Prediction of the residual flexural strength of fiber reinforced concrete using artificial neural networks

•This study proposes a database of hooked-end SFRC for an ANN model.•The ANN model well predicts the post-peak strength in bending based on EN 14651.•The model presented a particularly good fit with accuracy above 92%.•Peak loads reached by the ANN are in good fit with the FRC slabs experimental tests.•The model can help engineers in structural design without experimental tests. The work in hand proposes Artificial Neural Networks (ANN) to predict the residual strength of fiber reinforced concrete under bending load. A database containing experimental and synthetic data was constructed with 400 datasets, providing the following input data: fiber aspect ratio, matrix compressive strength, and steel fiber volumetric fraction. The parameter outputs were the flexural strengths fR,1, fR,3, and fR,4 of the composite. This study adopts the Bayesian regularization algorithm to predict the flexural parameters. Five different ANNs were trained and validated to assess the output datasets. The solution presented a particularly good fit for all tested networks, with regression values above 92 %. Once the networks are trained, mean square error analyses and blind tests with different experimental test sets check the solution accuracy, leading to the selection of the best network. Finally, an application regarding fiber-reinforced slabs-on-ground design considering the best ANN response is carried out. The resulting design load is compared to experimental slab tests performed by different authors from the literature and indicates a good agreement. In this sense, applying ANN methodology to slabs-on-ground design becomes an attractive tool for civil engineers.