Image based finite element modelling of steel microstructures

Recent progress in powerful non-destructive experimental techniques like X-ray diffraction contrast tomography (DCT) has given momentum to reconstructing the as-measured grain topologies and crystallographic orientations in finite element (FE) models. This offers new insights into material behaviour and enables advancements in used material models. Although grain structure can now be faithfully reconstructed in FE models, the issues of the constitutive models remain, e.g. the approach to the grain boundaries that are seldom explicitly modelled. An approach for explicitly accounting for them is the cohesive zone approach where the grain boundaries are modelled using the cohesive elements with zero physical thickness. The constitutive response is taken as a traction-separation curve while above certain separation threshold the load-carrying capability of the grain boundary is decreased. Such an approach can result in discrepancies between the expected and the numerically obtained grain boundary normal stresses. This is demonstrated first on a realistic case of a 0.4mm stainless steel wire where the grain topologies and crystallographic orientations are directly transferred into FE model. The model contains 363 grain and more than 1600 grain boundaries. Comparing the expected and numerically obtained grain boundary normal stresses in 3D Voronoi tessellation aggregates reveals similar discrepancies. These are observed at the triple lines between the neighbouring grains and have so far been observed only for the simple linear tetragonal FE. A simpler 3D Y model is then used to analyse the effects of using tetragonal and hexagonal FE with both linear and quadratic interpolation, as well as the conformal and non-conformal mesh between the grain-grain boundary-grain interfaces. It is shown that in all these cases the discrepancies persist no matter which combination of the FE is used. This points to an issue in the cohesive zone approach in such a complex geometries. An alternative approach would be to replace the cohesive FE with ordinary structural FE with of non-zero physical thickness. This requires increasing the grain boundary thickness in the current model. An early geometrical model of such a case is presented. Successfully increased grain boundary thicknesses are indicated with red colour. The algorithm however, needs further development as not all grain boundary thicknesses are increased successfully.

SIMONOVSKI Igor;  CIZELJ Leon; 

2013-04-05

University of Oxford, Mansfield College

JRC77188