Experiments in Fluids (2020) 61:75

https://doi.org/10.1007/s00348-020-2910-3

RESEARCH ARTICLE

Screened expanding turning‑vane concept

Mark Drela1 · Arthur Huang1 · David Darmofal1

Received: 6 November 2019 / Revised: 4 January 2020 / Accepted: 31 January 2020 / Published online: 21 February 2020
© Springer-Verlag GmbH Germany, part of Springer Nature 2020

Abstract
The paper presents the screened expanding corner vane concept, which can turn a channel flow by 90◦, while simultaneously
increasing the flow area by at least a factor of two and thus halving the mean velocity, all without incurring any significant
flow separation. The concept is demonstrated experimentally, and investigated with analytical and computational models. One
target application is an ultra-compact closed-circuit wind tunnel, whose overall length is roughly half that of a conventional
wind tunnel with the same test section.
Graphic abstract

Conventional Wind Tunnel

(for size comparison)

Screened Expanding
Turning Vanes Ultra−Compact Wind Tunnel

enabling
feature

screen

suppressed
BL separation
optional
honeycomb

List of symbols g BL area fraction

A Channel flow area H BL shape parameter = Δ∗ ∕Θ
c Vane chord K Screen pressure drop coefficient
Cp Pressure coefficient = (p−p0 )∕ 21 𝜌V02 𝐧̂ Unit vector normal to screen
CpT Total pressure coefficient = (pT −pT 0 )∕ 12 𝜌V02 p Static pressure
f BL velocity fraction pT Total pressure = p + 21 𝜌V 2
RA Outlet/inlet area expansion ratio = A3 ∕A0
* Mark Drela RΔ∗ Outlet/inlet blockage ratio = (Δ∗2 ∕A2 )∕(Δ∗0 ∕A0 )
[email protected] V Velocity magnitude = |𝐕|
1
Massachusetts Institute of Technology, Cambridge,
𝐕 Velocity vector
MA 02139, USA y, z Channel transverse coordinates

13
Vol.:(0123456789)
75 Page 2 of 11 Experiments in Fluids (2020) 61:75

𝛽 Screen open-area fraction

Δ∗ BL displacement area = ∫ (1− VV ) dA
e

Θ BL momentum area = ∫ (1− VV ) VV dA Conventional tunnel

e e
(50% scale NASA 14 x 22)
𝜂 Screen diffraction ratio = Vt2 ∕Vt1
9:1
𝜌 Density
( )0 Inviscid flow upstream of cascade
( )1 Flow immediately upstream of screen
( )2 Flow immediately downstream of screen
screened
( )3 Mixed-out flow far downstream of cascade wide−angle Traditional compact tunnel
diffuser (notional)
( )e Boundary layer (BL) edge quantity
8:1
( )n Component normal to screen
( )t Component tangential to screen
( )∞ Room ambient 2D diffuser

screened New WBWT

turning
vanes
1 Introduction 8:1

2D diffusers

Conventional closed-circuit wind tunnels have almost invari-

ably employed separate diffuser and turning-vane sections, Fig. 2  Size comparison of a conventional tunnel, a traditional com-
as reviewed by Bradshaw and Pankhurst (1964). Compact pact tunnel, and the New WBWT ultra-compact tunnel enabled by
the screened turning vanes in corner 4. Test-section volumes are the
tunnels have traditionally employed wide-angle diffusers in
same. The New WBWT’s 2D diffusers expand only vertically
which separation is prevented by screens, typically at the
cost of increased tunnel drive power (Squire and Hogg 1944;
Schubauer and Spangenberg 1947; Mehta 1977). experience, it was also expected that separation will typi-
One notable exception to these traditional practices is cally begin at the vane/wall intersections as discussed by Lei
the relatively recent wind tunnel design by Lindgren and et al. (2008). This indicates that further conservatism in the
Johannson (2004), where part of the diffusion task is car- vane area expansion ratio is needed.
ried out by “expanding” corner vanes which have a signifi- The screened-vane concept presented here considerably
cant area increase and hence velocity decreases. The task of increases the tolerable area expansion ratio by the use of
the diffusers is thereby reduced, so that they can be made a screen installed over the exit of each vane passage, per-
shorter, and for a given test-section size, the overall tun- pendicular to the outlet flow, as diagrammed in Fig. 1. The
nel size can be reduced. The experimental measurements of screen suppresses any upstream separation, or at least con-
Lindgren et al. (1998) indicate that the area increase ratio is tains the separation entirely inside the vane passage, so that
limited to roughly 1.4 or 1.5; else, the vane cascade losses the outlet does not exhibit any reversed flow. In effect, each
become excessive, which limits how much tunnel-size reduc- vane passage then not only turns the flow, but also functions
tion can be realized in practice. From compressor design as a wide-angle diffuser with zero added length. It should
also be mentioned that a simpler straight screen (without the
honeycomb (optional)
zig-zag shape) over the cascade outlet would not be suitable,
since the screen’s diffraction effect would then produce a
suppressed total pressure gradient across the entire channel, as described
BL separation
by Greitzer et al. (2004) and Elder (1959). To reduce outlet
streamwise vorticity, a honeycomb can also be incorporated
into the vanes as sketched in Fig. 1. However, this is seen as
screen secondary to the screen’s primary function of suppressing
flow separation. Hence, the effects of incorporating a hon-
A3
eycomb in the vanes will not be examined here.
The topic of this paper was motivated by the current (2020)
project to replace the 1935-vintage Wright Brothers Wind
Tunnel with an all-new tunnel (called the New WBWT) in the
A0 same site. Surrounding buildings put hard constraints on the
overall length and width of the tunnel, so in the design study,
Fig. 1  Screened expanding turning-vane cascade all possible means to minimize the tunnel circuit size were

13
Experiments in Fluids (2020) 61:75 Page 3 of 11 75

examined. The design objective was to give the largest possible V2 V3

test-section size for the fixed tunnel footprint and the chosen
8:1 contraction ratio. Tunnel layout studies indicated that the p1 p2
V0
most compact design (and hence the largest test section) is pro- p3
A2 = A3
duced if expanding corner vanes are used in corner 4, with the
p0
large area ratio of RA = 30∕16 = 1.875 enabled by the vane- A0
outlet screen. Also contributing to the tunnel compactness are p1 p2
g2 A 2
g0 A 0
2D-type diffusers with the width fixed and only the height f 2 V2
f 0 V0
increasing, which was also a feature in the novel tunnel design
of Lindgren and Johannson (2004). The third diffuser after the
Fig. 3  2D parallel-flow model within a passage of a screened expand-
fan is particularly aggressive, which is enabled by a conven- ing vane cascade. Station 3 is at hypothetical mixed-out state
tional fin-tube heat exchanger at its end which also functions
as a diffuser screen. Finally, the settling chamber and contrac-
tion are in effect combined into one progressively increasing which can be obtained for any given screen from pressure
contraction which starts at the last screen, and which is com- and velocity–direction measurements. The 3D CFD results
putationally designed for the shortest possible length without to be presented here were computed with Fluent ANSYS
incurring flow separation. (2018), in which the screen was modeled as a plane of
Figure 2 shows the extremely compact size of the New momentum sources with specified K and 𝜂 values. The local
WBWT which is enabled by these design features, and in par- source strengths depend on the local upstream screen-nor-
ticular the screened expanding vane concept. For comparison, mal dynamic pressure and flow direction. The screen-normal
Fig. 2 also shows the conventional-design NASA 14×22 tun- and tangential velocities are computed as:
nel (Gentry et al. 1990), scaled to 50% for the same test-sec-
tion volume, and a notional traditional compact tunnel using a
Vn1 = 𝐕1 ⋅ 𝐧̂ (3)
screened wide-angle diffuser.
The New WBWT has an estimated power factor of 0.35, 𝐕t1 = 𝐕1 − Vn1 𝐧,
̂ (4)
compared to an estimated 0.30 value for a conventional tunnel
of the same contraction ratio and the same number of flow- where 𝐕1 is the local velocity taken from the solution just
conditioning screens. This ∼17% power increase is deemed an upstream of the screen, and the screen-normal vector 𝐧̂ is
acceptable penalty for the large test section which is realized specified. The momentum source per unit volume (in effect
given the small tunnel footprint. Reduced construction costs a body force) is then computed by invoking the K and 𝜂
for a given tunnel capability are a clear side benefit. definitions:
The focus on this paper will be to identify the key param-
eters of the screened expanding turning vane, and also iden- 𝐅 = − 12 𝜌1 Vn2 K 𝐧̂ − 𝜌1 Vn1 𝐕t1 (1 − 𝜂). (5)
1
tify and characterize its design trade-offs. The performance
The 2D CFD results presented here were computed using
of the vane will be demonstrated through experiment and
the MISES viscous/inviscid code (Giles and Drela 1987),
3D CFD Reynolds-Averaged Navier–Stokes computations.
which was modified by the inclusion of the above momen-
tum source model into its Euler equation formulation. The
momentum sources were also incorporated into the MISES
2 Flow modeling
integral boundary layer model, as originally employed for
analyzing the aerodynamic impacts of boundary layer flows
2.1 Computational modeling
through heat exchangers (Drela 1996).
The screen’s effect on the flow is represented via its pressure
drop coefficient and refraction ratio:
2.2 Parallel‑flow model

K≡
p2 −p1 To examine the effect of the area expansion ratio and screen
1
𝜌 V2 (1) pressure drop on the ability of the vane row to suppress
2 1 n1
separation, we consider the simplified quasi-1D model dia-
grammed in Fig. 3.
Vt2
The tunnel wall boundary layer (BL) is idealized as a
𝜂≡ , (2)
Vt1 streamtube with a velocity which is a fraction f of the core
flow velocity, and with a thickness which is a fraction g of
the overall channel height. One rational way to set these for a

13
75 Page 4 of 11 Experiments in Fluids (2020) 61:75

V / V0 Cp
1.2 0
1.1 -0.2
1.2 1 0 -0.4
0.9
1 -0.2 -0.6
0.8
0.8 -0.4 -0.8
0.7
0.6 -0.6
0.6 -1
0.4 -0.8
0.5 -1.2
0.2 -1
0 0.6 -1.2 0.6
0.5 0.5
0.4 0.4
0.3 y 0.3 y
-0.5 -0.4 0.2 -0.5 -0.4 0.2
-0.3 -0.2 0.1 -0.3 -0.2 0.1
-0.1 0 -0.1 0
0.1 0.1
x 0.2 0.3 0 x 0.2 0.3 0

Fig. 4  Velocity and static pressure fields of a two-streamtube low-speed inviscid flow with f0 = 0.5 and g0 ≪ 1, interacting with a screen located
at x = 0, with parameters K = 1.0 and 𝜂 = 0.78

given wall BL is to match the actual known BL displacement This is an implicit equation for the screen-exit velocity
and momentum areas Δ∗ , Θ and channel area A. This gives: fraction f2 (RA , K, f0 , g0 ), which depends on the four param-
eters indicated. All the remaining ( )2 quantities can then be
Θ 1
f = = (6) computed as functions of RA , K, f0 , g0 using this f2 together
Δ∗ H
with (8)–(11).
We also determine the mixed-out state V3 , p3 , also as
Δ∗ 1 Δ∗ H functions of RA , K, f0 , g0 , using the mass and streamwise
g = = , (7)
A 1−f A H −1 momentum conservation equations for the downstream flow:
so that for a typical turbulent BL with H ≃ 1.4 , we have ( )
𝜌V3 = 𝜌V2 1−g2 + f2 g2 (13)
f ≃ 0.7 and g = 3.5 Δ∗ ∕A.
Low-speed inviscid calculation results for the two- ( )
streamtube model flow in Fig. 4 show that the flow distur- p3 + 𝜌V32 = p2 + 𝜌V22 1−g2 + f22 g2 . (14)
bance is almost entirely upstream of the screen, and the
upstream screen-face pressure has two distinct values in Also of interest will be the displacement areas of the slower
the two streamtubes, denoted here by p1 and p′1. Note also streamtube, and the blockage-fraction ratio across the vane
that p′1 < p1, which is the mechanism which accelerates the row:
slower moving streamtube into the screen and thus sup- Δ∗0 = A0 (1−f0 )g0 (15)
presses its separation. In contrast, the downstream flow is
almost parallel and at one pressure p2.
Applying mass conservation and Bernoulli equations
(with screen pressure drops) between stations 0 and 2 for
each streamtube, we obtain: 0
greater
separation
resistance
𝜌V0 A0 (1−g0 ) = 𝜌V2 A2 (1−g2 ) (8) -0.2
K = 0.5
lower 1.0
loss
𝜌V0 A0 f0 g0 = 𝜌V2 A2 f2 g2 (9) -0.4 1.5
C pT RA = 2.0 fixed
3 2.0
g0 = 0.2 fixed
p0 + 12 𝜌V02 = p2 + 21 𝜌V22 (1+K) (10) -0.6
2.5

3.0
p0 + 12 𝜌V02 f02 = p2 + 21 𝜌V22 f22 (1+K), (11) -0.8 f0 = 0.8
f0 = 0.7
f0 = 0.6
which can be combined to give: -1
0 1 2 3 4 5
( )2 R∆ ∗
1−f02 f0 g0
RA2 1 − = (1−f22 )(1+K).
(1−g0 )2 f0 g0 + f2 (1−g0 ) Fig. 5  Trade-off between downstream total pressure and blockage–
(12) fraction ratio, versus screen K and upstream BL velocity fraction f0

13
Experiments in Fluids (2020) 61:75 Page 5 of 11 75

0
greater
separation
resistance
-0.2

lower K = 0.5
loss
-0.4 1.0
C pT f0 = 0.7 fixed
3 1.5 g0 = 0.2 fixed
-0.6
2.0

-0.8 2.5 RA = 1.8

RA = 2.0
3.0 RA = 2.2
-1
0 1 2 3 4 5
R∆ ∗

Fig. 6  Trade-off between downstream total pressure and blockage–

fraction ratio, versus screen K and outlet/inlet area ratio RA

Fig. 7  Computed 2D surface Cp distributions of experimental rig

Δ∗2 = A2 (1−f2 )g2 (16) vanes, with specified K = 2.3 and 𝜂 = 0.58 set to match the observed
Cp2 = −0.05 value at the outlet. Gap between the vane metal (black)
and the adjacent colored line on each side is the displacement thick-
(1−f2 )g2 ness, which thins strongly ahead of the screen
RΔ∗ ≡ (Δ∗2 ∕A2 )∕(Δ∗0 ∕A0 ) = . (17)
(1−f0 )g0

A constant flow area with RA = 1 will always produce benefits of increased RA need to be traded off against the
RΔ∗ < 1, which is the usual BL-thinning action of a screen. increased demand on the flow screens and increased tun-
However, the present situation with RA > 1 can produce nel power.
RΔ∗ > 1, and, hence, a thicker downstream BL if the chosen Varying the remaining parameter g0 has almost no
K value is relatively small. Within limits this is considered effect, in that, the CpT vs. RΔ∗ curves for any reasonable
3

acceptable, since the primary objective of the screen here fixed RA and f0 will nearly overlay over a range of g0 val-
is to prevent BL separation, not to obtain a thinner BL. In a ues. Hence, this parameter does not need to be considered
wind tunnel, the latter task is left for the flow-conditioning in design trade-offs.
screens downstream of the expanding vanes.

3 Experimental design and setup

2.3 Design considerations
The success of the New WBWT critically depends on
A key parameter which must be selected in a given design the screened expanding vane row in corner 4 working as
application is the screen’s K value. Larger K values intended, specifically without significant flow separation. To
improve the screen’s ability to suppress flow separation
and make the outlet flow more uniform (decreased RΔ∗ ), at
the cost of a greater total pressure loss (more negative BL generator location
CpT ). The trade-off with fixed RA and g0 is shown in Fig. 5.
3 probe traverse
The loss significantly depends on f0 , although the K values with vanes
assembly
at the knee do not change much. Hence, the chosen K breaks

would be not be strongly affected by the specific f0 value,

i.e., by the H value of the upstream BL.
pitot−static probe probe traverse
The trade-off with fixed f0 and g0 , and with varying RA , tunnel flow
at channel midpoint without vanes
y
is shown in Fig. 6. Here, the best choice of K at the knee
of each curve is seen to strongly depend on the specific RA z x
value. The curves also quantify the rate of increase of the
losses and blockage fraction with RA , which is valuable
0 10 20 in

information during design when the tunnel-size reduction

Fig. 8  Top view of experimental rig. Vane-outlet screen not shown

13
75 Page 6 of 11 Experiments in Fluids (2020) 61:75

Figure 9 shows the rig attached to a 18×18 in. open-jet tun-

nel, via an adapter section which contracted the flow to the
12×18 in. cross section of the rig inlet.
A five-hole probe and a 30-port total pressure rake were
used in the flow surveys, using a 2D traverse system set up
to cover the diagonal survey planes, as shown in Fig. 8. Pres-
sures from the probes were measured using an ESP-32HD
differential pressure scanner from A-Tech Instruments, Ltd.,
with a 1 psi range and ±0.03% resolution. All readings were
referenced to room ambient pressure p∞ , and sampled at
80 Hz and averaged over 1 s to provide the raw pressure data.
Outputs from the upstream pitot-static probe shown in Fig. 8
were also measured together with the survey probe outputs,
which gave the reference p0 , pT 0 , and reference dynamic
pressure 12 𝜌V02 = pT 0−p0 simultaneously with√
the probe data.
Fig. 9  Experimental rig mounted on open-jet wind tunnel, with The probe Cp and corresponding V2 ∕V0 = 1 + CpT −Cp2
2
30-port pitot survey rake values were, therefore, very insensitive to tunnel speed drift.
At 75 mph inlet speed, the pressure sensor resolution is
±0.3% of the test inlet dynamic pressure, giving an uncer-
gain confidence in the design, an experimental rig simulat- tainty in the corresponding V2 ∕V0 values of roughly ±0.5%.
ing the entire corner 4 was constructed and tested. To obtain
some “confidence margin”, the rig geometry was intention-
ally made more aggressive, with a rig area ratio of RA = 2.0
which is 6.7% larger than the 1.875 ratio of the New WBWT 4 Survey measurements
corner 4. Also, the turning angle was designed to be 95◦,
compared to 90◦ in any conventional wind tunnel corner. 4.1 Tunnel velocities
Based on 2D MISES calculations, the vane airfoil shape
was chosen to be a circular arc of 108◦ subtended angle, Surveys were performed at inlet flows velocities of
1.0 in. radius, and a constant thickness of 0.035 in . This sim- V0 = 50 mph and 75 mph , to gauge the effect of Reynolds
ple geometry was chosen to enable the 42 vanes to be fabri- number. There were no significant differences in the meas-
cated by cutting available thin-wall aluminum tubing. The ured outlet V2 ∕V0 distributions between the two inlet speeds,
suction side of the leading edge was shaped into a roughly so only the V0 = 75 mph data will be presented. This inlet
4:1 ellipse, and the trailing edges were left square. The vanes speed gives roughly the same experimental vane chord
were oriented to have a −40◦ exit metal angle from the cas- Reynolds number V0 c∕𝜈 ≃ 105 as in the actual New WBWT
cade-plane normal, with a pitch of 0.656 in . The leading to at a 20 mph test-section speed, which is the lowest expected
trailing edge chord is c = 1.65 in., giving a pitch/chord ratio in typical operation. Hence, the present experiment has
of 0.40. Figure 7 shows the computed surface Cp distribu- Reynolds numbers representative of the full-scale article.
tion at the design inflow angle of arctan(2.0) = 63.43◦, and a
chord Reynolds number of 100 K. 4.2 Survey display
The rig is diagrammed in Fig. 8. The channel walls are
0.25 in. Lexan, and the vanes are glued into slots cut into All survey data, for cases with and without the
strips attached to the endwalls, flush with the interior sur- vanes, will be plotted in the y–z-plane, which is
face. The vanes also pass through cutouts in a 0.06 in. thick 12 in. × 18 in. = 7.27c × 10.9c for the entire channel. The
splitter plate at their midspan, which provides bending and geometry is top/bottom symmetric about the splitter plate,
torsion support for the vanes against the airloads. The screen and preliminary surveys indicated that the flow was also
is a 0.0037in. stainless steel wire mesh with 74×74∕ in. nearly symmetric. Hence, more extensive surveys were per-
pitch, giving an open-area fraction of 𝛽 = 0.54. The screen formed for the bottom half only. Preliminary surveys also
was bent into the required zig-zag form in a jig and glued indicated that the outlet velocities were lower over the out-
to the vanes with RTV silicone. There are no channel walls side of the corner than the inside, which is expected, since
over the outlet, so the outlet flow angle is entirely the result the outside inlet BL has a longer run and, hence, is thicker.
of loading on the vanes. The vane array can be removed Hence, surveys were focused on the lower left corner at
from the channel, to allow flow-survey measurement of y, z = (0, 0) of the display plane, where the flow is most sus-
the incoming wall BLs which the vane array must handle. ceptible to separation.

13
Experiments in Fluids (2020) 61:75 Page 7 of 11 75

11 inlet 1
outlet, no screen
10 channel outlet, with screen 0.9
inlet outline outlet avg.
9
0.8
8
0.7
7
vanes
0.6
6 V / V0
z /c inlet 0.5
5 survey line
0.4
4
0.3
3
outlet
0.2
2 survey lines
1 0.1

0 0
0 1 2 3 4 5 6 7 0 0.5 1 1.5 2 2.5
y/c y/c

Fig. 10  Inlet V1 ∕V0 velocity distributions without vanes and outlet V2 ∕V0 velocity distributions with no-screen and screened vanes, along the
indicated survey lines. The screen eliminates all separated flow from the vane passages, even at the tunnel walls where the inlet BL is present

0.06
measured
average
0.04 avg+/-std

0.02

Cp∞2 0

-0.02

-0.04

-0.06
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
y/c

Fig. 11  Measured static pressure coefficient, referenced to

freestream, of the screened vanes over a portion of the outlet flow Fig. 12  Spire BL generators installed at rig inlet to provide relatively
near lower left corner. The average with standard deviations is thick inlet wall boundary layers, which cover about 55% of the chan-
Cp∞ ≡ (p2 −p∞ )∕ 12 𝜌V02 = −0.0158 ± 0.0125 nel area. Vane array has been removed in the photo
2

4.3 Initial surveys
all the passages have significant separation near the bottom
Initial surveys were performed with a 0.6 mm diameter five- wall. In contrast, the presence of the screen makes the outlet
hole probe, both for the incoming BL without the vanes, and flow much more uniform, and in particular, the separated
also with the vanes without and with the screen installed. flow regions are completely eliminated.
The resulting velocity distributions along a few selected sur- Since the five-hole probe is outside its calibration range in
vey lines are shown in Fig. 10. The“peaks and valleys” in the reverse flow, the outlet flows were also qualitatively exam-
outlet surveys correspond to the passages and vanes which ined with a yarn tuft on a wand. In the no-screen vane case,
have a projected pitch of Δy∕c = 0.178, which was not well the tuft confirmed the presence of irregular and mostly stag-
resolved here, resulting in the ragged distributions. Regard- nant flow over the entire outlet perimeter, with the largest
less, the surveys show that without the screen, the entire two stagnant regions being at the outside corners. This flow sepa-
vane passages near the outer wall have separated flow, and ration was of course expected, since the thick wall-BL cannot

13
75 Page 8 of 11 Experiments in Fluids (2020) 61:75

with BL generators
unobstructed

V / V0 V / V0
1 1

0.8
0.8
0.6
0.6
0.4

0.2 0.4

0
2.5
3 0.2
2
2.5
1.5 2
z/c 1 1.5 0
1 y/c 0 0.5 1 1.5 2 2.5 3
0.5 0.5
0 0 y/c

Fig. 13  Measured inlet V1 ∕V0 velocity distributions without vanes in the BL generators. The uncertainty from pressure sensor resolution
the vicinity of the bottom left corner (see Fig. 14), comparing the thin is estimated to be ΔV∕V0 = ±0.5%, which is comparable to plotting
inflow BLs of the unobstructed inlet with the thick layers produced by resolution here

0.7
11
thick inlet BL
thin inlet BL
10 channel average 0.6
inlet outline
9
0.5
8

7
vanes
0.4

z /c
6 V / V0
5 rake survey 0.3
lines
4
0.2
3

2 0.1
1

0 0
0 1 2 3 4 5 6 7 0 0.5 1 1.5 2 2.5 3
y/c y/c

Fig. 14  Measured screened-vane-outlet V2 ∕V0 velocity distributions flow as “𝗈𝗎𝗍𝗅𝖾𝗍, 𝗐𝗂𝗍𝗁 𝗌𝖼𝗋𝖾𝖾𝗇” in Fig. 10, but here is measured in much
for thin and thick inlet BL flows, showing the absence of separa- higher resolution with a pitot rake rather than a five-hole probe
tion regardless of inlet BL thickness. The “ 𝗍𝗁𝗂𝗇 𝗂𝗇𝗅𝖾𝗍 𝖡𝖫” is the same

negotiate the sudden doubling of the flow area. The static in Fig. 11. The averaged pressure coefficient referenced to
pressure rise is quantified by the outlet static pressure coef- ambient static pressure is nearly zero, which is fully consist-
ficient, which was measured with a hand-held static probe to ent with the observed nearly parallel outlet flow.
be approximately Cp2 ≃ + 0.35 (about half of the ideal invis-
cid value of +0.75). Since inlet BL fluid with V1 ∕V0 < 0.8 4.4 Detailed surveys
has a dynamic pressure less than the observed static pressure
rise, some reverse flow through the wall-adjacent passages Since the initial surveys revealed that with the screen
is almost inevitable. In contrast, for the screened vanes, the installed, the outlet flow is nearly parallel, and at a nearly
tuft showed the flow direction to be the same everywhere constant static pressure, the five-hole probe was replaced
over the outlet plane to within several degrees, with no indi- with a 30-tube pitot rake with a 0.125 in. tube spacing. The
cation of reverse flow anywhere. The measured static pres- velocity was then obtained from the total pressure of the
sure values over the screened-vane-outlet flow are shown rake and the ambient static pressure. Another 30-tube rake

13
Experiments in Fluids (2020) 61:75 Page 9 of 11 75

z/c = 1
1.4
0.5
1.2
V/V 0
z/c Expt
1 CFD K=2.2
0
0 0.2 0.4 0.6 0.8 1
0.8
(a) Exp’t z/c = 0.5

0.6 0.5

V/V 0
0.4

0.2 0
0 0.2 0.4 0.6 0.8 1
z/c = 0.25
0
0 0.2 0.4 0.6 0.8 1
y/c
1.2 1.4 1.6 1.8 2 0.5
V/V 0
1.4

0
1.2
0 0.2 0.4 0.6 0.8 1
z/c y/c
1

0.8 Fig. 16  Line plots of outlet V2 ∕V0 velocities in Fig. 15, at three z/c
(b) CFD location slices.
0.6

0.4
The high-resolution measured inlet velocity distributions
0.2 with and without the generators are shown in Fig. 13. The
0
BL generators roughly quadruple the BL thickness, which
now covers about nine vane passages, and all the wall BLs
now cover a substantial 55% of the overall channel area. The
small waviness in the velocity distribution at the outer edges
Fig. 15  Comparison of experimental and CFD outlet velocity ratio of the BLs is from the individual spire wakes, which have
V2 ∕V0 in the vicinity of the bottom left-channel corner. The inflow
BLs are along the left and bottom plot edges. Red and blue horizontal not merged completely at the survey location. The “dent” in
lines indicate slices for line plots in Fig. 16 the unobstructed velocity distribution at y∕c = 1.2 appears to
with 0.080 in. spacing was also used in some cases for more be the remnant of a streamwise vortex originating from the
spatial resolution. corner of the tunnel adapter section. This is entirely elimi-
All the rake tubes were sampled simultaneously using nated by the BL generators.
the 32-port ESP sensor (the remaining two ports sampled The outlet velocity distributions for the thin and thick
the upstream pitot-static). The multiple-tube sampling and inlet BL flows are compared in Fig. 14. The most significant
the much faster settling time of the rake increased its data feature is that the screen acts to prevent stagnant outlet flow
collection rate by two orders of magnitude over the five-hole nearly equally well for the two cases. This corroborates the
probe, and allowed very-high-resolution flow surveys to be two-streamtube model results, which predict that g0 , which
performed in a reasonable time. All the data presented below is a measure of the inlet BL thickness, has little effect on the
were obtained using the rakes. outlet flow parameters for the same K ≃ 2.3 value.
As mentioned previously, the outlet flow direction for
the screened-vane cases was observed by the yarn tuft to
5 Thin and thick inlet boundary layers be nearly uniform over the entire outlet plane. The angle
estimated from a photo of the tuft is roughly 96◦ , which
In the surveys shown in Fig. 10, the inlet wall BL thickness is close to the arctan(2.0)−arctan(−0.6104) = 94.8◦ turning
covers roughly two vane passages. To investigate the ability angle predicted by the 2D MISES calculation, as shown in
of the screened-vane cascade to negotiate much thicker BLs, Fig. 7. Some discrepancy is expected from the substantial
spire BL generators were designed using the guidelines of 3D effects near the walls.
Irwin (1981), cut from 0.05 in. aluminum sheet, and installed
at the rig inflow location, as shown in Fig. 12.

13
75 Page 10 of 11 Experiments in Fluids (2020) 61:75

z/c =
1.4
0.5

1.2 V/V 0 K=2.2

z/c K=1.6
K=1.1
1
0
(a) K = 2.2 0 0.2 0.4 0.6 0.8 1
0.8
Cp2 = −0.04
z/c =
0.6 0.5

0.4
V/V 0

0.2
0
0 0.2 0.4 0.6 0.8 1
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
y/c
0.5 z/c = 0. 5
1.4
V/V 0
1.2
z/c
0
1 0 0.2 0.4 0.6 0.8 1
y/c
(b) K = 1.6 0.8
Cp2 = +0.10
0.6
Fig. 18  Line plots of computed outlet V2 ∕V0 velocities in Fig. 17, at
0.4
three z/c location slices

0.2

0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0
y/c
1D model + vane loss, g0 = 0.1
1.4
-0.1 1D model + vane loss, g0 = 0.4
-0.2
CFD, thin BL
1.2
z/c CFD, thick BL
1
-0.3

(c) K = 1.1 0.8 -0.4

Cp
Cp2 = +0.22 T3
0.6 -0.5

0.4 -0.6

0.2 -0.7

0 -0.8

-0.9
0 0.5 1 1.5 2 2.5
K
Fig. 17  Computed outlet V2 ∕V0 velocity contours for thick inlet BL,
for three K values, and corresponding computed outlet static Cp2 val- Fig. 19  Loss versus screen pressure drop coefficient, as predicted by
ues. Significant corner separation appears for the lowest K value. 1D model (with added estimated 2D vane loss), and as extracted from
Colored horizontal lines indicate slices for line plots in Fig. 18 CFD solutions

calculation was set, such that the computed outlet Cp closely

6 3D CFD simulations matched the Cp2 = −0.05 value which corresponds to the
measured outlet Cp∞ = −0.0158 shown in Fig. 11. The rela-
6.1 Case setup 2
tion between the two pressure coefficients is:
To investigate the effects of the design parameters, in par- Cp2 = Cp∞ − (p0 −p∞ )∕ 12 𝜌V02 , (18)
ticular the screen K value, the flow in the experimental rig 2

was simulated in ANSYS (2018) using the RANS equations. and the required last term was obtained directly from the
For computational economy, only the bottom outer quadrant p0 −p∞ and pT 0 −p∞ gauge pressures taken from the upstream
was simulated, using 20 vane passages of half span, with pitot-static probe.
inviscid walls representing the symmetry planes. The effect The 𝜂 = 0.58 value was specified based on the correlation
of the screen was implemented by the body force model of Livesey and Laws (1973), although this parameter will
defined by Eqs. (3)–(5). The K = 2.2 value in the initial have little effect in this case, since the flow is almost normal

13
Experiments in Fluids (2020) 61:75 Page 11 of 11 75

to the screen. In the calculation, the BLs start at the channel the effects of the key design parameters, and computational
inlet, so the inlet length was adjusted, so that the resulting simulations and experimental measurements confirm that the
BL profiles at the vane inlet location (with the vanes absent) concept works as intended. When implemented as the fourth
matched the experimentally measured profiles as closely as corner of a closed-circuit wind tunnel, it enables a dramatic
possible. reduction of the overall tunnel space footprint compared to
traditional wind tunnel designs of the same capability.
6.2 Calculated outlet flows
Acknowledgements The authors are grateful to Linda Hedges, Aaron
Bucher, and Scott Eberhardt of Amazon Web Services (AWS) for
The computed and measured outlet velocity contours are providing both computational resources and much assistance on the
compared in Fig. 15. Profiles along several z locations (see AWS computing platform. We are also grateful to Chris Hill, Dipankar
Fig. 8) are also shown in Fig. 16. The rather good match Choudhury, and Ryan Gordon of ANSYS for the provision of Fluent
gives some confidence that the 3D CFD calculations can be software for this study. Finally, we are grateful to David Robertson and
Annika Rollock of MIT for the assistance with the experimental setup.
relied upon to perform the parameter studies presented next.

6.3 Effects of screen pressure drop coefficient

References
The design trade-offs in the selection of the screen pressure
drop coefficient K, as predicted by the simple 2D inviscid ANSYS. Ansys-fluent. https​://www.ansys​.com/produ​cts/fluid​s/ansys​
model, were shown in Figs. 5 and 6. To verify these design -fluen​t. Accessed Apr 2018
trend predictions for viscous 3D flows, a sequence of 3D CFD Bradshaw P, Pankhurst RC (1964) The design of low-speed wind tun-
cases was run with the K value progressively decreased, until nels. Prog Aerosp Sci 5:1–69
Drela M (1996) Aerodynamics of heat exchangers for high-altitude
significant outlet separated flow was obtained. The computed aircraft. J Aircr 33(2):176–184
outlet velocity contours are shown in Fig. 17, and the corre- Elder JW (1959) Steady flow through non-uniform gauzes of arbirary
sponding profiles are shown in Fig. 18. shape. J Fluid Mech 5:355–368
These 3D CFD results corroborate and “calibrate” the Gentry CL, Quinto PF, Gatlin GM, Applin ZT (1990) The Langley
14- by 22-foot subsonic tunnel: description, flow characteristics,
trends predicted by the two-streamtube 2D model. Specifically, and guide for users. Technical Report NASA-TP-3008, NASA
the CFD results indicate that for this RA = 2.0 case, a signifi- Giles MB, Drela M (1987) Two-dimensional transonic aerodynamic
cant separation will start to appear for K < 1.5 roughly, which, design method. AIAA J 25(9):1199–1206
in Figs. 5 and 6, is seen to correspond to RΔ∗ > 2.0. For the Greitzer EM, Tan CS, Graf MB (2004) Internal flow—concepts and
applications. Cambridge University Press, Cambridge
New WBWT which has RA = 1.875, Fig. 6, in turn, then sug- Irwin HPAH (1981) The design of spires for wind simulation. J Wind
gests that a screen with K ≥ 1.25 will be needed to obtain the Eng Ind Aerodyn 7(3):361–366
same RΔ∗ ≃ 2.0 ratio, and the same separation-onset threshold. Lei VM, Spakovszky ZS, Greitzer EM (2008) A criterion for axial
Figure 19 shows the mixed-out total pressure coefficient CpT compressor hub-corner stall. J Turbomach 130(3):031006-1–
3 031006-10. https​://doi.org/10.1115/1.27754​92
for a range of K values, and for two inlet BL thickness ratios. Lindgren B, Johannson AV (2004) Evaluation of a new wind tunnel
Since the 1D model does not account for the losses of the vane with expanding corners. Exp Fluids 36:197–203
BLs, its CpT curves have been offset by −0.08, which is the Lindgren B, Osterlund J, Johannson AV (1998) Measurement and
3 calculation of guide vane performance in expanding bends for
additional vane total pressure loss estimated by 2D MISES wind-tunnels. Exp Fluids 24:265–272
calculations. Also shown are the CpT extracted from the 3D Livesey JL, Laws EM (1973) Flow through non-uniform gauze
3
screens. J Fluid Mech 59:737–743
CFD solutions, using the mass and momentum conservation Mehta RD (1977) The aerodynamic design of blower tunnels with
Eqs. (13) and (14), integrated over the CFD outlet flow plane. wide-angle diffusers. Prog Aerosp Sci 18:59–120
The model captures the basic trend of loss versus K reasonably Schubauer GB, Spangenberg WG (June 1947) Effect of screens in
well. wide-angle diffusers. Technical Report Report 949, NACA​
Squire HB, Hogg H (1944) Diffuser-resistance combinations in rela-
tion to wind tunnel design. Technical Report RAE Report No.
Aero 1933, RAE
7 Conclusions
Publisher’s Note Springer Nature remains neutral with regard to
jurisdictional claims in published maps and institutional affiliations.
This paper presented a new screened expanding vane con-
cept which enables simultaneous turning and diffusion in an
effectively negligible streamwise distance. A simple parallel-
streamtube model of the flow was developed to investigated

13