Close

A note on cookies

We use cookies to improve your experience of our website. If you want to find out more see our Privacy Policy

Menu

AboutFlow

Research menu
Home / Research / Technical Work Packages / WP 3 / 

WP 3: Automatic Differentiation Tools and Discrete Adjoint Solvers

Objectives

  • Advances with the source-transformation and operator-overloading AD tools
  • Further advances with discrete ACE+, extending it from its current tangent-linear to adjoint, including a larger set of the algorithm.
  • Discrete and Automatic differentiation of OpenFOAM

Methodology

QMUL will extend existing discrete adjoint codes to be suitable for unsteady flows. AD will be applied to OpenFOAM using the operator-overloading AD tool dco from partner RWTH. <> The functionality of dco will be extended to cope with the advanced C++ constructs of OpenFOAM. AD-differentiated code at the outer level will be complemented by hand-differentiated blocks where code proves too complex or AD code lacks the necessary performance.

Partners RWTH and INRIA will extend their tools to handle adjoints of MPI-parallel codes. Message-Passing primitives will be considered as primitive constructs of the application language.

The arbitrary-primal discrete adjoint of ESI developed in FlowHead will be improved w.r.t. robustness and performance. The advances in WP3 support the application WPs 4-6.

Lead: RWTH, participants: QMUL, INR, RR, ESI [months 1-36]

Participating Research Fellows

  • Jan Hückelheim (ESR 1 at QMUL): Application of AD for unsteady flows (WPs 3, 4, 5)
  • Mateusz Gugala (ESR 2 at QMUL): Multi-fidelity unsteady optimisation (WPs 3, 4, 5)
  • Siamak Abkarzadeh (ESR 3 at QMUL): Discrete Adjoint solver based on OpenFOAM (WPs 3, 5)
  • George Eleftheriou (ESR 5 at ESI): Discrete Adjoint Solvers and Parallelisation (WPs 1, 3, 4)
  • Mattia Oriani (ESR 6 at ESI): Discrete unsteady adjoint solver (WPs 3, 5)
  • Ala Taftaf (ESR 7 at INRIA): Source transformation AD tools (WPs 3, 5)
  • Christos Vezyris (ESR 8 at NTUA): Discrete and continuous adjoint methods: GPU-enabled and one-shot adjoints, for unsteady flow optimisation (WPs 2, 3, 4, 5)
  • Mehdi Ghavami Nejad (ESR 9 at NTUA): OpenFOAM-based continuous adjoint method for aerodynamic optimization of unsteady turbulent flows. (WPs 2, 3, 5)
  • Georgios Ntanakas (ESR 10 at RR) Regularisation of turbo-machinery flows with unsteady flow separations (WPs1, 3, 4, 5)
  • Zahrasadat Dastouri (ESR 11 at RWTH): Operator-overloading AD tools, application to ACE+ (WPs 3, 5)
  • Arindam Sen (ESR 12 at RWTH): Operator-overloading OpenFOAM (WPs 3, 5)
  • Nikolaos Magoulas (ESR 13 at VW): Application of continuous and discrete adjoint OpenFOAM to unsteady car aerodynamics, Adjoint-based optimisation workflows (WPs 1, 2, 3, 5)
  • Sheikh Razibul Islam (ESR 14 at WUT): Adaptation for unsteady adjoint optimisation (WPs 3, 4, 5)
^ Back to Top