A Hybrid Parameter Homogenization Workflow for Assessing the Mechanical Behavior of a Steel Fiber-Reinforced Concrete

Abstract

A novel generic workflow to predict homogenized material parameters of tensile behavior of a steel fiber-reinforced concrete (SFRC) using artificial neural networks (ANNs) was proposed. The neural network estimated the homogenized parameters of composite materials linked to finite-element (FE) models. An advantage of this approach is its flexibility in obtaining model parameters, which often have no physical interpretation. Moreover, the joint application of ANNs to estimate the model parameters and FE simulations to obtain the global mechanical behavior of SFRC is an innovation. An experimental database was constructed from the tests available in the literature and provided the ANN input data: the water-cement ratio, the fiber volume fraction, and the diameter and length of steel fibers. The outputs were Young’s modulus, the tensile strength, and the fracture energy of the composite. Three different networks were trained for each output dataset. The ANN configuration consisted of an input layer with four nodes and an output layer with one node. Blind tests with five experimental test sets checked the solution accuracy, presenting relative errors lower than 10%. Finally, a FE model of a direct tensile test was built, adopting the parameters obtained through the workflow. The load–displacement curve of the numerical solution showed a good agreement with the experimental curve, and peak–load errors were smaller than 5%.