[PDF]
Abstract: When monolithic approaches to solve multidisciplinary problems are not feasible, single-disciplinary solvers must be coupled. The NASA-developed Python framework OpenMDAO, which manages the communication between the single-disciplinary solvers, is used to implement various coupling methods (some of which support MPI-parallelism). The feasibility of using OpenMDAO on large-scale industrial problems is assessed, based on the experiences made with a small academic problem -- stationary natural convection of a fluid in a square cavity. Two deliberately simple single-disciplinary solvers -- one for the convection-diffusion equation and one for the NAVIER-STOKES equations -- are implemented in Python. The solutions returned from OpenMDAO match the reference, and the iterative convergence behaviors of the tested coupling methods match their theoretical expectations -- so there is no considerable drawback on convergence when using OpenMDAO. OpenMDAO allows to quickly test various coupling methods, but it should only be used if the single-disciplinary solvers are fixed and runtime performance is not a major issue.
/TeX: Contains the project paper
/Solvers: Contains the single-disciplinary solvers to couple
/OpenMDAO: Contains the OpenMDAO-components and the coupling scripts
/Examples: Contains various examples/tests
Simon Ehrmanntraut
Special thanks go to