This project is the result of a long-standing collaboration between IFPEN and Inria's TAU team. IFPEN is a French research institute dedicated to new energy and environmental technologies.

It has launched a new project, called ACAI (Acceleration of Computations through Artificial Intelligence), which coordinates several researchers in data science and applied research to combine state-of-the-art statistical learning with high-performance computing in CFD, computational mechanics or underground reactive transport simulations. Inria's TAU (TAckling the Underspecified) team is well known for its activities in the fields of statistical learning, stochastic optimization and, more generally, artificial intelligence. One of its main themes is the application of machine learning methods to scientific computing problems.

The aim of the project is to significantly speed up multiphase flow simulations by introducing approaches based on statistical learning.

The use of machine learning models in CFD simulations with complex physics (e.g. combustion, reactive phenomena, multiphase flows, etc.), can help speed up existing algorithms, for example by creating surrogate models for complex phenomena. The aim of this project is to study two approaches:

• Improving spatio-temporal patterns, preconditioning linear solvers and predicting constraining dynamics;

• Replacing demanding computations in the presence of spatial discontinuities, small-scale phenomena or extremely fast events.

These learning approaches could enable us to tailor spatio-temporal evolution step sizes more appropriately, taking into account estimates of future interactions.

Contacts : Jean-Marc Gratien (IFPEN) | Thibault Faney (IFPEN) | Michèle Alessandro Bucci (Inria) | Guillaume Charpiat (Inria) | Marc Schoenauer (Inria)