A.STASSINOPOULOS, P. ETIENNE, A. MURTHY

MM.JP. CHERET, WILLAIME, THIBEDORE

For need of French National Society for Railway, an interface has been developed to generate structured mesh around train from CAD file. Calculation is then carried out using PHOENICS 1.5. Predicted flow structure is quite good and comparison with experimental data is in good accordance.

Aerodynamic problems in railway industry increase as velocity of trains increases. SNCF (French National Society for Railway) is particularly concerned in such problems, using TGV (high speed train ) running at 300 km/h in commercial manner.

Used with experimental work, SNCF decided to test possibilities of a fluid dynamics numerical code. The object of the study is the new Transmache TGV, characterized by a step in front of the screen glass. Numerical work was achieved by TRANSOFT AND PHOENICS.

This paper reports the way followed to get the geometry and how the mesh was built. Then computational results are exposed and finally compared with experimental data get by SNCF in a wind tunnel.

An important feature here is to get the complex geometry of the train. The geometry was available in an IGES file created by a CAD code, CATIA (1), for structural analysis purposes. In this file, the surface of the train is described by a set of several hundred of Bezier's surfaces of order

3. The train was plotted on a VDU by way of DISPLAY (fig. 1), the NASA II preprocessor.(2) Then capabilities of DISPLAY were used to set gird nodes on the 3D surface of TGV, and arranged in such a way that a structured mesh could lie on these nodes. Finally, DISPLAY wrote number and coordinates of each node in a "neutral file" ( ASCII file ).

The second stage of the work was to write an interface between DISPLAY and PHOENICS. The interface read the neutral file and ask the user for some extra data :

- length in front of nose of TGV,

- height above top of TGV,

- number of cells front of TGV,

- number of cells above TGV, etc.

Then, the interface builds and writes a XYZ file. The last stage is done in SATELLITE, the PHOENICS preprocessor. The XYZ file written by the interface is read, using the READCO command in Q1 file. Then extra arrangements are done, including SETLIN, MAGIC(T) AND MAGIC(L). This permits the user to do some changes in the mesh, based for example on the results of a previous calculation.

It is to be noticed that the specific point of the mesh generation is the interface DISPLAY /PHOENICS. This interface was developed by TRANSOFT for this study, but is now used successfully for other ones.

From a fluid dynamics point of view, the aims of this study were the general structure of the flow and the effect of the step on the main flow. No special attention was made on wake, boggies or wheels. So the mesh used in the calculation is stopped in the middle of a wagon and the train lies directly on the ground.

The mesh is composed of 95 K planes disposed along the axis. 55 J planes disposed circumferentially and 27 I planes disposed radialy (fig. 2, 3, 4). The symmetry plane is used to reduce the grid dimensions.

15 cells are set in front of TGV, with sizes drastically decreasing so that boundary layer can be reasonably described. The length in front of the nose is 30 m. In the same manner, cells in other directions have rapidly decreasing sizes close to the surface. With this, the total number of cells is only 141,075.

The most difficult point is of course the region of the step. Despite a truly BFC grid such used in PHOENICS calculations can follow the geometry, 5 rows of cells in I direction are used, which enter the nose and are blocked (fig 5). This permits to avoid highly non orthogonal cells at the sharp edges, and therefore is very benefic for calculation purposes. Unfortunately, this kind of gird gives triangular cells at the end of the step. It will be see later that this prove not to cause trouble.

The flow field is determined by solving continuity equation and momentum equations. Of course, the flow considered may be turbulent so the viscosity used in Navier-Stokes equations is the effective one.

The turbulent viscosity is here calculated with K - e model. This needs to solve two extra transport equations for turbulent kinetic energy and for its dissipation rate. The K-e model modified by Chen and Kim (3) is used for the purpose of a better prediction of separation and re-attachment points.

Finally, the set of six transport equations is solved by PHOENICS, based on finite domain method and SIMPLEST algorithm.

At the inlet plane, a mass-flow corresponding to an upwind velocity of 50 m/s is specified. Although the Mach number is low here, the air is treated as compressible. The density is calculated from ideal gas law with constant temperature. This feature was used because it is though to improve convergence and also in prevision of future calculations (TGV has world speed record at more than 500 km/h ).

At the exit plane, a uniform pressure is specified. The same pressure is ascribe at the free stream plane.

On the ground and on the surface of the train, sources of momentum are set by way of generalized wall function to simulate friction. Train as well as ground are supposed to be at rest as in wind tunnel conditions.

The computation work has been carried out on IBM RS/6000 work station. 200 sweeps were done. Time per sweep on this machine was 15 mn. approximately. This explains because core memory needed (80 Mb) was not available.

On figure 6. Velocity and pressure in symmetry plane are plotted. Stagnation pressure is fairly well predicted in the region of the nose. Otherwise, an accelerated zone with low pressure stands on the roof. After this zone, a boundary layer of increasing depth develops.

On figure 7, lateral vortices at each side of TGV are shown. They correspond to perturbed zones going up and increasing in size from nose to rear of TGV. These perturbations occur from the front part of TGV, as can be seen in figure 8 showing velocity close to the surface.

Due to the step a vortex can be seen in the symmetry plane (fig. 9). Low pressures are associated with it, by contrast with higher pressures on the screen glass. It should be mentioned that only five cells are used to depict the step and that cells have a big length in the axial direction compared to the step's height. Therefore, it is not thought that accuracy of the predicted flow is very high.

Pressure close to the surface in the region of the step is plotted in figure 10. The view point is the same as in figure 5 showing triangular cells. It can be seen that no disturbance appears due to these cells (The isopressure lines are not solid lines at the end of the step only because of post-processing matters). Accordingly, the flow field (figure 11) is quite well calculated.

Experimental work has been done by SNCF and results were transmitted to TRANSOFT after calculation was achieved.

Computed pressure field on the surface of TGV (figure 12) is compared to experimental data by way of pressure coefficient Cp :

Cp =  P - P ¥
         1/2 r ¥ V ¥ 2

Special care was taken to values lying between -.5 and -.1 (figure 13). These regions correspond to accelerated zones and are of special interest in aerodynamic view point since separation may occur there.

In figure 13, two regions are to be distinguished : a lower region and a upper region.

In the lower region, experimental measures gave values lying between -.4 (letter D) and -.2 (letter B), while calculated values are not far from -.1. This Poor Agreement is to be explained by the fact that in the simulation, TGV has no wheel and lies directly on the ground. Therefore in this region affected by the boundary layer developing on the ground, results are not in good accordance.

In the upper region, this reason does not prevail and it is hoped agreement will be better. Indeed it is as well on the roof as on lateral side. All measured values from -.1 (letter A) to -.5 (letter E) are quite well located on the surface to TGV.

High speed trains such as TGV need now careful aerodynamic studies. In addition to usual experimental work in wind tunnel, numerical flow simulation can be of great help for this.

For French National Society for Railway, TRANSOFT has developed a tool able to give a mesh for calculation of the external flow from CAD file describing the surface of TGV. With this, the exact geometry is taken into account.

In spite of the dimensions of the calculation domain, the mesh has only 140,000 cells. A great care was taken to reduce size of the cells near the surface of TGV and to the region of the step in front of the screen glass. Nevertheless, this region need triangular cells.

For the calculation, the first positive point is that no trouble occurred from these cells. Second, structure of the flow seems to be quite well predicted, with accelerated zone on the roof, lateral vortices, etc.

Finally results of computation compared to experimental data give no good agreement in the lower part of TGV because the ground was not well located in the simulation. Elsewhere, pressure coefficients lying from -.1 to -.5 are quite well predicted, corresponding to zones where the fluid is accelerated.

REFERENCES :

(1) CATIA. from DASSAULT SYSTEMES.

(2) NISA II, from E.M.R.C.

(3) CHEN Y.S., KIM S.W.

"Computation of turbulent flows using an extended K-e turbulence model"

NASA CR - 17920, USA ( 1987 )