Neural network-aided prediction of post-cracking tensile strength of fibre-reinforced concrete

Structural fibres are an effective method to improve concrete post-cracking tensile strength (f(ctR)). Currently, the characterization of this property is mainly performed experimentally. This is a source of uncertainties at design stages, which hinders the development of new fibre type and/or optimization of those currently existing. This paper presents a multilayer perceptron neural network to predict f(ctR) of fibre-reinforced concrete (FRC) subjected to the Barcelona Test. The optimal architecture of the predictor is obtained by evaluating 9216 configurations of input dimension and number of hidden layers and neurons. The generalization performance is assessed using repeated random sub-sampling validation with 50 iterations. The final model can predict with high accuracy the f(ctR) of FRC for different cracking stages. A parametric analysis is performed to prove coherence between the results predicted by the model and the established understanding of the FRC behaviour. Finally, numerical expressions are provided as an alternative tool to traditional testing to predict the residual strength of the Barcelona Test for pre-design and quality control purposes based on fibre dosage, concrete strength, specimen type and height and fibre geometric characteristics. These type of approaches are found to be necessary for boosting the development of the FRC technology. (C) 2021 The Author(s). Published by Elsevier Ltd.