Unified model using artificial neural network for high strength fibrous concrete subjected to elevated temperature

The most interesting aim of this research is to assess the capability of artificial neural networks (ANN) to predict the post-fire residual stress-strain curve of unconfined plain and fibrous concretes under axial compression. In this study, the experimental variables are volume fractions of flat crimped steel fibers and polypropylene fibers, inclusion of hybrid fibers and temperature of exposure under natural cooling. A total number of 126 cylindrical specimens of different types of concrete were prepared. These specimens were then exposed to the elevated temperatures ranging from room temperature to 800 degrees C, and the mechanical properties were evaluated. Based on the test results, an ANN model is developed for the prediction of complete residual stress-strain responses of plain and fiber-reinforced concrete at elevated temperatures. The Levenberg-Marquardt (LM) algorithm has been used in the training. The performance parameters MSE and R values were obtained as 2.2944e-03 and 0.9885, respectively. The stress-strain curves of different samples were predicted and compared with the curves which were obtained experimentally. A good match between the predicted and experimentally obtained stress-strain curves can be observed. An equation based on the weights between the artificial neurons and biases of ANN model was also proposed in this study. The proposed ANN model is unified in nature as this single model is capable in predicting the stress-strain curves for all ranges of temperatures and various compositions of added fibers.