Computational & Technology Resources
an online resource for computational,
engineering & technology publications
PROCEEDINGS OF THE SIXTH INTERNATIONAL CONFERENCE ON PARALLEL, DISTRIBUTED, GPU AND CLOUD COMPUTING FOR ENGINEERING
Edited by: P. Iványi and B.H.V. Topping
Simulating train-tunnel aerodynamics with a parallel adaptive Cartesian method
R. Deiterding and J.M. Garro Fernandez
School of Engineering, University of Southampton, United Kingdom
R. Deiterding, J.M. Garro Fernandez, "Simulating train-tunnel aerodynamics with a parallel adaptive Cartesian method", in P. Iványi, B.H.V. Topping, (Editors), "Proceedings of the Sixth International Conference on Parallel, Distributed, GPU and Cloud Computing for Engineering", Civil-Comp Press, Stirlingshire, UK, Paper 9, 2019. doi:10.4203/ccp.112.9
Keywords: aerodynamics simulation, tunnel boom, Cartesian method, fluid-structure coupling.
As velocities of high speed trains increase, the loads created by pressure waves ahead and after a train have become an important design criterion. When two trains intersect, significant side forces are induced; the entry into a tunnel leads to a Mach compression wave that can eventually steepen into a shock wave (also known as “tunnel boom”). Predictive aerodynamic simulation of these transient phenomena requires an efficient approach to simulate the air flows around vehicles that move through a geometrically environment. While the use of body conforming meshes involves either expensive global regridding operations or complex interpolation schemes, which are both significant scalability obstacles, we have opted to apply a Cartesian embedded boundary method as implemented in our AMROC framework.
In AMROC geometrically complex objects represented by triangular surface meshes are represented on a Cartesian grid by a level set function that is re-evaluated in every time step and stores the signed distance to the nearest surface triangle. Based on this distance function, it is straightforward to implement moving wall boundary conditions in those near-surface cells that are inside the bodies. Subsequently, a time-explicit second-order-accurate Cartesian shockcapturing finite volume method is applied to simulate the flow field. Since forces and waves at the head of trains are pressure drag driven, the inviscid Euler equations are solved. In order to mitigate approximation errors by the use of Cartesian meshes, AMROC provides block-based structured mesh adaptation with recursive refinement in space and time. Refined cells follow the moving bodies as represented in the level functions as well as essential flow features, e.g., strong pressure gradients. AMROC is fully parallelised for distributed memory machines using MPI. A space filling curve is used for efficient re-partitioning and workload balancing as the hierarchical mesh evolves at run-time.
As a simplified validation configuration, experimental results from the 1:250 train-tunnelsimulator by Zonglin et al. (2002) - basically a cylinder fired into a cylindrical tunnel section with up 100m/s - are found to be in excellent agreement to our calculations. Full-scale 3D computations are carried out for a symmetric train head using the Next Generation Train 2 (NGT2) geometry by the German Aerospace Center (DLR). This train head is studied during high speed tunnel entry, when intersecting with another NGT2 train head in the open and inside a realistic double track tunnel, including previous tunnel entry. The latter computations were run on up to 96 cores of Intel 2.6 GHz Intel Sandybridge processors with 3 additional levels of dynamic refinement and used approximately 2300h CPU. Ease of the problem setup as well as scalable, quasi-automatic execution demonstrate the relevance of our approach as an efficient novel computational tool for the investigation of transient train aerodynamics.
purchase the full-text of this paper (price £22)