TRR 181 Seminar "Modelling and Optimization of Marine Engineering Two-Phase Flows" by Thomas Rung (Hamburg University of Technology)

The TRR 181 seminar is held every other week in the semester and as announced during semester break. The locations of the seminar changes between the three TRR181 locations, but is broadcastet online for all members of the TRR.

The TRR 181 seminar is held by our new TRR Project Leader Prof. Dr.-Ing. Thomas Rung (Hamburg University of Technology) on October 21, 11am.

Modelling and Optimization of Marine Engineering Two-Phase Flows


The presentation is devoted to modeling of two-phase flows in conjunction with the optimization of marine engineering shapes.We start from engineering approaches to two-phase flow models based on compressive interface capturing primal Volume-of-Fluid (VoF) schemes dedicated to two immiscible fluids, i.e. air and water. The strategy is of particular relevance due to its its dominance in marine and hydraulic engineering simulations. At the same time, it poses severe challenges to gradient based optimisations owing to the missing differentiability along the virtually sharp interface.

Using an engineering model problem, we outline that adjoint formulations to VoF-methods can not uniquely be solved, unless the dual consistency is compromised by a regularization, which in turn is required to obtain a numerically stable adjoint scheme for optimizing high Reynolds- and Froude-number applications. The interim conclusion is that the frequently employed two-phase flow model does not offer a (continuous) adjoint formulation.

In the spirit of "learning from the adjoint formulation", we conclude, that a more appropriate two-phase flow model should be derived from the suggested adjoint model along the route of a generalized Cahn-Hilliard-VoF method. It is seen that such attempts offer (a) an improved efficiency for simulations of engineering problems and (b) far more predictive realism for more involved/improved resolution studies of interface phenomena.

Moreover, we demonstrate, that related optimization efforts are no longer hampered by differentiability issues and could even address a free surface related objective functional.