Development of a hybrid regression model for predicting the mechanical behavior of macrosynthetic fiber reinforce concrete

This paper proposes a hybrid regression approach for the prediction of mechanical properties of materials within the framework of either numerical or analytical simulations. It demonstrates the application of this model to macrosynthetic fiber reinforced concrete, by using the results of an experimental campaign performed on groups of specimens characterized by varying nominal strength and fiber dosage. The proposed approach for developing the prediction model is based on a hybrid regression model that combines regression analysis with the results of a mechanical model, specifically a cracked hinge model. The proposed approach is general and can be extended to other materials and models. The parameters of the mechanical model are correlated with the nominal compressive strength of concrete and the fiber dosage, while the regression coefficients are estimated by minimizing a global error function, which directly depends on the outcomes of the cracked hinge model. This hybrid approach differs from traditional physics-inspired regression models, which consist of two separate and consecutive steps: the inverse analysis of the mechanical model and regression analysis. The developed prediction model demonstrates a strong correlation with the average experimental response and excellent extrapolation capacity, as tested on additional experimental data, outperforming the traditional two-step approach.