## Numerical Simulations of Flow around a Simplified Hull Form

##### Master thesis

##### Permanent lenke

http://hdl.handle.net/11250/2409483##### Utgivelsesdato

2016##### Metadata

Vis full innførsel##### Samlinger

- Institutt for marin teknikk [2289]

##### Sammendrag

Hydrodynamic flow around a simplified hull form has been investigated using numerical simulations in OpenFOAM. The flow in question is three-dimensional and turbulent at a Reynolds number of 26,400. The base geometry is a rectangular cylinder (or box) with an aspect ratio of 5:1, and flow in longitudinal (x-)direction (cross-flow width is 1 and in-flow length is 5). The height of the body is 1 in the z-direction, hence it is quadratic in the y-z plane. Simulations have been run in parallel on the supercomputer Vilje at NTNU.
Three different cases have been investigated. Two of the cases are concerned with double-body flow. This means the geometry is "doubled" (mirrored across the x-y plane), and fully submerged with fluid on all sides. The difference between these two cases is the turbulence modelling; one uses Reynolds Averaged Navier-Stokes (RANS), while the other uses Large Eddy Simulation (LES). The third case is called the floating body case. This is not a double-body, but a "single" body next to a simple free-slip approximated free surface boundary. The floating body case uses RANS.
Both the RANS and the LES approaches are based on decomposing the field variables. In RANS, variables are decomposed into a mean and a fluctuating part. In LES, they are decomposed into a filtered and a residual part based on a filter width. In short, RANS decomposition is based on statistical averaging, while LES decomposition is based on spacial filtering. LES is generally much more detailed, and requires a much finer grid. RANS is relatively simple, and much faster to run. Increased accuracy from LES must be paid for in increased computational effort.
For both RANS cases, the grid used was the same (only doubled in the double-body case). Wall functions were applied to reduce cell count near the wall and hence simulation time. The LES grid was much finer, as is required when the whole boundary layer is to be resolved. Applied turbulence models were the realizable k-epsilon model (RANS) and the Smagorinsky sub-grid scale model (LES). These models proved to be well suited to describe the flow in question.
The results showed some differences between the cases. One difference was higher force coefficients in the floating body case than in the double-body case. This was likely caused by the restrictive free-slip free surface boundary in the floating body case. A grid sensitivity study for the double-body RANS case showed that the applied grid is fine enough for this application.
Due to limited time and computational resources, the LES case could not be run long enough to achieve satisfying statistical convergence. Therefore, the comparison between RANS and LES in this project has some shortcomings. It was, however possible to draw some conclusions from it. LES is clearly better at capturing the vortex shedding, and maintaining it further downstream. The RANS simulations also captured vortex shedding, but not as accurately. Another result of the comparison was that RANS seemed to under predict the drag coefficient.