A Computational Approach to Predict the Bond Strength of Thin Steel Rebars in Concrete by Means of Artificial Neural Networks

Resumo

The bonding between steel rebars and concrete is one of the critical aspects of reinforced concrete
structures. As a phenomenon influenced by many variables, it is challenging to establish how the steel-concrete
adhesion can be described in the standards used for reinforced concrete design. This study used an experimental
data set of 190 pull-out specimens to develop an artificial neural network (ANN). The data used in the modeling
were collected from 8 different studies and were arranged as four input parameters: bar surface, bar diameter (φ),
concrete compressive strength (fc) and the anchorage length (Ld). The output result was the pull-out load.
Several scientific studies on this property have been performed since the 1940s, among many other
investigations in this field. Generally, these studies refer to bars with diameters greater than 12.0 mm. However,
few studies have evaluated the performance of reinforcing bars with diameters smaller than 10.0 mm, which
includes 6.0-, 6.3-, 8.0- and 9.5-mm diameters, usually used in reinforced concrete elements. This work uses
ANN to analyze and build a prediction model for the steel-concrete bond and its potential to deal with
experimental data. The root mean squared error (RMSE) found for the maximum pull-out load in the pull-out
test was 2.35 kN and the obtained R-squared was 0.94. Therefore, the pull-out load results found were compared
with the results obtained through the equation available in CEB 2010. Finally, it is possible to conclude that the
current model can satisfactorily predict the bond strength of thin bars.