See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/267591595

Computational Investigation of Backward-Facing Step Flow Preceding a Porous

Medium

Conference Paper · January 2009

DOI: 10.1115/IMECE2009-11228

CITATIONS READS
3 1,142

4 authors, including:

Chandramoulee Krishnamoorthy Frank Chambers

Abengoa Oklahoma State University - Stillwater
5 PUBLICATIONS 22 CITATIONS 34 PUBLICATIONS 309 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Chandramoulee Krishnamoorthy on 28 January 2016.

The user has requested enhancement of the downloaded file.

 Proceedings of IMECE2009
2009 ASME International Mechanical Engineering Congress and Exposition
November 13-19, Lake Buena Vista, Florida, USA

IMECE2009-11228

COMPUTATIONAL INVESTIGATION OF BACKWARD-FACING STEP FLOW

PRECEDING A POROUS MEDIUM

C. Krishnamoorthy K.C. Ravi

Burns and McDonnell Mechanical and Aerospace Engineering
Kansas City, MO 64114 Oklahoma State University
ckrishnamoorthy@burnsmcd.com Stillwater, OK 74078
krishna.c.ravi@okstate.edu

S. Yao F.W. Chambers

Advanced Research Systems, Inc. Mechanical and Aerospace Engineering
Macungie, PA 18062 Oklahoma State University
yshengh@gmail.com Stillwater, OK 74078
frank.chambers@okstate.edu

computational results did not compare well to no-filter cases or

ABSTRACT the experiments with filters at the lower Reynolds numbers. It
Optimal performance of air filters and heat exchangers is believed that the turbulence models were unsuitable for these
requires uniform inlet flow, but flow separation produces flows at transitional Reynolds numbers. Good agreement for
nonuniformity. The backward-facing step flow has a separation no-filter results and for the experiments with filters was
resembling those found in industrial flows. Flow resistance of observed for Re = 10,000. The CFD model seems to capture
the devices is a parameter which alters upstream pressure the physics of the separation better at the higher Reynolds
gradients, thereby affecting separation and device performance. numbers. The CFD velocity profiles at Re = 10,000 with the
Air filters often are modeled as porous media using an filters agree with those of the experiments. When the filter is
extended Darcy Law. The present work applied Computational placed at 4.25 step heights, the flow reattaches upstream of the
Fluid Dynamics (CFD) to examine the changes in the step flow filter with a reduction in recirculation area. When the filter is at
resulting from the resistance of a downstream air filter. 6.75 step heights, the separated flow tends to reattach and the
Computations were performed for a backward-facing step with opposite side tends to separate. At Re = 10,000 and the filter at
a 2:1 expansion ratio for a case without a filter (reattachment at 4.25 step heights, the variations of porous medium
~6 step heights) and for filters located 4.25 and 6.75 step permeability, inertial constant and the filter thickness have
heights downstream. FLUENT commercial CFD software was negligible effects on the recirculation region over the parameter
used and results were compared to many no-filter case results ranges considered.
in the literature and our own experimental studies for the step
with downstream filters. The simulations were performed for INTRODUCTION
Reynolds numbers based on approach channel mean velocity The backward-facing step flow is a fundamental flow with
and hydraulic diameter of 2000, 3750, 6550 and 10000. The a separation that resembles many found in industrial flows,
different turbulence models available in FLUENT were including housings for automotive air filters and headers for
evaluated and the Realizable k-ε model was used for the final compact cross flow heat exchangers. Filters and heat
computations. Grid independence studies were conducted. The exchangers provide a flow resistance that can alter the pressure
effects of different values of the filter modeling permeability, distributions and gradients in separated flow regions.
inertial constant and thickness also were investigated for Re = Automotive filters are fibrous filters that operate through depth
10000 with the filter at 4.25 step heights. It was found that the filtration as particles carried through the filter are deposited

1 Copyright © 2009 by ASME

upon the fibers. Flow through the filter is required for effective Previous Studies on Backward-Facing Steps
filtration. Thus, regions of flow separation incident on the There have been many experimental and numerical studies
upstream face of a filter provide little flow through that part of of backward-facing step flows. Eaton and Johnston [10] and
the filter, degrading performance. Biswas et al. [11] present reviews of experiments on backward-
facing step flows. The work reported by Armaly et al. [1] is
Armaly et al. [1] and Kaiktsis et al. [2] are among those
widely referenced. They used a single component LDA to study
who note that the step flow is a very good prototype for fluid
the flow over a backward-facing step for a Reynolds number
flows in complex geometries, as there is a large amount of
range which included laminar, transitional and turbulent flow.
experimental data available. It is a classic separating flow that
They defined Reynolds numbers less than 600 as laminar, and
has served as a challenging test case for numerical codes and
greater than 6600 as turbulent. Their experiment showed that
experimental techniques, as discussed by Badran and Bruun
the flow downstream of the step was two-dimensional at low
[3], Scarano and Riethmuller [4]. It has three important
Reynolds numbers and high Reynolds numbers, and was three-
properties: strong velocity shear at the edge of the step, strong
dimensional at intermediate Reynolds numbers. They identified
separation flow behind the step and downstream developing
for the first time a secondary flow separation on the flat side of
channel flow with boundary layer flow recovery from the
the channel downstream of the step, as shown in Figure 1.
separation. A typical flow domain for a backward-facing step
Two-dimensional numerical predictions were in good
flow is shown in Figure 1. In the low-Reynolds number regime,
agreement with the experimental results at low Reynolds
a unique relationship exists between the Reynolds number
numbers.
(Re), the expansion ratio ([S+D]/D), and the normalized length
of the circulation zone (Xr/S). In the high-Reynolds number Other significant experiments include those of Kim et al.
regime, the reattachment length (Xr) is determined by the [12], Durst and Tropea [13], Sinha et al. [14], Jovic and Driver
expansion ratio and weakly correlated with the Reynolds [15], and the PIV measurements of Piirto et al. [16]. Driver and
number. The large amount of information available on this Seegmiller [17] analyzed the effect of pressure gradient on
separation makes it an ideal flow to use to gain an reattachment length with experiments and numerical
understanding of the interaction between separation and a filter predictions performed with two-dimensional k-ε and algebraic-
placed within or downstream of the separation. The backward- stress turbulence models. They found that the numerical results
facing step flow results may serve as a guide to more complex under-predict the reattachment length. Lee and Mateescu [18]
configurations and problems. conducted experimental and numerical investigations on steps
with expansion ratios of 1.17 and 2.0 for laminar and
y
Umax x transitional Reynolds numbers, finding good agreement in the
X4 X5 results.
Low Reynolds number step flow was studied numerically
D Inflow U(y)
Downstream by Kim and Moin [19] who applied the Direct Numerical
Dividing Streamline 2h Boundary Simulation (DNS) approach. Their results were in good
S agreement with the experiments of Armaly et al. [1]. Le et al.
[20] also used DNS in their study of turbulent flow over a
backward-facing step at a Reynolds number of 5,100 based on
Reattachment Point
Xr the step height and the inlet velocity. The expansion ratio of
their step flow channel is 1:2, very close to that of Armaly et al.
Figure 1. Schematic of Backward-Facing Step Flow [1] and the same as that of the present study. The reattachment
location varies in the spanwise direction and oscillates about a
The fluid dynamic behavior of filters often is modeled by mean value of 6.28 step heights, in good agreement with the
treating the filter element as a porous medium. Research has experimental results of Jovic and Driver [21]. The velocity
been conducted examining the use of porous media mounted profiles in the recovery region fell below the universal log-law,
immediately downstream of a step to suppress separation and indicating the flow is not fully recovered at 20 step heights
do away with the recirculating flow zone (Assato et al. [5], beyond the step. Yoo and Baik [22] reported similar results.
Chan and Lien [6]), but little research has been performed on Chiang and Sheu [23] performed numerical studies for the
the macroscopic interactions between separating fluid flow same step geometry at Reynolds numbers between 100 and
outside porous media and the flow through the media. To gain a 1000. They found a significant effect of channel width and
better understanding of these problems, numerical and aspect ratio on the reattachment length.
experimental studies were conducted of the velocity fields in a
nominally two-dimensional duct with a backward-facing step Kostas et al. [24], Piirto et al. [16], Wengle et al. [25], and
with and without an air filter mounted a small distance others have performed experimental and numerical studies of
downstream providing flow resistance. Yao [7] performed the backward-facing step flow for cases in which the step
experiments (see Yao et al. [8]) and preliminary numerical height is a much smaller fraction of the channel height and the
studies, and Krishnamoorthy [9] extended the numerical boundary layer thickness upstream of the step is of the same
investigations. order of magnitude as the step height. Such studies report

2 Copyright © 2009 by ASME

reattachment lengths that generally are similar to those for The goal of the present work was to apply CFD to predict
channels with larger steps, as for Armaly et al. [1] and the changes in the velocity field and the separation produced by air
present study. filters, modeled as porous media, placed downstream of the
step.
Relevant Studies on Flow with Porous Media
Flows in porous media have received considerable
attention, but much of the research has been concerned with NOMENCLATURE
convection heat transfer or the microscopic flow through the Dh Inlet channel hydraulic diameter (Dh = 2S)
media, as considered by Chen and Chen [26] and Chwang and h Step height (h = S)
Chan [27]. Pedras and de Lemos [28] review advances in Re Reynolds number (Re = 2UmeanS/ν)
mathematical modeling of macroscopic flow through porous S Inlet channel height (S = h)
media. Turbulence models for flow in porous media have been Umax Maximum velocity, on inlet channel centerline
developed, as described by Antohe and Lage [29], Lee and Umean Average velocity across inlet channel
Howell [30], Masuoka and Takatsu [31], Nakayama and X Streamwise coordinate from step
Kuwahara [32], Pedras and de Lemos [33]. In addition to the Y Vertical coordinate
flow within the media, the interactions between the porous Z Spanwise coordinate
medium and the external flow is a topic fundamentally
important to the subject of improving air filter housing designs, NUMERICAL APPROACH
especially when separated flows are involved. The computations of the flows through the homogeneous
The resistance of an automotive air filter is not large pure fluid and the porous medium were performed
enough to make the flow through the filter uniform. Liu et al. simultaneously, using Reynolds-averaged Navier Stokes
[34] showed that the addition of the design load of filter test equations for the pure fluid and an extended form of Darcy’s
dust to a clean filter mounted in a laboratory test housing with a law for the porous medium. A schematic of the computational
non-uniform velocity distribution is not sufficient to render the domain is presented in Figure 2. Computations were performed
distribution uniform. They showed that for a clean filter with for cases with and without the downstream section of porous
an initial pressure drop of 361 Pa, test dust loads producing media. In order to compare the computational results to the
additional pressure drops of 2.49 kPa (terminal pressure drop – experiments of Yao et al. [8], the expansion ratio across the step
design dust loading) and 3.74 kPa (150 percent of terminal was 2:1, with the step height h equal to the approach channel
pressure drop) were sufficient to alter the flow, but not height, S = 25 mm. The porous media was positioned at either
sufficiently large to make the velocity distribution uniform. 4.25h or 6.5h downstream of the step. The entrance length to
the step was 10h. The length of the region downstream of the
Al-Sarkhi [35] and Al-Sarkhi and Chambers [36] describe step was 28h.
measurements of velocity distributions in a generic model
vehicular filter housing configured with rectangular cross Umax
sections with and without a real automotive filter present. In
this model, the flow approached the filter tangent to its inlet
Filter
plane and was required to make a 90 degree turn to pass
Inflow s
through the filter to the exit duct, which was perpendicular to Dividing
the entrance duct. The presence of the filter did change the flow Streamline Outflow
distribution from that of a separating corner flow, as measured
Y
without the filter present, but the resulting distribution was far h
from uniform, resembling the no-filter flow much more than a X
uniform flow. Yao [7] performed LDA measurements of the
velocity field inside a filter housing that very closely modeled a Re-attachment Length
production housing, finding that about 40% of the flow 13 mm (XR)
from the upstream filter surface is reversing and recirculating
Figure 2. Computational Domain of Backward-Facing Step
flow.
Flow with Downstream Porous Media
As noted earlier, Assato et al. [5], Chan and Lien [6]
FLUENT commercial software was used for the
performed CFD studies on step flows with segments of porous
computations. The numerical methods used in the clear fluid
media inserted in contact with the downstream side of the step.
and porous media regions are described below.
They examined the effects of the porous media in shortening
the reattachment length or completely eliminating separation
Pure Fluid Region
outside the porous medium. They considered the thickness of
Several turbulence models are available in FLUENT,
the porous medium segment and the porosity, permeability, and
ranging from one-equation to multi-equation Reynolds Stress
Forchheimer constants used to model it. They found that the
models. The performance of the models was evaluated in
porous media could reduce the reattachment length.
preliminary computations for the simple backward-facing step

3 Copyright © 2009 by ASME

flow without the porous medium. Fluent’s versions of the Boundary Conditions for the Clear Fluid and Porous Media
Spalart-Allmaras model, the Standard k-ε model, the Referring to Figure 2, the following boundary conditions
Renormalization Group k-ε model, the Realizable k-ε model, were applied for the computations. The walls all were no-slip
the Standard k-ω model, the Shear Stress Transport k-ω model, walls. The inlet flow was uniform velocity with 10%
and the Reynolds Stress model were tested on this flow, turbulence intensity. The outlet flow boundary condition was
comparing results to those in the literature. The Realizable k-ε continuative: fully-developed. FLUENT’s porous jump
model was chosen for the computations reported here as it boundary conditions were applied at the interface of the clear
provided good results for flow features such as the fluid and the porous media. Note that the boundary conditions
reattachment length while requiring less computational time at this interface are complex, for the clear fluid must suddenly
than the Reynolds Stress model. The Realizable k-ε model is adjust to flow through the porous media, which has a smaller
better-suited for flows with separation and recirculation than open area. The fluid velocity must increase at the interface.
the Standard k-ε model. It differs in the formulation of the
turbulent viscosity and in the model equation for the Turbulent Wall Function Approach
dissipation, ε. The momentum equation and the transport The grids used for the computations were not sufficiently
equations for k and ε were discretized using the second-order fine to resolve motion all the way to the sublayer at the wall.
upwind method. The relatively coarse grid was matched to the wall using
FLUENT’s Enhanced Wall Function option. This wall function
Flow inside the Porous Medium provides matching closer to the wall than standard log law-
FLUENT provides two options for modeling flow through based wall functions using an equation which blends the
porous media. They are (a) Porous Media boundary condition turbulent flow region into the sublayer.
and (b) Porous jump boundary condition. The latter is a one-
dimensional simplification of the former and is recommended Numerical Convergence
for modeling flows through filters, screens, etc. The porous For iterative convergence, the value of residual error
jump model is considered to be more robust and to yield better (magnitude of difference between the two sides of the
convergence. However, in the present study, two-dimensional difference equations) should be set to a very low value (Roache
features of the flow are important, so the Porous Media [38]). In the present study, it is set to 1E-6.
boundary condition option was used. The Extended Darcy Law
used in FLUENT to relate pressure drop and velocity used is Grids and Grid Independence
given in Equation 1. For the results reported here, the computational geometry
was fixed, but the flow Reynolds number was varied from 2000
⎛μ 1 ⎞ to 10,000. The Reynolds numbers were based on approach
Δp = −⎜ v + C2 ρ v2 ⎟b (1) channel mean velocity and hydraulic diameter, 2S, and inlet
⎝ α 2 ⎠
mean velocity, following Armaly et al. [1]. The different
Reynolds numbers placed somewhat different requirements on
The media permeability, α, pressure jump coefficient, C2,
the grid. The adequacy of the grid was evaluated through grid
and thickness, b, are user inputs through the pressure-jump
independence studies performed for the geometries of both
panel. Tebbutt [37] obtained values of permeability and
Armaly et al. [1] and Yao [7]. Details of the geometries are
pressure-jump coefficient by performing experiments on a
presented in Table 2. Reattachment length downstream of the
single 1 mm thick flat sheet of air filter paper. Yao [7] used a
step was used as a fundamental criterion for comparing the
pleated filter in his experiments. The values of parameters used
different turbulent models and the grids. The phenomena of
by Yao in his numerical studies are shown in Table 1.
flow separation and reattachment length are strongly dependent
on the correct prediction of the development of turbulence near
Table 1: Values for Porous Jump Boundary Conditions the walls. Grid density may result in excessive numerical
diffusion, enhancing viscous effects and leading to inaccurate
SI simulations. The grid independence studies enabled us to
Property Value observe the sensitivities of the dependent variable, the
units reattachment length to multiple refined spatial grids (AIAA
1.17 * 10- Editorial Policy Statement, AIAA [39]). Structured meshing
Permeability 9
m2 was employed for all cases in the present study.
Grid independence studies were conducted at Re = 7000
Thickness 15 * 10-3 m for the backward-facing step geometry of Armaly et al. [1].
Pressure Jump Their experiment found the reattachment length to be around
4.53 * 103 1/m 7h. It can be observed from Table 3 that Grid 1 with an X-grid
Coefficient spacing of 0.5 mm under-predicts the reattachment length while
Grid 3 predicts the reattachment length closest to the
experimental value. It was observed that any X-grid spacing

4 Copyright © 2009 by ASME

greater than 2mm (Grid 4) resulted in a further decrease of Figure 3 shows the structured grid generated for the
reattachment point. The Y-grid spacing of 0.143 mm is unique geometry of Yao et al. [8]. Figure 4 shows a clustered grid with
for Re = 7000 and has been chosen so that the first grid point a filter 4.25h downstream of the step. Note that the figures are
above the wall lies in the region of y+ = 11 to meet the criterion truncated: not all of the inlet and exit sections are shown.
for the Enhanced Wall Function. Conditions for fully
developed flow in the inlet channel were used to define this
point. Finally, a clustered grid was constructed (according to
the dimensions of Grid 3) that utilized a finer mesh near the
step and coarser mesh away from the step.
Table 2: Geometries of Experiments Considered

Geometry Expansio Step Channel Aspect

n Ratio Height Height Ratio
(h) mm (s) mm (AR)
Armaly et
1.94:1 4.9 5.2 36:1 Figure 3. Structured Grid for Geometry of Yao et al. [8] –
al. [1]
No-Filter Case
Yao et al.
2:1 25 25 20:1
[8]

Table 3. Grid Independence Study for Geometry of Armaly

et al. [1] with Realizable k-ε model at Re = 7000

Grid Reattachment Length

Case Grid Size (step
Δx, Δy (mm) (mm) heights)
1 0.5, 0.143 29.5 6.02 h
Figure 4. Clustered Grid for Geometry of Yao et al. [8] –
2 1.0, 0.143 31.5 6.43 h Filter at 4.25 Step Heights
3 1.5, 0.143 32.25 6.58 h
The Reynolds number was varied in computations for this
4 2.0, 0.143 32 6.53 h geometry for the case without a filter to check that the grids
provided results which compared well with experiments. Good
5 Clustered Grid 32.25 6.58 h agreement was observed for all Reynolds numbers except Re =
2000, as shown in Table 5. Other turbulence models were tried
The same approach was used for grid independence studies at this Reynolds number along with various combinations of
for the geometry of Yao [7]. The Realizable k-ε model was clustered and regular grids. However, it was found that the
used for Re = 6550 with a Y-grid spacing of 0.75 mm meeting expected reattachment length of approximately 8h, as in the
the requirements of the Enhanced Wall Function, again using experiment conducted at Re = 2802, was difficult to simulate in
conditions for fully developed flow in the inlet channel. Table 4 FLUENT.
shows good agreement with experimental result of 6.5h. This difficulty may be attributed to the transitional flow
Table 4. Grid Independence Study for Geometry of Yao et regime from Re = 1000 to 6600 described by Armaly et al. [1].
al. [8] with Realizable k-ε model at Re = 6550 The turbulence models in FLUENT are unable to simulate this
regime. Moreover, most of the numerical studies in the
Reattachment Length literature avoid this regime and concentrate either on the low
Grid Case Grid Size laminar regime Reynolds numbers or high Reynolds number
(step
Δx, Δy (mm) (mm) turbulence. No previous studies have reported results at the
heights)
1 0.5, 0.75 29.5 6.13 h transitional Reynolds numbers. Difficulties with the
computations at these Reynolds numbers are considered further
2 1.5, 0.75 31.5 6.42 h in the next section.
3 2.5, 0.75 32.25 6.32 h
4 3.5, 0.75 32 6.37 h
5 Clustered 32.25 6.32 h

5 Copyright © 2009 by ASME

 Table 5: Reattachment Lengths at Different Reynolds computational results in Figure 5 show very small effects of
Numbers Using Realizable k-ε Model for Geometry of Yao Reynolds number, with the separation zones similar in height
et al. [8] and with reattachments all close to X/h = 6. The reattachment
point from our computations for the no-filter case at Re = 6550
Reynolds Grid Size Reattachment Length is found close to 6.5h in good agreement with the experimental
Number Δx, Δy (mm) Experimental FLUENT results and the DNS results of Le et al. [20] at Re =5100. The
present studies provided no evidence of a nonstep side vortex
2000 3.5, 2 4.9h at Re = 6550, in agreement with the experimental results of
2802 8h both Yao and Armaly et al. [1].

3750 3.5, 1.15 6.23h 1.0

0.9
5549 6h
0.8

6550 3.5, 0.75 6.37h 0.7

0.6
10000 3.5, 0.5 6.6h
Y/h 0.5
10247 6.5h 0.4
Re = 2000
16156 7h 0.3
Re = 3750
0.2 Re = 6550

0.1 Re = 10000
RESULTS AND DISCUSSION
0.0
The computations were performed for conditions which 0 1 2 3 4 5 6 7
fell in the range of the experiments reported by Yao et al. [8]. X/h
For more details, see Yao [7]. The Reynolds numbers for the
computations were 2000, 3750, 6550, and 10,000, while those Figure 5. CFD Separation Lines with No Filter for
of the experiments were 2802, 5549, 10247, and 16156. An Different Reynolds Numbers
inconsistency discovered between the definition of Reynolds
number used for the experiments and the computations resulted 1.0

in this mismatch, shown in Table 5. For the experiments, the 0.9

Reynolds numbers based on inlet channel maximum velocity 0.8

and inlet channel height were 2000, 3750, 6550, and 10,000. 0.7
The experimental flow field behavior was determined by 0.6
measuring velocity profiles with a two-component Laser Y/h 0.5
Doppler Anemometer (LDA). The effects of the downstream 0.4
filter upon the backward-facing step flow were evaluated 0.3
Re = 2802

through the changes produced in the velocity profiles and in the Re = 5549
0.2 Re = 10247
extent of the recirculating, separated flow zone. Reattachment Re = 16156
0.1
points were not measured. Reattachment points, of course, are
not completely stationary, but exhibit some back and forth 0.0
0 2 4 6 8
motion, as observed in the DNS study of Le et al. [20]. The
X/h
measurement results presented are the mean locations. It is
difficult to measure the reattachment point, as one must either Figure 6. Experimental Separation Lines with No Filter for
extrapolate a velocity profile to the wall or measure the Different Reynolds Numbers (Yao et al. [8])
location of zero wall shear stress. Instead, Yao determined what
he called mean separation lines, the lines of zero stream-wise The very large difference between the experimental
mean velocity. The locations of zero velocity were found by separation regions and reattachment shown in Table 5 and in
linear interpolation between adjacent points of opposite sign in Figure 6 and the computational results of Figure 5 provides
the measured mean velocity profiles. These separation lines evidence that the turbulence model does not capture the physics
define the shape and extent of the separated region, providing of the flow at the lowest Reynolds numbers. These
more insight to the interaction between the filter and the discrepancies may be attributed to the fact that these Reynolds
recirculating flow than just the location of the reattachment number are in the transitional flow regime, as defined by
point. Armaly et al. [1], with the flow also losing its two-
dimensionality. Our literature review found no numerical
Flows with No Filter Downstream of Step studies which have attempted to simulate flow at the
The separation lines for the various Reynolds numbers of transitional Reynolds numbers between 2000 and 6600.
the present computations and the experiments of Yao et al. [8] Experimental observations of Yao et al. [8] and Armaly et al.
may be compared in Figure 5 and 6, respectively. The [1] agree in showing that the reattachment points at Reynolds

6 Copyright © 2009 by ASME

numbers in the 2000 to 3000 range for the no-filter case are before reattachment. In general, the experiments indicate
much greater than those for fully turbulent Reynolds numbers somewhat larger areas of separated, recirculating flow than thet
(Re > 6600). The unusual shape of the experimental separation computations.
line at Re = 5549 may be an effect of intermittent transition or 1.0
the three dimensionality of reattachment observed by Armaly et
0.9
al. in this range.
0.8
Additional evidence of the inability of the computations to 0.7
delineate the separation zone well in this Reynolds number 0.6
range is presented in Figure 7 for Re = 2800. Computed Y/h 0.5
velocity profiles at X/h = 4 and 5 are shown along with the 0.4
experimentally measured locations of the separation line, the Re = 2000
0.3
points at which the local velocity is zero. The experimental Re = 3750
0.2
locations are considerably higher than the computational Re = 6550
Re = 10000
results, suggesting later reattachment. In light of the preceding 0.1

discussion of transition phenomena, the computational results 0.0

at the lower Reynolds numbers and will not be considered in 0 1 2 3 4 5 6 7

detail. The computation results at a Reynolds number of 10,000 X/h

may be considered in the turbulent regime and will be Figure 8. Separation Lines for Current CFD with Filter at
examined in detail, as the turbulence model used in the 6.75h
computations is considered appropriate for these conditions.
2.0 1.0

1.8 0.9

0.8
1.6 Current CFD X/h = 4
Separation Line at X/h = 4 - Experiment [8] 0.7
1.4
Current CFD X/h = 5 0.6
1.2
Separation Line at X/h = 5 - Experiment [8] Y/h 0.5
Y/h 1.0
0.4
0.8 Re = 2802
0.3 Re = 5549
0.6
0.2 Re = 10247
0.4
Re = 16156
0.1
0.2
0.0
0.0
0 1 2 3 4 5 6 7
-0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0
X/h
U/Umax Figure 9. Separation Lines for Experiments of Yao et al. [8]
with Filter at 6.75h
Figure 7. Comparison of Current CFD Velocity Profiles
and Experimental Separation [8], No Filter, Re = 2800 The computed and measured separation lines with the filter
at X/h = 6.75 for Re = 10,000 are compared to the lines with no
Flows with Filter at X/h = 6.75 filter in Figure 10. Virtually negligible differences in the
The present computations and the experiments of Yao [7] reattachment points are evident and both cases show separation
were performed with and without the porous media or filter regions that are about the same in area. In both cases, the
placed at 4.25 and 6.75 step heights downstream of the step. computed separation lines are lower than the measured lines.
Let us first consider the case of the filter placed at X/h = 6.75, The computed and measured velocity profiles for the two
very close to the expected reattachment point for the higher cases at X/h = 6.25, just upstream of the filter location, are
Reynolds numbers. presented in Figure 11. The experimental profiles display
Figures 8 and 9 present the computed and measured virtually no effect of the filter, while the computational profiles
separation lines at the various Reynolds numbers for the filter show that the filter moves the location of the maximum
at X/h = 6.75. The lines agree in showing that the reattachment velocity slightly lower. This goes along with the earlier
occurs just upstream of the filter. The computed reattachment reattachment predicted by the computations. This shows that in
points appear a little earlier than the experimental ones. The the present numerical studies, the filter placed at 6.75h affects
pressure gradients imposed by the presence of the filter seem to the flow field slightly more than in the experiments.
have controlled the reattachment point. Note that the
computational results show almost no effect of Reynolds
number while at the lowest Reynolds number the experimental
separation region extends considerably higher above the wall

7 Copyright © 2009 by ASME

 1.0 1.0
0.9 0.9
0.8 0.8
0.7 0.7
0.6 0.6
Y/h 0.5 Y/h 0.5
0.4 0.4
Re = 2000
0.3 No Filter - Experiment [8]
0.3
Re = 3750
No Filter - Current CFD
0.2 0.2 Re = 6550
Filter at 6.75h - Experiment [8]
0.1 0.1 Re = 10000
Filter at 6.75h - Current CFD
0.0 0.0
0 1 2 3 4 5 6 7 0 1 2 3 4 5
X/h X/h
Figure 10. Comparison of Separation Lines for Current Figure 12. Separation Lines for Current CFD with Filter at
CFD and Experiment [8], No Filter and Filter at 6.75h, Re 4.25h
= 10,000
1.0
2.0
0.9
1.8
0.8
1.6 No Filter - Experiment [8]
0.7
1.4 No Filter - Current CFD
0.6
1.2 Filter at 6.75h - Experiment [8]
Y/h 0.5
Y/h 1.0 Filter at 6.75h - Current CFD
0.4
0.8 Re = 2802
0.3 Re = 5549
0.6
0.2 Re = 10247
0.4
Re = 16156
0.1
0.2

0.0
0.0
- 0.2 0.0 0.2 0.4 0.6 0.8 1.0 1.2
0 1 2 3 4 5

U/Umax X/h

Figure 11. Comparison of CFD Velocity Profiles to Figure 13. Separation Lines for Experiments of Yao et al.
Experiments of Yao et al. [8] at 6.25h with and without [8] with Filter at 4.25h
Filter at 6.75h, Re = 10,000 1.0
No Filter - Experiment [8]
0.9
Flows with Filter at X/h = 4.25 0.8
No Filter - Current CFD
Filter at 4.25h - Experiment [8]
The second case studied had the porous media or filter
0.7 Filter at 4.25h - Current CFD
placed at X/h = 4.25, well before the expected reattachment
0.6
point for all Reynolds numbers. Figures 12 and 13 present the
Y/h
computed and measured separation lines for this case at the 0.5

various Reynolds numbers. Again, the lines agree in showing 0.4

that the reattachment occurs just upstream of the filter; the 0.3

filter has changed the reattachment point. The computed results 0.2

also are similar to the computed results with the filter at 6.75 in 0.1

showing almost no effect of Reynolds number. The 0.0

experimental separation lines are similar except at the highest 0 1 2 3 4 5 6 7

Reynolds number, and again show somewhat larger areas of X/h

separated flow. The dramatic effects of the filter in controlling
Figure 14. Comparison of Separation Lines for Current
the reattachment point and reducing the separated flow region
CFD and Experiment [8], No Filter and Filter at 4.25h, Re
are shown in Figure 14, which compares the separation lines
= 10,000
with and without the filter at X/h = 4.25 for Re = 10,000. The
separated flow region does not penetrate the filter.
The velocity profiles at 3.75h and Re = 10,000 predicted
by our computations and measured in the experiments
described by Yao et al. [8] for the no-filter and filter at 4.25h
cases are shown in Figure 15. The computational and

8 Copyright © 2009 by ASME

experimental results agree in showing that the filter pushes the The inertial constant was varied from 4.53E01 to 4.53E05
maximum velocity region downwards toward the center of the m-1. The media thickness was varied from 3.5 to 25 mm. We
channel to present a more symmetrical profile, while the no- found virtually no changes in the computed separation lines
filter case exhibits a more skewed velocity distribution. The resulted from these changes in permeability, inertial constant,
presence of the filter clearly has a very large effect upon the and media thickness for this filter location and Reynolds
flow upstream. number. Similar results for variation in permeability were
obtained by Yao [7] in computations for low Reynolds number
2.0
for laminar step flows with the filter placed far downstream
1.8
from the step (20.55h). However, it was observed that the
1.6
No Filter - Experiment [8] velocities and the flow-field inside the filter and downstream of
1.4
No Filter - Current CFD the filter are affected strongly by the variation of these
1.2
Filter at 4.25h - Experiment [8] parameters. These results are considerably different from the
Y/h 1.0 Filter at 4.25h - Current CFD results of Chan and Lien [6] and Kuznetsov [40], who
0.8 examined the effects of porous inserts placed in contact with
0.6 the wall of the step. A major factor in the difference likely is
0.4 the location of the porous insert adjacent to the step.
0.2 Additionally, the order of the Darcy number for the present
0.0 study is from 10-9 to 10-5, while the referenced flows were at
-0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0 1.2 10-4 to 10-1. Finally, one must consider differences in the
U/Umax treatment of the boundary condition at the upstream interface
of the clear fluid and the porous medium in our computations.
Figure 15. Comparison of CFD Velocity Profiles to
Experiments of Yao et al. [8] at 3.75h with and without
CONCLUSIONS
Filter at 4.25h, Re = 10,000
Computations were performed for a backward-facing step
flow without a filter downstream and with filters placed at 4.25
Effects of Variation of Porous Media Permeability, Inertial
and 6.75 step heights downstream using the Realizable k-ε
Constant and Thickness on Separation
turbulence model in FLUENT and a modified Darcy law
The computed separation lines agree fairly well with the
porous medium. Conditions matched those of the experiments
experimental results reported of Yao [7] for the case of the filter
of Yao [7]. The step Reynolds number was varied from Re =
placed at 4.25h at Re = 10000, as may be observed in Figures
2000 to 10000. Filter porous media permeability, inertial
14 and 15. Hence, this case was chosen to investigate the
constant, and thickness were varied to investigate their effect
effects of the characteristics of the porous media on the
on the flow upstream of the filter. The following conclusions
computed upstream velocity field. Computations were
were drawn from this study:
performed for different media permeability, inertial constant,
and thickness. Figure 16 shows the results for the variation in • For Re = 2000 and 3750, the present computational
permeability. Permeability was varied over four orders of predictions do not compare well with the experimental
magnitude, from 1.17E-07 to 1.117E-11 m2. results in the literature or those of Yao [7] without the
1 filter. These differences may be attributed to three-
0.9
dimensionality and transition at these low Reynolds
0.8
numbers.
0.7 • Reattachment lengths both with and without the filter are
0.6 predicted well by the computations except for the case
with Re = 2000.
Y/h

0.5

0.4
a = 1.17e-7 m 2
• Good agreement between the present computations and the
0.3 a = 1.17e-9 m 2 experiments of Yao [7] is observed for Re = 10000. The
0.2 a = 1.17e-11 m 2 computations also compare well with the experimental
0.1 observation of Armaly et al. [1] for their case of Re =
7000. The computations appear to capture the physics of
0
0 0.5 1 1.5 2 2.5 3 3.5 4 the recirculation region to a better extent at higher
turbulent Reynolds numbers. The computed velocity
X/h
profiles at Re = 10000 also show the trends caused by the
Figure 16. Effects of Varying Permeability, α, on filters observed in Yao’s experiments.
Computed Separation Lines for Filter at 4.25h, Re = 10000. • When the filter is placed deep into the separation zone of
Inertial ConstantC2 = 4.533E03 m-1, Thickness b = 15 mm the non-filter flow, at 4.25 step heights, the flow reattaches
upstream of the filter, reducing the size of the separated

9 Copyright © 2009 by ASME

 flow zone. Separation does not appear to extend into the 12 Kim, J., Kline, S., and Johnston, J. P., 1980, "Investigation
filter. of a Reattaching Turbulent Shear Layer: Flow over a
Backward-Facing Step," Journal of Fluids Engineering, 102,
• With the filter placed at 4.25h and Re = 10000, the
pp. 302-308.
variation of permeability from 1.17E-7 to 1.17E-11 m2,
13 Durst, F., and Tropea, C., 1981, "Turbulent, Backward
inertial constant from 4.533E01 to 4.533E06 1/m, and
Facing Step Channel in Two-Dimensional Flows," Third
filter thickness from 3.5 to 25 mm, have no apparent effect
International Symposium on Turbulent Shear Flows, Davis,
on the recirculation region upstream of the filter.
California, USA, pp. 18.1-18.6.
14 Sinha, S. N., Gupta, A. K., and Oberai, M. M., 1981,
ACKNOWLEDGMENTS "Laminar Separating Flow over Backsteps and Cavities Part I:
The authors wish to thank the Oklahoma Center for the Backsteps," AIAA Journal, 19(12), pp. 1527-1530.
Advancement of Science and Technology Applied Research 15 Jovic, S., and Driver, D. M., 1995, "Reynolds Number
Program and Purolator Products, Inc. for their support. Effect on the Skin Friction in Separated Flows Behind a
Backward-Facing Step," Experiments in Fluids, 18(6), pp. 464-
REFERENCES 467.
1 Armaly, B. F., Durst, F., Pereira, J. C. F., and Schönung, 16 Piirto, M., Saarenrinne, P., Eloranta, H., and Karvinen, R.,
B., 1983, "Experimental and Theoretical Investigation of 2003, "Measuring Turbulence Energy with PIV in a Backward-
Backward Facing Step Flow," Journal of Fluid Mechanics, Facing Step Flow," Experiments in Fluids, 35, pp. 219–236.
127, pp. 473-496. 17 Driver, D., and Seegmiller, H. L., 1985, "Features of a
2 Kaiktsis, L., Karniadakis, G. E., and Orszag, S. A., 1991, Reattaching Turbulent Shear Layer in Divergent Channel
"Onset of Three-Dimensionality, Equilibrium and Early Flow," AIAA Journal, 23(2), pp. 163-171.
Transition in Flow over a Backward-Facing Step," Journal of 18 Lee, T., and Mateescu, D., 1998, "Experimental and
Fluid Mechanics, 231, pp. 501-528. Numerical Investigation of 2-D Backward-Facing Step Flow,"
3 Badran, O. O., and Bruun, H. H., 1999, "Comparison of Journal of Fluids and Structures, 12, pp. 703-716.
Flying-Hot-Wire and Stationary-Hot-Wire Measurements of 19 Kim, J., and Moin, P., 1985, "Application of a Fractional-
Flow over a Backward-Facing Step," Journal of Fluids Step Method to Incompressible Navier-Stokes Equations,"
Engineering, 121, pp. 441-445. Journal of Computational Physics, 59, pp. 308-323.
4 Scarano, F., and Riethmuller, M. L., 1999, "Iterative 20 Le, H., Moin, P., and Kim, J., 1997, "Direct Numerical
Multigrid Approach in PIV Image Processing with Discrete Simulation of Turbulent Flow over a Backward-Facing Step,"
Window Offset," Experiments in Fluids, 26, pp. 513-523. Journal of Fluid Mechanics, 330(1), pp. 349-374.
5 Assato, M., Pedras, M. H. J., and de Lemos, M. J. S., 2005, 21 Jovic, S., and Driver, D., 1994, "Backward-Facing Step
"Numerical Solution of Turbulent Channel Flow Past a Measurements at Low Reynolds Number, Reh = 5000," NASA
Backward-Facing Step with a Porous Insert Using Linear and TM 108807, NASA.
Nonlinear K-E Models," Journal of Porous Media, 8(1), pp. 22 Yoo, J. Y., and Baik, S. J., 1992, "Redeveloping Turbulent
13-29. Boundary Layer in the Backward-Facing Step Flow," Journal
6 Chan, E. C., and Lien, F.-S., 2005, "Permeability Effects of of Fluids Engineering, 114, pp. 522-529.
Turbulent Flow through a Porous Insert in a Backward-Facing- 23 Chiang, T. P., and Sheu, T. W. H., 1999, "A Numerical
Step Channel," Transport in Porous Media, 59, pp. 47–71. Revisit of Backward-Facing Step Flow Problem," Physics of
7 Yao, S. H., 2000, "Two Dimensional Backward Facing Fluids, 11(4), pp. 862-874.
Single Step Flow Preceding an Automotive Air Filter," Ph.D. 24 Kostas, J., Soria, J., and Chong, M. S., 2002, "Particle
Thesis, Oklahoma State University, Stillwater, Oklahoma. Image Velocimetry Measurements of a Backward-Facing Step
8 Yao, S., Krishnamoorthy, C., and Chambers, F. W., 2007, Flow," Experiments in Fluids, 33, pp. 838–853.
"Experiments on Backward-Facing Step Preceding a Porous 25 Wengle, H., Huppertz, A., Bärwolff, G., and Janke, G.,
Medium," 5th Joint ASME/JSME Fluids Engineering 2001, "The Manipulated Transitional Backward-Facing Step
Conference, San Diego, CA, ASME, FEDSM2007-37204. Flow: An Experimental and Direct Numerical Simulation
9 Krishnamoorthy, C., 2007, "Numerical Investigation of Investigation," European Journal of Mechanics - B/Fluids, 20,
Backward Facing Step Preceding a Porous Medium Using pp. 25–46.
Fluent," M.S. Thesis, Oklahoma State University, Stillwater, 26 Chen, F., and Chen, C. F., 1992, "Convection in
Oklahoma. Superposed Fluid and Porous Layers," Journal of Fluid
10 Eaton, J. P., and Johnston, J. P., 1981, "A Review of Mechanics, 234, pp. 97-119.
Research on Subsonic Flow Reattachment," AIAA Journal, 27 Chwang, A. T., and Chan, A. T., 1998, "Interaction
19(9), pp. 1093-1100. between Porous Media and Wave Motion," Annual Review of
11 Biswas, G., Breuer, M., and Durst, F., 2004, "Backward- Fluid Mechanics, 20, pp. 53-84.
Facing Step Flows for Various Expansion Ratios at Low and 28 Pedras, M. H. J., and de Lemos, M. J. S., 2001,
Moderate Reynolds Number," Journal of Fluids Engineering, "Macroscopic Turbulence Modeling for Incompressible Flow
126, pp. 362-374. through Undeformable Porous Media," International Journal
of Heat and Mass Transfer, 44(6), pp. 1081-1093.

10 Copyright © 2009 by ASME

29 Antohe, B. V., and Lage, J. L., 1997, "A General Two-
Equation Macroscopic Turbulence Model for Incompressible
Flow in Porous Media," International Journal of Heat and
Mass Transfer, 40(13), pp. 3013-3024.
30 Lee, K., and Howell, J. R., 1987, "Forced Convective and
Radiative Transfer within a Highly Porous Layer Exposed to a
Turbulent External Flow Field," Proceedings of ASME/JSME
Thermal Engineering Joint Conference, Washington, DC,
ASME, Washington, DC, 2, pp. 377-386.
31 Masuoka, T., and Takatsu, Y., 1996, "Turbulence Model for
Flow through Porous-Media," International Journal of Heat
and Mass Transfer, 39(13), pp. 2803-2809.
32 Nakayama, A., and Kuwahara, F., 1999, "Macroscopic
Turbulence Model for Flow in a Porous Medium," Journal of
Fluids Engineering, 121(2), pp. 427-433.
33 Pedras, M. H. J., and de Lemos, M. J. S., 2003,
"Computation of Turbulent Flow in Porous Media Using a
Low-Reynolds Number K-E Model and an Infinite Array of
Transversely Displaced Elliptic Rods," Numerical Heat
Transfer, 43(6), pp. 585-602.
34 Liu, G., Tebbutt, C. B., Duran, R., and Chambers, F. W.,
1996, "Filter Inlet Velocity Redistribution with Filter Loading,"
Particulate Science and Technology, 14(3), pp. 279-291.
35 Al-Sarkhi, A., 1999, "Optimization Technique for Design
of Automotive Air Filter Housing with Improved Fluid
Dynamic Performance," Ph.D. Thesis, Oklahoma State
University, Stillwater, Oklahoma.
36 Al-Sarkhi, A., and Chambers, F. W., 2004, "Optimization
Technique for Design of Automotive Air Filter Housings with
Improved Fluid Dynamic Performance and Filtration,"
Particulate Science and Technology, 22(3), pp. 235 - 252.
37 Tebbutt, C. B., 1995, "Cfd Model of Flow through Air
Filter Pleats," Oklahoma State University, Stillwater, OK.
38 Roache, P. J., 2002, "Code Verification by the Method of
Manufactured Solutions," Journal of Fluids Engineering, 124,
pp. 4-10.
39 AIAA, 1998, "Guide for the Verification and Validation of
Computational Fluid Dynamics Simulations," AIAA-G-077.

11 Copyright © 2009 by ASME

View publication stats