Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.

Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter.
Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 579

lengths, word breaks, heading styles, and other typesetting-specific formatting, however, cannot be retained, and some typographic errors may have been accidentally inserted. Please use the print version of this publication as
About this PDF file: This new digital representation of the original work has been recomposed from XML files created from the original paper book, not from the original typesetting files. Page breaks are true to the original; line
LARGE-EDDY SIMULATIONS OF TURBULENT WAKE FLOWS 579
Large-Eddy Simulations of Turbulent Wake Flows
Shaoping Shi, Andrei Smirnov, Ismail Celik
(West Virginia University, USA)
ABSTRACT
A numerical method for solving three-dimensional, time-dependent incompressible Navier-Stokes equations using
the large eddy simulation (LES) is briefly described. A new random flow generation technique (RFG) was used to provide
the turbulent inflow boundary conditions. The combined LES-RFG procedure was applied to simulate the wake of flat
plate and a ship model. The simulations of flat-plate wake were validated against the experimental data. In the case of a
ship-wake (ship model 5415), reasonable results were obtained.
1 INTRODUCTION
Most of the computational fluid dynamics efforts applied to flow past ships are based on Reynolds-Averaged Navier-
Stokes (RANS) equations utilizing various turbulence models (Sotiropoulos and Patel, 1995; Paterson et al., 1996;
Ratcliffe, 1998). The commonly used models include k-ε, k-ω and algebraic stress models. RANS is often quite adequate
for mean flow predictions, but provides only limited information about turbulence characteristics and almost no details on
large-scale unsteady structures of the flow field. LES technique, on the other hand, was designed to simulate the unsteady
behavior of at least the large coherent turbulent eddies. There is hardly any study in the literature where LES is applied to
flow around ship-hulls including the wake. The main reason obviously lies on the computer resource limitation which
increases almost exponentially with the Reynolds number. In LES, the energy containing eddies of the flow are computed
explicitly, while only the more universal (isotropic) small eddies are modeled. Thus very fine grids have to be applied in
order to resolve the boundary layer near the wall where the turbulence length scales tend to zero as the the wall is
approached. However, in some applications, like bubble dynamics modeling, it is still necessary to resolve coherent flow
structures—large turbulent vortices and eddies. In RANS simulations, especially those using two-equation models such as
k-ε model, these unsteady flow features are usually smeared out. The prediction of bubble population in the wake is
important in controlling the signature of surface ships.
Since it can be computationally prohibitive to include the ship hull and the wake in LES, it would be desirable for
the purposes of pure wake simulations to start the computations somewhere in the near wake excluding the ship-body.
This technique, called by other researchers the Initial Data Plane (IDP) approach (Hyman, 1998; Paterson et al., 1996;
Dommermuth et al., 1996) can introduce large errors as pointed out by Hyman (1998). By applying this procedure the
flow field in the far wake can be calculated. Then the wake computation can start form a plane in the near wake. This is
similar to our approach in which we call it inflow boundary instead of IDP. In IDP method, the data reflects the mean
flow, mean turbulence quantities and the scalar field at this plane. In our approach, a Random Flow Generation (RFG)
algorithm (Smirnov et al., 2000b; Shi et al., 2000b) is applied to generate the inflow turbulence based on the information
from experiments or RANS calculations. The appropriate time scale and length scale can be specified for each point at the
inflow boundary also by using the RANS results. In addition to that all six turbulent shear stress components can be
given, thus providing anisotropic and inhomogeneous turbulent inlet conditions. The generated velocity fluctuations
satisfy instantaneous continuity equation, and the turbulence statistics (Reynolds
the authoritative version for attribution.

OCR for page 579

lengths, word breaks, heading styles, and other typesetting-specific formatting, however, cannot be retained, and some typographic errors may have been accidentally inserted. Please use the print version of this publication as
About this PDF file: This new digital representation of the original work has been recomposed from XML files created from the original paper book, not from the original typesetting files. Page breaks are true to the original; line
LARGE-EDDY SIMULATIONS OF TURBULENT WAKE FLOWS 580
stresses) prescribed a priori from RANS or experimental results. Thus in some sense, although the LES starts at a plane
behind the body, the body is still seems virtually to be there. The features of the generated flow-field such as continuity,
anisotropy and inhomogeneity make it also well suited for setting the initial conditions for LES. The developing flow
field is calculated directly both spatially and temporally with appropriate boundary conditions. In particular we apply
anisotropic, inhomogeneous, unsteady IDP condition and let the flow develop according to the dynamics of Navier-Stokes
equations without forcing. In this case our approach differs from that of Dommermuth et al. (1996). In their approach, the
flow field was initialized with fully developed isotropic, homogeneous turbulence (IHT) superimposed on the measured
(or calculated via RANS) flow field. A forcing function (or a stirring force) is applied to the momentum equation to
provide a prescribed turbulent dissipation rate. It seems to us that Dommermuth et al.'s method relies too heavily on the
initialization using RANS results. Since, in its true sense, spatial development is not calculated but assumed to be related
to temporal development, the initial turbulent flow field must play dominant role in the subsequent development of the
turbulence. By imposing turbulent dissipation rate (as calculated from RANS) via a forcing function, this method become
even more dependent on RANS. Moreover, it is not clear if the periodic boundary conditions applied by Dommermuth et
al. in the axial direction where there is a significant flow development is appropriate for flows in the near field of a ship
wake.
In what follows we first briefly describe the LES scheme and the RFG algorithm. Then we present their applications
to the wake flow of a flat plate and a ship-wake. The inflow boundary of the wake of flat plate is calculated based on the
experiment results by Ramaprian, et al. (1981). Ship-wake calculations are based on the results of the RANS-solver
CFDSHIP-IOWA (Stern and Wilson, 2000). Comparison between LES and experiments or RANS have been made.
Conclusions and recommending for future work are given at the end.
2 MATHEMATICAL/NUMERICAL MODELS
2.1 Navier-Stokes Solver
Here we briefly summarize the mathematical formulation of the governing equations and the SGS models which are
used in this study. The LES code we use was originally developed by Zang et al. (1994). The spatially filtered flow
conservation equations are1:
(1)
(2)
where
(3)
(4)
(5)
In the above equations, uj is the filtered velocity vector, p is pressure, v is the kinematic viscosity and δij is
Kronecker delta. In Eq. 5 is the resolved strain rate tensor,
(6)
and is defined as
(7)
where the value of Cr is either 0 or 1 depending on the type of subgrid-scale (SGS) model being used. When Cr=0,
Eq. 5 represents the Smagorinsky model. When Cr=1 it represents the dynamic mixed model of Zang et al. (1993). The
dynamic mixed model is more stable than dynamic model. It has the capability to present the backscatter energy and near
wall features. However,
the authoritative version for attribution.
1The implied summation rule applies to repealed indices (i, j= 1, 2, 3)

OCR for page 579

lengths, word breaks, heading styles, and other typesetting-specific formatting, however, cannot be retained, and some typographic errors may have been accidentally inserted. Please use the print version of this publication as
About this PDF file: This new digital representation of the original work has been recomposed from XML files created from the original paper book, not from the original typesetting files. Page breaks are true to the original; line
LARGE-EDDY SIMULATIONS OF TURBULENT WAKE FLOWS 581
it requires a very small time step due to numerical instability. Moreover, from our experience, this model shows high
diffusion in a turbulent wake flow when used with relatively coarse grid.
The original code has capability of handling curvilinear coordinates but here we only use Cartesian coordinates.
As for numerics, Crank-Nicolson discretization scheme is applied for diagonal viscous and diffusive terms for time-
advancement while Adams-Bashforth scheme was chosen for the other terms. Spatially, QUICK scheme (Leonard, 1979)
or central differencing (CD) scheme is applied to discretize the convective terms. Central differencing scheme is used for
the other terms. With QUICK scheme, no subgrid-scale (SGS) model is used while for central differencing scheme
Smogarinsky model (Smagorinsky, 1963) is applied.
More detailed information can be found in Zang, et al. (1994; 1993) and Shi et al. (2000a).
2.2 Random Flow Generation (RFG) methodology
Here we provide a brief description of the method used for generating the inflow boundary. The algorithm proceeds
in the following sequence.
1. Given an anisotropic velocity correlation tensor (say from a RANS)
(8)
of a turbulent flow field find an orthogonal transformation tensor aij that would
diagonalize rij2
(9)
(10)
As a result of this step both aij and λn become known functions of space.
2. Generate a transient flow-field in a three-dimensional domain {vi(xj, t)}i-j−1..3 using the modified method of
Kraichnan (?)
(11)
(12)
(13)
where l, τ are the length and time-scales of turbulence, εijk is the permutation tensor used in vector product
operation (Spain, 1965), and N(M, σ) is a normal distribution with mean M and standard deviation σ.
Symbols represent a sample of n wavenumber vectors and frequencies of the modeled turbulence
spectrum
(14)
3. Apply a scaling and orthogonal transformations to the flow-field vt generated in the previous step to obtain a
new flow-field ui
(15)
(16)
The procedure described above takes as an input the correlation tensor rij of the original turbulent flow-filed
and information on length- and time-scales of turbulence (l, τ). The output of the procedure is the time dependent flow-
field ui(xj, t), which satisfies the anisotropy and inhomogeneity of the original flow-field ũ i, i.e. the shear-stresses of the
generated flow-field are equal to rij and length- and time-scales to l and τ respectively. As was shown by (Smirnov et al.,
2000b;
the authoritative version for attribution.
2 Repeated indexes imply summation, parentheses around indexes preclude summation.

OCR for page 579

lengths, word breaks, heading styles, and other typesetting-specific formatting, however, cannot be retained, and some typographic errors may have been accidentally inserted. Please use the print version of this publication as
About this PDF file: This new digital representation of the original work has been recomposed from XML files created from the original paper book, not from the original typesetting files. Page breaks are true to the original; line
LARGE-EDDY SIMULATIONS OF TURBULENT WAKE FLOWS 582
Shi et al., 2000c), the new flow-field ui is divergence-free in case of a homogeneous turbulence and to a high-degree
divergence free for an inhomogeneous turbulence. By virtue of Eq. (11), spatial and temporal variations of ui follow
Gaussian distribution with characteristic length and time-scales of l, τ, however, other distributions can be used to
simulate different turbulence spectra. More details of this method can be found in Smirnov et al. (2000b).
3 APPLICATIONS
The pseudo-random flow field generated by the RFG method is added to the mean flow to established the boundary
condition at the inlet plane. The overall procedure is applied to two significantly different cases, namely flat plate wake
and a ship model wake. Since we try to avoid solving for the flow around the bodies, we start the computational domain
from the inflow boundary (initial data plane) located immediately after the bodies in the wake where there is no flow-
reversal.
For the flat plate wake, we start the computational domain from 19.5mm (x/L=0.01) behind the edge of the plate,
which corresponds to a location where measurements were available. For the ship wake the domain starts from x/L=1.05
(x/L=1. is the end of the ship model). At these planes the RFG method described above (Smirnov et al., 2000a) is used in
conjunction with the the RANS calculations (Stern and Wilson, 2000).
For both cases, the inflow and outflow boundary conditions are applied in x direction. For outflow boundary both
convective and Neumann (free gradient) boundary conditions have been applied. A comparison (not shown here)
indicated that there is not much difference between the results obtained by applying these two conditions.
Symmetry conditions are used in y direction and periodic boundary conditions are used in the spanwise (z) direction.
At the free surface slip in x and z directions is allowed but the velocity component normal to the free surface is set to
zero. As such the free surface is approximated as a moving flat plane.
For flat plate wake, the domain size is 1.0m×0.2m× 0.6m in x, y and z direction, respectively. The grid size is
82×50×50. Non-uniform grid is used in both x and y directions with stretching not exceeding 3 percent. Note that, in our
study, x represents streamwise direction, y represents vertical direction and z represents spanwise direction, respectively
(see Fig. 1).
For the ship wake, the domain size is 1.5×0.2×0.6 (given in non-dimensional units in ship length) in x, y and z-
directions, respectively. The grid size is 162× 50×66. Non-uniform grid is used in both x and y directions.
The length scale and time scale are selected as constant in this paper. For the flat plate, the length scale is 4mm
which is chosen as 10% of the width of the wake, and time scale is 0.001s. For ship wake, the length scale is 0.02,
dimensionless of ship length, and the time scale is 0.01.
4 RESULTS
4.1 Flat Plate Wake
In Fig. 2, the velocity time histories are shown at different points in the wake. From these pictures it is seen that both
the amplitude and the frequency of the velocity decrease (decaying turbulence) along the streamwise direction, which is
consistent with the behavior of a turbulent wake. The predicted anisotropy is also noteworthy.
The logarithmic part of the mean velocity along streamwise direction at the center line can be expressed (Nakayama
and Liu, 1990; Ramaprian and Patel, 1982; Wygnanski et al., 1986; Andreopoulos and Bradshaw, 1980) by
(17)
where Uc is the streamwise mean velocity at the center line, uτ=0.853m/s is the friction velocity of the boundary
layer at the trailing edge, x is the distance from the plate edge, v is the kinetic viscosity and A and B are constants which
are 4.65 and 0.7 respectively (Andreopoulos and Bradshaw, 1980). In the transverse direction (direction normal to the
plate), the logarithmic part (if there exist one) of the mean velocity can be expressed as (Nakayama and Liu, 1990)
the authoritative version for attribution.

OCR for page 579

lengths, word breaks, heading styles, and other typesetting-specific formatting, however, cannot be retained, and some typographic errors may have been accidentally inserted. Please use the print version of this publication as
About this PDF file: This new digital representation of the original work has been recomposed from XML files created from the original paper book, not from the original typesetting files. Page breaks are true to the original; line
LARGE-EDDY SIMULATIONS OF TURBULENT WAKE FLOWS 583
(18)
where U+≡U/uτ and y+≡yuτ/v. κ is the Von Karman constant and equal to 0.41.
Figures 3 and 4 show the comparisons of our simulations with experimental results (Ramaprian and Patel, 1982).
The predictions are in good agreement both with experiments and the theoretical log-profiles. The mean velocity along x-
direction for our simulation has a wave-like oscillation. This phenomenon maybe due to a weak vortex shedding which
could exist because we only performed the statistical analysis at one line instead of the whole center plane.
Fig. 5 shows the decay of turbulent kinetic energy along the center line of the wake. We also depict the turbulent
kinetic energy along a vertical line at x=381mm in Fig. 6. Again the predictions agree well with the experiments. All of
the above results were obtained by using the QUICK scheme without any SGS model. This scheme has a built-in 4th
order dissipation which seems to act like a SGS model. Large eddy simulations without SGS are possible as elucidated by
Boris et al. (1992)
The spanwise vorticity contours are shown in Fig. 7(a). Central differencing (CD) with Smagorinsky model was
applied in this calculation. The results using the QUICK scheme were much smoother. In this Figure, large coherent
structures are clearly visible. These structures are similar in appearance to Karman vortex street because they seem to be
comprised of vortices of alternating sign of vorticity which are also visible in the experiments of Wygnanski et al. (1986).
As explained by Wygnanski et al., “Neither the varicose mode, which requires that the vortices appear in pairs distributed
symmetrically about the centerline, nor the sinuous mode, which requires vortices whose center coincides with the
centerline, dominate this flow”.
4.2 SHIP WAKE
The streamwise vorticity contours at different planes are shown in Fig. 8 and the contours of the vertical component
of vorticity are shown in Fig. 9. Fig. 8 shows that concentrated vorticity decreases with axial distance, and in the far wake
vorticity is only concentrated near the free surface. Some of this rapid decay may be partially due to the grid expanding
towards the outlet plane, while the size of the wake increases in axial direction of the flow. As a result of this, many
turbulence structures may have died out prematurely in simulations. It is interesting to note the two distinctly
concentrated vorticity streaks away from the center line of the wake (Fig. 9). The streamwise velocity contours at a plane
near the free surface are shown in Fig. 10. Due to the lack of experimental data for this wake, we can not make a
quantitative judgment of the predictions. However, we think that these results are reasonable. As the width of the wake
increases, turbulence decays. In the far wake only larger turbulence structures can been seen. This maybe due to two
reasons: (a) coarser grids are applied in the far wake; (b) the small turbulence structures contain significantly less energy
so that they can only last for shorter time compared to the larger turbulence structures.
To test the effects of the grid size on the flow field resolution, we doubled the grid number in x (streamwise)
direction and ran the simulations again. In Fig. 11 more detailed turbulence structures can been seen clearly as compared
to the coarser grid solution (Fig. 9). The resolved turbulence kinetic energy (TKE) for different schemes and different grid
sizes is shown in Fig. 12. The RANS solution is also shown in this picture for comparison. The resolved TKE from fine
grid is higher than that of the coarse grid as expected. Central differencing discretization with Smagorinsky model gives
better results than the other schemes. From this figure it also can be seen that there is not significant difference in the
results when the QUICK scheme is used with or without SGS model. Moreover, the resolved TKE is lower than that of
central differencing with Smagorinsky model. It means that QUICK scheme gives even higher numerical diffusion than
the Smagorinsky model. This is mostly due to grid resolution. For more detailed information about the comparisons of
numerical schemes and subgrid-scale models the reader is referred Shi et al. (2000a). One uncertainty in computations is
presented by sinusoidal-like distribution of TKE in the near wake. It may be because of the existing surface wave (not
accounted directively here). When the wave descends towards to the bottom of the domain, it seems to create a
constriction with flow passing through a small area. Thus both the velocity and TKE are higher at this re
the authoritative version for attribution.

OCR for page 579

lengths, word breaks, heading styles, and other typesetting-specific formatting, however, cannot be retained, and some typographic errors may have been accidentally inserted. Please use the print version of this publication as
About this PDF file: This new digital representation of the original work has been recomposed from XML files created from the original paper book, not from the original typesetting files. Page breaks are true to the original; line
LARGE-EDDY SIMULATIONS OF TURBULENT WAKE FLOWS 584
gion. We checked the wave profile of RANS calculation and the peak region of k-profile does indeed corresponds to the
descending wave. It is interesting that our calculations also provide a similar trend although no surface wave profile was
applied in our LES. This indicates that some wave information may be present implicitly in the inflow boundary3.
In figures 13–18 the velocity vectors on different vertical cross sectional planes are presented. Clearly, the large
scale turbulent eddies (vortices) are captured, which can play an important role in bubble dynamics. The prediction of
outward flow near free surface, and the streamwise evolution of the vortices are very encouraging indications of the
success of the present LES approach.
Currently, a case study with finer grids on the both vertical and spanwise direction is in progress. Here, we show the
velocity vectors in a cross-sectional plane at x/L=0.2 (Fig. 15). Comparison with the results from a coarser grid in Fig. 13
shows that more turbulence eddies are captured. The streamwise vorticity contours on the same plane are also shown in
Fig. 20 in oder to make the turbulent eddies more visible.
The vorticity contours on different vertical planes shown in Fig. 19, Fig. 21 and Fig. 22 are indicative of the degree
of resolution of the calculations. These structures are hard to see in Figures 13–18 due to relatively small magnitude of the
velocity vectors. However, to demonstrate the small weak turbulence structures, like those in Fig. 16(a), we enlarged the
velocity vectors where there is turbulence structure corresponding to the vorticity contour plot and depicted it in
Fig. 16(b). A corresponding area in the vorticity contour plot is also included in Fig. 22 for reference. The cross section is
at x/L=1.2. Having no interpolation, it seems that in the velocity vector plot the weak vortices can not be seen due to scale
difference, whereas the interpolated contours plots do show small structures. Figures 21 and 22 show again that much of
the vorticity is concentrated near the free surface and there are two large counter rotating vortices on two sides of the wake.
5 CONCLUSIONS AND FUTURE WORK
A combined LES scheme and an RFG approach suitable for wake flows is briefly described. The wake flow of a flat
plate is used as a validation case. Both the mean flow and turbulence intensities are compared with the experimental
results. A good agreement has been obtained. Regarding ship-wake simulations it is difficult to draw any definite
conclusions at this point in the absence of experimental data. However, the results are very reasonable and it means our
approach is a viable approach in that it is capable of capturing the most energetic unsteady, turbulent vortices and/or
eddies present in the wake. This study shows that with reasonable large grid nodes LES of ship wakes has very good
prospects. Nevertheless, full scale LES calculations of ship flow are still impossible because of the limitation of computer
resources and may require another 10 or 20 years (Larsson et al., 1998).
According to Hyman (Hyman, 1998), free-surface/turbulence interaction is important for the near-surface lateral
wake growth. So, including the effects of free surface will be the focus of our future work. Parallel computing (Osman et
al., 2000) is also an important factor since it can facilitate the usage of much finer grid resolution in LES.
ACKNOWLEDGEMENTS
This work has been performed under a DOD EPSCoR project sponsored by the Office of Naval Research (ONR),
Grant No. N000l4–98–1–0611. The program monitor is Dr. Edwin P.Rood. We thank Prof. Fred Stern and Dr. Robert
Wilson of University of IOWA for providing the RANS results for the ship model. Special thanks are due to Prof.
R.Street of Stanford University for providing us with the basic LES code. Prof. V.C.Patel prompt response with the report
that contained the experimental data for the flat plate wake is also appreciated.
the authoritative version for attribution.
3An animation of the ship wake flow is also available at http://cfd.mae.wvu.edu/shipwake

OCR for page 579

lengths, word breaks, heading styles, and other typesetting-specific formatting, however, cannot be retained, and some typographic errors may have been accidentally inserted. Please use the print version of this publication as
About this PDF file: This new digital representation of the original work has been recomposed from XML files created from the original paper book, not from the original typesetting files. Page breaks are true to the original; line
LARGE-EDDY SIMULATIONS OF TURBULENT WAKE FLOWS 585
REFERENCES
Andreopoulos, J. and Bradshaw, P.: 1980, Measurements of interacting turbulent shear laysers in the near wake of a flat plate, J.Fluid Mech. 100, 639
Boris, J., Grinstein, F., Oran, E., and Kolbe, R.: 1992, New insights into large eddy simulation. Fluid Dynamics Research 10, 199
Dommermuth, D., Gharib, M., Huang, H., Innis, G., Maheo, P., Novikov, E., Talcott, J., and Wyaatt, D.: 1996, Turbulent free shear flows: A
comparison between numerical simulations and experimental measurements, in 21st Symposium on Naval Hydrodynamics, pp 200–215,
Trondhiem, Norway
Hyman, M.: 1998, Computation of ship wake flows with free-surface/turbulence interaction, in 22nd Symposium on Naval Hydrodynamics, pp 11–32,
Washington, D.C.
Larsson, L., Regnstrom, B., Li, D., and Janson, C.: 1998, Failures, fantasies, and feats in the theoretical/numerical prediction of ship performance, in
22nd Symposium on Naval Hydrodynamics, pp 11–32, Washington, D.C.
Leonard, B.: 1979, A stable and accurate convective modeling procedure based on quadratic upstream interpolation, Comput. Methods Appl. Mech.
Engr. 19, 59
Nakayama, A. and Liu, B.: 1990, The turbulent near wake of a flat plate at low reynolds number, J.Fluid Mech. 217, 93
Osman, A., Ammar, H., Smirnov, A., Shi, S., and Celik, I.: 2000, Parallel analysis and implementation of large eddy simulations of ship wakes, in IEEE
International Conference on Cluster Computing CLUSTER 2000 Technische Universitt Chemnitz, Saxony, Germany
Paterson, E., Hyman, M., Stern, F., P., C., Bonetto, F., Drew, D., and Lahey, R.: 1996, Near- and far-field cfd for a naval combatant including thermal-
stratification and two-fluid modeling, in 21st Symposium on Naval Hydrodynamics, pp 102–117, Trondheim, Norway
Ramaprian, B. and Patel, V.: 1982, The symmetric turbulent wake of a flat plate, AIAA Journal 20, 1228
Ramaprian, B., Patel, V., and Sastry, M.: 1981, Turbulent wake development behind streamlined bodies, Technical Report IIHR Report No. 231, Iowa
Institute of Hydraulic Research, The University of IOWA
Ratcliffe, T.: 1998, Validation of free surface reynold's averaged navier stokes and potential flow codes, in 22nd Symposium on Naval Hydrodynamics,
pp 964–980, Washington, D.C.
Shi, S., Celik, I., and Smirnov, A.: 2000a, Comparison of different numerical schemes and sub-grid scale models in large-eddy simulation, in ASME
Fluids Engineering Division Summer Meeting, No. 11232 in FEDSM2000, Boston, Massachusetts
Shi, S., Smirnov, A., and Celik, I.: 2000b, An approach for generating time-dependent inflow boundary with application to large-eddy simulation of flat
plate wake, in International Mechancial Engineering Congress and Exposition, Orlando, FL
Shi, S., Smirnov, A., and Celik, I.: 2000c, Large eddy simulations of particle-laden turbulent wakes using a random flow generation technique, in ONR
2000 Free Surface Turbulence and Bubbly Flows Workshop, pp 13.1–13.7, California Institute of Technology, Pasadena, CA
Smagorinsky, J.: 1963, General circulation experiments with the primitive equations, part i: the basic experiment, Monthly Weather Rev. 91, 99
Smirnov, A., Shi, S., and Celik, I.: 2000a, Random Flow Generation Technique in Large Eddy Simulations and Particle-Dynamics Modeling, submitted
to Journal of Fluid Engineering
Smirnov, A., Shi, S., and Celik, I.: 2000b, Random flow simulations with a bubble dynamics model, in ASME Fluids Engineering Division Summer
Meeting, No. 11215 in FEDSM2000, Boston, Massachusetts
Sotiropoulos, F. and Patel, V.: 1995, Application of reynolds-stress transport models of stern and wake flows, Journal of Ship Research 39, 263
Spain, B.: 1965, Tensor Calculus, Oliver and Boyd
Stern, F. and Wilson, R.: 2000, RANS Results of Ship-Wake simulations, Private communication
Wygnanski, I., Champagne, F., and Marasli, B.: 1986, On the large-scale structures in two-dimensional, small-deficit, turbulent wakes, J.Fluid Mech.
168, 31
Zang, Y., Street, R., and Koseff, J.: 1993, A dynamic
the authoritative version for attribution.

OCR for page 579

About this PDF file: This new digital representation of the original work has been recomposed from XML files created from the original paper book, not from the original typesetting files. Page breaks are true to the original; line
lengths, word breaks, heading styles, and other typesetting-specific formatting, however, cannot be retained, and some typographic errors may have been accidentally inserted. Please use the print version of this publication as
the authoritative version for attribution.
LARGE-EDDY SIMULATIONS OF TURBULENT WAKE FLOWS
curvilinear coordinates. Journal of Computational Physics 114, 18
mixed subgrid-scale model and its applocation to turbulent recirculating flows, Phys. Fluids 5–12, 3186
586
Zang, Y., Street, R., and Koseff, J.: 1994, A non-staggered grid, fractional step method for time-dependent incompressible navier-stokes equations in

OCR for page 579

About this PDF file: This new digital representation of the original work has been recomposed from XML files created from the original paper book, not from the original typesetting files. Page breaks are true to the original; line
lengths, word breaks, heading styles, and other typesetting-specific formatting, however, cannot be retained, and some typographic errors may have been accidentally inserted. Please use the print version of this publication as
the authoritative version for attribution.
wake (x=0 is the inflow boundary)
Figure 1: The schematic of the flat plate wake
LARGE-EDDY SIMULATIONS OF TURBULENT WAKE FLOWS
Figure 2: Temporal history of the instantaneous velocity components at different points on the center line of the flat plate
587

OCR for page 579

About this PDF file: This new digital representation of the original work has been recomposed from XML files created from the original paper book, not from the original typesetting files. Page breaks are true to the original; line
lengths, word breaks, heading styles, and other typesetting-specific formatting, however, cannot be retained, and some typographic errors may have been accidentally inserted. Please use the print version of this publication as
the authoritative version for attribution.
Ramaprian, et al. (1981)
Figure 3: Centerline velocity in the flat plate wake
LARGE-EDDY SIMULATIONS OF TURBULENT WAKE FLOWS
Figure 5: Kinetic energy profile along the center line in the wake of flat plate
Figure 6: Kinetic energy profile at x=381mm (x/L= 0.2) in the wake of the flat plate
Figure 4: U-component of the mean-velocity profile in the wake of the flat plate The experimental results are from
588

OCR for page 579

About this PDF file: This new digital representation of the original work has been recomposed from XML files created from the original paper book, not from the original typesetting files. Page breaks are true to the original; line
lengths, word breaks, heading styles, and other typesetting-specific formatting, however, cannot be retained, and some typographic errors may have been accidentally inserted. Please use the print version of this publication as
the authoritative version for attribution.
LARGE-EDDY SIMULATIONS OF TURBULENT WAKE FLOWS
Figure 7: Comparison of the turnulence structures between LES and experiments
589

OCR for page 579

About this PDF file: This new digital representation of the original work has been recomposed from XML files created from the original paper book, not from the original typesetting files. Page breaks are true to the original; line
lengths, word breaks, heading styles, and other typesetting-specific formatting, however, cannot be retained, and some typographic errors may have been accidentally inserted. Please use the print version of this publication as
the authoritative version for attribution.
LARGE-EDDY SIMULATIONS OF TURBULENT WAKE FLOWS
Figure 8: The streamwise vorticity (ωx)contours on different y-z plane in the ship wake
Figure 9: The vorticity contours (ωy) on a x-z plane parallel to free surface at y/L=−0.01 in the ship wake
590

OCR for page 579

About this PDF file: This new digital representation of the original work has been recomposed from XML files created from the original paper book, not from the original typesetting files. Page breaks are true to the original; line
lengths, word breaks, heading styles, and other typesetting-specific formatting, however, cannot be retained, and some typographic errors may have been accidentally inserted. Please use the print version of this publication as
the authoritative version for attribution.
LARGE-EDDY SIMULATIONS OF TURBULENT WAKE FLOWS
Figure 11: The same as Fig. 9 but double grid number in streamwise direction
Figure 10: The streamwise velocity contours on a x-z plane parallel to free surface at y/L=−0.01 in the ship wake
591

OCR for page 579

About this PDF file: This new digital representation of the original work has been recomposed from XML files created from the original paper book, not from the original typesetting files. Page breaks are true to the original; line
lengths, word breaks, heading styles, and other typesetting-specific formatting, however, cannot be retained, and some typographic errors may have been accidentally inserted. Please use the print version of this publication as
the authoritative version for attribution.
LARGE-EDDY SIMULATIONS OF TURBULENT WAKE FLOWS
Figure 14: The velocity vectors on y-z plane at x/L= 0.6 in the ship wake
Figure 13: The velocity vectors on y-z plane at x/L= 0.2 in the ship wake
Figure 12: Comparison of resolved turbulence kinetic energy for different computing cases
Figure 15: The velocity vectors of finer grid solution on y-z plane at x/L=0.2 in the ship wake
592

OCR for page 579

About this PDF file: This new digital representation of the original work has been recomposed from XML files created from the original paper book, not from the original typesetting files. Page breaks are true to the original; line
lengths, word breaks, heading styles, and other typesetting-specific formatting, however, cannot be retained, and some typographic errors may have been accidentally inserted. Please use the print version of this publication as
the authoritative version for attribution.
LARGE-EDDY SIMULATIONS OF TURBULENT WAKE FLOWS
Figure 18: The velocity vectors on y-z plane at x/L= 1.4 in the ship wake
Figure 17: The velocity vectors on y-z plane at x/L= 1.0 in the ship wake
Figure 16: The velocity vectors and the enlargement of area A on y-z plane at x/L=1.2 in the ship wake
593

OCR for page 579

About this PDF file: This new digital representation of the original work has been recomposed from XML files created from the original paper book, not from the original typesetting files. Page breaks are true to the original; line
lengths, word breaks, heading styles, and other typesetting-specific formatting, however, cannot be retained, and some typographic errors may have been accidentally inserted. Please use the print version of this publication as
the authoritative version for attribution.
directions)
LARGE-EDDY SIMULATIONS OF TURBULENT WAKE FLOWS
Figure 22: The vorticity contours (ωx) on y-z plane at x/L=1.2 in the ship wake
Figure 21: The vorticity contours (ωx) on y-z plane at x/L=0.6 in the ship wake
Figure 19: The vorticity contours (ωx) on y-z plane at x/L=0.2 in the ship wake
Figure 20: The vorticity contours (ωx) on y-z plane at x/L=0.2 in the ship wake (finer grids are applied on both y and z
594

OCR for page 579

lengths, word breaks, heading styles, and other typesetting-specific formatting, however, cannot be retained, and some typographic errors may have been accidentally inserted. Please use the print version of this publication as
About this PDF file: This new digital representation of the original work has been recomposed from XML files created from the original paper book, not from the original typesetting files. Page breaks are true to the original; line
LARGE-EDDY SIMULATIONS OF TURBULENT WAKE FLOWS 595
DISCUSSION
M.Hyman
Coastal Systems Station, USA
This paper represents the next stage in an effort to obtain high resolution flow computations around and downstream
of a surface ship. Many parabolic algorithms exist to compute wake flow and more recently, Dommermuth (1994) applied
LES to the wake of a ship. That temporal simulation could, strictly speaking, begin some distance downstream of the
stern at a location where the flow was decaying fairly slowly. The present work reports a spatial simulation which can, in
theory, be applied very near the stern and extend as far downstream as computational resources permit. The method can
be initiated by applying the authors' random flow generation (RFG) methodology. The RFG allows one to take RaNS
simulation results and create a very plausible estimate of the instantaneous turbulent flow field. Once this flow field is
available, a LES calculation of the near wake can be undertaken. The authors show several interesting results from
applying their code to compute the wake of a flat plate and of a surface ship.
While it is clear that this work is in its' early stages, there are several issues that should be considered in the course of
future work. The near wake of a ship presumably contains a wide range of length scales from flow off the hull boundary
layer, separated stern flow and propulsor flow. This complex flow is further complicated by breaking stern waves at
higher Froude numbers. I would like to ask the authors how they would propose to characterize this flow by a single
length scale, what this length scale should be and how this length scale should relate to grid resolution for LES
calculations. In addition, I would like to ask if the authors' experience with LES calculations using the QUICK scheme
(and lack of SGS model) could be related to grid resolution.
AUTHOR'S REPLY
We appreciate Dr. Hyman's summary and comments on this paper. What follows is a brief response to his questions.
QUESTIONS
Question 1: “I would like to ask the authors how they would propose to characterize this flow by a single length-
scale, what this length-scale should be and how this length-scale should relate to grid resolution for LES calculations.”
Answer.
Though a constant was used in this study, in our method it is not necessary to use a single length-scale. In fact, in the
current version the algorithm is already implemented with the length scale parameter as a function of space which can be
computed from k and ε or k and ω. It adds practically no extra computational burden to the procedure. There can be a
slight loss of accuracy in satisfying the instantaneous continuity equation in a flow field generated with the RFG
procedure associated with using inhomogeneous turbulent length-scale. However, for the purposes of inlet plane
initialization a small violation of continuity can be tolerated since the flow solver will automatically adjust the velocities
in the next plane such that the flow field satisfies continuity. The length scale used in the RFG procedure is currently not
related to grid resolution. This might be a good idea to pursue in the future.
Question 2: I would like to ask if the authors' experience with LES calculations using the QUICK scheme (and lack
of SGS model) could be related to grid resolution.
Answer:
From Fig. 12 in the paper, it can be seen that there is no significant difference in the results when QUICK scheme is
used with or without SGS model. Moreover, the resolved turbulent kinetic energy is lower than that of central
differencing with Smagorinsky model. This means that QUICK scheme gives even higher diffusion than Smagorinsky
model. Theoretically, the numerical diffusion increases with increasing grid size, so the lower resolution of scale in the
LES calculations using the QUICK scheme (and lack of SGS model) is most probably related to grid resolution.
However, at this point we could not make assertion with regards to how strong the resolution depends on the grid size. A
new run with a finer grid resolution is underway to investigate this issue.
the authoritative version for attribution.

OCR for page 579

lengths, word breaks, heading styles, and other typesetting-specific formatting, however, cannot be retained, and some typographic errors may have been accidentally inserted. Please use the print version of this publication as
About this PDF file: This new digital representation of the original work has been recomposed from XML files created from the original paper book, not from the original typesetting files. Page breaks are true to the original; line
LARGE-EDDY SIMULATIONS OF TURBULENT WAKE FLOWS 596
DISCUSSION
L.Davidson
Charmers Univ. of Technology, Sweden
The paper presents LES of compressible wake flow. The method is applied to the wake after a flat plate and after a
ship model. The computations are started downstream of the bodies. At the inlet plane, turbulent inlet boundary
conditions are generated by specifying a turbulent spectrum; the procedure is called a “Random Flow Generation” (RFG)
technique.
The authors claim that the RFG technique is new. Are they aware of the method develop by the group at ECL [1]
used in Aeroacoustics?
How are the results presented in Figs. 3–6 compared to what can be achieved with RANS (either eddy-viscosity
models or second-moment closure)? Maybe the authors can include some comments?
On p. 4, col. 2, the authors comment upon the waviness of the velocity profile in Fig. 3. They say that this is because
they have averaged only along the centerline instead of in the whole plane. This is probably correct, but the same effect
would of course be achieved if they increased the time integration.
The results show in Figs. 3–6 have been obtained with the QUICK scheme and no SGS model. The authors say that
they also have carried out simulations with central differencing and the Smagorinsky model. What about these results? It
would be interesting to see a comparison.
A lot of contour plots are presented. I don't think this is very interesting. I would prefer to see more quantitative
results. Furthermore, I'm surprised that in the wake flow behind the ship model no comparison is made with experiments.
Why? Doesn't it exist any experimental data for the ship model wake? If so, why was it chosen in the first place?
REFERENCES
1. W.Bechara and C.Bailly and P.Lafon, “Stochastic Approach to Noise Modelling for Free Turbulent Flow”, AIAA J., volume 32, Number 3, pp. 455–
463, 1994
L.Davidson Chalmers Univ. of Technology, Sweden
2. How are the results presented in Figs. 3–6 compared to what can be achieved with RANS (either eddy-viscosity models or second-moment closures)?
3. The results shown in Figs. 3–6 have been obtained with the QUICK scheme and no SGS model. The authors say that they also have carried out
simulations with central differencing and the Smagorinsky model. What about these results? It would be interesting to see a comparison.
4. I'm surprised that in the wake flow behind the ship model, no comparison is made with experiments. Why?
AUTHOR'S REPLY
We would like to point out that there must be a typographical error in the first paragraph of reviewer's comments.
This paper presents LES of incompressible rather than compressible flow.
We agree with the reviewer that one can find numerous applications of Kraichnan's method pertaining to dispersed
phase modeling, turbulence-generated noise etc. Our RFG technique is also based on the Kraichnan method. However,
the original method was modified by us so as to handle the case of anisotropic turbulence while preserving the continuity
of the flow field. Another novelty is the way the technique is implemented in conjunction with LES, and Lagrangian
particle dynamics routines, which enable an efficient and flexible flow-field generation by matching all the Reynolds
stress components measured or obtained from RANS (see for details Smirnov, et al. 2000). In this respect we do not claim
RFG to be a new method, but rather a new technique used to augment the LES and particle dynamics computations.
The comment of the reviewer about averaging may have resulted from some misunderstanding. Probably we did not
make it clear enough in the paper that the velocity profile in Fig. 3 was not a result of space averaging along the
centerline, but rather of time averaging. In this context there is no disagreement with the reviewer's statement.
Answer to Question 2: Ramaprian, et al. (1981) performed RANS calculations with the k-ε model. Their results
agree fairly well with their measurements. However, no matter what the
the authoritative version for attribution.

OCR for page 579

lengths, word breaks, heading styles, and other typesetting-specific formatting, however, cannot be retained, and some typographic errors may have been accidentally inserted. Please use the print version of this publication as
About this PDF file: This new digital representation of the original work has been recomposed from XML files created from the original paper book, not from the original typesetting files. Page breaks are true to the original; line
LARGE-EDDY SIMULATIONS OF TURBULENT WAKE FLOWS 597
outcome of this comparison may be, LES technique provides extra information on unsteady large-scale component of the
flow-field, and thereby provides a more consistent and necessary approach for modeling Lagrangian bubble dynamics in
the wake which is the ultimate goal of the present work. This, in our opinion, gives LES a definite advantage over RANS
methods in this particular application area.
Answer to Question 3: Fig. 1 included here shows the streamwise velocity at the center line calculated by using
QUICK scheme without subgrid sale model and central differencing with Smagorinsky mode. As expect, the central
differencing scheme gives better results compared to the QUICK scheme. This is also true for the turbulent intensities
(not shown here).
Answer to Question 4: Measurements of the flow-field in the wake of Navy ship model 5415 is an on-going work
and the experimental results are not available at this time. In this respect our ship-wake results represent a prediction
rather than a post-diction. Our results will be compared with experimental data as they become available. On the other
hand our flat-plate computations presented in this paper were specifically done to validate our model and provide
justification for extending its application to the ship-wake. It must be pointed out that the LES predictions were done for a
shear free flat wall instead of wavy free surface. The influence of waves on turbulence is usually not negligible.
the authoritative version for attribution.

OCR for page 579

About this PDF file: This new digital representation of the original work has been recomposed from XML files created from the original paper book, not from the original typesetting files. Page breaks are true to the original; line
lengths, word breaks, heading styles, and other typesetting-specific formatting, however, cannot be retained, and some typographic errors may have been accidentally inserted. Please use the print version of this publication as
the authoritative version for attribution.
LARGE-EDDY SIMULATIONS OF TURBULENT WAKE FLOWS
number 11215 in FEDSM2000, Boston, Massachusetts, 2000.
REFERENCE:
Fig. 1. Modified comparison of predicted mean streamwise velocity at center line with measurements.
1. A.Smirnov, S.Shi, and I.Celik. Random flow simulations with a bubble dynamics model. ASME Fluids Engineering Division Summer Meeting,
598