Kunihide Ohashi

Numerical simulation of the free-running of a ship using the propeller model and dynamic overset grid method

ABSTRACT An unsteady Reynolds averaged Navier-Stokes (URANS) solver to estimate the trajectory on the free-running condition of a conventional ship is developed. Ship motions are obtained by solving the motion equations and accounted for by the moving grid technique. Propeller effects are accounted for by the body forces that are derived from the propeller model, which is based on the potential theory. The prescribed rudder motions of typical free-running conditions are accounted for using the dynamic overset grid method, in which the overset information is updated at each temporal step by implementing the existing overset grid method as the numerical library. The flow around the ship hull during the turning motions is analysed, and strong interactions between the ship hull and rudder in the propeller accelerated flows are observed. Through comparisons, the present method shows its applicability to compute the flow around a ship in free-running motion.

LARGE-SCALE FREE-SURFACE FLOW SIMULATION USING LATTICE BOLTZMANN METHOD ON MULTI-GPU CLUSTERS

Turbulent free-surface flows around ship strongly affect maneuverability and safety. In order to understand the details of the turbulent flow and surface deformation, it is necessary to carry out high-order accurate and large-scale CFD simulations. We have developed a CFD code based on LBM (Lattice Boltzmann Method) with a single-phase free-surface model. Since violent flows are turbulent with high Reynolds number, a LES (Large-Eddy Simulation) model has to be introduced to solve the LBM equation. The coherent-structure Smagorinsky model is a state-of-the-art sub-grid scale model. Since this model is able to determine the model constant locally, it is suitable for a large-scale calculation containing complicated solid bodies. Our code is written in CUDA and MPI. The GPU kernel function is tuned to achieve high performance on the TSUBAME 2.5 supercomputer at Tokyo Institute of Technology. We obtained good scalability in weak scaling test. Each GPU handles a domain of 192 × 192 × 192, and 27 components are defined at a grid by the D3Q27 model. The fairly high performance of 809 MLUPS(Mega lattice update per second) is achieved by using 1000 GPUs in single precision. By executing this high-performance computation, turbulent violent flow simulation with real ship data is performed, and details of turbulent flows and freesurface deformations will be simulated with much higher accuracy than ever before.

PERFORMANCE IMPROVEMENT OF FLOW COMPUTATIONS WITH AN OVERSET-GRID METHOD INCLUDING BODY MOTIONS USING A FULL MULTIGRID METHOD

A full multigrid method with an overset-grid method is developed to get a fast convergence. The weight values for an overstet-grid interpolation of flow variables can be determined on the fine grid level of multigrid, however, the weight value might be difficult to determine on coarser grid level. Therefore, the receptor cell value on coarser grid level is set and kept by the interpolation and correction of the multigrid method. Then, the flow variables on the coarse grid is interpolated to the fine grid as initial flows of the full multigrid method. The present method is applied to steady and unsteady simulations. The elapased time of steady simulations with and without free surafce is reduced about 35% of the time without the full multigrid method. The present method also succeeds to reduce the computational time remarkably on the unsteady simulation.

Flow Comparisons of DES, DDES and URANS for a Circular Cylinder

Abstract Detached eddy simulation (DES), delayed DES (DDES) and unsteady Reynolds averaged Navier-Stokes (URANS) simulations are numerically investigated for a circular cylinder. Both DES and DDES improve results for drag and Strouhal number for the cylinder flow compared to experiments, reproducing the complex flow structures behind the cylinder well. The pressure distributions on the cylinder are compared with measured data, demonstrating the effectiveness of DES and DDES for highly separating flows.