Journal of Fluids and Structures 44 (2014) 31 51

Contents lists available at ScienceDirect

Journal of Fluids and Structures

journal homepage: www.elsevier.com/locate/jfs

A study of the aerodynamics of a generic container freight wagon using Large-Eddy Simulation
Jan sth n, Sinia Krajnovi
Division of Fluid Dynamics, Department of Applied Mechanics, Chalmers University of Technology, SE-412 96 Gteborg, Sweden

a r t i c l e i n f o
Article history: Received 27 June 2012 Accepted 12 September 2013 Available online 29 October 2013 Keywords: Train aerodynamics Freight trains Container CFD Large Eddy Simulation Separated flow

abstract
In this work simulations using the Large Eddy Simulation technique have been made of the flow around a generic container freight wagon model. The model consists of one 11.8 m standard length container placed on a wagon. Details of the undercarriage such as wheels are included, but the container is generic and smoothed in comparison to a real freight wagon. The Reynolds number of the flow is 105 based on the container width of 2.354 m. Two cases have been considered in the study, one case where the wagon is standing alone and one case where it is submerged into a train set with wagons ahead and behind the wagon. The latter case is simulated using periodic boundary condition. Both the time-averaged and the instantaneous flow around the wagon for the two cases are described. For the single wagon case, it is found that the separation bubble formed on the roof of the container oscillates back and forth in the streamwise direction and that this oscillation is in phase with oscillations found in the upper shear layer of the ring vortex in the wake. The mechanism that is causing the synchronization of the oscillations of the separation bubble at the front and the upper shear layers in the wake is found to be waves of vorticity being shed from the separation bubble. The time-averaged ring vortex in the near wake of the single wagon is found to be inclined due to the disturbance of the undercarriage details on flow in the lower shear layer. The lower center of the ring vortex is located closer to the base face than the upper center. The drag coefficient of the wagon in the periodic case was found to be only 10% of that of the single wagon case. This is due to two symmetrical counter-rotating vortices found in the gaps which make the train set appear as a single body to the oncoming flow and shielding the wagon from any direct impingement of the flow. The counter-rotating vortices in the gap are found to inhibit periodic oscillations in the lateral direction. These oscillations cause vortical structures to form by the air that is pushed out from the gap and these flow structures cause a dominating oscillation of non-dimensional frequency St 0.12 in the side force signal. & 2013 Elsevier Ltd. All rights reserved.

1. Introduction The sum of the resistive forces acting on trains in the direction of travel is usually expressed as a second order polynomial (e.g. Davis, 1926; Rochard and Schmid, 2000; Raghunathan et al., 2002; Lukaszewicz, 2006): F A B V C V 2: 1

Corresponding author. Tel.: 46 31 7721390. E-mail address: ojan@chalmers.se (J. sth). URL: http://www.tfd.chalmers.se/  sinisa (J. sth).

0889-9746/$ - see front matter & 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.jfluidstructs.2013.09.017

32

J. sth, S. Krajnovi / Journal of Fluids and Structures 44 (2014) 31 51

It is assumed in Eq. (1) that the train does not accelerate or decelerate that it travels on flat ground and that the railway is straight. Otherwise, terms for the forces needed to overcome the resistance of acceleration, the gravitational force and mechanical curving resistance must be included in Eq. (1). The term A on the right-hand side contains mechanical resistances that are constant with respect to the speed of train, V, but whose magnitude is dependent on the number of axles, axle loads, track type and the length of the train. The second term contains resistances that vary with the length of the train, such as friction, and are linearly proportional to the speed. The last term contains the aerodynamic resistance of the train, which is proportional to the square of the speed. The values of the coefficients in Eq. (1) are highly dependent on each specific type of train under consideration and must be determined individually for each specific train (Rochard and Schmid, 2000; Lukaszewicz, 2001, 2006). Empirical equations that can be used to estimate the resistance of a general train configuration have been developed. See Rochard and Schmid (2000) for a review of empirical equations used to estimate resistances of trains and Tomasini and Cheli (2013) for a estimation of crosswind loads. A freight train normally consists of a large amount of wagons of different sizes, shapes and purposes. For a freight train, the coefficient C in Eq. (1) is the sum of the contribution to the aerodynamic drag from the locomotive and all wagons in the train. The size of the contribution to C of each wagon depends on the position of the wagon in the train (Engdahl, 1987) as well as on the spacing between the wagons (e.g. Hoerner, 1965; Engdahl et al., 1986). The operational speed of freight trains is typically much lower than speeds of passenger trains. Typical speeds for freight trains range from 50 to 130 km/h which correspond to Reynolds numbers between 2 106 and 6 106 based on the width of a standard cargo container. Fig. 1 shows the drag coefficient of a closed top gondola-type freight wagon depending on the position in the train. Results are shown for 01, 51 and 101 of yaw angle. It is seen in the figure that, after the initial 34 wagons, the drag coefficient will reach values that are some 2050% less than the drag coefficient of the first wagon. Thus, the majority of all the wagons in a freight train will experience an aerodynamic drag force slightly lower than that experienced by the first wagon in the train. The contribution to the total drag of the entire train from the locomotive will in turn be higher than the drag of the first wagon due to the stagnation pressure of the air on the front of the locomotive and the low pressure region in the wake of the locomotive. The low pressure in the wake of the locomotive will contribute to a smaller force in the axial direction on the first wagon. This effect is the same as the one on rectangular bluff bodies placed in tandem, e.g. Browand and Hammache (2004), and what is know as the platooning effect in road vehicle aerodynamics (Browand and Hammache, 1999). Platooning refers to when three or more vehicles are traveling in a convoy with small distances between the vehicles. The vehicles in the convoy having vehicles both upstream and downstream will experience dramatically reduced drag coefficient as compared to traveling alone. A study reported in Johansen (1936) suggests that the drag coefficient of the locomotive is in the order of 4 times of the drag coefficient of the first wagon. The order of this value is confirmed in the wind tunnel tests reported in Gielow and Furlong (1988), where the drag for two different types of locomotives was measured to be around 0.81.2. They based the drag coefficient on the units of drag area (ft2). Recomputing of the value of the first reference wagon in Fig. 1 using the same cross-sectional area as the one in the present study leads to the value of 0.3. The total drag of the generic configuration in Fig. 1 at zero yaw with one locomotive with drag coefficient 0.8 and the 11 wagons is around 3.7 as indicated in the figure when recomputing the values using the same cross-sectional area as in the present study. The total drag coefficient of a freight train consisting of a locomotive and 70 wagons of various sizes and shapes is estimated in Hoerner (1965) to be around 18. The distance between the wagons in a freight train significantly affects the drag coefficient of each wagon. Fig. 2 shows the relative change in drag of a freight wagon depending on the distance to the adjacent wagons. The data in the plot are taken from the extensive study on freight wagon aerodynamics reported by Paul et al. (2007). The data are a collection of

Fig. 1. The top figure shows a generic container freight train with a locomotive and 11 single stack wagons. The bottom figure shows the drag coefficient (y-axis) for a closed top gondola-type freight wagon depending on the position in the train (x-axis). All the values are normalized with the value of the drag coefficient of the wagon in the first position at zero yaw angle. The zeroth position corresponds to the locomotive. The results are redrawn from Fig. 54 in the report of the full-scale experimental study (Gielow and Furlong, 1988). The data for position 10 were left out in the study and are therefore missing in the above figure likewise.

J. sth, S. Krajnovi / Journal of Fluids and Structures 44 (2014) 31 51

33

10 5

% Change in Drag Coefficient

0 -5 -10 -15 -20 -25 -30 -35 0.5 1 1.5 2

Gap distance (m)

Fig. 2. The effect on the drag coefficient of one wagon (full-scale) when changing the distance between that wagon and adjacent cars. The baseline gap distance is 1.56 m. The figure is redrawn from data in Fig. 7 in Paul et al. (2007).

measurements from full-scale measurements, wind tunnel experiments and numerical simulations and should be understood as an indication of how the spacing affects the drag of a wagon. Fig. 2 shows that the drag of one wagon can be decreased by up to 30% by changing the inter-wagon gap from the baseline value of 1.6 m to 0.5 m (full-scale). As the gap size is decreased, the wagons behave more like a single body where the separating shear layers from the upstream wagon smoothly reattach on the roof and sides of the following wagon. This prevents high-momentum air around the wagons from impinging on the front face of the following wagon, and thus the flow in the gap is recirculating and does not interact with the surrounding air. As the gap size increases, the shear layers start to interact with the recirculating flow in the gap and more momentum is transferred from the surrounding air to the wagon and the drag thus increases. Cargo containers carried by freight trains are normally constructed with ribs on the sides of the container. Studies (Hoerner, 1965; Hucho, 1998) have indicated that the drag of the freight wagon can be reduced by as much as 15% by smoothing the sides of containers. A further strategy to decrease the total energy/fuel consumption of freight trains that has been developed is to use optimization algorithms to load the freight train as optimally as possible (Lai and Barkan, 2005; Lai et al., 2008) from an aerodynamic point of view, avoiding large spacings between wagons and wagons without any containers at all. An additional complexity concerning the operation of trains is the presence of tunnels along the tracks. Pressure variations in tunnels increase the drag of the train significantly in comparison to open air (Sakuma et al., 2008a) and can cause considerable ear discomfort for passengers in the train and riding discomfort due to large dynamic oscillation of the train (Sakuma et al., 2008b). Pressure waves radiate to the environment from the tunnel exit. The increased driving resistance inside tunnels increases the mechanical stress on the train. Some experimental wind tunnel studies on different types of freight wagons are reported in the literature. Wind tunnel simulations were made in Watkins et al. (1992) on 1:10 scale wagons. The objective of the study was to determine how a typical wagon in a train could be simulated in the wind tunnel. Typical means that the wagon should not experience effects in the measured global force quantities from the front or the rear of the train. In the study it was found that, for a wagon not to experience effects in the measured drag coefficient from the locomotive and/or the end of the train, one and a half dummy wagons were needed ahead of the wagon being studied. One half dummy wagon was needed downstream. In Saunders et al. (1993) measurements were made in a wind tunnel on two open top rail type wagons. It was found that, by dividing the cargo spacing into smaller subspaces, the drag of each wagon was reduced when the train was operating with empty wagons. The lateral stability of a single double-stacked container wagon subjected to crosswind was studied in Alam and Watkins (2007). They also studied Reynolds number dependence of the aerodynamic drag force when the wagon was subjected to a wind of zero yaw angle. A major finding was that for a Reynolds number larger than 0.8 105 based on the height from the ground of the double-stacked wagon, the measured drag coefficient showed little dependence on the Reynolds number for zero yaw. In the work reported by Paul et al. (2007) several aerodynamic issues of freight trains and locomotives were addressed by the authors using steady Reynolds Averaged NavierStokes (RANS) simulations. The issues addressed were drag characteristics of different wagons and train configurations, turn-over risk assessment and behavior of the diesel exhaust around the locomotive. A numerical simulation of the unsteady flow around a simplified freight wagon subjected to a crosswind at 901 yaw angle was reported in Hemida and Baker (2010). LES with the standard Smagorinsky model was used in the simulation, and the Reynolds number was 3 105 based on the height of the freight wagon from the ground and the crosswind speed. The study reported in Hemida and Baker (2010) is the only numerical study that uses unsteady simulation techniques to simulate the flow around any freight wagon model known to the authors of the present study.

34

J. sth, S. Krajnovi / Journal of Fluids and Structures 44 (2014) 31 51

The purpose of the present study is to explore the flow around a single stack container freight wagon. Two cases are investigated: one single standing wagon and one wagon submerged into an infinite long train set with one wagon ahead and one wagon behind. The method used to numerically simulate the unsteady flow around the wagons is LES with the standard Smagorinsky model. This method is the same as used in studies of flows around simplified train models in Hemida et al. (2005, 2010), Hemida and Krajnovi (2008, 2010), and Hemida and Baker (2010). Periodic boundary condition is used to simulate the flow around the wagon submerged in the infinite long train. Section 2 describes the geometry of the simplified freight wagon model. Section 3 presents the governing equations and the employed numerical technique to solve the equations. The results from the simulations are presented and discussed in Section 4 and the paper is concluded in Section 5. 2. Description of the freight wagon model In this section we present the wagon model that we have used in the present work. As mentioned in the Introduction, we have considered two cases, one single standing wagon and one case where the wagon is a part of an infinite long train. For the latter case we have used the same geometric model for the wagons ahead and behind the middle wagon. A three-dimensional view of the model is shown in Fig. 3 and the model dimensions are presented in Fig. 4. The model represents a single-stacked container freight wagon. The width of the container is W 2.354 m and the length is 5W. A view of the front of the model is shown in Fig. 4(a) including some dimensions of the model. The total height of the model from the ground to the roof of the container is 1.6W. The origin is located 0.56W above ground. The width between the inner edges of the wheels is 0.63W, corresponding to the Normal Gauge of the width between the rails. The height from the ground to the container is 0.49W. The origo of the coordinate system used in the present work is located on the front of the wagon as shown in Fig. 4(a). The x-axis corresponds to the axis parallel but opposite to the direction of travel of the wagon. This direction is called the streamwise direction. The z-axis corresponds to the axis parallel with the direction of action of the gravitational force but in the opposite direction to it. This axis is denoted the transversal direction to the flow. The y-axis is the axis perpendicular to both the x- and z-axes and corresponds to the spanwise direction of the flow. Fig. 4(a) and (b) shows the freight wagon model from a side view (top) and a view from below (bottom), respectively. The radius of the wheels is 0.24W. 3. Numerical set-up In this section the governing equations and the numerical techniques and details that we have used to solve the instantaneous flow around the wagon models are presented. 3.1. Governing equations The filtered continuity and momentum equations read u i 0; xi u i 1 p 2 u i ij uu : xi t xj i j xj xj xj 2 3

The influence of the smallest scales (the sub-grid scales) on the resolved ones appears in the sub-grid stress tensor on the right side of Eq. (3), ij ui uj u i u j . The Standard Smagorinsky model (Smagorinsky, 1963; Pope, 2000) is used in the present work to the model sub-grid stress tensor as 2 ij 2sgs S ij ij ksgs ; 3 4

Fig. 3. The container wagon model used in the present work.

J. sth, S. Krajnovi / Journal of Fluids and Structures 44 (2014) 31 51

35

Fig. 4. Geometry of the wagon.

where sgs C s f 2 jS j is the SGS viscosity and the characteristic length is taken to be x y z 1=3 in the present work, where i are the local computational cell sizes in the three coordinate directions. The kinetic energy of the sub-grid scales, ksgs , is ksgs 1 2kk . S ij   1 u i u j 2 xj xi 5

is the resolved rate-of-strain tensor and jS j 2S ij S ij 1=2 is the magnitude of the resolved rate-of-strain tensor. The value of the Smagorinsky constant CS 0.1 previously used in studies of similar types of unsteady flows around simplified trains and bluff bodies (Hemida and Krajnovi, 2008, 2010; Krajnovi, 2009; Cheng et al., 2012) and for other applications (Breuer et al., 2012; Lam et al., 2011, 2012) is used in the present work. f in the expression for the SGS viscosity is the van Driest damping function   n f 1 exp ; 25 6

where n is the wall normal distance in viscous units and exp is the exponential function. An alternative to using the Standard Smagorinsky model is to use a dynamical model where the value of the Smagorinsky constant CS is not set fixed, but is computed from the flow variables each time step. The dynamical model of Germano et al. (1991) and similar offsprings are not suitable, however, because for such complex geometries as in the present study there is no homogeneous direction that could be used for averaging when calculating the value of CS. Local averaging in a volume of neighboring cells is not a good solution either since this could lead to negative values of the computed Smagorinsky constant. A dynamical model that combines more accurate physical modeling than the Standard Smagorinsky model while not being much more computationally costly is the Coherent Structure Model (LES-CSM) proposed by Kobayashi (2005). LES-CSM is based on an eddy-viscosity representation of the sub-grid stress as the Standard Smagorinsky model (Eq. (4)) but the constant CS is modeled by C S C 1 jF CS j3=2 F with F CS Q =E , F 1 F CS and C1 1/22. Q is the second invariant of the filtered velocity 2 1 gradient tensor, Q 1 2 u i u j =xi xj and E 2u j =xi is the magnitude of the filtered velocity gradient tensor. The LESCSM model was used by Han and Krajnovi (2012) to simulate the flow around a D-shaped two-dimensional body with a passive device to control to vortex shedding from the body. In Krajnovi and Lrusson (2012) the LES-CSM model was used to simulate the complex flow around a pyramid.

36

J. sth, S. Krajnovi / Journal of Fluids and Structures 44 (2014) 31 51

Fig. 5. (a) The blocking structure around the wheels and the front of the wagon used to construct the computational grid. (b) A cut of the grid at y 0 showing the front of the wagon.

3.2. Numerical method Eqs. (2) and (3) are discretized using the commercial finite volume solver AVL Fire (AVL, 2010). The discretization is done using a colocated grid arrangement where all the variables that are solved for are stored in the same grid and position, in opposition to staggered arrangements where different grids are used for different variables. The convective fluxes are approximated by a blend of 95% linear interpolation of second-order accuracy (Central Differencing Scheme) and of 5% upwind differences of first-order accuracy (Upwind Scheme). The diffusive terms containing viscous plus sub-grid terms are approximated by a central differencing interpolation of second-order accuracy. The time marching procedure is done using the implicit second-order accurate three-time level scheme:   d 3n 4n 1 n 2 ; t n t t n 1 t n 1 t n 2 ; 7 dt n 2 t n where indices n and n 1 denote the new and old time levels, respectively. To determine the pressure, the discrete form of the filtered continuity equation (Eq. (2)) is converted into an equation for the pressure correction which is solved by using the SIMPLE algorithm (Patankar and Spalding, 1972). 3.3. Computational grid Three computational grids were used in the present work: one coarse and one fine grid for the single wagon case and one fine grid for the periodic wagon case. The fine and coarse grids for the single wagon case are used to quantify the dependency of the results on the grid. Only one fine grid was used in the periodic wagon case due to the high computational costs of the simulations. The grids are block-structured hexahedral and were constructed using the grid generator software Ansys ICEM CFD. The coarse grid contains nine million cells and the fine grid contains 17 millions and 15 millions for the single and periodic wagon cases, respectively. In Fig. 5(a) the blocking structure around the front of the wagon is shown for the fine grid. The whole wagon is enclosed in an O-type blocking structure, and the rest of the domain consists of H-type blocks. The number of grid points on the edge in the normal direction of the O-grid is 80. This large number of grid points in the normal direction of the O-grid is required to resolve the geometrical details of the undercarriage. The blocking structure around the wagon is symmetric with respect to the yz plane and the xz plane. The distribution of cells is not symmetric in the xz plane as more cells are located behind the wagon than in front of it in order to accurately resolve the flow structures in the wake. The grid is shown in Fig. 5(b) around the front of the wagon and the undercarriage in a cut of the grid at y 0. It should be noted that, for complex geometries using non-equidistant computational grids as in the present work, a commutation error is introduced since the formulation of filtered equations (Eqs. (2) and (3)) assumes that the filtering operation and spatial derivation commute. This is, however, not true for nonequidistant grids. The commutation error introduced is of the same order as the truncation error of the second-order linear interpolation scheme (e.g. Ghosal and Moin, 1995). Fig. 6 shows the grid on the surface of the wagon. Hyperbolic stretching of the grid is used in order to concentrate grid points near the walls. Thus, the sizes of the cells are non-uniformly distributed and the cells may also be skewed. 3.4. Computational domain and boundary conditions The computational domains for the two cases are presented in Fig. 7. The cross sectional area is 7.1W 6.1W giving a blockage area in the numerical wind tunnel of 3.7%. The distance from the inlet to the front of the wagon is 8W for the single wagon case. The distance from the aft of the wagon to the outlet is 20W. These distances are standard distances used in the

J. sth, S. Krajnovi / Journal of Fluids and Structures 44 (2014) 31 51

37

Fig. 6. Two figures showing the computational grid on the surface of the wagon.

Fig. 7. The computational domain around the wagon and the boundary conditions: (a) single wagon case, (b) periodic wagon case.

studies mentioned in the Introduction of similar types of vehicle bluff body flows (Krajnovi and Davidson, 2003, 2005; Krajnovi and sth, 2010; sth and Krajnovi, 2012). For the periodic case, one half wagon is placed upstream of the typical wagon and one half wagon is placed downstream. Periodic boundary condition is set that maps the outlet flow conditions to the inlet. Thus, the mid-wagon represents a wagon in an infinitely long train. For the single wagon case, uniform and in time constant velocity profile of U 1 1:1 m=s was used on the inlet. This corresponds to a Reynolds number of 105 based on the width of the wagon, inlet velocity U 1 and the kinematic viscosity of air at temperature 20 1C, 1:5 10 5 m2 =s. For the periodic wagon case the mass flow at the inlet was set fixed. On the lateral sides and top of the domain a symmetry condition was used (u =y w =y v 0 on the sides and u =z v =z w 0 on the top). The homogeneous Neumann condition was used (u i =xi 0 on the outlet in the single wagon case. The no-slip condition was used on the ground in conjunction with setting the velocity component in the streamwise direction equal to the free streaming velocity

38

J. sth, S. Krajnovi / Journal of Fluids and Structures 44 (2014) 31 51

(u U 1 ; v 0 and w 0). This is done in order to accurately simulate the effects on the flow due to the movement of the wagon over the ground (Krajnovic and Davidson, 2005). 4. Results This section presents the results from the three simulations. Two simulations are for the single wagon case where a coarse and a fine grid have been employed. For the periodic wagon case only the fine grid was used. The temporal and spatial resolution are presented first. Then the pressure coefficient on the wagon is compared for the two grids for the single wagon case in order to establish the degree of grid dependency of the results. The global force quantities are presented followed by the instantaneous and mean flow around the wagons for the two cases. All the results of the flow field are from the fine grid simulations if not stated otherwise. 4.1. Temporal and spatial resolution The flow field was averaged over 100 000 time steps in all three cases. This corresponds to 225 convective time units t n tU 1 =W (W is the width of the wagon). The size of the time step in physical time was 0.005 s and in convective time units t n tU 1 =W 0:00235. This time step size kept the CFL number below one in more than 99% of the cells all the time. It exceeded one only in a very small volume fraction of cells near the front edges of the wagon. The conceptual idea behind a successful LES simulation employing an explicit model for the sub-grid stress, as in the present work, is to use a computational grid that is fine enough to resolve all the motion down to the inertial subrange (Geurts, 2008; Pope, 2000). After carrying out an LES simulation, the spatial resolution obtained in the simulation must be evaluated in order to assess the quality of the results obtained. In attached boundary layer flow, the size of the grid cells adjacent to the wall must be small enough to resolve the low and high speed streaky structures in the viscous region of the boundary layer (y o 10). y y= is the wall-normal distance measured in the viscous length unit =un , where un is the wall friction velocity. The guidelines in the literature say that the size of the cells adjacent to the wall should be n 2 , x 100 and s 2030 in order to resolve the streaky structures (Davidson, 2009). n; x and s are the sizes of the cells in the normal, the streamwise and the spanwise direction, respectively. The mean spanwise spacing between low speed streaks is reported in experimental studies to be around 100120 and the spanwise width of the lowspeed streaks is reported to range from 33 to 100 . The mean length of the streaks is reported to be around 1000 but much shorter and longer streak lengths can be observed (Kline et al., 1967; Smith and Metzler, 1983; Zacksenhause et al., 2001). Fig. 8 shows low and high speed streaky structures in the viscous layer from the experimental study (Zacksenhause et al., 2001). For a comprehensive review and description of streaks and other flow structures occurring in the turbulent boundary layer and the present understanding of them, we refer the reader to Adrian (2007). In other flow situations than attached boundary layer flow, such as separated flow, there exist no general guidelines for how the resolution in the simulation should be evaluated (Davidson, 2010). In scientific studies where LES is used, some flow properties (e.g. forces, velocity profiles, Reynolds stress profiles, pressure distribution) are normally compared to experimental wind tunnel data in order to assess the quality of the results of the simulation. The present work is entirely a numerical investigation. To show that the computational grids provided a sufficiently fine spatial resolution in the simulations, the resolution is calculated after the simulation. On the whole wagon, the timeaveraged resolution obtained in the normal direction gives a location of the first grid points within 0 o t o 2 on the wagon. The subscript t refers to the time-averaged value. The spatial mean of t on the container of the wagon is 0.5 for both the fine and coarse grids for the single wagon case. The values are somewhat higher for the underbody of the wagon. A boundary layer is formed on the side and the roof of the container after the initial separation from the leading edges on the wagon. The largest cell size in the spanwise direction of the grid is s 12:5t for the fine grid and s 30t for the coarse grid. In the streamwise direction, the maximum resolution is x 30t for the fine grid and x 80t for the coarse grid. The largest values of x occur at the center of the wagon in the streamwise direction.

1400+

1800+

Fig. 8. Thermal image from the experimental study of spatial characteristics of near-wall streaks in boundary layers. Picture is taken from Zacksenhause et al. (2001). Dark regions correspond to high-velocity regions and bright regions are low-velocity regions.

J. sth, S. Krajnovi / Journal of Fluids and Structures 44 (2014) 31 51

39

Fig. 9. Planes that are used to plot time-averaged streamlines to visualize the flow around the wagon.

0.3

Fig. 10. A plane showing the instantaneous streamwise velocity component, u , on the roof from the fine grid simulation. The cut is taken at a distance of approximately 5 from the wall. The flow is from left to right. Dark regions correspond to high velocity regions and bright regions are low-velocity regions.

Fig. 11. Ratio of sgs-viscosity (sgs to molecular viscosity (mol ): (a) the coarse grid, (b) the fine grid.

One cut showing the instantaneous streamwise velocity component at approximately n 5 from the wall on the second half of the roof is presented in Fig. 10. Streaky structures of low (white regions) and high (dark regions) speed can be identified on the right part of the plane. The location of the plane is indicated in Fig. 9. The white areas of low speed fluid on the left part are due to the separated flow on the front of the roof. The value of the calculated mean viscous length, t , on the second half of the roof where the flow is attached is 0.0016. The size of these streaks visible in Fig. 10 is in the range of the sizes reported of streaks by, e.g. Kline et al. (1967) and Smith and Metzler (1983). The shape of the streaks is very similar in the simulations to the ones visualized in Fig. 8. The low-speed streak denoted by S1 in Fig. 10 has a length of approximately 1100 and a width of approximately 100 . Fig. 11 shows the ratio between the sgs-viscosity, sgs , and the molecular viscosity, mol , in the symmetry plane y 0 for the coarse grid and the fine grid. The values of the ratio are up to 10 in both simulations but it is evident that overall the ratio is less in the fine grid simulation than in the coarse grid simulation. The highest values of the sgs-viscosity occur in the shear layers of the separation bubble on the top. Close to the surface of the container the values are close to one or below.

40

J. sth, S. Krajnovi / Journal of Fluids and Structures 44 (2014) 31 51

4.2. Grid comparison of the pressure coefficient Fig. 12 presents the pressure coefficient defined as C p pt p1 =0:5U 1 ; 8

plotted along three lines on the surface of the wagon for the single wagon case. The lines are shown in Fig. 12(a). In Fig. 12 (b), Cp is plotted along line ly starting on the front face of the wagon and then goes along the y-symmetry line of the wagon in a counter clockwise manner. It is seen in the figure that the results from the coarse and fine grid simulations collapse on the rear, top and front faces of the container but a deviation is present along the undercarriage. Fig. 12(c) plots Cp along line lz in the xz plane cutting the wagon at half of the height of the wagon from the ground. Here, the results from the coarse and fine grid simulations collapse everywhere. Fig. 12(d) shows Cp along line lx which lies in the yz plane and cuts the wagon at half its length. Here, the results collapse on the side and the tops while once again a deviation is present along the undercarriage. 4.3. Global quantities The time-averaged forces from the simulations are presented in Table 1 for the single wagon case for both the fine grid and the coarse grid. The forces for the periodic wagon are presented in Table 2. The aerodynamic coefficients are defined as CD Fx ; 1 2 U 1 Ax 2 CS Fy ; 1 2 U 1 Ax 2 CL Fz : 1 2 U 1 Ax 2 9

1:18 kg=m3 is the constant density of air at 201 and Ax 1W 1:6W is the cross sectional area of the freight wagon projected onto the yz plane. Thus, the same cross sectional area is used to normalize all three force components. The timeaveraged value of the drag coefficient for the single wagon is 0.90 on the fine grid. In the fine grid simulation 0.23 (74%) of the drag comes from the container and the rest comes from the undercarriage details. For the coarse grid the value of the drag from the undercarriage is 0.21 which is a 9% difference from the value in the fine grid (0.23). The presence of this rather large discrepancy can be understood by the difference in values of the sgs-viscosity for the fine and coarse grids shown in Fig. 11. Near the front part of the undercarriage, the difference in the levels of SGS-viscosity is rather large between the two grids. This might affect the solution of the flow so that we obtain the discrepancy in the global quantities as observed in
Undercarriage
Rear Top Front

1.5 1 0.5
Coarse grid Fine grid

Cp

0 -0.5 -1 -1.5

10

12

14

ly/W

Front

Side

Rear

Side

Front

Undercarriage

Side

Top

Side

Undercarriage

1.5 1 0.5
Coarse grid Fine grid

0.1 0.05 0 -0.05

Cp
0 -0.5

Cp
-0.1 -0.15 -0.2 -1 0 2 4 6 8 10 12 -0.25 0
Coarse grid Fine grid

lz/W

lx/W

Fig. 12. Pressure coefficient, C p , plotted on three lines on the wagon comparing values from the fine and the coarse grid simulations: (a) the location of the profiles, (b) C p along line ly , (c) C p along line lz , (d) C p along line lx .

J. sth, S. Krajnovi / Journal of Fluids and Structures 44 (2014) 31 51

41

Table 1 Aerodynamic forces and root mean square (rms) values of the signals on the container wagon from the single wagon simulation. See text for definition of the force coefficients. The percentage shows difference between coarse grid results and fine grid results. Aerodynamic coefficients single wagon Whole wagon (fine) Container (fine) Undercarriage (fine) Whole wagon (coarse) Container (coarse) Undercarriage (coarse) CD 0.904 0.67 0.23 0.895 1% 0.68 1% 0.21 9% rms(CD) 0.013 0.016 19% CS 0.0006 0.0004 33% rms(CS) 0.047 0.058 23% CL 0.05 0.056 11% rms(CL) 0.029 0.035 21%

Table 2 Aerodynamic forces and root mean square (rms) values of the signals on the container wagon from the single wagon simulation. See text for definition of the force coefficients. Aerodynamic coefficients periodic Whole wagon (fine) Container (fine) Undercarriage (fine) CD 0.095 0.040 0.055 rms(CD) 0.0085 CS 0.0016 rms(CS) 0.042 CL 0.044 rms(CL) 0.018

Table 1. Since the grid size is incorporated in the SGS model there is no guarantee that convergence of results will ever be obtained with grid refinement. In fact, the only time we could demand exactly similar results from two different grids for complex geometries is if both grids provide a DNS-like resolution which would then be indicated by very low levels of sgsviscosity (sgs o 1). The drag coefficient of the single wagon that has been obtained in the present study can be compared to the value of the drag coefficient reported in other studies. For a similar type of wagon but with a smoother undercarriage, a drag coefficient of around 0.8 at a Reynolds number of 106 is reported in Hammit (1976). In Peters (1993), drag coefficients of around 0.9 are reported for Reynolds numbers based on length, ranging from 5 106 to 30 106, for wind tunnel measurements on a single 1:3 scale geometrically detailed container wagon model. Splitting up the contribution to the drag force in the present simulation into the friction part and the pressure part of the fluid force exerted on the wagon shows that only 3% of the total drag on the wagon can be attributed to the fluid friction force of the drag. This is due to the areas of separated flow around the front of the wagon, as will be seen in the figures showing the flow field in the coming subsections. The drag coefficient for the periodic wagon is 0.09, a value which is 90% less than that of the single wagon. This low value will be explained by inspection of the flow field in the coming subsections. Two counterrotating vortices are found in the gap that effectively carry over the gap the separating shear layers from the wagon upstream to the wagon downstream, thus making the train set appear to the flow as an elongated body. The value of 0.09 is significantly less than the values of E 0.2 (70% of 0.3) as was estimated in the Introduction from the full-scale study reported in Engdahl (1987). However, the closedtop container that was used for the full-scale measurements contained ribs on the sides of the container leading to flow separation and an increase in drag. They also tested to smooth the sides of the container when it was placed in both positions 8 and 12 in the train. The drag coefficient of the container was reduced from  0:22 to 0.09 for both positions, close to the value obtained for the periodic wagon in the present study. Thus, the present study confirms what was found in Engdahl (1987) that a major drag reduction can be made by smoothing the sides of the container on freight trains. For the periodic wagon, 0.05 of the drag coefficient comes from the underhood and 0.04 from the container. Thus, the high value of 0.23 (fine grid) for the single wagon case on the undercarriage can be attributed to the high pressure on the front due to the impinging flow. This high pressure is not present for the periodic case, and the drag drops significantly. Fig. 13 presents the time histories of the drag of side forces for both cases. The sizes of the oscillations in the drag forces are significantly smaller than the oscillations in the side forces for both cases. The standard deviation of the drag force is 0.013 in the single wagon case and 0.0085 for the periodic wagon case. The standard deviation of the side force is 0.047 and 0.042 for the single and periodic wagon cases, respectively. Power spectral densities of the drag and side force signal from the single wagon case reveal some frequencies but no dominating ones. The side force signal from the periodic wagon case reveals one dominating frequency as can be seen in the time-history of the signal in Fig. 13(b) and in the PSD of the signal in Fig. 14. The Strouhal number of the dominating frequency is St 0.12. It will be shown in the following sections where instantaneous flow is presented for the periodic case that this frequency stems from lateral oscillations of the quasi-stable vortices formed in the gap.

42

J. sth, S. Krajnovi / Journal of Fluids and Structures 44 (2014) 31 51

0.95

CD

0.9

0.1

CS

-0.1 0 10 20 30 40

t =Ut/L

CD

0.1

CS

-0.1

10

20

30

40

t =Ut/L
Fig. 13. Time signals for drag and side force coefficients. L in the dimensionless time tn is the length of the wagon: (a) the signal wagon case, (b) periodic wagon case.

1y

0.3

0.2

0.1

0.1

0.2

0.3

0.4

0.5

Fig. 14. Power Spectral Density of the time signals of the side force coefficients from the periodic wagon case.

4.4. The flow around the wagons In the preceding subsection the drag and side forces were presented. In this section the flow around the wagon for the two cases is presented and effort is put on trying to highlight the differences in the flow topology for the two cases that

J. sth, S. Krajnovi / Journal of Fluids and Structures 44 (2014) 31 51

43

cause the difference in the drag force. Three planes are used to visualize the flow by using streamlines of the time-averaged and the instantaneous velocity around the wagon. The locations of these planes are shown in Fig. 9.

4.4.1. y-Symmetry plane In this subsection the flow in the symmetry plane is presented for the single wagon case. Figures showing streamlines in the y 0 plane are not presented for the periodic wagon case as that would not contribute to further understanding than the figures that will be shown in z-cuts for the periodic case. The flow in the gap and the difference between the single wagon case and the periodic wagon case will be thoroughly investigated in the coming subsections. Fig. 15 presents streamlines in the symmetry plane y 0 for the single wagon case. Fig. 15(a) shows streamlines of the time-averaged velocity. The flow is from left to right. The background in all the figures presented in the subsection is colored by the velocity magnitude, where gray represents high velocity and white represents low velocity. It is seen that the flow impinging on the front face separates from the front edge, forming the time-averaged separation bubble called Vfront in the figure. This type of impinging leading edge separation but from quasi-two-dimensional elongated bluff bodies with blunt leading edges has been studied by researchers because of the induced vibrations that the separation cause (Parker and Welsh, 1983; Naudascher and Wang, 1993). The induced vibrations are closely connected to the shedding of vortices from the separation bubbles (Nakamura and Nakashima, 1986; Nakamura et al., 1991). Several studies have shown that vortices are shed from the separation bubble, but some have found that this shedding is not periodic (Parker and Welsh, 1983; Cherry et al., 1984) and others have found that shedding occurs at certain Strouhal numbers for low Reynolds numbers (e.g. Nakamura et al., 1991). In the study by Taylor et al. (2011) the flow around a two-dimensional elongated bluff body with a length-to-thickness ratio of 7 was investigated. The Reynolds number in their experiments based on the thickness (b) of the bluff body was 3 104. The length of the separation bubble for this quasi-two-dimensional case was found to be 4.44t. This value is close to the value of 4.42b that was found for an infinite long plate at similar Reynolds number by Cherry et al. (1984). One study concerning the size of the separation bubble on a blunt axi-symmetric, three-dimensional bluff body is reported by Rastgou and Saedodin (2013).

1.75W 0.35W 1 0.5 Vfront 0 Vwake

1 Vwake 1 Vfront 2 Vfront 2 Vwake

1 Vwake 1 Vfront 2 Vfront 2 Vwake

1 Vwake 2 Vfront 1 Vfront

2 Vwake

Fig. 15. Streamlines projected on the plane y 0 for the single wagon case: (a) streamlines of the time-averaged flow, (b) streamlines of the instantaneous velocity at one realization of the flow, (c) streamlines of the instantaneous velocity at another realization, (d) streamlines of the instantaneous velocity at a third realization. The time between the realizations is 0.45 convective time units (tn).

44

J. sth, S. Krajnovi / Journal of Fluids and Structures 44 (2014) 31 51

The length of the separation bubble is reported to be around 1.6 diameters of the axi-symmetric bluff body for this case. The lengths of the separation bubbles from the quasi-two-dimensional and axi-symmetric three-dimensional studies are interesting to compare to the length of the separation bubble on the roof and the sides of the container in the present study. The length of the separation bubble on the roof of the container was found to be 1.75W in the streamwise direction (see Fig. 15). The height of the container is 1.02W so it is relevant to compare the width W in the present geometry to the bluff body thickness b in the quasi-two-dimensional studies. The height of the separation bubble was found to be 0.35W in the present case. In the study by Taylor et al. (2011) the height of the separation bubble was found to be 0.44b. Although the Reynolds number in the present study is higher (105) than the Reynolds number in Taylor et al. (2011), it is evident that the length and the height of the separation bubble are significantly smaller as a result of the three-dimensional shape of the container. The flow in the near wake of two-dimensional bluff bodies is characterized by reversed flow, low pressure and flow instabilities. The near wake is bounded by the free shear layers which is formed by the boundary layer separating from the sides of the body. The most well-known instability formed in the wake behind quasi-two-dimensional bluff bodies is the periodic vortex shedding which consists of large scale coherent structures being shed at a regular frequency from alternating sides of the body. The coherent vortex shedding results in fluctuation of the drag and lift force of the body with the frequency of the shedding. The flow in the wake of three-dimensional bluff bodies is more complex due to the additional two shear layers emanating from the lateral sides of the body and more degrees of freedom. The flow behind threedimensional bluff bodies is often more irregular than the flow around two-dimensional bodies, but weak vortex shedding can sometimes be found behind three-dimensional bodies (Bearman, 1997). The time-averaged flow in the near wake of a three-dimensional bluff body consists of a ring vortex which is formed by the four separating shear layers. Although the time-averaged structure looks very simple, the instantaneous flow is characterized by large unsteadiness and the timeaveraged flow structure is made up of a substantial number of instantaneous flow structures that move randomly in time and space (Krajnovi and Davidson, 2003). Two instabilities that are responsible for the unsteadiness are instabilities in the separating shear layers and pumping of the wake (Baker, 2001). An additional factor that adds to the complexity is the presence of the ground plane in the proximity of one of the sides for bluff bodies representing simplified road vehicle models. Although the presence of the ground not necessarily changes the flow structures it affects the pressure distribution on the base face and thus affects the forces on the model (Krajnovic and Davidson, 2005). In Fig. 15(a) showing time-averaged streamlines in the symmetry plane y 0 the lower and upper parts of the ring vortex are seen. The lower part of the ring vortex is located considerably closer to the base face of the container than the upper part. This is not the case for vehicle bluff bodies such as those studied in, e.g. Krajnovi and Davidson (2003) and Wassen et al. (2010). In both the mentioned studies a simplified three-dimensional vehicle bluff body with a square back in the proximity of the ground has been studied and in both studies the streamwise location of the upper and lower parts of the ring vortex is approximately the same. However, both these bluff bodies from Krajnovi and Davidson (2003) and Wassen et al. (2010) had a smooth underhood. Thus, we can assume that the location of the lower part of the ring vortex Vwake in Fig. 15(a) found behind the container in the present study is due to the details of the undercarriage which has been included in the freight wagon model. The undercarriage disturbs the flow under the wagon so that the flow coming from beneath the wagon has less momentum than the air coming from the top of the container. The air with less momentum coming from underneath is not able to push the lower center of the ring vortex as far in the streamwise direction as the air coming from the top is. Fig. 15(b)(d) shows streamlines of the instantaneous velocity at three realizations. The time between the realizations is 2.25 convective time units (tn), corresponding to 1000 computational time steps. Comparing the position of the two centers of Vwake in the three instantaneous realizations, it is seen that the position of the lower center (V 2 wake ) is more or less stable. The position of the upper part (V 1 ) is, however, not stable but oscillates back and forth in the wake streamwise direction. This is the pumping effect of the shear layers. Pumping shear layers in the near wake was found in Krajnovi and Davidson (2003) as well. This pumping is caused by shedding of vorticity from the separation bubble on the roof at the front of the container. In the figures showing the instantaneous realizations it is seen that the front 2 separation bubble also is split into two parts, the front part, V 1 front , and the rear part, V front . The front part appears to be oscillating back and forth with a small amplitude (  0:1W ). The second part is oscillating as well but with a 2 considerably larger amplitude. The amplitude of V 2 front is around 0.25 0.4W. It appears as vorticity is shed from V front when it is at its rightmost position. The shedding of vorticity can be seen in Fig. 15(c) and (d). The oscillations of V 2 front and the upper part of the ring vortex in the wake, V 1 wake , are in phase, while the oscillations of the front part of the 2 separation bubble, V 1 front , and the rear part, V front , are 180 1 out of phase. The oscillations back and forth in the streamwise direction of the vortices forming the time-averaged separation bubble contributes to the large variation ( 7 50%) of the instantaneous reattachment point that has been observed in the separation from the leading edge for two-dimensional quasi-bluff bodies (Cherry et al., 1984).

4.4.2. Vortical structures Fig. 16 presents instantaneous flow structures around the single wagon in a view from behind. The structures are visualized by iso-surfaces of positive values of the second invariant of the velocity gradient tensor, Q. For incompressible

J. sth, S. Krajnovi / Journal of Fluids and Structures 44 (2014) 31 51

45

Fig. 16. Instantaneous flow structures (iso-surfaces of the Q-criteria) around the wagon viewed from behind the wagon. The iso-surfaces are colored by the pressure coefficient. The vector arrows in the figures are drawn proportional to the instantaneous velocity at the symmetry plane on the roof and in the wake.

flows, Q is defined as Q 1 1 2 S2 2 2  u i x i 2 u i u j xi xj

1u i u j ; 2 xi xj

10

where is the rate-of-rotation tensor and S is the rate-strain-tensor of the filtered velocity variables. Iso-surfaces of positive values of Q thus reveal locations in the flow where rotation dominates over strain and hence indicates vortical structures in the flow (Jeong and Hussain, 1995). Q is related to the pressure in the fluid by a Poisson equation: 2 p 2Q : 11

The iso-surfaces in Fig. 16 show areas of Q 12.5 at three consecutive time steps and vector arrows that are proportional to the instantaneous velocity. The time between the three pictures is t n 0:9. The iso-surfaces are colored by the pressure coefficient, CP. In the first instance in Fig. 16(a), it is seen how the flow separates from the leading edges on the roof and the sides forming the separation bubble which is filled with vortical structures and low pressure. From the separation bubble vortical structures are shed occasionally downstream. From the shear layers of the separation bubble waves of vorticity appear to be shed quite regularly, but no frequency of shedding was identified. Some waves are formed already at the first shear layers than they appear to be magnified by the impingement of the separated shear layers on the roof. This process can be seen in the snapshot in Fig. 16(a). One wave is being born at the attachment of the separation bubble. This wave is denoted W1. Further downstream in the same figure another wave denoted W2 is visible. Following the time evolution of these waves, they have moved downstream in Fig. 16(b) and in Fig. 16(c) the wave W2 has entered the wake while W1 has grown and been shed from the separation bubble and is propagating downstream. These waves synchronize the oscillations 1 of the second part of the separation bubble, V 2 front , and the upper part of the instantaneous ring vortex in the wake, V wake , that were found in the preceding section. 4.4.3. z-Planes Fig. 17 presents streamlines in the plane z 0.24W for the single wagon case. Along the sides, the separation bubble Vfront is formed. The bubble's size in both the streamwise and lateral directions is the same as the size of similar structure on the roof presented in the previous subsection. The instantaneous realizations in Fig. 17(b)(d) show that the structure of the instantaneous separation bubble does not consist of distinct parts as the separation bubble on the roof did. The separation bubble oscillates and the oscillations of bubbles on the two sides appear to be in phase. The right and left parts of the

46

J. sth, S. Krajnovi / Journal of Fluids and Structures 44 (2014) 31 51

1.75W 0.35W 1 Vfront 0.5 0 Vwake

1.02W

Fig. 17. Streamlines of the time-averaged velocity field (u ) projected onto the plane z 0.24W: (a) streamlines of the time-averaged flow, (b) streamlines of the instantaneous velocity at one realization of the flow, (c) streamlines of the instantaneous velocity at another realization, (d) streamlines of the instantaneous velocity at a third realization. The time between the realizations is 2.25 convective time units (tn).

Vgap,front

Vgap,base

Fig. 18. Streamlines projected onto the plane z 0.24W for the periodic wagon case: (a) time-averaged velocity, (b) one instantaneous realization.

time-averaged ring vortex, Vwake, in the wake are shown in Fig. 17(a). The streamwise locations of both centers are the same which we would expect due to symmetry. The instantaneous realizations show that the unsteadiness is large and that the time-averaged structures are made up of a large number of instantaneous vortices. Having considered the flow around the single wagon we now turn the attention to the periodic wagon case. Streamlines projected onto the plane z 0.24W are presented in Fig. 18. The time-averaged streamlines in Fig. 18(a) reveal two symmetric vortex pairs in both the front gap and the rear gap. The width of the gap, g, is g/W 0.5. Gap flow between an upstream body and a downstream body has been investigated for simplified tractor-trailer models by, e.g. Arcas et al. (2004) and sth and Krajnovi (2012). The flow in the gap plays a crucial role for tractor-trailers as it can be a dominating source of

J. sth, S. Krajnovi / Journal of Fluids and Structures 44 (2014) 31 51

47

Fig. 19. Vector arrows proportional to the instantaneous velocity field at three time steps around the gap in the periodic wagon case. The cut is placed at z 0.32W. The iso-surfaces are of Cp 0.05.

48

J. sth, S. Krajnovi / Journal of Fluids and Structures 44 (2014) 31 51

Fig. 20. Streamlines of the sampled mean velocity field (u ) projected onto the plane z 0.32W: (a) streamlines of the time-averaged flow for the single wagon case, (b) streamlines of the instantaneous velocity at one realization of the flow for the single wagon case, (c) streamlines of the time-averaged flow for the periodic wagon case, (d) streamlines of the instantaneous velocity for the periodic wagon case.

drag if not treated properly. Previous investigations by Allan (1981) and Arcas et al. (2004) of the flow around simplified tractor trailer models (consisting of one front box representing the tractor which is separated by a gap width, g, from one rear box representing the trailer) showed that for small gap widths (g =W o 0:40:5) the drag coefficient is small, but for a critical gap width of around g =W  0:5 the drag suddenly increases by up to 100%. The difference in the flow structures around the tractor and in the gap and their influence of the drag for large and small gap widths was thoroughly described in sth and Krajnovi (2012). Arcas et al. (2004) showed that for gap widths of g =W o 0:5, the flow in the gap consists of symmetric stable counter-rotating vortices. The flow in the gap starts to become unsteady in the gap for g =W  0:5 and for the regime of 0:5 o g =W o 0:7 the flow alternates between a symmetric state and an anti-symmetric state. The vortices in the gaps effectively shield the gaps from the separating shear layers from the upstream wagon and the shear layers are carried over the gap by the gap vortices and reattach on the next wagon again. This is the reason why the drag coefficient of the part coming from the container is only 0.04 in the present study. The periodic train acts almost like one connected body. The instantaneous realization shown in Fig. 18(b) shows that the time-averaged gap vortices are present in the instantaneous flow as well. It was found in Section 4.3 that the side force signal in the periodic case is dominated by one frequency of St 0.12. One event showing that this oscillation originates from the upstream flow entering the gap is shown in Fig. 19. The time between the realizations is t n 3:7. The cut in Fig. 19 is placed at z 0.32W which is beneath the container. The iso-surfaces seen in the figures are of the pressure coefficient CP 0.05. In Fig. 19(a) it is seen that the upstream shear layers are impinging on the rightmost part of the second wagon's wheels. This is denoted F1 in the figure. In the next time step in Fig. 19(b) the upstream shear layers on the left side are drawn into the gap (F2). In the third realization the shear layers are impinging on the leftmost part of the second wagon's wheels and areas of low pressure are convected downstream. Fig. 20 shows streamlines in the plane cutting the undercarriage at z 0.20W for both cases. The time-averaged flow is presented in Fig. 20(a) and (c) for the single wagon and periodic wagon cases, respectively. One major difference between the cases is that for the single wagon case flow enters the undercarriage and then spreads outwards to the sides under the

J. sth, S. Krajnovi / Journal of Fluids and Structures 44 (2014) 31 51

49

wagon before the rear wheels. This spread of flow outwards is not the case for the periodic wagon case. In the latter the flow under the wagon is parallel to the global streamwise direction both on the front and on the rear of the wagon. One realization of the instantaneous flow is presented for each case in Fig. 20(b) and (d). For the single wagon case in Fig. 20(b) it is seen that vortical structures origin at the front wheels which are responsible for the push of air to the sides further downstream. For the periodic wagon case shown in Fig. 20(d) it is seen that the instantaneous vortical structures are confined to the space between the wheels, at least in this realization. The flow is also more parallel aligned to the global streamwise direction. 5. Conclusions 5.1. Drag of the wagons In this work, the unsteady flow at zero yaw angle around a container freight wagon has been simulated successfully. Two cases have been considered: one single wagon and one wagon submerged in a train set with a wagon in front of it and one wagon behind. The numerical technique that was used in the study was LES with the standard Smagorinsky model. The container wagon model consisted of a container of length 11.8 m placed on an eight wheel freight wagon. The Reynolds number in the simulation was 105 based on the width of the container. The work was purely numerical, and no direct comparisons could be made to any wind tunnel experiments or other numerical studies. However, the value of the drag coefficient of the single wagon (0.90) and the periodic wagon (0.09) in the simulations correlated well with values from wind tunnel studies and full-scale studies where similar types of container freight wagons were studied. The spatial resolution obtained by the grids in the simulations was fine enough to be considered to be well resolved for both the coarse and the fine grid. Analysis of the flow close to the surface of the container revealed that the streaky structures in the boundary layer were resolved. Comparison of the pressure coefficient along three different lines on the wagon between the coarse and fine grids showed a collapse of the results everywhere except on the undercarriage. The periodic wagon was found to have a drag coefficient of 0.09 which is only 10% of that of the single wagon. This value was equally distributed on the container and the underhood. The container in the present study is smooth (thus, no ribs on the sides) and findings from full-scale studies (Engdahl, 1987) have shown that removing ribs from containers can reduce the drag of a periodic wagon some 5060% to the same levels as in the present study. The conclusion from this is that the highest potential to reduce the drag on container wagons is to construct containers with smooth sides. The present study has only considered the wind direction of zero yaw. For zero yaw conditions the gap between containers for gap widths of g/W 0.5 as in the present study does not constitute a large contributor to the drag. 5.2. Flow around the wagons Along the side of the wagon in the single wagon case, a longitudinal vortex is formed which emanates from flow being pushed to the sides under the wagon. This push of the flow outwards can be seen in the streamlines in the plane z 0.32 in Fig. 20. The instantaneous flow is more complex than what the time-averaged flow pictures reveal. However, the swirling component of the velocity of the longitudinal vortex along the side of the wagon is only some 1020% of U 1 . The instantaneous behavior of the separation bubble on the roof of the single wagon was found to be complex. It was found that the separation bubble on the roof is split into two parts which oscillate in the global streamwise direction 1801 out of phase. From the separation bubble vortices are shed irregularly and waves of these vortices propagating downstream from the separation bubble were shown on the roof. Eventually these waves enter the shear layers forming the near wake of the single wagon. The mean flow in wake was found to consist of a ring vortex whose lower center in the symmetric plane was located closer to the base face than the upper center. This is due to the influence on the flow of the geometric details of the undercarriage that was included in the freight wagon model in the present work. The location of the lower center of the ring vortex in the wake was found to be more or less stable while the upper center was found to oscillate in the streamwise direction in phase with the most downstream part of the separation bubble on the roof. The mechanism that governs the synchronization of the ring vortex in the wake and the separation bubble consists of waves of vorticity being shed from the separation bubble. A similar synchronization has been found in the flow over two-dimensional bluff bodies with sharp leading edges and length-to-thickness ratios that make the leading edge separation reattach on the surface of the body again before entering the trailing edge (Parker and Welsh, 1983; Cherry et al., 1984). The flow around the periodic wagon was found, as expected, to be very different from the flow around the single wagon. The presence of a wagon upstream and a wagon downstream removes all the areas on the wagon where flow impinge directly on the front of the wagon and thus the areas of massive separated flow on the roof and on the side of the wagon. The gap between the wagons was found to contain two counterrotating vortices that are symmetric in the mean. These vortices effectively shield the gap from the separating shear layers from the upstream wagon and carries them over the gap where they re-attach on the roof and sides of the container on the next wagon. It was found that the flow in the gap oscillates in the lateral direction which causes the shear layers to impinge on the outermost parts of the face of the downstream wagon. This oscillation causes a dominating frequency of St 0.12 in the side force signal. It appears as the oscillations are stronger on the lower part of the wagon near the undercarriage than on the upper part of the container.

50

J. sth, S. Krajnovi / Journal of Fluids and Structures 44 (2014) 31 51

Acknowledgments This project is supported financially by Trafikverket (Swedish Transport Administration). Software licenses were provided by AVL List GMBH. Computations were performed at SNIC (Swedish National Infrastructure for Computing) at the Center for Scientific Computing at Chalmers (C3SE), Center for High Performance Computing at KTH (PDC) and National Supercomputer Center (NSC) at LiU. References
Adrian, R.J., 2007. Hairpin vortex organization in wall turbulence. Physics of Fluids 19, 041301. Alam, F., Watkins, S., 2007. Lateral stability of a double stacked container wagon under crosswinds. In: Proceedings of the International Conference on Mechanical Engineering 2007 (ICME2007), Dhaka, Bangladesh, pp. 225242. Allan, J., 1981. Aerodynamic drag and pressure measurements on a simplified tractor-trailer model. Journal of Wind Engineering and Industrial Aerodynamics 9, 125136. Arcas, D., Browand, F., Hammache, M., 2004. Flow structure in the gap between two bluff bodies. AIAA Paper 20042250. AVL, 2010. CFD Solver. AVL Fire Manual, v2010.1, edition 04/2010. Baker, C., 2001. Flow and dispersion in ground vehicle wakes. Journal of Fluids and Structures 15, 10311060. Bearman, P., 1997. Near wake flows behind two- and three-dimensional bluff bodies. Journal of Wind Engineering and Industrial Aerodynamics 6971, 3354. Breuer, M., De Nayer, G., Munsch, M., Gallinger, T., Wuchner, R., 2012. Fluidstructure interaction using a partitioned semi-implicit predictorcorrector coupling scheme for the application of large-eddy simulation. Journal of Fluids and Structures 29, 107130. Browand, F., Hammache, M., 1999. Aerodynamic forces experienced by a 3-vehicle platoon in a crosswind. SAE Paper 1999-01-1324. Browand, F., Hammache, M., 2004. The limits of drag behavior for two bluff bodies in tandem. SAE Paper 2004-01-1145. Cheng, S., Tsubokura, M., Nakashima, T., Okada, Y., Nouzawa, T., 2012. Numerical quantification of aerodynamic damping on pitching of vehicle-inspired bluff body. Journal of Fluids and Structures 30, 188204. Cherry, N.J., Hillier, R., Latour, M.E.M.P., 1984. Unsteady measurements in a separated and reattaching flow. Journal of Fluid Mechanics 144, 1346. Davidson, L., 2009. Large eddy simulations: How to evaluate resolution. International Journal of Heat and Fluid Flow 30 (5), 10161025. Davidson, L., 2010. How to estimate the resolution of an les of recirculating flow. Quality and Reliability of Large-Eddy Simulations II 16 (5), 269286. Davis, W.J., 1926. The tractive resistance of electric locomotives and cars. General Electric Review 29, 224. Engdahl, R., 1987. Full-scale rail car testing to determine the effect of position-in-train on aerodynamic resistance. In: Publication R-705. Association of American Railroads. Engdahl, R., Gielow, R.L., Paul, J.C., 1986. Train resistance aerodynamics volume II, open top car application. In: Proceedings, Railroad Energy Technology Conference II. Association of American Railroads, Atlanta, GA, pp. 225242. Germano, M., Piomelli, U., Moin, P., Cabot, W.H., 1991. A dynamic subgrid-scale eddy viscosity model. Physics of Fluids A: Fluid Dynamics 3 (7), 17601765. Geurts, B., 2008. Regularization modeling for LES of separated boundary layer flow. Journal of Fluids and Structures 24, 11761184. Ghosal, S., Moin, P., 1995. The basic equations for the Large Eddy Simulation of turbulent flows in complex geometry. Journal of Computational Physics 118, 2437. Gielow, M., Furlong, C., 1988. Results of wind tunnel and full-scale tests conducted from 1983 to 1987 in support of the Association of American Railroads' Train Energy Program. In: Publication R-685. Association of American Railroads. Hammit, A.G., 1976. Aerodynamic forces on freight trains: volume 1 wind tunnel tests of containers and trailers on flatcars. In: Report No. FRA/ORD-76295.I. U.S. Department of Transportation. Han, X., Krajnovi, S., 2012. Numerical investigation of passive flow control around a d-shaped bluff body. In: 7th International Symposium on Turbulence, Heat and Mass Transfer, Palermo, Sicily, Italy. Hemida, H., Baker, C., 2010. Large-eddy simulation of the flow around a freight wagon subjected to a crosswind. Computers and Fluids 39, 19441956. Hemida, H., Gil, N., Baker, C., 2010. LES study of the slipstream of a rotating train. Journal of Fluids Engineering 132. 05110319. Hemida, H., Krajnovi, S., 2008. LES study of the influence of a train-nose shape on the flow structures under cross-wind conditions. Journal of Fluids Engineering 125. 091101112. Hemida, H., Krajnovi, S., 2010. LES study of the influence of the nose shape and yaw angles on flow structures around trains. Journal of Wind Engineering and Industrial Aerodynamics 98, 3446. Hemida, H., Krajnovi, S., Davidson, L., 2005. Large eddy simulations of the flow around a simplified high speed train under the influence of cross-wind. In: 17th AIAA Computational Dynamics Conference, Toronto, Ontario, Canada. Hoerner, S., 1965. Efficiency of railroad trains. In: Fluid Dynamic Drag. Hoerner Fluid Dynamics, Brick Town, New Jersey, pp. 1210 to 1214. Hucho, W.-H., 1998. Aerodynamics of Road Vehicles, 4th ed. Society of Automotive Engineers Inc. ISBN 0-7680-0029-7. Jeong, J., Hussain, F., 1995. On the identification of a vortex. Journal of Fluid Mechanics 285, 6994. Johansen, 1936. Air resistance of trains. Proceedings of the Institution of Mechanical Engineers 134, 91. Kline, S.J., Reynolds, W.C., Schraub, F.A., Runstadler, P.W., 1967. The structure of the turbulent boundary layer. Journal of Fluid Mechanics 30, 741773. Kobayashi, H., 2005. The subgrid-scale models based on coherent structures for rotating homogeneous turbulence and turbulent channel flow. Physics of Fluids 17, 045104. Krajnovi, S., 2009. LES of flows around ground vehicles and other bluff bodies. Philosophical Transactions of the Royal Society A 367 (1899), 29172930. Krajnovi, S., Davidson, L., 2003. Numerical study of the flow around the bus-shaped body. Journal of Fluids Engineering 125, 500509. Krajnovi, S., Davidson, L., 2005. Flow around a simplified car, part 1: large eddy simulation. Journal of Fluids Engineering 127, 907918. Krajnovic, S., Davidson, L., 2005. Influence of floor motions in wind tunnels on the aerodynamics of road vehicles. Journal of Wind Engineering and Industrial Aerodynamics 93, 677696. Krajnovi, S., Lrusson, 2012. Large eddy simulations of the flow around pyramids. In: Proceedings CFD Society of Canada Conference, Canmore, Canada. Krajnovi, S., sth, J., 2010. LES study of breakdown control of A-pillar vortex. International Journal of Flow Control 2 (4), 237257. Lai, Y.-C., Barkan, C.P., 2005. Options for improving the energy efficiency of intermodal freight trains. Transportation Research Record 1916, 4755. Lai, Y.C., Barkan, C.P., nal, H., 2008. Optimizing the aerodynamic efficiency of intermodal trains. Transportation Research Part E: Logistics and Transportation Review 44, 820834. Lam, K., Lin, Y.F., Zou, L., Liu, Y., 2011. Numerical study of flow patterns and force characteristics for square and rectangular cylinders with wavy surfaces. Journal of Fluids and Structures 28, 159377. Lam, K., Lin, Y.F., Zou, L., Liu, Y., 2012. Numerical simulation of flows around two unyawed and yawed wavy cylinders in tandem arrangement. Journal of Fluids and Structures 28, 135151. Lukaszewicz, P., 2001. Energy Consumption and Running Time for Trains (Ph.D. thesis), Royal Institute of Technology. Lukaszewicz, P., 2006. Running resistance results and analysis of full-scale tests with passenger and freight trains in Sweden. Proceedings of IMechE, Part F: Journal of Rail and Rapid Transit 221, 183193. Nakamura, Y., Nakashima, M., 1986. Vortex excitation of prisms with elongated rectangular, h and [vertical, dash] cross-sections. Journal of Fluid Mechanics 163, 149169.

J. sth, S. Krajnovi / Journal of Fluids and Structures 44 (2014) 31 51

51

Nakamura, Y., Ohya, Y., Tsuruta, H., 1991. Experiments on vortex shedding from flat plates with square leading and trailing edges. Journal of Fluid Mechanics 222, 437447. Naudascher, E., Wang, Y., 1993. Flow-induced vibrations of prismatic bodies and grids of prisms. Journal of Fluids and Structures 7 (4), 341373. sth, J., Krajnovi, S., 2012. The flow around a simplified tractor-trailer model studied by Large Eddy Simulation. Journal of Wind Engineering and Industrial Aerodynamics 102, 3647. Parker, R., Welsh, M., 1983. Effects of sound on flow separation from blunt flat plates. International Journal of Heat and Fluid Flow 4 (2), 113127. Patankar, S., Spalding, D., 1972. A calculation procedure for heat, mass and momentum transfer in three-dimensional parabolic flows. International Journal of Heat and Mass Transfer 15, 17871806. Paul, J., Johnson, R., Yates, R., 2007. Application of CFD to rail car and locomotive aerodynamics. In: The Aerodynamics of Heavy Vehicles II: Trucks, Buses, and Trains, vol. 1, Lake Tahoe, USA, pp. 305315. Peters, J.L., 1993. Effect of Reynolds number on the aerodynamic forces on a container model. Journal of Wind Engineering and Industrial Aerodynamics 49, 431438. Pope, S.B., 2000. Turbulent Flows, 1st ed. Cambridge University Press, Cambridge. Raghunathan, R.S., Kim, H.-D., Setoguchi, T., 2002. Aerodynamics of high-speed railway train. Progress in Aerospace Sciences 38, 469514. Rastgou, H., Saedodin, S., 2013. Numerical simulation of an axisymmetric separated and reattached flow over a longitudinal blunt circular cylinder. Journal of Fluids and Structures 42, 1324. Rochard, B.P., Schmid, F., 2000. A review of methods to measure and calculate train resistances. Proceedings of IMechE, Part F: Journal of Rail and Rapid Transit 214, 185199. Sakuma, Y., Paidoussis, M.P., Price, S.J., 2008a. Dynamics of trains and train-like articulated systems travelling in confined fluid part 1: modelling and basic dynamics. Journal of Fluids and Structures 24, 932953. Sakuma, Y., Paidoussis, M.P., Price, S.J., 2008b. Dynamics of trains and train-like articulated systems travelling in confined fluid part 2: wave propagation and flow-excited vibration. Journal of Fluids and Structures 24, 954976. Saunders, J., Watkins, S., Cassar, R., 1993. Vortex optimisation of slotted tops and cavities of two different open rail wagons. Journal of Wind Engineering and Industrial Aerodynamics 49, 421430. Smagorinsky, J., 1963. General circulation experiments with the primitive equations. Monthly Weather Review 91 (3), 99165. Smith, C.R., Metzler, S.P., 1983. The characteristics of low-speed streaks in the near-wall region of a turbulent boundary layer. Journal of Fluid Mechanics 129, 2754. Taylor, Z., Palombi, E., Gurka, R., Kopp, G., 2011. Features of the turbulent flow around symmetric elongated bluff bodies. Journal of Fluids and Structures 27 (2), 250265. Tomasini, G., Cheli, F., 2013. Admittance function to evaluate aerodynamic loads on vehicles: experimental data and numerical model. Journal of Fluids and Structures 38, 92106. Wassen, E., Eichinger, S., Thiele, F., 2010. Simulation of active drag reduction for a square-back vehicle. In: King, R. (Ed.), Active Flow Control II, Notes on Numerical Fluid Mechanics and Multidisciplinary Design, vol. 108. , Springer, Berlin, Heidelberg, pp. 241255. Watkins, S., Saunders, J., Kumar, H., 1992. Aerodynamic drag reduction of goods trains. Journal of Wind Engineering and Industrial Aerodynamics 40, 147178. Zacksenhause, M., Abramovich, G., Hetstoni, G., 2001. Automatic spatial characterization of low-speed streaks from thermal images. Experiments in Fluids 31, 229239.