You are on page 1of 5

feature

Numerical Simulation of Snow Entrainment with Application to Train Undercarriage Design


M. Trenker, W. Payer, C. Krenn, G. Haider, M. Mann Austrian Research Centers, Austria

This article is based on the paper which was first presented at the NAFEMS World Congress 2005 in May 2005.

The paper and its author received the congress award for :

"Most Innovative Use of Simulation Technology"

now accretion plays an important role for the operation of vehicles in countries with frequent snowfall. In the case of a train running on snow covered tracks, snow is being dispersed from the ground and accumulates on parts of the undercarriage. This can result in blockage of air inlets and outlets of the engine cooling system and various other essential devices, finally leading to vehicle failure. A straightforward approach to model particle accretion for engineering applications will be introduced. A detailed solution of the flow field serves as a starting point for the injection of particles. On walls a criterion has to be included in the boundary conditions whether particles will be trapped or reflected. In the literature, detailed investigations can be found for snow accretion on cylindrical shapes. Poots (1996) analysed ice and snow kinetics and snow accretion on overhead transmission lines. However these models would be too complex to study accretion on train undercarriage aggregates. In one of the first publications about saltation Bagnold (1941) identified three distinct transport modes for the transport of particles by an air stream: surface creep, saltation and suspension. Surface creep describes grains which make short hops just above the surface, having been ejected by a saltating particle. Bagnold defined two wind threshold

velocities given by equation 1: The impact threshold friction velocity u*t which must be exceeded for saltation to be maintained. At friction velocities higher than the impact threshold velocity particles entering the airstream move along parabolic trajectories before returning to their bed. After exceeding the impact threshold velocity the next threshold is reached which is referred to as the fluid threshold friction velocity u*tf. For wind speeds greater than the fluid threshold velocity, particles are being

Nomenclature
Atf,Af C F d ds g u* u*t u*tf u*c Pn,Pt Pn,Pt constants dimensionless coefficient mass flux per unit width in [kg/m-s] grain diameter in [m] standard grain diameter in [m] gravitational acceleration in [m/s2] friction velocity in [m/s] impact friction velocity threshold in [m/s] fluid friction velocity threshold in [m/s] critical fluid friction velocity threshold in [m/s] momentum in normal and tangential direction before reflection in [kgm/s] momentum in normal and tangential direction after reflection in [kgm/s] density of air in [kg/m3] density of particle material in [kg/m3] wall shear stress in [N/m2]

t t t

July 2005

29

feature

Sorensen (1991) modified Bagnolds formulation, equation 2, and derived an equation,


Equation 4

for the mass flux in (c.g.s) units, which is an approximately cubic function of u*. The critical friction velocity u*c is set to u*tf in the following. Although these models were developed for sand, they are in good agreement with experiments also for snow particles and are therefore widely used in the literature, e.g. Nishimura (2000), Doorschot (2002). Figure 1 shows the mass flux per unit width versus the friction velocity for the two models under consideration. For quantifying snow saltation several differences in grain characteristics compared to sand particles are obvious. Newly fallen snow has a dendritic (greek. dendron tree) shape rather than a spherical shape which will influence the aerodynamic behaviour during suspension and saltation. Due to their distinct shape, snow particles will as well differ in their structure-elastic behaviour compared to sand particles. This is important for finding correct values for the threshold velocities. Snow particles being transported by an airstream will also change in shape due to collisions with neighbour particles and in that way alter the overall snow properties. (Doorschot 2002). The simplicity of the aforementioned models together with their use for a wide range of engineering applications concerning particle saltation found in the literature confirms the use of these models to simulate snow accretion. Due to the lack of any experimental data the model introduced by Sorensen is used. This model gives higher mass fluxes compared to the model by White and is therefore the conservative choice for the calculation of accretion on critical surfaces.

Figure 1: The particle mass flux plotted versus the friction velocity

transported away from their bed by turbulent eddies thus entering the airstream and in this way are able to move far downstream. This phenomenon is commonly referred to as suspension. Since suspension is supposed to be the main reason for the accretion of snow on train undercarriage devices only the fluid threshold friction velocity u*tf will be used to determine the onset of snow particle release.

Equation 1

From measurements (Bagnold) it is known that for particles larger than about 200 m Atf is 0.1. To quantify the saltating mass flux depending on the local friction velocity a number of mostly empirical models can be found in the literature. Bagnold gives an equation in (c.g.s.) units for the mass flux per unit width

Equation 2

with the friction velocity and C a dimensionless coefficient which is set to 1.5, indicating a nearly uniform distribution in grain size. The standard grain size diameter ds set to 480 m. White (1979), who initially investigated soil transport by winds on Mars, found a correlation, borne out of microphysical saltation models as well as wind tunnel studies. The model also holds for conditions on earth and is given by
Equation 3

Numerical Simulation
The flow field is simulated by solving the steady-state Reynolds averaged Navier-Stokes equations employing the standard k- SST model with standard wall functions, (Fluent 2004). In addition to solving the transport equations for the continuous phase, a discrete second phase is introduced in a Lagrangian frame of reference, following the EulerLagrange approach. This second phase consists of spherical particles dispersed in the continuous phase. Trajectories of these discrete phase entities are computed. A fundamental assumption made in this model is that the dispersed second phase occupies a low volume fraction, even though high mass loading is acceptable. It is assumed that the snow concentration in the air is small enough to have no significant influence on the solution

in (c.g.s.) units per unit width with cs= 61.2

30

July 2005

feature

Figure2: The CAD-model of the train

of the flow field. The particle trajectories are computed individually after the fluid phase calculation has reached the convergence criterion. The particle mass flux is calculated using equation 4 with u*tf = 0.21 m/s, according to Nishimura (2000). The velocity on the inlet boundary is set to 22.2 m/s. The turbulent intensity is 5% with a turbulent length scale of 0.01 m. At the outlet the static pressure is set to the ambient value of 101325Pa. The gravitational acceleration is set to the usual value. The fluid is simulated as an incompressible fluid with a constant density of 1.225 kg/m3 and a dynamic viscosity of 1.79 E-5 Pa-s. As particles are being transported by the flow, they are either trapped or reflected on walls. The trapping of particles depends on the angle between the particle path and the surface normal vector as well as the local fluid threshold friction velocity. It is assumed that particles hitting the surface at an angle smaller than 45 together with a local friction velocity smaller than u*tf = 0.21 m/s (Nishimura 2000) will be trapped. For particles being reflected on a wall the momentum of the particle after the impact is reduced by using pn= 0.01pn and pt= 0.2pt. This assumption is set to ensure that reflected particles travel alongside the wall instead of an elastic impact. For an average snow particle the diameter was set to 0.48 mm (Nishimura 2000). The density of snow particle is set to 366 kg/m3. This value results from the density of ice which is 915 kg/m3 and the assumption that the particle consists of approximately 40% ice. It is assumed that the wetted surface of a snow particle is a hundred times larger than the surface of a ball shaped particle of similar volume, which is taken into account by setting the shape-factor (Fluent 2004) to a value of 0.01.

Figure 3: A detailed view of the surface mesh of the train

Figure 4: Snow accretion in kg/m2/s on wall boundaries

Application
The train has a total length of 54 meters. Its CADmodel, already simplified for CFD, is depicted in figure 2. In the simulation, the tracks and the ground are moving with the inlet velocity of 22.2m/s, the wheels and axels of the train are kept non-rotating, because it has only marginal effects for the given problem. As can be seen from figure 2, specific care was taken in reproducing the underfloor devices in highest detail. The size of the mesh is 6.6 million volume cells. A structured/paved mesh approach using hexahedral and prism cells was used where ever possible to save mesh cells. A close-up view of the surface mesh is shown in figure 3. A finer mesh is set on those devices where snow accretion is assumed to be critical. The simulation was initialized with the inlet conditions. The flow simulation was performed in parallel on 4 CPUs in approximately 10hrs. About 1500 iterations were necessary to achieve the converged solution.

Figure 5: Snow accretion in kg/m2/s on in- and outlets

July 2005

31

feature
For the following calculation of particle trajectories 10 iterations were performed consuming about 6hrs of CPU-time. Figure 4 shows the accretion of snow on wall boundaries. These parts of surfaces which are exposed to the main flow show higher accretion values than surfaces hidden or protected by others. Concerning the accretion on the axle one has to keep in mind that the axle is kept non-rotating during the simulation. Accretion of snow on inlets and outlets is shown in figure 5. The volume flux on the air inlet is given by 0.35 m3/s which results in an average inlet velocity of 3.71 m/s. Almost no accretion is found on the outlet, since the averaged velocity of 4.85m/s prevents particles from being adhered. A velocity of 0.32 m/s is set on the ventilation inlet which results from the given volume flux of 0.44m3/s.
Figure 6: Snow accretion rate in kg/m2-s on the compressor inlet

The compressor inlet is shown in figure 6. This inlet is located on the upper side of the compressor box and is assumable well hidden from the main flow. The compressor volume flux is 0.3 m3/s. This inlet shows higher accretion rates compared to the air inlet in figure 5, which is much more exposed to the main flow. The rather narrow gap between the compressor inlet and the underfloor of the train leads to local pressures below the ambient pressure, which is the main reason for particles being sucked into the compressor inlet. Figure 7 reveals that the location of the compressors mounting strut above the inlet has great influence on the local flow field and is partly responsible for these high accretion rates. Particles tracks are post-processed in reverse. This ensures that the paths of all particles which will finally stick on the compressor inlet are tracked from their beginning.
Figure 7: Tracks of particles which are collected by the compressor inlet, colour depicts the local particle velocity in m/s

Figure 8 shows a global view of the particle tracks from a viewpoint behind the train. The train is moving away from the observer. Colours correspond to the residence time of particles in the flow domain. It is clearly seen that particles trapped under the train exhibit longer residence times compared to particles passing beside the train. The vortices in the trains wake further lift particles off the ground.

Conclusions
The discrete phase model of a commercial flow solver was adapted to study snow release, reflection and accretion for a railroad engineering application. Bagnolds theory was found most useful for calculating a threshold velocity which has to be exceeded before particles are lifted from the ground by the mainstream flow. Several models for the mass flow rate of particles depending on the local friction velocity were investigated. Sorensens model which is an

32

July 2005

feature
updated version of Bagnolds theory was used due to its simplicity and its validation found in the literature. Results gained from these simulations give a good idea about accretion on a complex geometry such as the undercarriage of a train and lead to design modifications. Accretion rates on critical surfaces are examined by integrating particle tracks which collide with these surfaces. Visualization of particle tracks shows the origin of particles which accumulate on critical surfaces. It is analysed how they find their way through the complex undercarriage geometry. Consequently design changes are made by mounting critical devices in areas less exposed to accretion. By including simulation tools in the early design process of trains, potential shortcomings and expensive redesign can be circumvented, thus reducing development risk and overall costs of train development. Arsenal research likes to thank Siemens Transportation Systems for their cooperation in the current project. BM

References
BAGNOLD, R.A., 1941, The Physics of Blown Sand and Desert Dunes, Methuen, London DOORSCHOT, J.J.J., LEHNING, M., 2002, Equilibrium Saltation: Mass fluxes, Aerodynamic Entrainment and Dependence on Grain Properties, Boundary-Layer Meteorology, Kluwer, 104(1), 111-130 Fluent 6.1.22, 2004, User Manual NISHIMURA, K., HUNT, J.C.R., 2000, Saltation and incipient suspension above a flat particle bed below a turbulent boundary layer, J. Fluid Mech., 417, 77-102 POOTS, G., 1996, Ice and Snow Accretion on Structures, John Wiley & Sons Inc, Research Studies Press Ltd, England SORENSEN, M., 1991, An analytic model of windblown sand transport, Acta Mechanica Suppl. 1, 67-82 WHITE, B.R., 1979, Soil Transport by winds on Mars, J. Geophys. Res., 84, 4643-4651

Contact
Markus Trenker, Arsenal Research markus.trenker@arsenal.ac.at

- July 2005

33

You might also like