Empowering pre-exascale computers for Darcy-Brinkman simulation of wormhole growth based on X-CT data - can we recover experiments?

Advent of pre-exascale and exascale computers opens possibility for much higher resolution simulations of porous media flows. During the launch phase of the LUMI supercomputer, a number of simulations of wormhole growth commenced with an aim to use as much spatial information as possible with up to 1e9 DOFs. The goal was to investigate if properties of growing wormholes could be recovered if sufficient resolution is assured. Samples used in this study underwent experimental studies. They were scanned before and after the experiment, as well as during the dissolution. This 4D tomographic data provided necessary input for high-res simulations as well as validation framework.

Fig: Wormhole groth patter with branching and rapid direction change observed in evperiment

As the LUMI computer, as well as most of the newly built HPC machines, is based on GPUs we decided to use the Lattice Boltzmann code as main flow and transport solver. LBM has significant number-crunching performance thanks to its intrinsic parallelization properties which was paramount for this study. Based on an open-source, highly parallel multi-GPU TCLB solver, we design the model capable of handling Darcy - scale simulations with the initial porosity fields constructed based on X-ray microtomography images. In particular, we analyze the reactive-infiltration instabilities, which lead to the  formation of dissolution fingers (wormholes), in which both the flow and reactant transport become spontaneously localized.

Since dissolution fingers dramatically increase permeability of the rock, wormholing is important both for industrial applications and in hydrogeological studies. The main problem in modeling of wormholing is a multi-scale character of this process, with flow and transport near a wormhole tip strongly coupled to the macroscopic geometry of the emerging structures. The ability to perform large scale parametric and sensitivity studies of wormholing constitutes thus an important addition to experimental studies, hence the need for a high throughput simulator. 

We test our numerical predictions against the data from time-lapse dissolution experiments in an aim of constructing a predictive model capable of recovering time evolution of 3D wormhole shape based on the initial X-ray tomography data.