2002 - 1-S2.0-S0045793002000671-Main PDF

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 22

Computers & Fluids 32 (2003) 953–974

www.elsevier.com/locate/compfluid

Numerical study of the flow over shallow cavities


P.S.B. Zdanski, M.A. Ortega, Nide G.C.R. Fico Jr. *

Division of Aeronautics, Technological Institute of Aeronautics, SJ Campos 12228-900, Brazil


Received 29 August 2001; received in revised form 4 June 2002; accepted 17 June 2002

Abstract
This paper reports on a series of numerical simulations of both laminar and turbulent flows over shallow
cavities. For the turbulent case the influences of the following parameters were considered: (i) cavity aspect
ratios, (ii) turbulence level of the oncoming flow, and (iii) Reynolds number. Several important results and
conclusions are reported. We have found that for the turbulent case the external flow touches the floor of
the cavity, and this depends on a specific value of each of these parameters. This condition has an important
impact upon convective effects inside the cavity. The mathematical model corresponds to the incom-
pressible, Reynolds-averaged, Navier–Stokes equations plus a high-Reynolds j–e model of turbulence, and
the numerical computation is performed using the SIMPLER algorithm.
Ó 2002 Elsevier Science Ltd. All rights reserved.

Keywords: Shallow cavities; Incompressible flow; Numerical methods

1. Introduction

The upper surface of solar collectors, normally used for water heating purposes, is made of flat
glass pieces. If we introduce vertical wind barriers along the perimeter of the collector (Fig. 1) the
efficiency of the heat absorption improves accordingly, because wind convection is inhibited. This
has been shown experimentally by Gomes [1]. This work has inspired the present research effort.
The main goal here is to simulate numerically the flow along the two-dimensional cavity, which is
the first geometry used to model the collector with wind barriers, in order to explore the fluid
dynamical details. Besides, one is interested in assessing the influence of the main parameters upon
the topology of the flow inside the cavity formed by the surface of the collector and the wind

*
Corresponding author. Permanent address: CTA/ITA/IEA, SJ dos Campos, SP, 12228-900, Brazil. Tel.: +55-12-
3947-5830; fax: +55-12-3947-5824.
E-mail address: nide@aer.ita.br (N.G.C.R. Fico Jr.).

0045-7930/03/$ - see front matter Ó 2002 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 4 5 - 7 9 3 0 ( 0 2 ) 0 0 0 6 7 - 1
954 P.S.B. Zdanski et al. / Computers & Fluids 32 (2003) 953–974

Fig. 1. Sketch of the flow over a solar collector with wind barriers.

barriers. This influence is decisive for the level of heat transfer inside the cavity. In the most
general case the flow over the solar collector (imagined mounted on the roof of a building) is
three-dimensional. Besides, the angle of attack between the wind direction and the plane of the
collector can take any value from zero up to ninety degrees. In order to start with a tractable
problem we have made the following simplifying assumptions: the flow is two-dimensional and
the angle of attack of the wind is equal to zero (Fig. 1).
The aspect ratio of the cavity, AR, is defined as the ratio of its length by its height, L=h (Fig. 2).
Gomes [1] has investigated several values of the aspect ratio and has found that 8 is a good value.
For AR equal to 8 (turbulent flow) there is a good compromise between convection inhibition and
shadowing effects (of the wind barrier upon the absorption area of the collector). For two-
dimensional cavities with AR of order one the solution is classical, and there is a great number of
papers in the literature referring to it. The dynamics inside the cavity is dominated by a great
vortical structure whose dimension is equal to the dimension of the cavity. At the corners, there
appear smaller structures that serve to accommodate the flow to the confining walls [2]. Sinha et al.
[3] present experimental results for laminar flows over deep (small values of AR) as well as shallow
(large values of AR) cavities. However, the data are not sufficient for the understanding of the
various aspects of such flows, especially the very important point relative to the general classifi-
cation of a cavity as ‘‘open’’ or ‘‘closed’’ [4]. Aung [5] reports experimental heat transfer data for

Fig. 2. Cavity geometry with the main dimensions.


P.S.B. Zdanski et al. / Computers & Fluids 32 (2003) 953–974 955

cavities of AR equal to one and four, but little attention is devoted to fluid dynamical details.
Bhatti and Aung [6] simulated numerically the laminar flow over a cavity and correlated the heat
transfer inside the cavity as a function of the aspect ratio and Reynolds number. Some turbulent
results are presented by Metzger et al. [7], but only some streamline maps are shown. Besides, the
geometry studied is that of a shrouded cavity. Richards et al. [8] have measured turbulent heat
transfer rates from bottom heated cavities of small aspect ratios (AR less than 1). More recently,
Matos et al. [9] presented turbulent heat transfer results for two-dimensional cavities, but data for
the free cavity (AR equal to 2) restrict to the presentation of a pressure distribution along the floor
of the cavity.
The main aim of this work is to investigate shallow cavities of large aspect ratios. We have
obtained numerical solutions for laminar as well as turbulent flows over two-dimensional shallow
cavities of various aspect ratios. A very careful parametric study was conducted for both laminar
and turbulent flows. In the laminar case the parameters of influence were selected as the entrance
velocity profile, the Reynolds number (based on the depth of the cavity) and the cavity aspect
ratio; for the turbulence case the parameters were defined as the cavity aspect ratio, the level of
turbulence of the incoming flow and the Reynolds number. We address now the reason for this
parametric study. To the best of our knowledge, published material related to cavities of large
aspect ratio is very scarce. So, the parametric study had the very important aim of revealing any
general trend that could be explored in some way. The result was the discovery that the mean
turbulent flow inside the cavity, for some ranges of those parameters, ‘‘encapsulates’’, that is, the
mean external flow does not touch the floor of the cavity. The ‘‘encapsulating’’ phenomenon is
discussed in detail in Section 3.4. This is very important for the heat-transfer regimes of the solar
collector. The capsule acts as a thermal insulator and this is the ultimate reason why one would be
using the wind fences. The main aim of this research effort is to assess, numerically, to what extent
the wind barriers really avoid the convection effect on top of the solar collector.
Turbulence is predicted by means of a high-Reynolds-number j–e modeling method [10]. Ours
here is essentially an engineering approach, and, therefore, the interest lies on the steady mean
flow. One wants to have a quick answer to the question: is it worth using the wind barriers, and, if
it is, what is the optimum aspect ratio? We are aware that the dynamics inside the cavity is ex-
tremely complicated, and that the instantaneous flow play a very important role. In order to learn
about this and compare results in the future we are working the same problem with our groupÕs
DNS/LES capability. But, the emphasis of this paper is on the engineering aspect of the problem.
On the other hand, results presented herein refer only to the fluid dynamics of the flow. Heat
transfer simulation data will be published subsequently.

2. Mathematical model and numerical solution algorithm

The flow is modeled by the two-dimensional, Cartesian, incompressible, Reynolds-averaged,


Navier–Stokes equations. To these, a two-equation model of turbulence is added. The j–e model
in its high-Reynolds-number traditional format [10] is used. The equations are discretized fol-
lowing a finite-volume procedure and a stretched staggered grid is used. The resulting system of
algebraic equations is solved using the SIMPLER algorithm of Patankar [11]. The flowing me-
dium is always air considered as thermally and calorically perfect.
956 P.S.B. Zdanski et al. / Computers & Fluids 32 (2003) 953–974

Fig. 3. View of the corner cell.

Boundary conditions were enforced according to the following: At the entrance plane (Fig. 2)
distributions of velocity, turbulent kinetic energy and turbulent dissipation are specified. Typical
values of the turbulent kinetic energy were considered between one to ten percent of the mean flow
kinetic energy. The pressure at the entrance plane was extrapolated from internal cells. At the exit
and upper frontiers parabolic conditions were established for all variables. At solid walls the
condition of zero velocity was enforced and the shear stress is obtained from the law of the wall.
The values of pressure and turbulent kinetic energy at the wall are obtained by a zero-order
extrapolation from the values at the first cell. Values of the turbulent dissipation at the first cell
away from the wall are not obtained by solving the complete e equation, but, rather, those figures
result from a balance between production and dissipation at the cell [12]. Corner volumes were
specially treated (Fig. 3). For the staggered control volume, there is mass crossing at the south
face only at the right half of the face area. The velocity that determines this flux is to be considered
the one at the external cell and not a mean between this value and that at the wall.

3. Results and discussion

3.1. Code validation and grid-independence study

Initially, a series of reference cases were treated with the specific purposes of code validation. In
the laminar regime we have considered the flat plate, the two-dimensional constant-cross-section
duct and the backward-facing step, whereas in the turbulent case we have treated the two-
dimensional duct, first with a constant cross-section and then with a sudden expansion. The
overall agreement with literature data was in general very good.

3.1.1. The laminar case


Flow along a flat plate: The height of the calculation domain is equal to h and the length is
considered as L ¼ 2h. The grid is uniform with 200  200 nodes, and the Reynolds number is
ReL ¼ 66 000. Fig. 4 shows the comparison between our laminar solution and that of Blasius. As
the reader can observe agreement is quite good, with the maximum error not exceeding 2%.
P.S.B. Zdanski et al. / Computers & Fluids 32 (2003) 953–974 957

Fig. 4. Comparison between numerical results and Blasius analytical solution.

The two-dimensional constant-cross-sectional duct: The height of the channel is equal to h, and
its length is equal to 10h. The grid used here is uniform with 141  61 nodes, and the Reynolds
number is Reh ¼ 184. At the entrance section the velocity profile is uniform and therefore both the
initial region where the flow develops as well as the final part with fully developed flow were
predicted. At the exit section the calculated maximum velocity at the mid-height section was
ðu=ui Þ ¼ 1:49, an error of less than 1% (Fig. 5).
The flow over a backward-facing step: Longitudinal distances of 20s and 40s upstream and
downstream the corner, respectively, with a transversal height relative to the horizontal entrance

Fig. 5. Laminar velocity profile for the exit section of the two-dimensional duct.
958 P.S.B. Zdanski et al. / Computers & Fluids 32 (2003) 953–974

Fig. 6. Reattachment length, xr =s, as a function of Reynolds number, Res .

level equal to 40s were considered, where s is the step height. The grid is stretched with a total
number of 3976 volumes, 21  39 and 41  77 volumes at the upstream and downstream regions,
respectively. Grid clustering was established from the step outwards, horizontally, and from y ¼ s
upwards, vertically; in the region of the step, i.e., from y ¼ 0 to y ¼ s the grid is uniform. The flow
over the step is a free flow and consequently boundary conditions for the upper frontier were
considered parabolic. In this specific example, the most important physical parameter for code
validation purposes is the reattachment length, xr . Fig. 6 compares present values with literature
data. Our results are close to the numerical data of Fletcher [13] and the general trend of both
solutions is the same (the greatest error is about 8% at both extremes of the Reynolds interval). On
the other hand the calculated solutions falls consistently below the experimental values of Sinha
et al. [14].

3.1.2. The turbulent case


The two-dimensional constant-cross-sectional duct: The height of the channel is equal to h and its
length was taken as L ¼ 60h. The length was chosen so as to guarantee a fully developed flow. For
this test, the following values were adopted: h ¼ 0:015 m, ui ¼ 18:68 m/s, ki ¼ 0:02ðui Þ2 =2:0, where
ui and ki are, respectively, mean velocity at the inlet section (considered uniform along the cross-
section) and turbulent kinetic energy (also considered uniform); the Reynolds number based on h
is therefore Reh ¼ 16 482. The grid is uniform with a total of 2400 volumes––120 and 20 partitions,
respectively, along the length and the height of the channel. In Fig. 7, where u stands for the
friction velocity, the kinetic energy distribution is compared with literature data. Next to the wall
our solution is closer to the DNS simulation of Mansour et al. [15], when compared to the
j–x model of Wilcox [16]. Far from the wall our solution practically coincides with the reference
cases.
P.S.B. Zdanski et al. / Computers & Fluids 32 (2003) 953–974 959

Fig. 7. Cross-sectional profiles of turbulent kinetic energy for the two-dimensional duct.

Fig. 8. Profiles of turbulent kinetic energy for different mesh sizes.

For the case of the turbulent duct flow a grid-independence study was conducted, and the
solutions for three different grid resolutions appear in Fig. 8. As the reader can observe, the code
behaved very well under this test and the results are basically grid independent.
The two-dimensional duct with a sudden expansion: This case is commonly used in the validation
of codes whose scope is basically the calculation of flows with large recirculation regions. Besides,
960 P.S.B. Zdanski et al. / Computers & Fluids 32 (2003) 953–974

Fig. 9. Sudden expansion duct geometry with main dimensions.

there is a great number of literature references, both numerical and experimental. The geometry is
depicted in Fig. 9. The dimensions of the computational domain are: l ¼ 8s, L ¼ 22s, h ¼ 2s and
H ¼ 3s. A total of 5560 volumes were used, with the following distribution: 40  40 in the en-
trance region and 60  66 in the region after the step. The grid is stretched with grid clustering
next to the walls, and greater concentration of points in the region of the expansion. The Reynolds
number based on the step height is Res ¼ 7860, and the inlet mean velocity is ui ¼ 6:68 m/s
(considered to be uniform along the section). Let us also point out that the Cartesian coordinates
x and y are measured from the lower step corner.
The calculated reattachment length, which, for the turbulent flow, does not depend on the
Reynolds number, is predicted as ðxr =sÞ ¼ 5:98. Experimental data due to Eaton and Johnston
[17] show that ðxr =sÞ ¼ 7:0  1:0. Keeping in mind the well-known fact that the standard j–e
model underestimates the reattachment length [12,18]), the present result can be considered as
good. In Fig. 10(a) and (b) profiles of turbulent kinetic energy are compared, for values of
ðx=sÞ ¼ 7:7 and ðx=sÞ ¼ 10:3, with experimental data of Kim et al. [19] and numerical results of
Mansour et al. [12]. It is worth to remember that the high-Reynolds j–e model that was used by
Mansour et al. [12] is modified to take into account rotation effects, and the model used in this
work is the standard one with no corrections. A comparison of velocity profiles is made in Fig. 11

Fig. 10. Turbulent kinetic energy profiles at the positions: x=s ¼ 7:7 (a), 10.3 (b).
P.S.B. Zdanski et al. / Computers & Fluids 32 (2003) 953–974 961

Fig. 11. Velocity profiles at the station x=s ¼ 5:3.

at the station ðx=sÞ ¼ 5:3. In this particular instance our solution compares very well with ex-
perimental values for values of (y=s) in excess of about 0.5. Closer to the wall there is a pro-
nounced deviation due to the underestimation of xr , but at this region our results are practically
coincident with the numerical reference.
By observing the results presented above, which span all the main aspects of the cases to be
tackled in the sequel, the reader can appreciate the good overall agreement between our solutions
and former published data. This fact corroborates the validation of the computational code that is
being used as a valid numerical tool.

3.2. Computational grid for the cavity studies

A typical computational grid for cavity simulations is shown in Fig. 12. Points clustering is used
close to solid walls and to the horizontal plane connecting the two corners, that is, where the
strongest gradients are expected to happen. In spite of the parabolic character of the upper
boundary, numerical experiments showed that, for the laminar case, a minimum value of
H ¼ 10h, was necessary in order to avoid spurious interference on the numerical solution. As
indicated in Fig. 2, H is the height of the computational domain above the horizontal basic plane.
For turbulence simulations a minimum value of H ¼ 5h was sufficient. Any grid specifics for
different situations will be indicated in the text.

3.3. Laminar studies

The interest here is to investigate the influence of three important parameters upon the flow in
the cavity. These parameters are: (i) entrance velocity profile, (ii) Reynolds number based on the
depth of the cavity, Reh and (iii) cavity aspect ratio.
962 P.S.B. Zdanski et al. / Computers & Fluids 32 (2003) 953–974

Fig. 12. Typical computational grid with points clustering.

3.3.1. Influence of the inlet velocity profile


For this study, the following geometric and physical characteristics were assumed: (i) inlet
length, li ¼ 4h, (ii) outlet length, le ¼ 6:8h, (iii) aspect ratio, AR ¼ 28, (iv) cavity depth, h ¼ 0:625
cm, (v) inlet free stream velocity, ui ¼ 1:8 m/s, (vi) Reynolds number, Reh ¼ 662. These values
were selected because there were experimental data available [3]. Many numerical experiments
were done, and the most representatives are shown in Fig. 13. The reader can observe the plot of
the pressure coefficient, Cp , along the floor of the cavity for three different cases: (a) uniform
velocity profile at the entrance section, (b) a Blasius profile at the entrance section, such that the
boundary layer thickness at the cavity upstream corner (just before separation) is equal to 0.54
cm, (c) a Blasius profile such that the boundary layer thickness at the separation corner is equal to
1.4 cm. For comparison purposes the experimental Cp distribution is also plotted. Case (c) is the
one that best reproduced the experimental data, and this is because the measured value of the
separation boundary layer thickness is exactly 1.4 cm. This is, evidently, an expected result, be-
cause the vorticity rate that is shed at the upstream corner, which depends upon the characteristics
of the boundary layer at the separation station, is instrumental to the development of the shear
layer emanating from the corner [18]. Therefore, this result helped to firmly consolidate the code
validation.

Fig. 13. Pressure distribution on the cavity floor.


P.S.B. Zdanski et al. / Computers & Fluids 32 (2003) 953–974 963

3.3.2. Influence of the Reynolds number


For this part of the investigation a cavity having an aspect ratio of 12 was used and at the
entrance plane an uniform velocity profile was considered (for convenience, because, here, there
were no experimental data to compare to). The height h was given the value of 0.625 cm and the
Reynolds number, Reh , was made to vary between 147 and 662 (different values of the Reynolds
number corresponded to different values of the entrance velocity). The flow topologies appear in
Fig. 14(a)–(d). In the range of Reynolds number investigated it was found that two recirculating
bubbles were always present inside the cavity. Nevertheless, their position along the cavity as well
as their shape varied with Reh .
For Reh ¼ 147, what corresponds to ui ¼ 0:4 m/s, the oncoming flow penetrates the cavity
touching its bottom at about x=h  8:5 (Fig. 14(a)). The two bubbles are well defined and the flow
is reversed along parts of the cavity floor. The bubble closer to the upstream vertical wall has its

Fig. 14. Streamlines for the cases: Reh ¼ 147 (a), 294 (b), 442 (c), and 662 (d).
964 P.S.B. Zdanski et al. / Computers & Fluids 32 (2003) 953–974

Fig. 15. Distance between the centers of the two bubbles as a function of Reynolds number.

center at x=h  2:4, while the one near the downstream wall is centered at x=h  11:6. As the
Reynolds number increases the upstream bubble elongates while the downstream one gets bigger
but maintains its rather rounded form. Also, for larger values of Reh the external flow does not
reattach anymore and the center of the upstream bubble moves downstream, while the center of
the downstream bubble practically does not move. For the highest Reynolds number investigated,
Reh ¼ 662, the center of the bigger recirculating bubble was found to be at x=h  8:9. On the other
hand the smaller one barely moved, its center appearing at x=h  11:1. Fig. 15 gives the distance
between the bubble centers as a function of the Reynolds number.

3.3.3. Influence of the cavity aspect ratio


The Reynolds number, Reh , was kept constant and equal to 662. The aspect ratios simulated
were equal to 9.6, 10 and 28. Details of the streamline maps of the flow for AR equal to 9.6 and 10
are presented in Fig. 16, while for AR ¼ 28 a sketch is shown in Fig 17. In the latter case a sketch
is used because the proportion of scales is such that a clear representation of the streamlines is not
possible. The flow is very different for each of the three aspect ratios. The shortest cavity presented
a flow pattern in which only one recirculating bubble appeared. It is important to emphasize that
the numerical code is being very accurate in reproducing the flow topology. This is confirmed by
the two-bubble formation for the slightly greater aspect ratio, AR ¼ 10, as reported in the lit-
erature [3]. The much longer cavity, AR ¼ 28, tends to behave like a sequence of a backward-
facing step followed by a flat plate and ending with a forward-facing step. Thus the flow separates
abruptly at the upstream corner, reattaches somewhere on the cavity floor and separates again
when approaching the downstream vertical wall.

3.4. Turbulent studies

Most flows of interest are turbulent. To the authors knowledge very little work has been done
on high aspect ratio turbulent shallow cavities. Therefore, careful attention will be devoted to this
P.S.B. Zdanski et al. / Computers & Fluids 32 (2003) 953–974 965

Fig. 16. Streamlines for the cases: AR ¼ 9:6 (a), and 10 (b).

Fig. 17. Sketch of the flow for AR ¼ 28.

type of flow. At first we shall discuss the case of the ‘‘typical cavity’’, for which an assortment of
results will be presented. After this, a parametric study of the turbulent cavity will be conducted,
following the same general guidelines already pursued in the laminar case.

3.4.1. Typical cavity


The aspect ratio of the typical cavity is taken as equal to 8. As we have already called attention in
the introduction, this is an optimized value obtained by Gomes [1] in his investigation of solar
collectors with wind barriers. The depth was taken as equal to 4 cm and the entrance velocity equal
to 8 m/s, what gives Reh ¼ 18 823. The oncoming turbulent kinetic energy is considered to be 4% of
the mean free-stream kinetic energy. The calculation domain is such that li ¼ 3h, le ¼ 4h, and
H ¼ 5h (Fig. 2). The grid, part of which is shown in Fig. 18, is not uniform along the domain. The
total number of nodes is: 41  41 upstream of the first corner, 81  91 in the region of the cavity,
and 41  41 downstream of the second corner. The maximum stretching factor was equal to 9%.
Fig. 19 shows the topology of the flow inside the cavity. Two structures are formed, a big one
resulting from the main separation at the upstream corner and a small one in front of the
downstream vertical wall. The external stream does not reattach and the flow is reversed along all
the floor of the cavity. This point will be latter discussed in more detail.
The pressure variation in an incompressible flow field is in general quite small. One can observe
in Fig. 20 a region of high pressure corresponding to stagnation conditions in front of the
966 P.S.B. Zdanski et al. / Computers & Fluids 32 (2003) 953–974

Fig. 18. Aspect of the grid at the first corner region.

Fig. 19. Streamlines for the typical cavity (AR ¼ 8).

downstream vertical wall. Two regions of low pressure appear. One near the upstream corner, at
x=h  2, corresponding to the center of the big vortical structure inside the cavity and the other
just after the downstream corner (x=h  8:5). The latter structure appears due to flow separation
at the second corner. Values of Cp plotted in Fig. 20 are referred to a mean value of pressure at the
entrance section.
The field distribution of turbulent kinetic energy, 2k=ðui Þ2 , is shown in Fig. 21. After the up-
stream separation there is an increase in the level of k due to the development of the shear layer
emanating from the corner. Inside the cavity the level of turbulence is relatively high. There is a
P.S.B. Zdanski et al. / Computers & Fluids 32 (2003) 953–974 967

Fig. 20. Pressure isolines for the typical cavity (AR ¼ 8).

Fig. 21. Turbulent kinetic energy.


968 P.S.B. Zdanski et al. / Computers & Fluids 32 (2003) 953–974

Fig. 22. Turbulent kinetic energy profiles: x=h ¼ 1 (a), 3 (b), 5 (c), and 7 (d).

Fig. 23. Velocity profiles for the typical cavity: x=h ¼ 1, 3, 5 and 7.
P.S.B. Zdanski et al. / Computers & Fluids 32 (2003) 953–974 969

maximum of k close to the second corner due to local separation and a consequent increase in
turbulence production. Profiles of k=ðui Þ2 are shown in Fig. 22, and the reader can observe that the
point of maximum of k follows approximately the evolution of the shear layer along the upper
regions of the cavity. Fig. 23 presents the evolution of the velocity profile and we call attention for
the reverse character of the flow all along the floor of the cavity.
We pass now to a parametric study of the turbulent cavity. The following influences will be
investigated: (i) aspect ratio, (ii) entrance level of turbulence, and (iii) Reynolds number, Reh . In
the discussions to follow we shall concentrate mainly on topological results, because our main
objective here is to understand the conditions for minimum convective activities alongside the
floor of the cavity.

Fig. 24. Streamlines for the cavity with aspect ratios: AR ¼ 6 (a), 10 (b), and 12 (c).
970 P.S.B. Zdanski et al. / Computers & Fluids 32 (2003) 953–974

3.4.2. Influence of the cavity aspect ratio


We have considered cavities with AR equal to 6, 8, 10, and 12. All the other parameters defining
the flow, including Reynolds number and level of turbulence of the incoming stream, were con-
sidered equal to the values of the typical case (Section 3.4.1). Fig. 24(a) gives the streamlines map
for AR ¼ 6. There is only one recirculating bubble whose center is located approximately at
x=h  2:2. For AR ¼ 8 there is already two bubbles (Fig. 19), and the same two bubbles pattern is
also obtained for AR equal to 10 and 12 (Fig. 24(b) and (c)). The most interesting aspect that can
be observed here is related to the velocity distribution along the cavityÕs floor. For AR equal to 6,
8, and 10 the velocity is reversed all the way; and this is confirmed by Fig. 25(a), where the
distribution of horizontal velocity is plotted against the distance along the floor for AR ¼ 10. For
AR ¼ 12 the distribution of velocity is not all reversed, and this can be confirmed by Fig. 25(b).
(The plotting in Fig. 25(a) and (b) correspond to the horizontal component of the mean velocity at
the center of the first volume close to the wall.) We understand that this result is of utmost im-
portance in relation to heat transfer effects on the floor of the cavity (i.e., the solar collector plate
surface). The reader can observe by inspecting Figs. 24(a), 19, 24(b) that the bubbles, one for AR
equal 6 and two for AR equal to 8 and 10, are, say, ‘‘encapsulated’’ by an enveloping streamline
along all the space of the cavity. This ‘‘capsule’’ serves the purpose of a thermal insulating
mechanism, because the convective action due to the external flow is isolated from the floor of the
cavity. This situation will change dramatically when the capsule is ruptured, what can be seen to
have happened for AR ¼ 12. In this case the external flow ‘‘touches’’ the bottom of the cavity and
convective effects will certainly increase to a very high level. It is important to realize that the
point of rupture can be recognized as that situation for which the velocity along the floor first
becomes zero. For the case in hand, this situation most certainly happens for a value of AR
between 10 and 12.
The important conclusion here is that convective effects do not depend necessarily upon the
number of bubbles present in the cavity, but, rather, if all the bubbles are encapsulated or not.

Fig. 25. Velocity distribution along the cavity floor for aspect ratios: AR ¼ 10 (a), and 12 (b).
P.S.B. Zdanski et al. / Computers & Fluids 32 (2003) 953–974 971

3.4.3. Influence of the entrance turbulence level


All basic parameters are equal to those of the typical cavity. The oncoming turbulence level
varied and the values of turbulent kinetic energy investigated corresponded to 4%, 7%, and 10% of
the mean flow kinetic energy. In all three cases investigated there were two bubbles present inside
the cavity. Fig. 26(a)–(c) show the details of the flow field topology in the region between the two
bubbles for the three levels of turbulence. We verified that the same phenomenon of bubbles
encapsulation appears also here. That is, for the levels of 4% and 7% the flow along the cavity
floor is always reversed, and for the level of 10% this condition is lost. Fig. 27(a) and (b) show this
behavior, i.e., for 10% of inlet turbulence level the velocity becomes positive at some station on the

Fig. 26. Streamlines for the cavity with turbulence levels: (a) 4%, (b) 7%, and (c) 10%.
972 P.S.B. Zdanski et al. / Computers & Fluids 32 (2003) 953–974

Fig. 27. Velocity distribution along the cavity floor for turbulence levels: (a) 7%, and (b) 10%.

cavity floor. Therefore, for a certain level of turbulence between 7% and 10% a very sharp increase
of convective effects will take place.

3.4.4. Influence of the Reynolds number


As in the other two studies the reference here is the typical cavity. The variation of the Reynolds
number was a consequence of the variation of the entrance velocity. Values of ui equal to 5, 8, and
12 m/s, corresponded to values of Reh of 11 765, 18 823, and 28 325, respectively. For these three
values of Reh the two bubbles pattern was established inside the cavity. The topologies of the flows
for these three cases are represented in Figs. 26(a), 28(a) and (b). Inspection of the figures suggests

Fig. 28. Streamlines for the cavity: Reh ¼ 11 765 (a), and 28 235 (b).
P.S.B. Zdanski et al. / Computers & Fluids 32 (2003) 953–974 973

Fig. 29. Velocity distribution along the floor of the cavity: Reh ¼ 18 823 (a), and 28 235 (b).

one more time the existence of two ‘‘regimes’’. For low values of Reh vortices encapsulation occurs
and heat losses by convection from the bottom of the cavity shall be probably very low. As Reh
increases and gets larger than a certain critical value convective effects will increase accordingly.
Observing the figures one immediately verifies that this critical value is located between
Reh ¼ 18 823 and 28 235 (Fig. 29).

4. Concluding remarks

Comparison of numerical data reported above suggests striking contrasts between laminar and
turbulent results. The most obvious is the fact that, for increasing Reynolds numbers, the center of
the great laminar vortical structure tends to approximate the small structure. For the turbulent case
the centers of both large and small structures are practically immobile. For low values of Reh , and for
laminar streams, the external flow reattaches at the floor of the cavity; as Reh increases, the external
flow does not touch the bottom. For turbulent flows the effect is exactly the contrary. But, by far, the
most important result of this research work is related to the turbulent case. The parametric study has
indicated clearly that there are two regimes of flow inside the cavity. For low values of certain
parameters, in general two recirculating bubbles are present inside the cavity, but the flow along the
bottom is always reversed. This characterizes a situation that we have called vortexes encapsulation.
When encapsulation occurs convective effects are probably inhibited. For larger values of these
parameters (larger than certain critical values), the external flow reattaches and convective effects
along the cavityÕs floor will become effective. Here we have tested as parameters of influence, the
cavityÕs aspect ratio, the oncoming flow turbulence level and the depth Reynolds number.

Acknowledgements

This work was supported in part by the Brazilian agency, CNPq, National Council of Scientific
and Technological Development, through grant 522413/96-0.
974 P.S.B. Zdanski et al. / Computers & Fluids 32 (2003) 953–974

References

[1] Gomes DG. Optimization of flat plate solar collectors. Master Thesis Dissertation, TIA––Technological Institute
of Aeronautics, S~ao Jose dos Campos, SP, Brazil, 1998 [in Portuguese].
[2] Ghia U, Ghia KN, Shin CT. J Comput Phys 1982;48:387–411.
[3] Sinha SN, Gupta AK, Oberai MM. Laminar separating flow over backsteps and cavities part II: Cavities. AIAA J
1982;20:370–5.
[4] Sarohia V. Experimental investigation of oscillation in flow over shallow cavities. AIAA J 1977;15:984–91.
[5] Aung W. An interferometric investigation of separated forced convection in laminar flow past cavities. J Heat
Transfer 1983;105:505–12.
[6] Aung W, Bhatti A. Finite difference analysis of laminar separated forced convection in cavities. J Heat Transfer
1984;106:49–54.
[7] Metzger DE, Bunker RS, Chyu MK. Cavity heat transfer on a transverse grooved wall in a narrow flow channel.
J Heat Transfer 1989;111:73–9.
[8] Richards RF, Young MF, Haiad JC. Turbulent forced convection heat transfer from a bottom heated open surface
cavity. Int J Heat Mass Transfer 1987;30:2281–7.
[9] Matos A, Pinho FAA, Silveira-Neto AS. Large-eddy simulation of turbulent flow over a two-dimensional cavity
with temperature fluctuations. Int J Heat Mass Transfer 1999;42:49–59.
[10] Launder BE, Spalding DB. The numerical computation of turbulent flows. Comput Meth Appl Mech Eng
1974;3:269–89.
[11] Patankar SV. Numerical heat transfer and fluid flow. New York, 1980.
[12] Mansour NM, Kim J, Moin P. Computation of turbulent flows over a backward-facing step. Nasa Technical
Memorandum 1983;85851.
[13] Fletcher CAJ. Computational techniques for fluid dynamics. Berlin: Springer-Verlag; 1988.
[14] Sinha SN, Gupta AK, Oberai MM. Laminar separating flow over backsteps and cavities part I: backsteps. AIAA J
1981;19:1527–30.
[15] Mansour NM, Kim J, Moin P. Reynolds stress and dissipation rate budgets in turbulent channel flow. J Fluid
Mech 1988;194:15–44.
[16] Wilcox DC. Turbulence modeling for CFD. La Can~ ada, 1998.
[17] Eaton JK, Johnston JP. A review of research on subsonic turbulent flow reattachment. AIAA J 1981;19:1093–100.
[18] Silveira-Neto A, Grand D, Metais O, Lesieur M. Numerical investigation of the coherent vortices in turbulence
behind a backward-facing step. J Fluid Mech 1993;256:1–25.
[19] Kim J, Kline SJ, Johnston JP. Investigation of a reattaching turbulent shear layer: flow over a backward-facing
step. J Fluid Eng 1980;102:302–8.

You might also like