood and does not require complicated parameters analysis. Furthermore, its calculation time is negligible compared with the numerical simulation as the output can be obtained in a second from the input penetration features. Moreover, tests are conducted to validate the feasibility and accuracy of the proposed model.