AAMPS: Adaptive and AI-assisted model-selection and parallelisation in subsurface flow applications

In modern times, some of the most pressing issues facing humanity involve energy, storage, and pollutants. 
As a possible solution, scientists have turned towards the subsurface which in its vastness hides oil and gas, geothermal energy, and clean water to extract;
 and hydrogen, nuclear waste, and excess environmental CO2 to store. 
The question of accurately modelling such processes comes down to our understanding of multiphase flow through porous domains. 
Some of the main challenges in this regard stem from fractures (specially for applications such as fracking), 
hysteresis (the history dependence of the flow-behaviour), and heterogeneous nature of the abyss. 
Although the tools required to simulate such real-life applications are PDE-based,
 recent advances in machine learning (ML) algorithms and parallel-computing (PC) have the potential to significantly improve the computational performance, 
robustness, and accuracy of such classical methods. This is what we aim to achieve in this project.

The objectives of the project are to apply ML and PC in two separate contexts.

  1. Adaptive a-posteriori estimate and AI-based model-selection and parallel computation
  2. ML-based flux selection in reservoir scale problems using parallelly computed training data