Skip to content
forked from mdolab/dafoam

DAFoam: Discrete Adjoint with OpenFOAM for High-fidelity Gradient-based Design Optimization

License

Notifications You must be signed in to change notification settings

XrzCactus/dafoam

 
 

Repository files navigation

DAFoam: Discrete Adjoint with OpenFOAM

tests codecov

DAFoam implements an efficient discrete adjoint method to perform high-fidelity gradient-based design optimization with the MACH-Aero framework. DAFoam has the following features:

  • It uses a popular open-source package OpenFOAM for multiphysics analysis.
  • It implements an efficient discrete adjoint approach with competitive speed, scalability, accuracy, and compatibility.
  • It allows rapid discrete adjoint development for any steady and unsteady OpenFOAM primal solvers with modifying only a few hundred lines of source codes.
  • It supports design optimizations for a wide range of disciplines such as aerodynamics, heat transfer, solid mechanics, hydrodynamics, and radiation.

Documentation

Refer to https://dafoam.github.io for installation, documentation, and tutorials.

Citation

Please cite the following papers in any publication for which you find DAFoam useful.

  • Ping He, Charles A. Mader, Joaquim R.R.A. Martins, Kevin J. Maki. DAFoam: An open-source adjoint framework for multidisciplinary design optimization with OpenFOAM. AIAA Journal, 58:1304-1319, 2020. https://doi.org/10.2514/1.J058853

  • Ping He, Charles A. Mader, Joaquim R.R.A. Martins, Kevin J. Maki. An aerodynamic design optimization framework using a discrete adjoint approach with OpenFOAM. Computer & Fluids, 168:285-303, 2018. https://doi.org/10.1016/j.compfluid.2018.04.012

License

Copyright 2019 MDO Lab

Distributed using the GNU General Public License (GPL), version 3; see the LICENSE file for details.

About

DAFoam: Discrete Adjoint with OpenFOAM for High-fidelity Gradient-based Design Optimization

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Languages

  • C 62.2%
  • C++ 18.2%
  • Python 18.2%
  • Other 1.4%