Machine Learning based prediction of detailed chemistry phenomena in CFD computations

CFD simulations are currently used for the design and optimization of propulsion and industrial devices involving chemically reactive flows. They facilitate the research of new solutions that allows to reduce CO2 and pollutant emissions. Recently, the use of Machine Learning in physical sciences and engineering has attracted a lot of attention, with the goal of accelerating numerical simulations of physical processes. In this context, the main challenge is to reach a significant reduction in the computational costs while maintaining a high accuracy with respect to standard solvers. The objective of the present thesis is to evaluate the ability of Machine Learning surrogate models to replace CPU-expensive chemistry solver to accelerate CFD computation of industrial applications. The work will pursue the strategy recently developed, both at IFPEN and at CORIA laboratory, combining chemistry reduction and the use of an Artificial Neural Networks surrogate model as chemistry solver. It has been recently demonstrated, in an academic configuration, that this approach leads to significant gains in terms of CPU cost, while preserving the accuracy in describing complex chemistry effects. The objective of this thesis is to bridge the gap between academia and industry by extending the methodology to cases of practical interest. In particular, the optimal definition of the ANN training database will be investigated in depth. It will be based on previous work where the chemical manifold of the simulated system is approached by a set of simple reactors interacting with each other through a prescribed diffusion law. The final validation cases will involve applications of interest at IFPEN, such as industrial safety, H2 production or CO2 capture processes.

Keywords: Machine Learning, Chemistry reduction, CFD simulation

  • Academic supervisor Pr., VERVISCH Luc , Laboratoire CORIA / INSA Rouen , https://orcid.org/0000-0003-0313-2060
  • Doctoral School ED591 PSIME, https://ed-psime.normandie-univ.fr/
  • IFPEN supervisor Dr., AUBAGNAC-KARKAR Damien, Numerical Modelling of Energetic Systems department, damien.aubagnac-karkar@ifpen.fr, https://orcid.org/0000-0003-3995-4920
  • PhD location IFP Energies nouvelles, Rueil-Malmaison, France
  • Duration and start date 3 years, starting in fourth quarter 2023
  • Employer IFP Energies nouvelles, Rueil-Malmaison, France
  • Academic requirements University Master degree in relevant disciplines
  • Language requirements Fluency in French or English, willingness to learn French
Contact
Encadrant IFPEN :
Dr., AUBAGNAC-KARKAR Damien
PhD student of the thesis:
Promotion 2023-2026