Speaker
Description
The microstructure of a radioactive waste confinement barrier strongly determines its flow and transport properties. Numerical flow and transport simulations for these porous media at the pore-scale are necessary for in-depth understanding of the pore-scale processes, and might for instance allow making predictions beyond current experimental timescales. Such simulations of course require input data that describe the microstructure as accurately as possible. This data can be deduced from various imaging techniques, but for complex porous materials with heterogeneities at different spatial scales, a combination of imaging techniques is required. To merge all available information into a single synthetic but realistic microstructure, and to lower the need for e.g. costly 3D imaging, high quality numerical reconstruction algorithms are needed.
Simulated annealing (SA), which is based on structural descriptors, is one of the oldest approaches and has recently received renewed attention. The method is typically only applied to binary images and is computationally very demanding without optimized algorithms and inventive computational approaches. We extended the state-of-the-art of the SA approach to multiphase reconstructions in a multiresolution-multiphase hierarchical approach, which allow for decreasing the computational burden. The algorithm reconstruct first a binary image at a coarse resolution with refinement afterwards. Once the simulation of the first phase is finished, the corresponding pixels are frozen and the next phase is reconstructed. This new methodology reduces the computational time by at least 50% compared to single grid simulations while additionally improving the reconstruction quality. In reconstruction with a high-particle-size-to-simulation-grid-dimension ratio, the speed up can be several orders of magnitude.
Compared to Direct Sampling (DS), the new SA algorithm does not show the problem of honouring the histogram and the occurrence of verbatim copies. Additionally, the new SA algorithm shows an improvement in the long range connectivity of the different phases within the training image. We show this using two case studies: cement paste and Boom Clay.
| Acceptance of Terms and Conditions | Click here to agree |
|---|
