Machine learning adaptation for laminar and turbulent flows: applications to high order discontinuous Galerkin solvers

  • Tlales, Kenza (Universidad Politécnica de Madrid)
  • Otmani, Kheir-eddine (Universidad Politécnica de Madrid)
  • Ntoukas, Gerasimos (Universidad Politécnica de Madrid)
  • Rubio, Gonzalo (Universidad Politécnica de Madrid)
  • Ferrer, Esteban (Universidad Politécnica de Madrid)

Please login to view abstract download link

This study proposes a machine learning-based method for mesh refinement in steady and unsteady flows. The regions to be refined are marked by using a clustering technique used to identify viscous and inviscid regions in a flow past a cylinder at different Reynolds numbers. After the regions have been marked, we apply $p$-refinement within these clustered regions to achieve a similar level of accuracy compared to uniform mesh while reducing the computational cost. The data used in this work have been generated using the high-order spectral element CFD solver HORSES3D.