Armaly 1983
Armaly 1983
Armaly 1983
4 7 3 4 9 6 473
Printed in &eat Britain
1. Introduction
The phenomena of flow separation of internal flows caused by sudden changes in
test-section geometries is well known. The importance of such flows to engineering
equipment has been stressed in many publications (e.g. see Abbot & Kline 1962 ; Seban
1964; Goldstein et al. 1970), and attempts have been made to develop advanced
experimental and theoretical techniques in order to study carefully flows with
separation regions (e.g. see Durst & Whitelaw 1971; Gosman & Pun 1974; Kumar
& Yajnik 1980). However, it is only recently that these techniques have reached the
required state of development so that they can be of immediate use in fluid-mechanical
studies of internal flows with regions of recirculation. I n the past, most experimental
studies relied on flow-visualization techniques and/or heat- and mass-transfer
f Present address : Mechanical Engineering Department, University of Missouri-RolIa, Rolla,
Missouri 65401, U.S.A.
$ Present address : Lehrstuhl fur Stromungsiehre, UniversitLt Erlangen-Nurnberg, Egerland-
straBe 13, D-8520 Erlangen, F.R.G.
474 B, F . Armaly, F. Durst, J . C. F . Pereira and B . Schonung
measurements to obtain fluid-mechanical information in separated flows. Although
the obtainable information was limited to what one might call integral flow
parameters, several publications have become available for various flow geometries.
Among these, the two-dimensional backward-facing step has received particular
attention owing to its geometrical simplicity (e.g. see Denham & Patrick 1974;
Etheridge & Kemp 1978; Wauschkuhn & Vasanta Ram 1975a, b). Relying on
information for boundary-layer flow behaviour, i t was believed that this simple flow
geometry should also yield a simple flow pattern showing a single separation region
attached to the step, as sketched in figure 1 . I n most of the available work the length
of the separation region was thought to be only dependent on the Reynolds number,
on the step height and on the momentum thickness of the oncoming flow (e.g.
Etheridge & Kemp 1978; Wauschkuhn & Vasantaram 1975a, 6). The results were
thought to scale with the dimensions of the recirculation region attached to the step
(e.g. see Gersten & Wauschkuhn 1978; Eaton & Johnstone 1980), and other regions
of detached flow were not measured and/or have not been reported so far. Most of
the existing work also concentrates on laminar and turbulent flows only, leaving out
the region of transitional flows. I n addition, very few data exist for laminar and
turbulent flows in the same test-section geometry, and hence very little is known on
the variation of flow structure with Reynolds number without any changes in
test-section geometry.
The present research work was carried out t o add t o the existing knowledge of
backward-facing step flow and to deepen the understanding of internal flow with
separation by measurements of velocity distributions over a wide Reynolds-number
range covering laminar, transitional and turbulent flows. A laser-Doppler anemometer
was employed to define quantitatively the variation of separation length with
Reynolds number and t o obtain detailed information on the velocity proflies
downstream of the step. The experimental results show mutiple regions of separation
downstream of the backward-facing step on both sides of the channel walls. Such
regions will strongly influence heat and mass transfer, as shown by Kottke, Blenke
& Schmidt (1977), Sparrow & Kaljes (1977) and Armaly, Durst & Kottke (1980).
Understanding the details of internal flows with multiple regions of separation will
also help t o deepen the insight into heat and mass transfer associated with them.
The present experimental study yielded two-dimensional flows only a t Reynolds
numbers Re < 400 and Re > 6000. I n between these Reynolds numbers the flow was
found to be strongly three-dimensional, but always maintained its symmetry to the
centreplane of the test section. This fact is clearly documented in this paper,
providing an insight into the complex structure of the flow.
Two-dimensional numerical predictions of mean-velocity distributions of
backward-facing step flow were also carried out by the authors, and examples of
predictions are presented in this paper. Good agreement is obtained between
experimental and theoretical results for Reynolds numbers smaller than 400, where
the required two-dimensionality of the flow could be maintained. Although numerical
prediction procedures encounter false diffusion, the good agreement between the
predicted and measured flow field for Re < 400 demonstrates that truncation errors
due to ‘false diffusion’ could be kept very low. For Reynolds numbers larger than
400, two-dimensional predictions were also obtained, and results are provided that
also show multiple regions of recirculating flow. The longitudinal extensions of these
regions are given in diagrams and are compared with those obtained experimentally.
Some of the differences are explained as being caused by the three-dimensionality
of the flow. These conclusions are supported by additional information drawn from
heat- and mass-transfer measurements carried out in the same test section.
Backuiard-facing step $OW 475
Flow
Expected separation / 7
I
region
d to 3D (x, y , z )
traversing table
FIGURE
1. Schematic of air tunnel and test section (dimensions in mm).
2. Experimental investigation
2.1. Apparatus and experimental procedures
The open air-driven flow channel that was used for this study is shown schematically
in figure 1 . It incorporated a two-dimensional backward-facing step that provided
an expansion ratio of 1: 1.94. The larger channel, downstream of this step, had a height
of 1-01 cm and an aspect ratio of 18:1 . The tunnel and the test section were
constructed from aluminium and all parts were machined to very close tolerances
regarding parallelity of walls, surface roughness, manufacturing of step corners, etc.
The two sidewalls were made of glass of 1 cm thickness to support the rigidity of the
test section and also to facilitate laser-Doppler measurement with forward-scattered
light. The air flow into the channel contained scattering particles of 2 p m mean
diameter provided by a silicone-oil particle generator described by Cherdron, Durst
& Whitelaw (1978). The air flow with particles was passed through a large settling
chamber and was fed through five, 6 mm diameter, bored tubes into the first part
of the flow channel. This part consisted of an expansion section packed with steel
wool to smoothen the flow and t o prevent input disturbances from affecting the
measurements. The flow was then passed through a section with flow straighteners
and was afterwards guided into a smooth contracting nozzle with a n inlet-to-outlet
area ratio of 30 : 1, The outlet of the nozzle was connected to the inlet of the channel
test section, which was 0 5 2 cm in height and 20 cm in length up to the backward-facing
step. These dimensions ensured a two-dimensional fully developed flow a t the
cross-section where the step was located for the entire Reynolds-number range
studied by the authors. The upper and the lower part of the test section and the glass
sidewalls were held parallel together by machine screws and were attached t o the
nozzle exit by location pins. The assembly of channel and inlet section was placed
on the top of a three-dimensional traversing table which allowed the measuring
I6 ELM 127
476 B. F . Armaly, F . Durst, J . C. F. Pereira and B. Schonung
Transparent sides (glass)
Laser
I II
1
- 2
I
Analog transient
processing
Hewlett-
tracker Packard
21 16 C
output
U
Time-domain
output
analyser
Printer
FIGURE
2. Block diagram of data acquisition and processing system for LDA measurements.
position t o be located t o within 0.1 mm in the x- and z-directions and t o within 0.01
mm in the y-direction. This coordinate system is indicated in figure 1 together with
the dimensions of the test section and some details of the construction.
The laser-Doppler anemometer used in the experiments was operated in the so-called
fringe mode, with an optical arrangement similar to that used by Cherdron et al.
(1978). It was employed in the forward-scattering mode and was set up to measure
only the x-velocity component, that is, parallel t o the channel walls and along the
channel axis. The optical system consisted of a 1 5 m W He-Ne laser, integrated,
although modular, transmission optics with Bragg cells, light-collecting optics with
photomultiplier, and the signal-processing equipment corresponding to the block
diagram of figure 2. Two frequency-tracking demodulators (BBC-Goertz LS301 and
Cambridge Consultants CC04) were initially used to process the Doppler signals. These
two trackers proved to be incapable of properly tracking the Doppler signals in the
neighbourhood of the reattachment zone for Reynolds numbers larger than 2000. This
was due to the high turbulence levels and the low particle rate in that region, as well
as the observed deterioration of signal quality very near t o the wall. For that reason,
a Biomation transient recorder was used t o digitize and store the signals from the
photomultiplier. The stored signals were transferred to a Hewletb-Packard Computer
( H P 21 16) and processed t o yield the required Doppler frequency of individual bursts.
Time-averaged velocities and turbulence intensities were computed from these
individual Doppler frequencies. All the results presented in this paper were processed
Backward;facing step $ow 47 7
with the signal-processing system based on the transient recorder and computer. This
is discussed in more detail in a report by Durst & Tropea (1977).
I n order to obtain information on velocity profiles, the optical system was spatially
fixed and the test section was moved in steps, with the abovementioned three-
dimensional traversing table to record the velocity information a t fixed positions in
space.
The velocity profile U(x = 0, y) a t the step was close to a parabolic profile, but
showed systematically small deviations due to the pressure gradient enforced by the
step. The inlet profile was carefully recorded for all Reynolds numbers. Flow rates
were varied by adjusting the regulating valves available in the air supply. For each
flow rate, the silicone-oil particle rate was also adjusted to insure a sufficiently high
particle concentration for each flow rate to permit proper processing of the Doppler
signals, e.g. to provide individual Doppler bursts a t a signal rate yielded approxi-
mately 100 signal bursts per second.
To measure the reattachment length at a given flow rate, the inlet velocity was
measured to determine the Reynolds number, and then the lower and the upper walls
were scanned in the x-direction and at constant and known y-position. To determine
the reattachment length of the primary separation region and the locations of
detachment and reattachment of additional regions of recirculating flow behind the
step, the position of the zero-mean-velocity line was measured. The points of
detachment and reattachment were taken as the extrapolated zero-velocity line down
the wall, e.g. xR is determined from the following relationship :
1.5
U
(mis) 1.4
1.3
1 .;
I
1.1
I 8.55 mm
I
ooooooooocDooooooooooo 0 O c
1.o I
0.9 I
0.8 I 5 6 mm
0.7 ~ ~ ~ ~ I o o ~ o 0 o ~i
- ~ o ~ ~ o o o
0.6 I
0.5 1
0.4
0.3 I
0.: I
0.1
( *
6 0.4 0.2 0.0
I
0.2 0.4 0.6 0.8
ZlL
FIGURE
3. Examples of velocity traverses to check two-dimensionality of the inlet flow.
25
20
15
-X
S
10
0
1 2 ReX10T3 4 5 6 7
FIGURE
4. Location of detachment and reattachment of the flow at the centre of the test section;
variation of locations with Reynolds number.
480 B. F . Armaly, F . Durst, J . C. F . Pereira and B. Schonung
to a constant level that characterizes the turbulent-flow regime. An additional
recirculating-flow region was measured downstream of the primary region of separ-
ation, and its extension is shown in figure 4 as x2 and x3.This additional region of
circulating flow has not been previously reported in the literature. It is very thin,
approximately 0.4mm, with an x-extension given in figure 4. It originates in the
Reynolds-number range corresponding to the early part of the transition region (see
figure 4),where the reattachment length experiences a sharp drop in its magnitude.
Cherdron et al. (1978) and Sparrow & Kaljes (1977) believed these strong variations
in flow properties to be caused by vortex shedding from the edge of the step. These
vortices were thought to approach the wall, and the second region of recirculating
flow might then be due t o the sharp change of flow direction which the eddies
experience. Figure 4 shows that the length of this secondary recirculation region
decreases rapidly with Reynolds number, and it disappears for Reynolds numbers
larger than 2300. Such a second separation bubble might have been the cause of the
second maximum observed by Sparrow & Kaljes (1977) in their mass-transfer
measurements.
The turbulent range (Re > 6600) is characterized by a constant reattachment
length. The measured value in that region is in agreement with the one measured by
Abbot & Kline (1962) in a similar test-section geometry at higher Reynolds numbers.
It is interesting to note that in the turbulent region de Brederode & Bradshaw (1972)
for H / h = 1.2 measured L i R / 8 = 6, also Moss, Baker & Bradbury (1979)for H / h = 1-1
measured XR/X = 5 5 . These data suggest that, as the expansion ratio H l h decreases,
a lower reattachment-length to step-height ratio XR/X is obtained. This is confirmed
by the detailed measurements of Durst & Tropea (1981),who studied the dependence
of the reattachment length of the primary separation region as a function of Reynolds
number and expansion ratio, but only in the larger-Reynolds number range.
An additional recirculating-flow region was measured at the upper wall downstream
of the expansion, as shown in figure 4.To the authors’ knowledge, this separation
region has also not been previously reported in the literature. It develops in the
laminar range and remains in existence throughout the transition region. I t s
appearance is due t o the adverse pressure gradient created by the sudden expansion,
and its existence is largely dependent on the expansion ratio of the backward-
facing-step flow geometry. As figure 4 shows, this separated-flow region exists for
Reynolds numbers higher than approximately 400, and i t disappears in the Reynolds-
number range that corresponds t o the end of the transitional-flow regime. Figure 4
shows the longitudinal extension of the third region of separated flow which existed
in parts of the laminar flow regime and the regime of laminar-turbulent transition.
The results indicate that the length of this separated-flow region initially increases
with Reynolds number and thereafter decreases and disappears above a Reynolds
number of approximately Re > 6600.
Figure 4 summarizes the locations of the three circulation regions. It clearly
demonstrates that, for most of the investigated Reynolds-number range, the begin-
ning of the recirculation region a t the upper wall is upstream from the reattachment
point of the primary recirculating flow region and its end is downstream from it.
Detailed measurements of the velocity profile in this flow geometry a t different
x-positions are presented in figures 5 and 6 for Reynolds numbers 1095 and 1290
respectively. Further measurements are provided in $ 3 together with flow compu-
tations. All extensive velocity measurements carried out by the authors were of the
x-velocity component U , and were taken at the centreplane of the channel, z = 0,
where the flow was expected to be two-dimensional. The measurements demonstrated
Backward-facing step flow 48 1
0 80 160 240 0 80 160 240 0 80 160 240 0 80 U(cm/s)
Y (mm)
FIGURE
5. Velocity profiles for Re = 1095.
FIQURE
6. Velocity profiles for Re = 1290.
that in the laminar range the velocity profile before before the step, e.g. at X I S = - 4,
was parabolic, and only a small deviation from that profile was measured at the step,
X I S = 0. The velocity profiles indicate that the flow separated a t the step, resulting
in one, two or three recirculation regions behind the step, and then redeveloped to
a fully developed parabolic velocity profile in the larger channel. A velocity scan at
a distance of 0 6 mm away from the upper wall is presented in figure 7 for a Reynolds
number of 1095 to demonstrate more clearly the occurrence of the recirculating flow
region at the wall opposite to the step. Along that plane 0.6 mm away from the upper
wall the velocity decreases with distance away from the step, and it becomes negative
at X = 76 mm. At X = 120 mm the region of negative velocity ends and positive
velocity occurs again beyond that point. The thickness of this recirculation region
for that Reynolds number and at X I S = 19.04 can be seen in figure 5 to be 0.22 em.
Similar results were also obtained for Reynolds numbers in the transitional-flow
regimes where both secondary separation regions existed. These measurements
indicated that, in the transitional-flow regime, the recirculating-flow region on the
upper wall moves upstream as the Reynolds number increases, and the shape of the
velocity profile within it is skewed towards the downstream end. I n the Reynolds-
number range 1200 < Re < 2300 the second recirculating-flow region on the lower
wall exists, as seen from figure 4. It occupies a relatively small volume and the
velocities within it are small.
2.2.2. Spanwise flow structure. The present experimental investigations were
carried out for inlet and outlet conditions set up to yield nominally two-dimensional
482 B. F . Armaly, F. Durst, J . C . F . Pereira a,nd B. Schonung
-40
FIGURE
7 . Scan of upper wall at a wall distance of 0.6 mm and Re = 1592.
,y 0.5-
~ e = 3 9 7 y=7-5rnm I x/S = 18.88
+ + +t/ + + + + + +
(mls)
0.4- I
32.35
0 0 0 0 0 0
0.3 - 0 8 8:s
O O 1 O O 14.3
0.2 - I
i 6.22
n a na a a
i n a a a
01
-0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0
ZlL
FIGURE
8. Example of mean-velocity profiles to demonstrate the two-dimensionality
of the flow for Re = 397.
flow at the step location and very far downstream of the step. Irrespective of this,
the flow downstream, in the immediate vicinity of the step, was found to be
three-dimensional within the Reynolds-number range 400 < Re < 6600. At Reynolds
numbers smaller than 400 the flow maintained its two-dimensionality, as indicated
in figure 8 for a Reynolds number Re = 397. This figure gives spanwise velocity
profiles at various x-locations and at a fixed y-position and shows the constant
velocity values obtained for various z-locations. The two-dimensionality of the flow
for this Reynolds number is also demonstrated by the spanwise location of the
reattachment line shown in figure 9. The same figure also provides the location of
the reattachment line for the primary separated-flow region for a Reynolds number
Re = 648, and indicates the three-dimensionality of the flow a t Reynolds numbers in
the range where multiple regions of separated flow occur. For Re = 648 more details
of the spanwise flow structure are obtained from figures 10(a, b ) , providing examples
of measured spanwise velocity profiles a t various x-locations and for two fixed
483
20 - I
-
XI o Re=648 I
S 0 Re = 3 9 7
16 - I
I
12-
8
I :t--:-:- I \
\
4-
I \
I I
I
0 - I I I 9 ~ 1 I I I I
Re = 648 I
1.5 -
y = 7.5 mm
I
U I XIS = 0.0
(m/s)
1.0-
0.5 -
0 - , , , , j , ~, I , , , , , , ,
0
-
1'0 Re = 648
- y=2,35rnm
0.8 -
0.6 -
0.4 -
0.2 -
0
-0.1 -
FIGURE
10. Examples of spanwise mean velocity profiles t o demonstrate
three-dimensionality of the flow.
484 B. F . Armaly, F . Durst, J . ( 7 . F. Pereira and B. Schonung
-
9
I2-
10-
8- x,/s ’
-
FIGURE
11. Variation of spanwise location of line of detachment and
reattachment for Re = 2690.
ZIL
FIGURE
12 (part). For caption see facing page.
U
3.0
(ds)
1 A
+
Backward-facing step JEow
y=0,25mm
y = 8.25 m m
= 9 6 5 mm
1 485
2.0 1 I
2.0
0.5
-0.5
+---TT-r-I I , , , , , I , , I
z/z
FIQURE
12. Spanwise velocity profiles for various x- and y-locations for Re = 2690.
y-positions. These clearly indicate the three-dimensionality of the flow for Reynolds
numbers in excess of approximately 400.
The three-dimensionality of the flow indicated in figure 10 is enhanced with
increasing Reynolds number. This was quantified by detailed velocity measurements
for various Reynolds numbers, and figures 11 and 12 show a data set for a Reynolds
1 - -.
^ ^ ^ ^ . . " ..
number K e = ZtiYU. 'lhe spanwise locations 01 the lines ot separation and reattachment
1
are shownin figure 11and indicate that the longitudinal extensions of the recirculating-
flow regions vary across the test-section width. Whenever the longitudinal extension
of the separation region opposite to the step increases, the primary region of separated
flow attached to the step decreases and vice versa. The extensions of both recirculating-
flow regions are strongly coupled, as can be seen from figure 11.
486 B. F. Arrnaly, F . Durst, J . C. F. Pereira and B. Schonztng
Figure 11 clearly demonstrates the three-dimensionality of the flow, and this is
further supported by the spanwise velocity profiles provided in figures 1%(a-c). The
momentum of the main flow ‘deforms’ the dividing streamline of the upper
separated-flow region, yielding streaklike flow patterns in the direction of the flow
and located across the channel test section. It is worth noting that the flow maintains
its two-dimensionality up t o x-positions where the separation region opposite to the
step starts. The two-dimensionality of the flow is again re-established at x / s z 19.
Between these 2-locations the flow is strongly three-dimensional.
3. Theoretical investigations
3.1. Governing equations and solution procedure
Predictions of the laminar flow field in a geometry equivalent to the case used for
the experiments were obtained by numerically solving the governing elliptic
differential equations. The flow was considered to be two-dimensional and stationary.
The governing equations are as follows :
continuity equation a a
-((pU)+ -@V) = 0;
ax aY
momentum equations
au
exit a t x = b : - = o ,
ac.
- -- 0.
ax ax
The length L measured from the step to the end of the calculation dbmain was selected
to be equivalent to a t least four times the experimentally measured reattachment
length of the primary recirculation region, IJ = 4X,, and the boundary condition at
that cross-section was taken as that of a fully developed flow, e.g. a U / a x = 0,
a V / a x = 0. The chosen distance L has been shown to be sufficient t o make the
reattachment length independent of the length of the calculation domain.
The solution of the above partial differential equation was obtained by using a
finite-difference numerical scheme, embodied in the computer code TEACH (see
Gosman & Pun 1974). The solution procedure starts by supplying initial guesses of
the velocity and pressure fields and then computes a converged solution by iteration.
Convergence of the solution was considered satisfactory when the normalized
residuals of each equation, summed over the whole calculation domain, were smaller
than lo-*.
Backward-facing step $ow 487
The grid distribution in the calculation domain was non-uniform in both the
longitudinal and the cross-flow coordinate directions. A large number of grid points
were placed in the areas where steep variation in velocities were expected. This
information was deduced from experiments and also from preliminary calculations
using equally spaced numerical grids. Solutions were performed with different grid
densities, and the resulting reattachment length of the primary separated-flow region
was studied as a function of grid number. The required grid number to ensure grid
independence increased with Reynolds number. It was found that a grid density of
N , x N y = 45 x 45 is sufficient to provide a primary reattachment length that is
independent of the grid density up to Reynolds number of 400. Based on this, all
the numerical results presented in this study up to Reynolds numbers of Re = 400
were obtained with a grid density of N , x N y = 45 x 45. For a Reynolds number of
100, a solution utilizing a grid density of 45 x 45 requires 400 iterations for
convergence. The number of iterations increased strongly with Reynolds number up
to values where the additional recirculation zones occur. For Re = 450 the required
number of iterations was 1200, using the same grid density. This required 75 min of
computing time on the UNIVAC 1108 of the University of Karlsruhe.
As already mentioned, the two-dimensional flow predictions also yielded multiple
regions of separated flow. I n order t o compute the flow inside these regions accurately,
the grid distribution would require further refinements, i.e. more grid points would
be needed inside the computational domain, especially in the region where the
additional separated-flow regions occur. However, the available UNIVAC 1108
computer did not permit enough grid points for grid independence to be obtained,
and hence, with the occurrence of more than one region of separated flow, complete
grid independence could not be achieved even with the highest-possible grid number
of 2600. This should be borne in mind for assessing fully the validity of the numerical
results in 53.2.
3 . 2 . Numerical results
Employing the computer code TEACH, flow predictions were carried out to deepen
insight into the flow structure of two-dimensional backward- facing step flow. These
predictions were performed up t o Reynolds numbers Re = 1250, a t which the flow
was found experimentally t o turn from the laminar- into the transitional-flow regime.
The predicted flow was two-dimensional but still showed multiple regions of
separated flow, as can be seen in figures 13(a,b). These figures show the computed
locations of separation and reattachment. They show that the length of the primary
separated-flow region is predicted t o increase nonlinearly with Reynolds number up
to Re x 420. The ensuing decrease with Reynolds number is caused by the additional
region of separated flow that occurs on the wall opposite to the step and a t Reynolds
numbers larger than Re > 420. The longitudinal dimensions of this additional region
of separated flow are predicted t o increase with Reynolds number up t o Re x 980,
above which two more regions of recirculating flow are predicted. Both are located
on the channel wall with the step; one inside the primary region of separated flow
and the other downstream of its reattachment line. These predicted locations of
separated-flow regions are shown in figures 13 ( a , b ) .
Information on the entire flow field was also deduced from the predictions. These
confirmed most of the experimental findings about the flow. For smaller Reynolds
numbers, the flow shows only the primary region of separated flow attached to the
backward-facing step. The additional separated-flow regions occur, however, for the
488 B. F . Armaly, F . Durst, J . C. F . Pereira and B . Schonung
-S
X
14
12
10
2
I
I
- Measured
......... ... Computed (b)
0 1 ReX10-3 1
- -
x3
I
3
14-
12-
-
- Predicted
Measured
10-
a-
6-
01 1 2 3 4 5 6 7 8
Re x
FIGURE
14. Comparison of experimental and theoretical results for the
reattachment length up t o Re = 800.
with the occurrence of more than one separated flow region, the flow in the experiments
becomes three-dimensional in the region downstream of the step, and this prevents
direct comparison between the experimental and theoretical results. Nevertheless, it
is still interesting t o make the comparison between the computed and measured
separation lengths x3/S. This information is provided in figure 14.
Figure 15 shows, for a Reynolds number Re = 100, comparisons of measured and
predicted mean-longitudinal-velocityprofiles. These numerical results were obtained
for a 45 x 45 grid for which grid independence was demonstrated for higher Reynolds
numbers. Up to Re x 400 the experimental and numerical results did not yield any
additional regions of separaton apart from the primary one attached to the step. Good
agreement between experiments and predictions is obtained under these conditions.
For a Reynolds number Re = 389, figure 16 shows, for the first sixteen x-stations,
the measured Uvelocity profiles and the corresponding predictions. At x-stations
between 7 and 12 step heights downstream of the step, small deviations are indicated
between the experimentally obtained velocity profiles and those deduced from
two-dimensional numerical predictions. Although experimentally and numerically no
additional region of separated flow is found on the channel side opposite to the step,
the flow inside the experimental test rig showed small deviations from two-
dimensionality. At higher Reynolds numbers a t which multiple regions of separated
flow are found experimentally and numerically, the experimental flow loses its
two-dimensionality completely. This explains the discrepancies between experimental
and numerical results indicated in figure 17.
I n the published literature only a few velocity and/or reattachment length
measurements exist for internal flow downstream of a single two-dimensional
backward-facing step. The measurements reported by Abbot & N i n e (1962) were
obtained for turbulent flow and in a test-section geometry similar to the one employed
in this study. Their measurements of reattachment length agree very well with the
authors' data for high Reynolds number, where xR/Sz 8 was found for Re z 6000.
Buckumrd-facing step jioui 491
U (cmls)
0 1020 10 20 10 20 1020 10 20 10 20 10 70 10 20 10 20 10 10 10'0 1020 10 20 10 20
x/S=O.O 2.55 3.06 3.57 4.18 4.80 5.41 6.12 7.76 12.04 16.33 20.92 30.31 44.90
FIGURE
15. Experimental and theoretical velocity profiles for Re = 100
a t different s/s-locations.
0 80 80 80 80 80 80 80 80 80 80 80 80 80 80 80
x/S=?I 2.55 3.06 3.57 4 18 4.80 5.41 6.12 7.14 7.76 8.52 9.18 9 7 4 11.07 11.84 13.57
U (cm/s)
0 180 180 180 180 180 180 180 180 180 180 180 180 180 180 180
x/S=O 2.55 4.00 6.00 8.00 10.00 12.00 14.00 15.53 16.12 20.00 22.45 23.88 30.20 49.59
FIGURE
17. Experimental and theoretical velocity profiles for Re = 1000
a t different x/S-locations.
All other reattachment lengths reported in the literature for turbulent backward-
facing step flows were smaller than 8. Existing information suggests this to be due
to differences in test-section geometries. A review of available results, provided by
Durst & Tropea (1981), clearly indicates a decrease in reattachment length with
decreasing expansion ratio H / h = 1+ S / h a t a constant Reynolds number.
I n the laminar-flow region, measurements of detailed velocity distributions were
carried out by Denham & Patrick (1974) and of reattachment length by Denham &
Patrick and Goldstein et al. (1970). Figures 18(a,b ) indicate that both sets of data
compare well with the authors' separation-length measurements when plotted versus
a Reynolds number defined on the diagrams. This would suggest that a Reynolds
number based on the step height might be used as a single parameter that defines
the reattachment length in a laminar two-dimensional backward-facing step flow.
This conclusion was examined by utilizing numerical solutions for reattachment
492 B. F . Armaly, F . Durst, J . C. F . Pereira and B. Xchonung
10 - 0
V
V
8- 0
00
6-
V
0
4-
0
e
2-
I I I 4
01
100 200 300
Re, = US/,.
xR
12
80(2:
10 1 O0
.O
3 00
0
4
0 Goldstein et al. (1 970)
Present
measurements
FIGURE
18. ( a )comparison with Denham & Patrick’s (1974) results. (b) Comparison with
the results of Goldstein et al. (1970).
length obtained by TEACH for laminar flows a t Re 5 380. The reattachment length
for a fixed step Reynolds number in channels with different expansion ratios were
computed, and it was shown that the conclusion drawn from figures 18(a, b ) was
incorrect. This indicates that the good agreement of the measured values shown in
figures 18(a, b ) is quite accidental. This confirms that the reattachment length in
laminar two-dimensional backward-facing step flows is not a function of a single
variable but is more likely to be a function of three or more variables, including the
expansion ratio, the step or inlet-section Reynolds number and the slope a t the wall
of the inlet velocity profile.
Measurements in the laminar-turbulent transition region of a backward-facing step
flow do not exist for comparison with the present values. I n addition, computational
schemes have not yet been developed for numerically treating this regime of the flow.
It is clearly the more complex region for predictions or measurements because of the
Ba,ckujard-facing step Jlow 493
FIQURE
19. Mass-transfer measurements on the stepsid? wall, by Armaly et al. (1980)
REFERENCES
ABBOTT,D. E. & KLINE,S. J. 1962 Experimental investigations of subsonic turbulent flow over
single and double backward-facing steps. Trans. A.S.M.E. D: J . Basic Engng 84, 317.
ARMALY, B. F., DURST,F. & KOTTKE, V. 1980 Momentum, heat and mass transfer in backward-
facing step flows. Sonderforschungsbereich 80, llniversitat Karlsruhe.
BACK,L. H. & ROSCHKO, E. J. 1972 Shear layer flow regimes and wave instabilities and
reattachment lengths downstream of an abrupt circular channel expansion. Trans. A .S.M.E.
E: J . Appl. Mech. 39, 677.
BREDERODE, V. DE & BRADSHAW, P. 1972 Three-dimensional flow in nominally two-dimensional
separation bubbles. Flow behind a rearward-facing step. I.C. Aero-Rep. 72-19.
CASTRO, I. P. 1978 The numerical prediction of recirculating flows, Proc. Int. Conf. on Numerical
Methods in Laminar and Turbulent Flow, London.
CHERDRON,W . , DURST,F. & WHITELAW, J. H. 1978 Axisgmmetric flows and instabilities in
symmetric ducts with sudden expansions. J . Fluid Mech. 84, 13.
DENHAM, M. K. & PATRICK, M. A. 1974 Laminar flow over a downstream-facing step in a
two-dimensional flow channel. Trans. Inst. Chem. Engrs 52, 361.
DURST,F. & TROPEA, C. 1977 Processing of laser-Doppler signals by means of a transient recorder
and digital computer. S F B 80/E/ 118, Sonderforschungsbereich 80, Universitat Karlsruh,e.
DURST,F. & TROPEA,C. 1981 Turbulent, backward-facing step in two-dimensional ducts and
channels. In: Proc. Turbulent Shear Flow 3 Symp. Datvk, September 1981.
DURST,F. & WHITELAW, J. H. 1971 Aerodynamic properties of separated gas flows: existing
measurements techniques and new optical geometry for the laser-Doppler anemometer. Prog.
Heat Mass Transfer 4, 311.
EATON, J. K . & JOHNSTONE, J. P. 1980 An evaluation of data for the backward-facing step flow.
Report prepared for 1980/81 Conference on Complex Turbulent Flours, Stanford llnieiersity.
ETHERIDQE,D. W. & KEMP,P. H. 1978 Measurements of turbulent flow downstream of a
backward-facing step. J . Fluid Mech. 86, 545.
FILETTI,W. G. & KAYS,W. W. 1967 Heat transfer in separated, reattachment and redevelopment
regions behind a double step at the entrance to a flat duct. Trans. A.S.M.E. C: J . Heat Transfer
89, 163.
GERSTEN, K. & WAUSCHKUHN, P. 1978 Uber die abgeloste turbulente Stromung bei endlichem
Ablosegebiet. Bericht No. 63/1978, Lehrstuhlf u r Stromungsmechanik, Ruhr-Uniaersitat Bochum.
GOLDSTEIN, R. J., ERIKSEN, V. L., OLSON,R. M. & ECKERT,E. R. G. 1970 Laminar separation
reattachment, and transition of flow over a downstream-facing step. Trans. A.S.M.E. D: J .
Basic Engng 92, 732.
GOSMAN, A. D. & PUN,W. M. 1974 Lecture notes for course entitled: ‘Calculation of recirculating
flow ’. Heat Transfer Rep. HTS/74/2, Imperial College, London.
KOTTKE,V., BLENKE,H. & SCHMIDT, K. G. 1977 EinfluB von Anstromprofil und Turbulenz-
intensitat auf die Umstromung liingsangestromter Platten endlicher Dicke. Warme- und
Stoffubertragung 10, 9.
KUMAR, A. & YAJNIK,K . S. 1980 Internal separated flows at large Reynolds number. J . Fluid
Mech. 97, 27.
LEONARD, B. P., LESCHZINER, M. A. & MCGUIRK,J . 1978 Third-order finite-difference method
for steady two-dimensional convection. Proc. Int. Conf. on Numerical Methods in Laminar and
Turbulent Floui, London.
496 B. F . Armaly, F . Durst, J . C. F . Pereira and B. Schonung
LEONARD, P. B. 1979 A stable and accurate convective modelling procedure based in quadratic
upstream interpolation. Comp. Meth. i n Appl. Mech. Engng 19.
LESCHZINER, M. A. 1980 Practical evaluation of three finite difference schemes for the computation
of steady-state recirculating flows. Comp. Meth. Appl. Mech. Engng 23, 293.
LESCHZINER, M. A. & RODI,W. 1981 Calculation of annular and twin parallel jets using various
discretization schemes and turbulence-model variations. Trans. A.S.M.E. I : J . Fluids Engng
103, 352.
MACAGNO, E. 0. & HUNG,T. K. 1967 Computational and experimental study of a captive annular
eddy. J . Fluid Mech. 28, 43.
MOSS,W. D., BAKER,S. & BRADBURY, L. S. J. 1979 Measurements of mean velocity and Reynolds
stresses in some regions of recirculating flows. In Turbulent Shear FZows 1 (ed.
F. Durst, B. C. Launder, F. W. Schmidt & J. H. Whitelaw). Springer.
RAITHBY, G. D. 1976 Skewed upwind differencing schemes for problems involving fluid flow. Comp.
Meth. Appl. Mech. Engng 19, 153.
SEBAN,R . A. 1964 Heat transfer to the turbulent separated flows of air downstream of a step in
the surface of a plate. Trans. A.S.M.E. C: J . Heat Transfer 86, 259.
SPARROW, E. M. & KALJES,J . P. 1977 Local convective transfer coefficients in a channel
downstream of a partially constructed inlet. Int. J . Heat Mass Transfer 20, 1241.
VAHLDAVIS,G. DE & MALLINSON,G. D. 1976 An evaluation of upwind and central difference
approximations by a study of recirculating flow. Comp. Fluids 4, 29.
WAUSCHKUHN, P. & VASANTARAM, V. 1975a Die turbulente Grenzschicht hinter einem
Ablosegebiet. Z. Flugwiss. 23, 1 .
WAUSCHKUHN, P. & VASANTA RAM,V 1975b Die turbulente Grenzschicht unmittelbar hinter dem
Wiederanlegen eines Ablosegebietes. 2. angew. Math. Mech. 55, T166.