Estudo do comportamento à fadiga de cabos condutores de energia com alma de alumínio utilizando redes neurais artificiais

Abstract

The Brazil continental dimensions and distance from the power generating units (usually hydroelectric) of the consumers centers are factors that raise the importance of the distribution lines, since an unexpected failure due to fatigue at the conducting cables, can reach tens of millions of consumers involving therefore, very high costs. In order to estimate the remaining cable fatigue life of these transmission lines, sophisticated laboratories and long tests are necessary to obtain S-N curves with the same level of mean stress applied in the field. Thus, it becomes desirable a model capable of predicting the resistance to fatigue of conductive cable effectively, simplifying the need for testing and the costs, in addition, of course, the risks of blackout. At this way, this study seeks to create Artificial Neural Networks (ANNs) architectures capable of estimating the fatigue strength of aluminum conducting cables. By varying the cable’s structural parameters such as the linear weight (W), the flexural stiffness module (EI) and the Poffenberger-Swart constant (K), taking into account the mean stress (stretching load) and the number of cycles to which the cables are subjected. The training and testing of the ANNs were made through a set of data obtained from fatigue tests carried out in the cable laboratory of the Fatigue, Fracture and Materials Group (GFFM) of the Federal University of Brasília (UnB) for different types of aluminum alloys cable. ANNs will be used to construct constant life diagrams for this family of cables, and compare the results obtained experimentally and by ANNs using the root mean square (RMS). The results obtained by the ANNs proved to be adequate, based mainly on the low RMS values obtained for the different ANN architectures and on the plotted constant life curves, thus attesting to the good generalization of the fatigue resistance for the aluminum conductor cables analyzed.