Sie sind auf Seite 1von 9

J. Wind Eng. Ind. Aerodyn.

144 (2015) 7078

Contents lists available at ScienceDirect

Journal of Wind Engineering


and Industrial Aerodynamics
journal homepage: www.elsevier.com/locate/jweia

LES modelling of unsteady ow around the Silsoe cube


Peter Richards n, Stuart Norris
Department of Mechanical Engineering, University of Auckland, Auckland 1142, New Zealand

art ic l e i nf o a b s t r a c t

Large Eddy Simulation is used to model the unsteady ow over the Silsoe 6 m cube with the wind
Keywords: perpendicular to one face. The mean, standard deviation, maximum and minimum pressure coefcients
Large Eddy Simulation are shown to be in reasonable agreement with published data. In regions affected by building induced
Cube pressure uctuations the standard deviation coefcient is higher than in full-scale due to the lower
Vortex dynamics turbulence intensity in the LES model. Conditional averaging is used to highlight the sequence of pres-
sure changes that occur around strong suction spikes on both the sidewalls and roof. The LES model is
shown to reproduce the pattern observed in full-scale. Flow visualisations reveal that these events are
associated with the formation of a strong vortex on the windward half of the sides or roof, which is then
shed and carried downstream.
& 2015 Elsevier Ltd. All rights reserved.

1. Introduction the symbols are the data from one 12 min run, while the solid
lines are short Fourier series tted to the data by using a least
The Silsoe 6 m cube, see Fig. 1, was constructed in order to squares method. The three dashed lines represent quasi-steady
provide a facility for fundamental studies of the interactions expectations which are derived from the curve tted to the mean
between the wind and a structure. This shape was chosen since it data. For full details of the tting techniques and quasi-steady
represents a simplied building, has multiple planes of symmetry modelling see Richards and Hoxey (2012a, b). It may be observed
and in spite of its simplicity still exhibits many of the complex ow that the standard deviation and maximum pressure coefcients
phenomena found on more complex building shapes. Richards are approximately equal to that expected from the quasi-steady
et al. (2001) was the rst of a series of papers which have provided theory; however when the wind direction is in the range 34545
full-scale data together with in-depth analysis of the pressure and the minimum pressure coefcient is consistently more negative
ow elds. Although the 2001 paper only contained limited mean than predicted. In fact the quasi-steady line almost forms a lower
pressure data, it has been used for verication of CFD techniques bound to the measured data. It may be noted that this range of
including Large Eddy Simulations (LES) by Lim et al. (2009) and angles encompasses those where the ow separates from the
Detached Eddy Simulations (DES) by Haupt et al. (2011) and for nearby windward vertical edge and probably reattaches to the face
evaluation of experimental facilities such as the Wall of Wind (Aly containing H2 at some point. Richards and Hoxey (2012b) suggest
et al., 2011). More recently Richards and Hoxey (2012a, b) have that these high suctions are due to the dynamic response of the
provided standard deviation, maximum and minimum pressure separating and reattaching ow which periodically rolls up into an
coefcient data. The research reported in this paper uses LES to intense vortex. A similar pattern is observed with roof Tap V8 with
understand some of the unsteady phenomena observed at full- wind directions around 90. The form of the pressure coefcients
scale. used here follows Richards and Hoxey (2012a), who recommend

Cp () =
p
, C p~ () =
p
, C
p ( ) =
p
, p () =
and C p

q q q q (1)
2. Unsteady pressure data
where p is the surface pressure and q is the reference dynamic
Fig. 2 is a typical example of the full-scale pressure data from pressures measured at cube height in the approach ow. sp is the
the Silsoe cube for one of the mid-height wall taps. In each graph standard deviation of pressure, while p,
p and p are the mean,
maximum and minimum values of pressure respectively, with
n
Corresponding author. similar meanings when applied to the reference dynamic pressure.
E-mail address: pj.richards@auckland.ac.nz (P. Richards). This scaling was chosen by Richards and Hoxey (2012a) since these

http://dx.doi.org/10.1016/j.jweia.2015.03.018
0167-6105/& 2015 Elsevier Ltd. All rights reserved.
P. Richards, S. Norris / J. Wind Eng. Ind. Aerodyn. 144 (2015) 7078 71

coefcients are all of comparable magnitude and are much less The majority of calculations were made on a 15 h  10 h  6.67 h
sensitive to changes in conditions in comparison to the more usual computational domain (i.e.: 90 m long, 60 m wide and 40 m high),
coefcients, where all pressure statistics are normalised by the using a 246  123  186 mesh with a resolution varying from 0.02 m
mean dynamic pressure. Further explanation can be found in that at the wall, to 0.5 m in the far eld (see Fig. 3). The cube was
paper. modelled as being aerodynamically smooth, while the ground was
modelled with a roughness of z0 0.01 m. The top of the domain had
a free-slip boundary condition applied, and the two side boundaries
were periodic. The velocity was prescribed at the upstream bound-
3. Computational technique
ary, with a mean velocity of 6 m/s at a height of 6 m, and a pre-
scribed pressure outlet boundary condition was applied at the
The simulations were performed using an in-house massively
downstream boundary.
parallel Large Eddy Simulation code, SnS (Armeld et al., 2002;
The inlet boundary condition at the upstream boundary required
Norris, 2001), which uses an incompressible non-staggered nite
the denition of an atmospheric boundary layer, including both the
volume formulation based on a structured Cartesian mesh. Second
mean prole and the temporal and spatial uctuations of the ow.
order central differences were employed for approximating the
This was generated using a precursor calculation of the ow in a
advective and diffusive uxes in the momentum equations, and an 720 m  40 m  60 m empty domain with periodic boundary con-
AdamsBashforth fractional step solver was employed, which ditions in the streamwise direction, the ow being driven by a
gives a solver that has been shown to be second order accurate in pressure gradient in the streamwise direction. The ow at the low x
both space and time (Armeld and Street, 2002). The fractional boundary was sampled at 0.05 s intervals for a period of 75 min. This
step method negated the need for iterative coupling at each time was broken into 12.5 min blocks of data, which were used for 6 runs
step, allowing the efcient calculation of transient ows. Subgrid modelling the ow around the cube. The ow was allowed to settle
scale turbulence is modelled using the standard Smagorinsky for the rst 30 s of each run, and then 12 min of data was recorded.
model (Smagorinsky, 1963) with Cs 0.18, damped at the wall Fig. 4(a) shows the velocity prole created by the precursor cal-
using the method of Mason and Thompson (1992), with a rough culation, which up to a height z4 h is approximately equal to a
wall function (Mason and Callen, 1986) being applied at the simple log law with the prescribed roughness length z0 0.01 m,
ground boundary. which is typical for the full-scale site. The slightly higher velocities at

Fig. 1. (a) The Silsoe 6 m cube with the metal plates around each pressure tap clearly visible and (b) the pressure tap numbering system used in this paper.

Fig. 2. Full-scale pressure coefcients for Tap H2, at mid height and 0.24 h from one vertical edge, (a) mean and standard deviation and (b) maximum and minimum.
72 P. Richards, S. Norris / J. Wind Eng. Ind. Aerodyn. 144 (2015) 7078

larger heights result from the decreasing shear stress with height. constant, because the continuity equation requires that any var-
Since the precursor calculation is an example of a pressure driven iation in the total inlet ow is matched by an instantaneous
boundary layer the Reynolds shear stress decreases almost linearly change in the ow rate throughout the domain. Therefore large
with height. scale gusts with a length scale greater than the domain size cannot
Fig. 4(b) shows the non-dimensional spectra at z h. It can be be modelled in the inlet turbulence. As a result the cube height
seen that while the spectral densities approximately match at longitudinal turbulence intensity is only 11% in the LES model
mid-range frequencies there is signicant ltering at both high whereas in full-scale it is typically 20%.

Fig. 3. (a) The 6 m cube within the 90 m long, 60 m wide and 40 m high ow domain, and (b) a detail showing the mesh on the cube.

Fig. 4. (a) Velocity prole generated by the precursor calculation, and (b) the three component spectra at cube height generated by LES and measured in full-scale (FS).

and low frequencies. The high frequency ltering results from the The code is parallelised using MPI, and was run on the NeSI Intel
LES simulation being unable to represent eddies smaller than the Xeon based computational cluster based at the University of Auck-
grid, while the low frequency end is limited by the size of the land. The momentum equations were solved using a Jacobi solver,
domain. It should be noted that the length scale of perturbations with the pressure correction equation being solved using a Bi-Con-
in the inlet ow is limited by the use of an incompressible for- jugate Gradient Stabilised (BICGSTAB) solver (van der Vorst, 1981)
mulation. The mass ow rate through the inlet boundary has to be preconditioned using block SIP (Stone, 1968). When running on 48
P. Richards, S. Norris / J. Wind Eng. Ind. Aerodyn. 144 (2015) 7078 73

cores, 12.5 min of simulation time was calculated in approximately behind the cube is very similar to the full-scale measurement of 1.4h
4 days of wallclock time. from the rear surface (Hoxey et al., 2002).

4. Pressure coefcients and mean ow elds 5. Short duration pressure spikes

Fig. 5 shows the pressure coefcient for both the vertical and 5.1. Full-scale observations
horizontal ring of pressure taps determined from full-scale mea-
surements and from LES for a wind direction of 90. The agree- As noted in Section 2 for Tap H2 when the wind angle is near 0,
ment between the experimental and simulated pressure coef- which is equivalent to Tap H8 with wind angle 90 as modelled here,
cients is not perfect, but displays similar trends. The main differ- the minimum pressure is more negative than might be expected
ences are as follows: from the quasi-steady analysis. Observation of time histories for the
horizontal taps along the side, as illustrated in Fig. 7(a), shows that
 The mean positive pressures on the windward face are some- pressure peaks appear to be short duration spikes.
what larger in the CFD model, however the maximum pres- Expansion of the time history around the time of the lowest
sures are a better match. The reason for this discrepancy is recorded suction, Fig. 7(b), which occurred at Tap H8 at time 601.7 s,
currently unknown. suggests a pattern of pressure changes where the pressures at Taps
 The minimum pressures on the windward face are not as low H9-11 initially go positive and then rapidly decrease consecutively. A
as observed in full-scale. This is probably due to the smaller similar pattern is shown in Fig. 8 for the roof centreline pressures,
range of wind directions in the LES runs. where the expansion (Fig. 8(b)) shows a clearer sequence of events.
 The maximum pressures on the roof and side faces are not as The form of these spikes can be claried by using conditional
positive as observed in full-scale. This may also be related to averaging. This data analysis process seeks to highlight the under-
the range of wind directions occurring during each 12 min run. lying systematic patterns by averaging out random variations. This
In the LES modelling the lateral turbulence is equivalent to a was implemented by selecting a threshold level for pressure at the
standard deviation of wind directions of only 3 in contrast to trigger tap (either H8 or V8), which was usually set at the mean
the typical full-scale value of 10. pressure coefcient times the peak dynamic pressure observed
 The standard deviation coefcients for both the roof and side during the whole record. In this way the identied events are those
faces are signicantly greater in the LES runs than in full-scale. which are more negative than expected from simple quasi-steady
Since the form of pressure coefcient used is the ratio of the processes. In addition the trigger times were required to be the point
standard deviation of the surface pressure to the standard of lowest pressure within a 72 s interval. This process typically
deviation of the reference dynamic pressure, this coefcient is identied 1030 spikes in a record. Once these have been identied
sensitive to the lower level of turbulence that exists in the LES the signals are averaged at the peak of each spike and at equal
models. This means that any building induced turbulence increments in time ahead and after the peak. Fig. 9(a) shows these
contributes a greater proportion of the total standard deviation averages 2 s either side of all pressure spikes at Tap H8 with a peak
and hence tends to increase this coefcient. below 340 Pa. A clear sequence of events can be discerned. Just
before each peak at H8 a weaker suction peak is observed at H7,
The observed ow patterns are also similar to those observed in while following the peak at H8 the other sidewall taps experience a
experiment. Fig. 6 shows the mean stream lines around the cube. It low suction followed by a rapid change to higher suction. In some
may be observed that both the ow over the roof and around the cases the pressures at Taps H1012 are positive for short periods
sides detaches at the windward edge but does reattach. The reat- during these events, as seen in Fig. 7(b), but the conditional averages
tachment point in both cases is around 90% of the way across the remain negative throughout. This gure shows that the series of
faces, which is somewhat larger than the 57% for the roof, reported events lasts about 1.0 s, from about 0.33 s before the peak until 0.66 s
by Hoxey et al. (2002), and the 5083% for the sides, reported by after. The conditionally averaged reference dynamic pressure curve
Richards and Hoxey (2002). The streamlines show reversed ow suggests that these peak events may be weakly linked to a gust but
near the ground upstream of the cube extending about 0.9h (where do not appear to be immediately triggered by a gust of similar
h is the cube height) upstream, which is similar to the 0.75h obser- duration. However the fact that the conditionally averaged dynamic
vation by Hoxey et al. (2002). The length of the separation bubble pressure at the time of the spike, at around 200 Pa, is well above the

Fig. 5. Pressure Proles for (a) the vertical ring and (b) the horizontal ring of pressure taps, for wind direction 90.
74 P. Richards, S. Norris / J. Wind Eng. Ind. Aerodyn. 144 (2015) 7078

12 min average of 126 Pa, suggests that the strongest peaks occur taps in Fig. 11(b) where the trigger used was a minimum pressure
during periods of stronger wind. peak at Tap V8 below 38 Pa. A more negative threshold was used
Fig. 9(b) reveals a very similar sequence for the roof centreline taps because of the slightly lower pressures on the roof, but this still
however these are of lower magnitude and longer duration than with meant that the output is averaged over 44 events.
the sidewall taps since the wind speed was considerably lower during The general forms of both sequences in Fig. 11 are very similar
the recording period. For these records the threshold level was around to each other and to that revealed in the full-scale results of Fig. 9.
61 Pa. Once again the mean dynamic pressure around the time of The two most noticeable differences are as follows:
these spikes, at 28.8 Pa, is well above the full-record mean of 22 Pa.
1. The duration of the spike in the LES model is markedly longer
5.2. Pressure spikes in the LES simulations than the full-scale wall tap results in Fig. 9(a) and somewhat
longer than the full-scale roof tap data in Fig. 9(b). This is due
Fig. 10(a) shows the time histories for the sidewall taps from to the higher wind speed around the time of the spikes in full-
one LES run. The general behaviour is similar to that in Figs. 7(a) and scale (16.8 m/s for the wall taps and 6.9 m/s for the roof taps) in
8(a) although the dynamic pressure is more uniform due to the lower comparison with the LES model value (5.6 m/s). The observed
levels of low frequency turbulence in the model. As a consequence sequences are thought to be caused by the formation of a tight
the suction spikes at the sidewall taps are not as variable in magni- vortex near the windward edge of the roof or sidewall, which is
tude. Expansion of the time series around the time of the most then swept downstream. The speed of movement of the
negative peak pressure at Tap H8 (146.4 s) in Fig. 10(b) reveals a released vortex is likely to be directly proportional to the wind
sequence of pressure variations very similar to that seen by condi- speed.
tional averaging in Fig. 9. Again it may be observed that this extreme 2. The conditionally averaged dynamic pressure around the time of
spike occurred during a period when the dynamic pressure was the peaks in the LES model is close to the 6 run mean of 19 Pa.
above the 12 min average of 19 Pa. This may suggests that the strength of the spikes in the LES
The pressure sequence is even clearer when conditional averaging modelling are almost independent of the simultaneous dynamic
is applied across all six LES runs. For the sidewall taps in Fig. 11(a) the pressure whereas the full-scale results suggest that stronger
trigger used was a pressure peak at Tap H8 below  35 Pa which spikes occurred during periods of stronger wind. However the
occurred 38 times. A very similar pattern is revealed with the roof fact that the strongest spike in Fig. 10 occurred when the dynamic

Fig. 6. LES mean streamlines around the cube, (a) isometric view and (b) plan view.

Fig. 7. Full-scale pressure data from one 12 min run during which the wind direction was 90.3; (a) time histories for q and the side pressures and (b) the same time
histories expanded around the time of the lowest suction which occurred at Tap H8 at time 601.7 s.
P. Richards, S. Norris / J. Wind Eng. Ind. Aerodyn. 144 (2015) 7078 75

pressure was higher that 19 Pa may mean that there is still some  In this non-dimensional form the duration of the sequence of
relationship but it is less obvious. This observation is probably events is similar in all cases.
related to the inlet turbulence not modelling the long length scale  The time between the negative peak at Tap H7 (or V7) to the
energetic gusts present in the atmospheric boundary layer. positive going peak at Tap H12 (or V12) is about one unit of
normalised time, supporting the idea that the sequence is
associated with a vortex forming and being carried along the
5.3. Normalising the conditionally averaged results sidewall (or across the roof) by the wind.
 In both the full-scale and LES data, the peak pressures on the
As noted in the previous sections both the strength of the peak roof at Tap V8 are slightly stronger than on the sidewall at Tap
pressure and the duration of the sequence of events appear to be H8. This may be related to the slightly higher mean pressure
related to the wind strength. These relationships are illustrated in coefcients for the roof.
Fig. 12 where the pressures are expressed as pressure coefcients
 The full-scale results at Taps H8 and V8 show a somewhat
and the time normalised in the form tU/h, where h is the height of sharper peak than calculated by LES. As noted in Section 4 the
the cube. Investigation of various options showed that the most length of the mean separation on both the sidewall and roof is
relevant dynamic pressure was the conditionally averaged refer- larger in the LES modelling than observed in full-scale. The
ence dynamic pressure at the time of the peak pressure q(0) and smaller recirculating ow in full-scale can be expected to create
a more intense but shorter duration peak.
that the appropriate wind speed to use in calculating the nor-
malised time was the associated speed calculated from

U= 2q (0) / (2)
6. Flow visualisation
where is the air density. It may be noted that in this form one
unit of normalised time is that required for the reference wind to Fig. 13 shows the difference in ow pattern between an instant
move one building height. when the pressure at Tap V8, centre of the windward half of the
From the data presented in Fig. 12 it may be noted: roof, is near its minimum in contrast to that when it is near its
maximum. It can be seen in Fig. 13(a) that in the rst case there is
 The general pattern of events is similar, though not identical, in a strong vortex over the windward half of the roof resulting in low
all cases. pressures on this region, as illustrated by the light coloured area

Fig. 8. Full-scale pressure data from one 20 min run during which the wind direction was 93.5; (a) time histories for q and the roof centreline pressures and (b) the same
time histories expanded around the time of the lowest suction which occurred at Tap V8 at time 1079.7 s.

Fig. 9. (a) Conditionally averaging of the full-scale horizontal sidewall taps around pressure peaks at H8 o  340 Pa and (b) conditional averaging of the full-scale roof
centreline taps around pressure peaks at V8 o  61 Pa.
76 P. Richards, S. Norris / J. Wind Eng. Ind. Aerodyn. 144 (2015) 7078

Fig. 10. LES pressure data from one 12 min run with wind direction 90; (a) time histories for q and the sidewall pressures and (b) an expanded view around the time
when the lowest pressure occurred at Tap H8.

Fig. 11. Conditional averaging of the LES pressures for (a) the sidewall Taps H7H12 triggered by pressure peaks at H8 o  35 Pa and (b) the roof Taps V7V12 triggered
by pressure peaks at Tap V8 o  38 Pa.

Fig. 12. Conditionally averaged pressure coefcients as a function of normalised time for (a) the sidewall taps and (b) the roof centreline taps from both full-scale (dashed
lines) and LES (solid lines).

on the windward half of the roof surface. In both Figs. 13 and 14 where the reference dynamic pressure is the mean dynamic pressure
the instantaneous pressure coefcient is calculated from at the inlet at cube height, zh. At the same time the pressure colour
on the leeward half of the roof is a medium grey indicating neutral
p and possibly slightly positive pressures. In contrast Fig. 13(b) shows
Cp =
q (3) the vortex is more of a horseshoe shape extending from the sides
P. Richards, S. Norris / J. Wind Eng. Ind. Aerodyn. 144 (2015) 7078 77

Fig. 13. Surface pressure coefcients and streamlines for two instances: (a) an instant when the pressure at V8 was near its minimum value and (b) when it was near
its maximum. (For interpretation of the references to colour in this gure, the reader is referred to the web version of this article.)

Fig. 14. Streamlines and pressure coefcient contours on the central vertical plane: (a) an instant when the pressure at V10 was near its maximum value and (b) when it was
near its minimum. (For interpretation of the references to colour in this gure, the reader is referred to the web version of this article.)

over the leeward half of the roof. The roof pressure contours are to have lifted away from the surface the surface pressures are not
more uniform showing weak suctions on most areas with the as low as occurs in Fig. 14(a).
exception of some light areas about half way across on both edges. These ow visualisations support the concept that the pressure
The low pressure region on the near side appears to link with a low sequences observed, both in full-scale and in the LES modelling,
pressure line on the sidewall which runs down towards the wind- result from an intense vortex forming on the windward half of
ward lower corner at the ground level. either the roof or sidewall. This not only creates high suctions near
Since roof Tap V10, which is just past the mid-point of the roof, its core but also weak suctions, and in some cases positive pres-
has a negative mean and minimum pressure coefcient, but a sures, around its leeward edge. This vortex then breaks free
positive maximum, the ow elds have also been extracted for sweeping over the surface creating a positive going pulse followed
instants when the pressure at V10 is near its maximum and when by a rapid change to a suction pulse. As the vortex moves down-
it is near its minimum. In Fig. 14(a) it may be noted that the yellow stream it lifts away from the surface and so the surface pressures
colour represents a weak positive pressure and there is therefore a are weaker further downstream.
small area of positive pressure on the leeward half of the roof at
this instant in time. Also at this instant the suction on the wind-
7. Conclusions
ward half is very low and approaches a pressure coefcient of
almost  2.0, which is obviously linked to the intense vortex that The unsteady ow around the Silsoe cube for a wind direction
exists at this time. It appears that the positive pressure is the result of 90, perpendicular to one face, has been modelled using Large
of ow around the vortex impinging onto the surface. In contrast Eddy Simulation. The mean, standard deviation, maximum and
Fig. 14(b) shows an instant when the vortex has broken away from minimum pressure coefcients for 42 points on the building sur-
the windward edge and is passing over Tap V10, making the local face have been compared to published full-scale data and gen-
pressure quite low. However since the core of the vortex appears erally reasonable agreement is achieved. In some regions where
78 P. Richards, S. Norris / J. Wind Eng. Ind. Aerodyn. 144 (2015) 7078

building induced pressure uctuations are expected the standard Haupt, S.E., Zajaczkowski, F.J., Peltier, L.J., 2011. Detached eddy simulation of
deviation coefcient is higher than in full-scale, probably as a atmospheric ow about a surface mounted cube at high Reynolds number.
J. Fluids Eng.: Trans. ASME 133 (3) (Article number: 031002).
result of the small solution domain which limits the large scale, Hoxey, R.P., Richards, P.J., Short, J.L., 2002. A 6 m cube in an atmospheric boundary
low frequency turbulence. layer ow, Part 1. Full-scale and wind tunnel results. J. Wind Struct. 5 (24),
The modelling has been used to investigate the suction spikes 165176.
Lim, H.C., Thomas, T.G., Castro, I.P., 2009. Flow around a cube in a turbulent
that occur on the roof and sidewalls. Conditional averaging was boundary layer: LES and experiment. J. Wind Eng. Ind. Aerodyn. 97 (2), 96109.
used to highlight the sequence of pressure changes which occurs Mason, P.J., Callen, N.S., 1986. On the magnitude of the subgrid-scale eddy coef-
on the sidewall or roof around these high suction events. It is cient in large-eddy simulations of turbulent channel ow. J. Fluid Mech. 162,
439462.
shown that the LES recreates the same phenomena, which when Mason, P.J., Thomson, D.J., 1992. Stochastic backscatter in large-eddy simulations of
normalised using the mean dynamic pressure at the time of the boundary layers. J. Fluid Mech. 242, 5178.
peak and the corresponding wind speed exhibit similar patterns Norris, S.E., 2001. A Parallel NavierStokes Solver for Natural Convection and Free
Surface Flow (Ph.D. thesis). University of Sydney.
and sequences. Flow visualisation shows that the associated
Richards, P.J., Hoxey, R.P., 2002. Unsteady ow on the sides of a 6 m cube. J. Wind
velocity eld can be characterised as the formation of a strong Eng. Ind. Aerodyn. 90 (12), 18551866.
vortex on the windward half of the sidewall or roof, which is then Richards, P.J., Hoxey, R.P., Short, L.J., 2001. Wind pressures on a 6 m cube. J. Wind
shed and carried downstream. Eng. Ind. Aerodyn. 89 (1415), 15531564.
Richards, P.J., Hoxey, R.P., 2012a. Pressures on a cubic buildingPart 1: full-scale
results. J. Wind Eng. Ind. Aerodyn. 102, 7286. http://dx.doi.org/10.1016/j.
jweia.2011.11.004.
References Richards, P.J., Hoxey, R.P., 2012b. Pressures on a cubic buildingPart 2: quasi-steady
and other processes. J. Wind Eng. Ind. Aerodyn. 102, 8796. http://dx.doi.org/
10.1016/j.jweia.2011.11.003.
Aly, A.M., Gan Chowdhury, A., Bitsuamlak, G., 2011. Wind prole management and Smagorinsky, J., 1963. General circulation experiments with the primitive equa-
blockage assessment for a new 12-fan wall of wind facility at FIU. Wind Struct. tions. Mon. Weather Rev. 91, 99164.
14 (4), 285300. Stone, H.L., 1968. Iterative solution of implicit approximations of multidimensional
Armeld, S.W., Norris, S.E., Morgan, P., Street, R.A., 2002. Parallel non-staggered partial differential equations. SIAM J. Numer. Anal. 5, 530558.
NavierStokes solver implemented on a workstation cluster. In: Proceedings van der Vorst, H.A., 1981. Iterative solution methods for certain sparse linear sys-
of the Second International Conference on Computational Fluid Dynamics. tems with a non-symmetric matrix arising from PDE-problems. J. Comput.
Sydney, Australia, pp. 3045. Phys. 44, 119.
Armeld, S.W., Street, R., 2002. An analysis and comparison of the time accuracy of
fractional-step methods for the NavierStokes equations on staggered grids.
Int. J. Numer. Methods Fluids 38, 255282.

Das könnte Ihnen auch gefallen