A Probabilistic Approach for Predicting the Fatigue Life of Concrete

Abstract

The fatigue failure of structural elements subjected to repeated cyclic loading may reduce the life of infrastructures. Heterogeneous nature of concrete and random factors in fatigue testing lead to great variability in fatigue life of concrete. As deterministic approach depends on certain parameters and initial conditions, it is not reliable for the prediction of fatigue life of concrete. In this study, a probabilistic approach using artificial neural network is utilised to predict the fatigue life of plain concrete. An artificial neural network predictive model was developed utilising the data from fatigue tests conducted on plain concrete beams of three different sizes mainly small, medium and large. The model is trained using the available experimental data of small and medium specimen and is validated using available experimental data reported on large specimens. The developed model is able to predict the number of cycles of failure of concrete by considering material and fracture mechanics properties responsible for the softening behavior of concrete as input. This approach is advantageous over other methods as it includes the randomness in the fatigue of concrete and will be able to predict the fatigue life of concrete with reasonable accuracy.