Flow Turbulence Combust (2008) 80:489506

DOI 10.1007/s10494-008-9133-7
Mixing Measurements in a Supersonic
Expansion-Ramp Combustor
Aristides M. Bonanos Jeffrey M. Bergthorson
Paul E. Dimotakis
Received: 17 July 2007 / Accepted: 3 January 2008 /
Published online: 25 January 2008
Springer Science + Business Media B.V. 2008
Abstract This paper reports results on molecular mixing for injection via an
expansion-ramp into a supersonic freestream with M
1
= 1.5. This geometry produces
a compressible turbulent shear layer between an upper, high-speed air stream and
a lower, low-speed fuel stream, injected through an expansion-ramp at = 30

to
the high-speed freestream. Mass injection is chosen to force the shear layer to attach
to the lower guide wall. This results in part of the ow being directed upstream,
forming a recirculation zone. Employing the hypergolic hydrogen-uorine chemical
reaction and pairs of ip experiments, molecular mixing is quantied by measuring
the resulting temperature rise. Initial experiments established the fast-chemistry limit
for this owin terms of a Damkhler number (Da). For Da 1.4, molecularly mixed
uid effectively reacts to completion. Parameters varied in these experiments were
the measurement station location, the injection velocity of the (lower) fuel stream,
the stoichiometry for the ip experiments, and the density ratio of the fuel and air
streams. As expected, mixing increases with increasing distance from the injection
surface. The mixed uid fraction increases by 12% when changing the fuel-to-air
stream density ratio from 1 to 0.2. Comparisons with measurements at subsonic
(high-speed) air stream velocities show that the trend of decreasing mixing with
increasing speed documented in free-shear layer ows is also encountered in these
ows. The current geometry produces higher mixing levels than do free shear layers.
Keywords Molecular mixing Compressible turbulent shear layer
Supersonic combustion
A. M. Bonanos (B) P. E. Dimotakis
Graduate Aeronautical Laboratories, California Institute of Technology,
1200 E. California Blvd, MC 301-46, Pasadena, CA 91125, USA
e-mail: abonanos@tyrvos.caltech.edu
J. M. Bergthorson
Department of Mechanical Engineering, McGill University, 3480 University Street,
Montreal Quebec, Canada H3A 2A7
490 Flow Turbulence Combust (2008) 80:489506
1 Introduction
Hypersonic air-breathing vehicle engines require, from a cycle efciency point
of view, that the internal ow should remain supersonic through the combustor.
Further, to shorten the combustor length and extract the benet for the unavoidable
penalty paid to internal friction losses, the rate of heat release should ideally
be limited by the mixing rate and not by chemical reaction rates [1]. The same
requirements also dictate that mixing should be rapid, since this ensures compactness
of the propulsive device, minimizing internal losses and engine weight.
These requirements have led to studies of various mixing enhancement methods
in compressible ows. Several techniques for fuel injection into supersonic cross-
ow exist, such as transverse jets, compression and expansion ramps, cavities, steps,
strut injectors, and shear layers, to name a few [25]. The success of these designs
is evaluated based on performance characteristics, such as fuel penetration and
dispersion rates, mixing efciency, and the minimization of total pressure losses.
Such assessments require measurement methods to quantify the efciency of
the mixing mechanism. Non-intrusive laser-based techniques for measuring fuel
dispersion and oweld velocities, such as laser Doppler velocimetry, laser induced
uorescence, and diode laser absorption spectroscopy have been developed [6].
Mixing measurements rely on the scattering of light off a passive tracer introduced
into one of the freestreams, with the signal related to tracer concentrations. Other
studies have employed aspirating probes to sample the ow, typically limited to
that of a binary gas, to deduce mixture concentrations [7, 8]. Such techniques,
however, lack the spatio-temporal resolution required to measure molecular mixing,
especially in the high Reynolds number and Mach number environments of interest.
To address some of these limitations cold-chemistry techniques that employ
uorescence quenching to provide a direct measurement of uid unmixedness have
been developed [9].
Exploiting that chemical product formation in the fast-kinetic regime is limited
by mixing, a technique called the ip experiment was developed to measure
molecular mixing [1012]. In this technique, the temperature rise from an exothermic
fast-kinetic reaction is used as a tracer for molecular mixing and statistics are
computed from a pair of chemically reacting experiments with ipped stoichiome-
tries. Although heat release itself affects mixing, its effects have been studied and
understood [1214]. In an extension of the original implementation, the heat release
is tailored in each case such that the two ip experiments are matched, as indicated
by the total pressure proles of the runs. The details are further discussed below.
This paper presents measurements of molecular mixing in an expansion-ramp
combustor. A supersonic air stream is expanded over a ramp inclined at 30

to the
high-speed freestream ow, with a fuel stream injected through perforations in the
ramp. The two streams are seeded with hydrogen and uorine reactants, respectively.
The consequent temperature rise provides the tracer of mixing. The terms air and
fuel streams are used here in a non-conventional sense. In these experiments,
hydrogen is actually carried in the high-speed stream that would typically be the
oxidizer, while uorine is carried in the low-speed stream that would typically carry
the fuel. This geometry entails many attributes of a scramjets design. However, it is
designed to provide an archival environment for design studies and code validation
and is not intended to represent any sense of scramjet design optimization. In
Flow Turbulence Combust (2008) 80:489506 491
particular, a goal of the greater effort, of which the experiments reported here
are a part, is capturing the results of such experiments in numerical simulation of
ows characterized by attributes relevant to scramjet propulsion. The attempt to
provide a well-characterized ow environment has led to many of the choices in the
experiments reported here.
2 Facility Description
Experiments were conducted in the supersonic shear layer (S
3
L) laboratory at
Caltech, whose test section was developed for these experiments and is schematically
shown in Fig. 1. The facility is a two-stream blow-down wind tunnel capable of
delivering mass uxes up to m
1
=10 kg/s in the upper stream and m
2
=2.5 kg/s in the
lower stream, with a nominal run time of 26 s. The facility is designed to handle
fast-kinetic reactants. Specically, the top stream can be seeded with a mixture of
hydrogen (H
2
) and nitric oxide (NO), and the bottom stream with uorine (F
2
). The
remainder of the gas in both streams is comprised of helium, argon and nitrogen
inert diluents, such that the desired ratio of density (molar mass) and specic (molar)
heat capacities of the two streams can be set. Nitric oxide is added to the hydrogen
stream to generate radicals on contact with uorine to initiate the hydrogen-uorine
reaction [10, 15]. The reaction then proceeds hypergolically, obviating external
ignition sources.
Briey, the top stream gas is supplied from a 1.2 m
3
pressure vessel, typically
lled to 3.8 MPa (550 psi). Gas mixtures in this vessel are created using the partial-
pressure technique. The ow to the top stream is controlled with a proportional-
integral-derivative (PID) control algorithm applied to a servo-motor-actuated sonic
valve, set to deliver a constant pressure to the top stream plenum for the experiments
described here. The ow to the lower stream is supplied from a Teon bag within a
smaller pressure vessel (0.57 m
3
), to minimize the amount of hazardous gas handled.
The Teon bag is also lled with the desired gas mixture using the partial-pressure
technique. During a run, the outside of the bag is pressurized with nitrogen from
Fig. 1 Schematic of the expansion-ramp geometry
492 Flow Turbulence Combust (2008) 80:489506
a 12.7 m
3
surge tank, creating a very nearly constant supply pressure for the lower
stream for the duration of a run without the need for active control. The bottom
stream ow is metered using a calibrated sonic valve.
The top streamis accelerated through a converging-diverging nozzle to the desired
Mach number (M
1
= 1.5 in the present study). The lower stream is accelerated
through a converging nozzle and injected through a perforated expansion ramp
angled at = 30

with respect to the horizontal, with an open area fraction near

65%, chosen to avoid the jet-coalescence instability [16].
The test section height is h = 8.38 cm (3.30 in.), with a top-stream height of
h
1
= 3.30 cm (1.30 in.). The lower-stream mass ux, which is set by a choked metering
valve, is calibrated without the ramp installed at the splitter plate tip, where the
height is 5.97 cm (2.35 in.), and the lower-stream velocity, U
2
, is computed from this
measurement. A measurement rake is located downstream of the start of the expan-
sion ramp, at either L = 38.1 or 46.7 cm (15.0 or 18.4 in.). The measurement rake
consists of K-type thermocouples, static pressure probes, and total pressure probes.
Pressure taps are included along the top and bottom guide walls, as well as the top
and bottom stream plena. During a run, temperature and pressure data are recorded
at 1 kHz/channel through a National Instruments LabView hardware/software data
acquisition and control system.
Schlieren ow visualization is also utilized. The schlieren setup employs a few-
nanosecond duration light source (MiniStrobokin power supply with KL-L lamp)
and a high-speed CMOS camera (Phantom v7.1). The schlieren setup is mounted
on rails to allow imaging of both the forward and rearward portions of the test
section. Composite schlieren images are assembled by performing two experiments
at identical conditions, imaging the upstream and downstream portions of the test
section in separate runs. The images are then aligned using test section features
common to both images as ducials.
The two gas streams form a shear layer that reattaches to the lower guide-wall
some distance from the ramp. The reattachment generates a re-entrant jet, similar to
that formed by a backward facing step [17, 18], which brings reacted uid upstream
into a recirculation ow region where it mixes with the bottom stream and is re-
entrained into the shear layer (see Fig. 1). Beyond the measurement rake, the ow
expands into a large duct and is diverted into a chemically resistant catch bag. The
products are scrubbed down and neutralized prior to release in the atmosphere.
3 Data Reduction Techniques
To measure mixing, one would integrate the probability density function (PDF)
of mixed uid over the domain of interest. Since the PDF is unknown a priori, a
method of obtaining statistics about its form is required. In these experiments, this
is implemented using the hydrogen-uorine reaction and the ip experimental
technique that relies on two experiments: one performed with the upper-stream
reactant rich and the other with stoichiometries ipped such that the lower stream
is reactant rich. The overall reaction is H
2
+ F
2
2 HF. NO is premixed with the
hydrogen stream to initiate the overall reaction, via the release of atomic uorine,
by the reaction, NO + F
2
NOF + F. The stoichiometric mixture ratio, , dened
Flow Turbulence Combust (2008) 80:489506 493
as the number of moles of upper-stream uid required to fully consume one mole of
lower-stream uid, is given for the reactants used here by,
=
[F
2
]
[H
2
] +
1
2
[NO]
, (1)
where brackets denote freestream (molar) concentrations. If Z is the mole fraction
of high-speed uid in a mixture, a stoichiometric mixture ratio of =4 implies that
four parts (moles) of upper-stream uid per part of lower-stream uid must mix for
complete consumption of all reactants, or Z

=0.8.
In computing the probability of mixed uid, the ip experiment technique ex-
cludes contributions from the edges of the PDF. These are dominated by unmixed
uid and as such do not contribute to the probability of mixed uid calculation. This
is expressed as
P
m
(y) =
_
1

p(Z, y) dZ, (2)

where p(Z, y) is the PDF of mixed uid and Z

/2 [11, 12].
The ip experiment is used to measure the total amount of molecular mixing in
the ow. This can be expressed as a fraction of the mixed uid in the duct,
m
/h,
or as a fraction of the mixed uid in the mixing layer,
m
/
t
. The details of the ip
experiment have been previously documented [1013]. A brief description of the
data reduction steps is included in the Appendix for completeness.
A key assumption in this technique is that chemical reactions are kinetically fast.
For a chemical reaction to be mixing limited, the reaction time scale,

, must be
shorter than the uid dynamic mixing time scale,
m
. The later is proportional to the
convection time,
c
= L/U
c
, where U
c
is the convective velocity, given by,
U
c
U
1
=
1 +r s
1/2
1 +s
1/2
, (3)
where r U
2
/U
1
is the freestream speed ratio, s
2
/
1
is the freestream density
ratio [12] and the subscripts 1 and 2 refer to upper- and lower-stream quantities,
respectively. This is expressed in terms of the Damkhler number dened here
in terms of the outer (convective) time scale, Da =

/
c
. Other characteristic
mixing times that could be used are proportional to this choice, with a constant of
proportionality that depends on known ratios of large-scale variables.
The chemical time scale,

, is estimated using the balloon reactor model [19].

In this model, the chemical and thermodynamic state of a uid parcel is solved as a
function of time at constant pressure. Fluid from the two streams is entrained into
the shear layer mixed-uid core volume, at rates governed by shear layer growth
and entrainment, where reactions proceed, with a rapid increase in temperature. The
maximum slope of this temperature rise vs. time is extrapolated to the asymptotic
temperature to determine the chemical time scale,

. Calculations are performed

using the Cantera software package [20]. Cantera is also used to calculate the adi-
abatic ame temperature rise, T
f
, computed as the difference between the temper-
ature of the stoichiometric mixture at constant enthalpy and pressure and the initial
ambient temperature.
494 Flow Turbulence Combust (2008) 80:489506
An assumption of the ip technique is that the ow remains unchanged between
the high- and low- experiments. This assumption has been relaxed over that used by
Mungal and Dimotakis [10], who required that the heat release be minimal in order
for its effect on the ow to be negligible, i.e., the reacting and a corresponding non-
reacting runs should have identical total-pressure proles at the rake location. In the
present study, the magnitude of the heat released was allowed to be relatively high,
so long as the ow in the two reacting experiments was matched, as indicated by the
total-pressure proles.
4 Results and Discussion
Figure 2 presents two composite schlieren visualizations of the oweld: (a) for non-
reacting ow and (b) for reacting ow. In both cases, M
1
= 1.5 and U
2
= 45 m/s. In the
non-reacting case, the mismatch in the pressure of the two streams causes the shear
layer to curve downwards and a shock train beginning with an expansion fan is set up
at the tip of the splitter plate. However, with heat release, the shear layer growth rate
is reduced [13, 14], and, as a result of the change in angle of the dividing streamline,
the shock train begins with a compression wave. A change in the gradient of the heat
release, registered as the transition fromdark to light in the schlieren image, is seen in
the lower portion of the reacting shear layer, and indicates the location of maximum
heat release. It is somewhat shifted toward the lean reactant, as expected [10, 21].
The recirculation zone captures unmixed high-speed upper-stream uid as well as
partially reacted lower-stream uid and pumps it upstream towards the ramp (see
Fig. 2 Schlieren visualization of a M
1
= 1.5 and U
2
= 45 m/s shear layer with the rake at L = 38.1 cm,
visible at the right edge of the gures. Each image is a composite from two separate experiments.
a Non-reacting, top and bottom streams are both are 100% [N
2
]; b reacting with = 1/6, reactant
compositions for top stream are 17.8% [H
2
], 0.4% [NO] and for bottom stream is 3% [F
2
]
Flow Turbulence Combust (2008) 80:489506 495
Table 1 Stoichiometric run
( = 1) conditions for
U
1
430 m/s (M
1
= 1.5)
and U
2
= 25 m/s
Upper-stream Lower-stream T
f
Da
[H
2
] (%) [NO] (%) [F
2
] (%) (K)
0.96 0.10 0.98 98 1.03 0.09
1.95 0.10 1.97 197 1.01 0.59
2.95 0.10 2.97 294 0.99 1.44
3.96 0.10 3.93 389 1.02 2.52
5.95 0.10 5.88 573 1.02 4.65
Fig. 1). A secondary, low strain-rate shear layer is established at the intersection
of the ramp with the lower guide wall that can serve as a ameholding region in a
scramjet combustor. The instrumentation rake, located at L = 38.1 cm, is visible at
the far right edge of the schlieren images.
Flows examined to validate that the runs used to determine mixing were in the
fast kinetic regime are summarized in Table 1. Subsequently, ip experiments are
compared, measuring mixing and highlighting differences between various points
in the test matrix. A list of experimental conditions, namely, lower-stream speed,
stoichiometry, reactant composition, adiabatic ame temperature rise, Damkhler
number, and mixed uid fractions are listed in Table 2.
Figure 3 presents the normalized temperature-rise proles, T/T
f
, from
each case of Table 1. T is computed as the difference between the measured
temperatures of a reacting and a corresponding non-reacting run. Insight can be
gained as to the minimum Da, above which reactions are mixing limited, by
comparing these curves. Normalization by the adiabatic ame temperature rise
further removes any bias of different molar concentrations on overall temperature
rise and allows a direct comparison of relative effective kinetic rates [15]. Beyond
convergence of the normalized temperature rise proles, further increases in kinetic
rate do not increase overall chemical product formation and mixing-limited reacting
ow is established. A maximum of 80% of T
f
is seen to be attained, as limited by
the entrainment of unreacted uid into the mixing layer.
Table 2 Summary of run conditions
Case U
2
[H
2
] [NO] [F
2
] L T
f
Da
m
/
t

m
/h
a 45 6 1.30 0.1 8.0 0.38 223 6.74 0.54 0.44
1/6 17.80 0.4 3.0 0.38 495 2.48
b 45 6 1.30 0.1 8.0 0.47 223 8.16 0.66 0.58
1/6 17.80 0.4 3.0 0.47 495 3.01
c 15 4 1.45 0.1 6.0 0.47 234 6.23 0.61 0.45
1/4 11.95 0.1 3.0 0.47 465 2.05
d 45 4 1.45 0.1 6.0 0.47 234 5.97 0.66 0.54
1/4 11.90 0.2 3.0 0.47 464 2.49
e 45 4 1.40 0.2 6.0 0.47 291 9.67 0.74 0.54
1/4 7.90 0.2 2.0 0.47 397 1.89
Note that M
1
= 1.5 (U
1
430 m/s), U
2
in m/s, L in m and T
f
in K
496 Flow Turbulence Combust (2008) 80:489506
Fig. 3 Normalized
temperature rise proles
for stoichiometric runs with
increasing heat release.
M
1
= 1.5, U
2
= 25 m/s
T / T
f
y
/
h
0 0. 2 0. 4 0. 6 0. 8 1
-0.5
-0.4
-0.3
-0.2
-0.1
0. 0
0. 1
0. 2
0. 3
0. 4
0. 5
1% [H
2
; F
2
]
2% [H
2
; F
2
]
3% [H
2
; F
2
]
4% [H
2
; F
2
]
6% [H
2
; F
2
]
Transition to mixing-limited kinetics in this ow occurs at reactant concentrations
above 3%, corresponding to Da 1.4. This is a conservative estimate, as the
balloon reactor model was formulated for a free shear layer and does not account
for the benets of the recirculation zone that makes hot products and radicals
available for entrainment, increasing the overall kinetic rate. Note that, for the 3%
reactant concentration case, a higher fraction of the adiabatic ame temperature
rise is reached near the lower guide wall than in the other fast cases. This
is attributed to the benecial secondary effect of the recirculation zone. In the
3% case, the recirculation bubble re-entrains partially mixed uid into the lower
portion of the shear layer, allowing more time for reactions to occur. Increasing
reactant concentrations, and hence heat release, leads to a reduction in shear layer
entrainment and growth rates [13]. Consequently, the attachment location is pushed
further downstream and so the probability of nding mixed uid at the measurement
location near the lower guide wall decreases, more-closely resembling the behavior
of a free shear layer.
One way of quantifying the effect of the heat release on the oweld is through
the pressure coefcient, dened as:
C
p
=
p
e
p
i
1
2

1
U
2
1
=
2
M
2
1
_
p
e
p
i
1
_
, (4)
where p
e
and p
i
are the upper guide wall exit and inlet (static) pressures, respectively,

1
is the upper-stream density, and the ratio of specic heats (matched in the
two streams). Figure 4 shows the overall pressure coefcient of the device for inert
and reacting mass injection through the lower stream. Increasing the (inert) mass
ux through the lower stream deects the upper stream upwards and increases the
pressure coefcient within the test section, eventually recovering the (near) zero
Flow Turbulence Combust (2008) 80:489506 497
Fig. 4 Overall pressure
coefcient with M
1
= 1.5 for
reacting (open symbols -
U
2
= 25 m/s) and non-reacting
(closed symbols) ow
r = U
2
/ U
1
C
p
0 0.05 0.1 0.15 0.2 0.25
-0.5
-0.4
-0.3
-0.2
-0.1
0.0
0.1
0.2
Inert (M
1
= 1.5)
1% [H
2
; F
2
]
2% [H
2
; F
2
]
3% [H
2
; F
2
]
4% [H
2
; F
2
]
6% [H
2
; F
2
]
value typical of a free shear layer. As mentioned previously, heat release reduces the
shear layers growth rate and entrainment requirements, and this dilatation effects
the ow by increasing the effective pressure coefcient. Therefore, increasing heat
release in the ow acts in a similar fashion to increasing mass injection, albeit for
different reasons.
4.1 Variable measurement location
Table 2 summarizes the velocity, composition, adiabatic ame temperature rise, and
mixing results for the ip experiments. Figure 5 presents results from the series of
ip experiments performed, showing the normalized temperature rise for the two
cases as well as the probability of mixed uid. For each composition, the location of
maximum heat release is shifted toward the stream that is lean in reactants, however
the peak values are almost equal, indicating a near unity entrainment ratio for
this ow.
Normalized temperature proles from a pair of ip experiments with M
1
= 1.5,
U
2
= 45 m/s, and = 6 and 1/6 are presented in Fig. 5a. The measurement station
is at L = 38.1 cm. The stoichiometric adiabatic ame temperature rise, T
f
, for
the low- experiment was approximately twice that of the high- experiment, to
match the two ows. The ows are considered matched when the total pressure
proles for the two cases are very nearly matched, as discussed in the Appendix and

Fig. 5 Normalized temperature rise and probability of mixed uid proles for M
1
= 1.5. For Cases
ad, s = 1, whereas for Case e, s = 0.2. a U
2
= 45 m/s, = 6 and 1/6, L = 38.1 cm; b U
2
= 45 m/s,
= 6 and 1/6, L = 46.7 cm; c U
2
= 15 m/s, = 4 and 1/4, L = 46.7 cm; d U
2
= 45 m/s, = 4 and 1/4,
L = 46.7 cm; e U
2
= 45 m/s, = 4 and 1/4, L = 46.7 cm
498 Flow Turbulence Combust (2008) 80:489506
T/T
f
; P
m
y
/
h
0 0.2 0.4 0.6 0.8 1
-0.5
-0.4
-0.3
-0.2
-0.1
0.0
0.1
0.2
0.3
0.4
0.5
= 1/6; T
f
= 495 K
= 1/6 fit
= 6; T
f
= 223 K
= 6 fit
P
m
(from raw data)
P
m
(from fits)
T/T
f
; P
m
y
/
h
0 0.2 0.4 0.6 0.8 1
-0.5
-0.4
-0.3
-0.2
-0.1
0.0
0.1
0.2
0.3
0.4
0.5
= 1/6; T
f
= 495 K
= 1/6 fit
= 6; T
f
= 223 K
= 6 fit
P
m
(from raw data)
P
m
(from fits)
T/T
f
; P
m
y
/
h
0 0.2 0.4 0.6 0.8 1
-0.5
-0.4
-0.3
-0.2
-0.1
0.0
0.1
0.2
0.3
0.4
0.5
= 1/4; T
f
= 464 K
= 1/4 fit
= 4; T
f
= 234 K
= 4 fit
P
m
(from raw data)
P
m
(from fits)
T/T
f
; P
m
y
/
h
0 0.2 0.4 0.6 0.8 1
-0.5
-0.4
-0.3
-0.2
-0.1
0.0
0.1
0.2
0.3
0.4
0.5
= 1/4; T
f
= 464 K
= 1/4 fit
= 4; T
f
= 234 K
= 4 fit
P
m
(from raw data)
P
m
(from fits)
T/T
f
; P
m
y
/
h
0 0.2 0.4 0.6 0.8 1
-0.5
-0.4
-0.3
-0.2
-0.1
0.0
0.1
0.2
0.3
0.4
0.5
= 1/4; T
f
= 397 K
= 1/4 fit
= 4; T
f
= 291 K
= 4 fit
P
m
(from raw data)
P
m
(from fits)
a b
c d
e
Flow Turbulence Combust (2008) 80:489506 499
illustrated in Fig. 6. Figure 5a also plots the probability of mixed uid at any location
within the mixing region [see (2)]. Fits are computed using a matched-exponential
function, characterized by a single peak but different widths for the two branches
(see Appendix for more details). The t to the probability of mixed uid is computed
as a sum of the ts of the normalized temperature rise proles and not as a t to the
probability of mixed uid data.
Figure 5b presents data from a set of ip experiments at identical conditions,
with the measurement location shifted downstream to L = 46.7 cm. The mixed uid
fraction in the duct,
m
/h, dened in the Appendix, has increased by 32%, with the
shear layer continuing to mix previously and freshly entrained uid from the free
streams. The upper portion of the normalized temperature proles, up to the location
of maximum heat release, is very similar, indicating that little or no additional heat
release occurs in the high-speed section of the ow. However, there is increased
mixing in the lower portion of the proles, with Case b yielding much higher mixing.
This is attributed to the resulting re-entrant jet into the recirculation zone. The re-
entrant jet is rich in upper-stream uid and also contains unreacted lower-stream
uid. These have ample time to react in the relatively slow moving recirculation
zone. This causes an increase in the mixed uid fraction within the shear layer,
m
/
t
,
of 22%.
In the core of the mixing layer, a probability of mixed uid of unity is achieved, in-
dicating that all uid in that region has mixed on a molecular level by the downstream
measurement station. In some locations, a probability larger than unity is estimated
as a result of compounded measurement errors from the pair of experiments, as
discussed in the Appendix.
Fig. 6 Flow matching between
low- and high- runs of Fig. 5d
500 Flow Turbulence Combust (2008) 80:489506
In gas-phase free shear layers, both the shear layer thickness,
t
, and the mixed-
uid thickness,
m
, are linear functions of the distance from the splitter plate, and
their ratio has been found to have the constant value of
m
/
t
0.49 [12]. Therefore,
the increase in mixed uid fraction with increasing distance of the measurement
station indicates a higher mixing efciency of the expansion-ramp geometry over
that in a free shear layer.
4.2 Variable injection
Figure 5c and d presents the normalized temperature rise proles from a pair of ip
experiments and the calculated probability of mixed uid in a ow with M
1
= 1.5,
= 4 and 1/4, and L = 46.7 cm. The difference between the two cases is the
lower stream injection velocity, namely U
2
= 15 and 45 m/s, respectively. Case c is
interesting. In the low- case ( = 1/4), the low injection speed of the lower stream,
U
2
, and the lower mass ux, limit the available reactants. The reaction is completed
shortly after the two streams meet, fully consuming the lower-stream reactants. Once
the lower stream has been entrained (and consumed), the shear layer attaches to
the lower guide wall and the recirculation zone is established. The hot products are
further diluted by the entrained upper-stream uid leading to a lower maximum
temperature rise. However, the temperature rise prole is wider, indicating that
good mixing extends throughout the reaction zone. For the high- case ( = 4), the
recirculation zone causes the location of the maximumheat release to shift downward
(towards the lower stream).
Comparing these ip experiments, the maximum value of the probability of mixed
uid, P
m
, increased from 0.9 for Case c, to 1 for Case d, indicating that the shear layer
interior is fully mixed on a molecular scale in this region. Increasing the lower-stream
speed by a factor of 3 produces a 20% increase in the mixed uid fraction,
m
/h. The
increase in mixing with increased injection can be related to the increased availability
of lower-stream reactants and the enhancement of the recirculating ow associated
with the chosen geometry. Further, the location of the maximum is shifted towards
the high-speed uid. In the context of a scramjet engine, for example, this implies
that more high-speed uid has mixed and reacted, and so a higher ux of reacted,
thrust-producing uid exits the combustor.
4.3 Variable stoichiometry
To assess the error introduced by neglecting the edge contributions of the PDF [see
(2)], ip experiments were conducted at the same owconditions at two different sto-
ichiometries. Figure 5b and d presents the normalized temperature rise proles and
the probability of mixed uid for M
1
= 1.5, U
2
= 45 m/s, L = 46.7 cm, with = 1/6 and
6, and = 1/4 and 4, respectively. The high- and low- normalized temperature rise
proles are quite similar, both in the maximum T/T
f
achieved and in their overall
shape. Slight differences may be attributable to the different levels of heat release
in the two cases and are localized in the low-speed region of the ow (y/h <-0.3).
The mixed-uid fraction within the mixing layer,
m
/
t
, is identical for the two cases
and equal to 0.66. A 7% difference in the mixed uid fraction within the duct,
m
/h,
in the two cases can be attributed to the differences mentioned above and in the
different resolution in the temperature rise data. Good agreement in mixing between
Flow Turbulence Combust (2008) 80:489506 501
the two cases validates the assumptions in the ip experiment, despite the relatively
large value of corresponding to the selected equivalence ratios (for = 1/4, 0.1
and for = 1/6, 0.07).
4.4 Variable density ratio
Previous research has focused on freestream density ratio effects on mixing of a free
shear layer [7, 22]. These studies showed that the mixed uid fraction is...essentially
independent of the density ratio [12]. The mean mixed uid composition, Z
m
, is
dened as
Z
m
=

p
(1 Z

p
(Z

) +
p
(1 Z

)
(5)
where
p
is the chemical product thickness and Z

is the stoichiometric mixture

fraction, as dened in the Appendix.
The mixing region, formed between the upper- and lower-stream uids, entrains
at a certain ratio, which can be solved for in terms of the mean mixed uid
composition [23],
E
n
=
Z
m
1 Z
m
=

p
(1 Z

p
(Z

)
(6)
In contrast to the mixed-uid fraction, both the mean mixture composition and the
entrainment ratio depend strongly on the density ratio [12].
Flip experiments conducted at M
1
= 1.5, U
2
= 45 m/s, = 4 and 1/4, and s = 1 and
0.2, are presented in Fig. 5d and e, respectively. Considering the difference in density
of the two streams, care was taken in the mixture preparation to match molar specic
heats. In both cases, the mixed uid fraction within the duct,
m
/h, remains constant,
and the probability of mixed uid proles are quite similar, both in the location
and width of the curve, in good agreement with the results for a free shear layer.
However, the mixed uid fraction within the mixing layer,
m
/
t
, increases by 12%for
the ow with s = 0.2 (Case e). Both cases exhibit signicantly higher levels of mixing
over the value of
m
/
t
= 0.49 quoted for a free shear layer [12], indicating a higher
mixing efciency for the expansion-ramp injection geometry. The entrainment ratio,
which is near unity for Case d, increases to 1.4 for Case e, indicating increased upper-
stream uid entrainment in the latter case.
5 Summary and Conclusions
Molecular mixing was measured using the ip experiment technique in a two-stream
blow-down wind tunnel. The air stream was supersonic (M
1
= 1.5) and expanded
over a 30

ramp. Fuel was injected through perforations in the ramp forming a

mixing layer between the two streams. Once the lower stream is fully entrained, the
mixing layer attaches to the lower guide wall, establishing a recirculation zone. This
geometry is characterized by many desirable features of a scramjet combustor, such
as high mixing rates, low pressure losses, and low strain rate regions for ameholding.
502 Flow Turbulence Combust (2008) 80:489506
The fast chemistry limit for this ow is established by increasing reactant concen-
tration in a stoichiometric ( = 1) reaction. The reacting ow is found to be mixing-
limited for Da 1.4, a value accepted as the minimum Da required for the ip
experiments.
A signicant increase in mixing is observed when the measuring station is moved
from L=38.1 to 47.6 cm, indicating the effectiveness of the recirculation zone as a
mechanism to enhance mixing in compressible high-speed ows. Such an increase in
mixing is not seen in free shear layers, which maintain a constant amount of mixed
uid, independent of the measurement station.
Mixing is found to depend on the velocity ratio, r, for a xed top stream velocity.
Low injection levels lead to full entrainment and consumption of lower stream uid.
The mixed uid is then further diluted until the measurement station, yielding low
total mixing due to limited available reactants. For large lower stream velocities,
a self-similar free shear layer ow is recovered, where the mixing level becomes
independent of the measurement position. However, an intermediate range of
velocity ratios exists that results in higher levels of mixing than for free shear layers.
This mixing enhancement is attributed to the action of the recirculating uid.
Molecular-mixing measurements were insensitive to the selected ip experiment
stoichiometric ratios chosen. This indicates that edge-effects of the PDF do not
signicantly contribute to the mixed uid fraction measurements and that there exists
a region in Z-space where mixing is very nearly independent of the particular choice
of stoichiometry selected for the pair of experiments.
For reduced density ratios, s, more high-speed uid is entrained into the mixing
layer. The overall mixing within the duct (
m
/h) achieved is independent of the
density ratio, however the particulars of the mixing region change, and the mixing
within the shear layer (
m
/
t
) is found to increase over the density-matched case.
Overall, the expansion ramp geometry produces higher levels of mixing than
those exhibited by free shear layers and appears to be a promising combustor ow
geometry element for high-speed propulsion systems.
Acknowledgements The authors would like to acknowledge constructive discussions with C. Bond
and G. Matheou. D. Lang assisted with the computers and electronics associated with the facility
control and data acquisition. The experiments were conducted with the expert assistance of E. Dahl.
This work was funded by the AFOSRunder Grants FA9550-04-1-0020 and FA9550-04-1-0389, whose
support is gratefully acknowledged.
Appendix: Flip-Experiments, Flow Matching and Error Estimation
Let the mixture mole fraction be dened as Z = n
1
/(n
1
+n
2
), where n
1
and n
2
represent the number of moles of upper- and lower-stream uid. Accordingly, the
complete consumption of all reactants will occur at a stoichiometric mixture mole
fraction [12],
Z

=

+ 1
(7)
and a maximum amount of chemical product, (Z; Z

), will be formed. At off-

stoichiometric conditions and with complete consumption of the lean reactant, the
amount of chemical product decreases to zero at Z = 0 and Z = 1, yielding a
Flow Turbulence Combust (2008) 80:489506 503
triangular normalized product function with a peak of 1 at Z = Z

. Exploiting the
triangular shape of the chemical product function for off-stoichiometric mixtures,
the integral of the PDF of mixed uid can be estimated by adding the normalized
temperature rise values from a low-,
1
, and a high-,
2
, experiment at any location
in the ow (see discussion in [12] and references therein).
For a chemically fast (high-Da) system accompanied by heat release, the mean
normalized temperature rise, T/T
f
, at any location in the ow, equals the integral
of the product of the probability density function (PDF) of mixed uid, p(Z, y), and
the normalized chemical product function, (Z; Z

) [12]. Therefore, the probability

of mixed uid at a location in the ow is given by [13],
P
m
(y) =
_
1
0
(Z; =
1
) + (Z; =
2
)
1/(1 Z

1
)
p(Z, y) dZ, (8a)
or
P
m
(y) = [1 Z(
1
)]
_
T(y)
T
f

=
1
+
T(y)
T
f

=
2
_
. (8b)
A measure of the total amount of mixing in the ow is the mixed uid fraction,

m
/h, given by

m
h
=
_
1/2
1/2
P
m
_
y
h
_
d
_
y
h
_
, (9)
yielding an estimate of the total (mole) fraction of uid within the duct that has mixed
on a molecular scale. It takes into account the effects of changing growth rate and
amount of mixing in the layer with increasing upper stream velocity. The fraction of
mixed uid within the mixing layer,
m
/
t
, is given by

t
= (1 Z
=
1
)
_

=
1
+

p

=
2
_
, (10)
where
p
is the product thickness,

p
=
_
h/2
h/2
T(y)
T
f
dy (11)
is the integral of the normalized temperature rise prole, and
t
is the distance from
the lower guide wall at which point the normalized temperature prole reaches 1%
of its maximum value [10, 12, 13]. A pair of ip experiments yields the probability of
mixed uid, P
m
, by measuring the resulting temperature rise. The amount of mixed
uid is then obtained through the mixed uid fraction,
m
/
t
, or
m
/h.
The T/T
f
proles were t using the (asymmetric) exponential function
T(y)
T
f
=
T
T
f ,i
+
_
T
T
f,max

T
T
f ,i
_
exp
_
_
4

j=2

i, j
( y y
0
)
j
_
_
, (12)
where y = y/h and T/T
f,max
is the single characteristic peak located at y
0
. The t
on either side of the peak has different slopes, tted by the coefcients
i, j
, where i = 1
504 Flow Turbulence Combust (2008) 80:489506
for y < y
0
and i = 2 for y > y
0
and j = 24. Exploiting the fact that the temperature
proles asymptote to zero in the top stream, the parameter (T/T
f
)
,2
was set to
zero and not included in the optimization in that portion of the ow.
An assumption of the ip experiment is that the ow remains unchanged with the
changing location of the temperature rise between the low- and the high-equivalence
ratio run. The criterion for ow matching is taken to be the total pressure prole,
p
0
(y). The ows in a pair of ip experiments are said to be matched when the total
pressure proles are in good agreement. Example total pressure proles from Fig. 5d
are shown in Fig. 6. The total pressure proles are in good agreement if the adiabatic
ame temperature of the stoichiometric mixture is chosen to be twice as high for
the low- experiment as for the high- experiment. Even at moderate levels of heat
release, changes in the amount of heat release do not signicantly affect the ow [13].
In calculating the probability of mixed uid utilizing the ip experiment technique,
two sources of measurement errors affect the determination of P
m
: the temperature
data, and the mixture composition of each stream, which affects the adiabatic ame
temperature rise, T
f
. Temporal temperature measurements are averaged over a
period of time in which the ow has achieved steady state. The averaging time
is chosen to allow sufcient ow-through times, based on the convective velocity
of the shear layer. Statistical convergence in this ow typically requires averaging
times of the order of 0.5-1 s, while typical ow through times are on the order of
1-2 ms. The thermocouple response time was measured to be less than 3 ms. Prior to
each run, tare reference temperature values are obtained for each probe, which
are subtracted as offsets from measured values. For a typical run, uncertainties in
temperature due to ow unsteadiness are typically less than 3% F.S., with 95%
condence, which translate to a 4.5% error in the mixed uid measurement,
m
/h.
Errors are slightly higher in the recirculation zone, because lower speeds result in
Fig. 7 Figure 5b including
error estimates. Temperature
probes are sampled at 1 kHz
with data averaged over 1 s.
Error bars increase in the
region near the lower guide
wall due to longer ow time
scales in the recirculation zone
T/T
f
; P
m
y

/

h
0 0.2 0. 4 0.6 0. 8 1
-0.5
-0.4
-0.3
-0.2
-0.1
0. 0
0. 1
0. 2
0. 3
0. 4
0. 5
= 1/6; T
f
= 495 K
= 1/6 fit
= 6; T
f
= 223 K
= 6 fit
P
m
(from raw data)
P
m
(from fits)
Flow Turbulence Combust (2008) 80:489506 505
longer characteristic time scales for a given averaging period. Radiative losses were
calculated to be less than 1% for these conditions, and so no further corrections
where applied to the measurements. Sample error estimates in the ame temperature
rise and the probability of mixed uid for the case of Fig. 5b presented previously are
shown in Fig. 7.
The mixtures for each stream are prepared with the partial pressure technique.
Thus, the accuracy of the mixture is limited by the accuracy of the pressure gage
used. The gages were Druck type PDCR-920 with a reported combined non-linearity,
hysteresis and repeatability of 0.1% full scale (best straight line). Errors while
preparing the mixtures typically resulted in a deviation of less than 1%in the targeted
values of the mixtures density and specic heat, and thus are considered negligible.
References
1. Ferri, A.: Mixing-controlled supersonic combustion. Annu. Rev. Fluid Mech. 5, 301338 (1973)
2. Bogdanoff, D.W.: Advanced injection and mixing techniques for scramjet combustors. J. Propuls.
Power 10(2), 183190 (1994)
3. Gutmark, E.J., Shadow, K.C., Yu, K.H.: Mixing enhancement in supsersonic free shear ows.
Annu. Rev. Fluid Mech. 27, 374417 (1995)
4. Ben-Yakar, A., Hanson, R.K.: Cavity ame-holders for ignition and ame stabilization in scram-
jets: an overview. J. Propuls. Power 17(4), 869877 (2001)
5. Seiner, J.M., Dash, S.M., Kenzakowski, D.C.: Historical survey on enhanced mixing in scramjet
engines. J. Propuls. Power 17(6), 12731286 (2001)
6. Bonnet, J.P., Grsillon, D., Taran, J.P.: Nonintrusive measurements for high-speed, supersonic,
and hypersonic ows. Annu. Rev. Fluid Mech. 30, 231273 (1998)
7. Brown, G.L., Roshko, A.: On density effects and large structure in turbulent mixing layers.
J. Fluid Mech. 64, 775816 (1974)
8. Ninnemann, T.A., Ng, W.F.: A concentration probe for the study of mixing in supersonic shear
ows. Exp. Fluids 13, 98104 (1992)
9. Clemens, N.T., Paul, P.H.: Scalar measurements in compressible axisymmetric mixing layers.
Phys. Fluids 7(5), 10711081 (1995)
10. Mungal, M.G., Dimotakis, P.E.: Mixing and combustion with low heat release in a turbulent
shear layer. J. Fluid Mech. 148, 349382 (1984)
11. Koochesfahani, M.M., Dimotakis, P.E.: Mixing and chemical reactions in a turbulent liquid
mixing layer. J. Fluid Mech. 170, 83112 (1986)
12. Dimotakis, P.E.: Turbulent free shear layer mixing and combustion. In: Murthy, S.N.B.,
Curran, E.T. (eds.) High-Speed Propulsion Systems of Progress in Astronautics and Aeronautics,
vol. 137, pp. 265340. AIAA, Washington, DC (1991)
13. Hermanson, J.C., Dimotakis, P.E.: Effects of heat release in a turbulent, reacting shear layer.
J. Fluid Mech. 199, 333375 (1989)
14. Dahm, W.J.A.: Effects of heat release on turbulent shear ows. part 2. turbulent mixing layers
and the equivalence principle. J. Fluid Mech. 540, 119 (2005)
15. Hall, J.L.: An Experimental Investigation of Structure, Mixing and Combustion in Com-
pressible Turbulent Shear Layers. Ph.D. thesis, California Institute of Technology (1991)
http://resolver.caltech.edu/CaltechETD:etd-09232005-141544
16. Loehrke, R.I., Nagib, H.M.: Experiments on management of free-stream turbulence. Technical
Report AGARD-R-598 (1972)
17. Knapp, R.T., Daily, J.W., Hammitt, F.G.: Cavitation. McGraw-Hill, New York (1970)
18. Bradshaw, P., Wong, F.Y.F.: The reattachment and relaxation of a turbulent shear layer. J. Fluid
Mech. 52, 113135 (1972)
19. Dimotakis, P.E., Hall, J.L.: A simple model for nite chemical kinetics analysis of supersonic
turbulent shear layer combustion. 23rd AIAA, SAE, ASME and ASEE Joint Propulsion
Conference. San Diego, CA, June 29July 2, AIAA-1987-1879 (1987)
20. Goodwin, D.G.: An open-source, extensible software suite for CVD process simulation. In:
Proceedings of CVD XVI and EuroCVD Fourteen, Electrochemical Society, pp. 155162 (2003)
506 Flow Turbulence Combust (2008) 80:489506
21. Konrad, J.H.: An Experimental Investigation of Mixing in Two-Dimensional Turbulent Shear
Flows with Applications to Diffusion-Limited Chemical Reactions. PhD thesis, California
Institute of Technology. http://etd.caltech.edu/etd/available/etd-10132005-105700/ (1977)
22. Frieler, C.E., Dimotakis, P.E. Mixing and reaction at low heat release in the non-homogeneous
shear layer. First National Fluid Dynamics Congress, Paper 88-3626 (1988)
23. Dimotakis, P.E.: Two-dimensional shear-layer entrainment. AIAA J. 24, 17911796 (1986)