Extension of a fast method for 2D steady free surface flow to stretched surface grids
Demeester, T.; van Brummelen, E.H.; Degroote, J. (2019). Extension of a fast method for 2D steady free surface flow to stretched surface grids, in: Bensow, R. et al. MARINE 2019 - Computational Methods in Marine Engineering VIII, 13-15 May, 2019, Gothenburg, Sweden. pp. 235-246 In: Bensow, R.; Ringsberg, J. (Ed.) (2019). MARINE 2019 - Computational Methods in Marine Engineering VIII, 13-15 May, 2019, Gothenburg, Sweden. CIMNE: Barcelona. ISBN 978-84-949194-3-5. 836 pp., more |  |
| Available in | Authors | | Document type: Conference paper
|
| Author keywords | free surface flow; fitting method; surrogate model; quasi-Newton; convolution theorem |
| Authors | | Top | - Demeester, T., more
- van Brummelen, E.H.
- Degroote, J., more
| | |
| Abstract | Steady free surface flow is often encountered in marine engineering, e.g. for calculating ship hull resistance. When these flows are solved with CFD, the water-air interface can be represented using a surface fitting approach. The resulting free boundary problem requires an iterative technique to solve the flow and at the same time determine the free surface position. Most such methods use a time-stepping scheme, which is inefficient for solving steady flows. There is one steady technique which uses a special boundary condition at the free surface, but that method needs a dedicated coupled flow solver. To overcome these disadvantages an efficient free surface method was developed recently, in which the flow solver can be a black-box. It is based on quasi-Newton iterations which use a surrogate model in combination with flow solver inputs and outputs from previous iterations to approximate the Jacobian. As the original method was limited to uniform free surface grids, it is extended in this paper to stretched free surface grids. For this purpose, a different surrogate model is constructed by transforming a relation between perturbations of the free surface height and pressure from the wavenumber domain to the spatial domain using the convolution theorem. The method is tested on the 2D flow over an object. The quasi-Newton iterations converge exponentially and in a low number of iterations. |
|
