Important global challenges related to energy, groundwater and environment are associated with
the simultaneous flow of multiple fluids in porous rocks and sediments, e.g. water, air, pollutants
and CO2. This process, called multi-phase flow, is controlled by the microstructure of the pores in
these materials. However, detailed computer simulations of multi-phase flow through these pores
are computationally intensive and suffer from a wide range of uncertainties. This research
proposal presents a new modelling paradigm which fully acknowledges these uncertainties, and
attempts to reduce them by directly incorporating experimental data through machine learning
techniques. The experimental data is obtained from time-resolved X-ray microtomography (micro-
CT), allowing to follow the fluid distribution in a rock over time during flow experiments. In the
first phase of the project, a methodology to validate pore-scale models on a pore-by-pore basis
using this data will be established. In the second phase, new data-assisted models trained on the
micro-CT experiments will be developed, using machine learning methods in hybrid combinations
with physics-based models. Comparison to pure physics-based models allows new and exciting
opportunities to better understand multi-phase flow and to improve the modelling by enhanced
validation, calibration, and extrapolation. This in turn results in better constraints for large-scale
simulations of flow in the sub-surface.