An approach to predict ultimate transverse tensile strength based on mixed ANN models for a composite lamina

This research work approaches the prediction of the ultimate transverse strength behavior of unidirectional composite laminas (Yt) using mixed artificial neural network (ANN) models. Thereby, two empirical relations, presented by Teanu and Barbero, were used for comparison purposes and to construct the mixed ANNs. These equations differ in the mechanical properties used to obtain Yt. For the ANN training, a dataset with 85 laminas was used, all obtained from the literature and manufactured with different types of fibers (aramid, glass and carbon) and polymer matrices. The best model obtained was the one based on the relation presented by Barbero, where the void fraction volume (Vv) and the transverse modulus of the fiber (Eft) are input parameters. Another aspect was the discovery of a region in the dataset that drastically changes its behavior, forming a valley in the prediction curves with values far lower than those forecast by empirical equations. This approach can be seen as a new methodology to estimate Yt using the same variables encountered in the classical equations but obtaining values closer to experimental ones.