Decreasing computation time in finite element simulations coupled with polycrystalline plasticity
Résumé
In this work, finite element (FE) modeling is used to simulate uniaxial compression of two different aluminium specimens with a known crystallographic texture. The first specimen has a cubic shape discretized by 48 tetrahedral finite elements, while the second specimen has a cylindrical shape described by 1085 elements. The anisotropy of the metal is, in both cases, simulated using polycrystalline plasticity under the Taylor full-constraints (FC) model. The metal is represented by a sampling of 1200 crystallographic orientations that are statistically representative of the global texture. The simulation is repeated several times for each specimen, using the same finite element mesh, but assigning a different number of grains to each finite element. The grains are assigned to each element by using two different methods of distribution. The compression test is simulated without friction for the cubic specimen, and with a significant friction for the cylindrical specimen. A reference simulation is given each time by the case where the complete sampling of 1200 grains is assigned to each element. Subsequent simulations are computationally "cheaper", even though all of them rely on the same set of 1200 orientations : each element is represented by a number n of crystals smaller than 1200. The n crystal orientations are randomly selected, and the selection differs from one element to the other. The simulation results are compared for different values of n in terms of computation time, and accuracy of the mechanical anisotropy and textural predictions.