The Prediction of Bond Strength Between Thin Steel Bars and Concrete Using an Adaptative Neuro-Fuzzy Inference System

Resumo

In reinforced concrete, the solidary behaviour between steel and concrete means that tensile, compressive, bending, or torsional stresses are transferred from one material to another due to adhesion. Tests such as pull-out and beam tests, proposed by the EN:10080 standard, can be used to verify this behaviour; however, the difficulties inherent in destructive tests, as well as the non-linearity and number of variables involved in this issue, make it pertinent to use alternative methods. Using a database with experimental results from pull-out tests, the aim is to create an alternative technique to these tests. Hence, this work proposes the use of an Adaptive Neuro-Fuzzy Inference System (ANFIS) to predict the maximum adhesion load in pull-out tests. An ANFIS is a hybrid model made up of two techniques: Artificial Neural Networks (ANN) and Fuzzy Logic. The central idea in a Fuzzy Inference System is that the techniques work in parallel, with the neural network used to adjust fuzzy logic parameters. The combination of ANN and Fuzzy Logic is a useful approach to obtain the benefits of these systems in a single model, namely Neuro-Fuzzy. The Neuro-Fuzzy Inference System implemented in this work consists of four input variables: the surface of the bar, the diameter of the bar (ϕ), the compressive strength of the concrete (fc), and the length of the anchor (Ld). The output is the maximum adhesion force at the steel-concrete interface. The performance metrics obtained in this work will be evaluated against those obtained by other computational methods carried out earlier by our research group. The study indicates an alternative to the destructive tests that are widely carried out, overcoming their limitations considering the safety coefficients used in engineering.