Artificial neural networks for the prediction of mechanical behavior of metal matrix composites

In this paper we demonstrate the power of artificial neural networks in predicting strengthening in the transverse direction of metal matrix composites by regularly arranged strong fibers. A neural network is trained in different ways based on a numerical study in which the fiber volume fraction and the matrix hardening ability was studied systematically for fibers in a hexagonal arrangement loaded at 0 and 30° transverse direction and for a square arrangement of fibers loaded at 0 and 45° transverse directions. Strengthening predictions are then made for hardening cases of both fiber arrangements which were not covered by the finite element calculations as well as for arbitrary loading directions not achievable by simple finite element unit cell calculations in the case of square fiber arrangements.