PARAMETER SENSITIVITY ANALYSIS IN STEEL FIBER REINFORCED CONCRETE

Resumo

In steel fiber reinforced concrete, the random dispersion of discrete fibers in the cementitious
matrix and the distinct lengths, cross sections and volumetric contents cause deviations in the global
behavior of a given structure. Traditionally, the association of probabilistic and numerical approaches
reproduces the variability effects in the material mechanical behavior. However, the accomplishment
of these simulations is strongly dependent on statistical study of experimental tests. In recent years,
computational intelligence techniques emerge as powerful tools for parameter identification and
calibration in engineering. However, parameter sensitivity analysis is a predecessor stage required for
applying these new concepts into an Artificial Neural Network (ANN). In this sense, this paper presents
a sensitivity study of macroscale parameters in steel fiber reinforced concrete by using the Design of
Experiments (DOE) methodology. This technique is an important method to analyze the influence of one
or more parameters on the given output of the ANN. An experimental database available in literature
provides the input data for the ANN: water-cement ratio, volumetric content, diameter and length of
steel fibers. The parameter outputs are the Young modulus (E), tensile strength (ft) and fracture energy
(G) of the material. Response surface plots are provided in order to identify the relevant experimental
parameters for the description of the mechanical behavior of the composite predicted by the ANN.
Thereby, sensitivity analysis in Artificial Intelligence methods becomes an attractive approach to verify
the mechanical behavior of concrete, establishing the most suitable parameters that will generate a
reliable neural network.