Metodología para el ajuste de leyes de daño en modelos de elementos finitos a través del uso de redes neuronales y metamodelos

Abstract

The most commonly obtained variables in material characterization campaigns are the force and displacement values found at the actuators. Regardless of other experimental measurements which can be registered (e.g., crack mouth opening (CMOD), local deformations through image correlation techniques, etc…), force and displacement at the actuators are the variables which must be used to validate the results obtained through the numerical modelling of the material´s behaviour under test conditions. However, when facing the numerical modelling of experimentally observed damage, many parameters have a significant impact on the force-displacement relation. For this reason, the superposition of numerical and experimental data on the same graph is not a trivial matter. In this article, a methodology based on the use of neural networks is presented, to obtain the optimum damage parameters of a material. This guarantees that the differences between the numerical and experimental force-displacement curves are minimised. By using neural networks, previous knowledge related to the damage parameter values is no longer needed, allowing non-expert users the opportunity to easily develop a correct damage law.