Performance-Based Prediction of Shear and Flexural Strengths in Fiber-Reinforced Concrete Beams via Machine Learning

The accurate and precise prediction of shear and flexural strengths in reinforced concrete (RC) and fiber-reinforced concrete (FRC) beams necessitates advanced computational techniques. This study pioneers the application of an Artificial Neural Network (ANN) to model these strengths and to classify failure modes in beams. Leveraging a dataset of 116 experimental tests on ultimate strengths from extensive literature, the ANN was meticulously trained, tested, and validated, revealing that the optimal neuron count for the modeling task was 15. This configuration achieved a root mean square error (RMSE) of 0.096 MPa and a coefficient of determination (R-2) of 0.95, outperforming traditional design models. The study further explored an independent variable importance analysis, revealing that the beam width and effective depth were paramount in predicting strengths, findings that are congruent with established structural engineering principles. The analysis also highlighted the significance of post-cracking resistance parameters, particularly the residual flexural strength at 2.5 mm deflection, in enhancing the predictive model. The ANN classification successfully differentiated between shear and flexural failure modes, achieving an impressive accuracy of 96.5% with 25 neurons. This dual strength to model and classify underscores the ANN's robustness, offering a comprehensive tool that surpasses conventional model codes in both accuracy and precision. The results advocate for the integration of ANN techniques in structural design, promising a future where machine learning not only informs but also transforms engineering practices.