45th AIAA Aerospace Sciences Meeting and Exhibit AIAA 2007-115

8 - 11 January 2007, Reno, Nevada

Numerical Simulations of a Scramjet Isolator Using RANS and

LES Approaches

Jeffrey B. Allen∗ and Thomas Hauser† C-J John Tam‡

Utah State University, Logan, Utah, 84322, USA Taitech, Inc., Beavercreek, OH

The internal, compressible, turbulent flow through a scramjet-isolator configuration is presented, with the
primary goal being to better determine the shock train leading edge location of a typical Mach 2 nozzle-isolator
configuration. Both 2D and 3D approaches are utilized, in conjunction with a variety of different turbulence
models taken from both RANS and filtered models. The effects of inlet turbulence, as well as the use of grid
adaption techniques are evaluated under the 2D assumption and render certain simplifying assumptions valid
for the 3D cases. Experimental comparisons reveal that the RANS approach best conform to experimental
observations, while the LES approach showed the most degree of disparity. Further LES simulations are
warranted however, particularly since these were performed without the aid of density based solvers, which
greatly facilitates the prior to the inception of a new FLUENT release, (version 6.3) which contains several
improvements over previous releases with respect to compressible, turbulent flows.

I. Introduction
In light of current interest levels in high altitude, high Mach number, airbreathing technologies, improving scramjet
isolator performance has become a necessity.1 Numerical simulations used independently and in conjunction with
empirical studies2 have greatly enhanced our understanding of these flows, and given us better insight into the inherent
difficulties associated with modeling the complicated flow physics.
The present work represents an effort to determine the shock train leading edge of a scramjet-isolator. The isolator’s
primary function is to contain the shock train and prevent interaction between the scramjet combuster and the inlet.3
The length of the scramjet is thus a function of the pressure ratio between combustor and inlet. An increase in this
ratio during operation will cause the shock train length to increase and potentially interact with the inlet, leading to
inlet and engine unstart. Because the shock train consists of a series of normal and/or oblique shocks, the leading edge
will either be normal or oblique. It is well established, that these compressible effect can significantly influence the
turbulent boundary layer, often causing separation, and create a potentially high degree of flow unsteadiness.4 The
measurement of the static pressure along the isolator wall thus represents a straightforward and reliable way to monitor
these changes in boundary layer structure. The present work, therefore, will utilize Computational Fluid Dynamics
(CFD) in an effort to better determine the shock train leading edge and compare these results with experimental
observations. The study will utilize both the RANS and space filtered (LES) turbulence models, as well as evaluate the
effects of several different inlet conditions, and grid adaption techniques to the overall solution. Both two-and three
dimensional scramjet geometries are considered, the former, being used to arrive at simplifying assumptions prior to
the use of the more computationally expensive three-dimensional applications.

II. Numerics and Modeling

Both 2D and 3D geometries were considered in this study, the former, due to the scramjet symmetry in the z-plane,
the relatively significant cross-stream coordinate length (see Figure 1), and the need to test various inlet conditions
and grid adaptation techniques. FLUENT 6.35 was used for all cases with double precision accuracy. The RANS
simulations were conducted in steady-state using the Spallart-Allmaras6 and standard k-ε7 turbulence models. All
RANS solutions incorporated density-based solvers, coupled with implicit formulation. All RANS flow and transport
∗ Post-Doctoral Researcher, Center for High Performance Computing, Utah State University, Studet Member AIAA.
† Director,Center for High Performance Computing, Utah State University, Member AIAA.
‡ Sr. Research Scientist, Contractor, AFRL/PRAS/Taitech, Inc., Associate Fellow of AIAA.

1 of 9

American
Copyright © 2007 by Jeffrey B. Allen and Thomas Hauser. Published Institute ofInstitute
by the American Aeronautics and Astronautics
of Aeronautics and Astronautics, Inc., with permission.
quantity discretizations were performed using either the second order upwind or the third order MUSCL approach as
specified. The inviscid, convective fluxes were evaluated using the Roe flux difference splitting8 method.
The LES simulations utilized the dynamic, Smagorinsky-Lilly sugrid-scale turbulence model9, 10 along with a
coupled, explicit solver. Space and time discretizations were accomplished using Second-order, up-wind, and the
Non- Iterative Time Advancement (NITA) scheme employing a second order fractional step method,11 respectively.
In all cases, unless noted otherwise, the wall adjacent cells corresponded to the log-law layer (30 < y + < 500).
The law of the wall was thus employed according to:
 
u 1 ρut y
= lnE (1)
ut κ µ
where, u is the velocity parallel to the wall, ut is the shear velocity, y is the distance from the wall, κ is the von Karman
constant (0.4187), and E = 9.793.
The simulations were performed using up to 16 processors (for LES cases) on the Utah State University “Uinta”
supercomputer: a Linux Networx cluster composed of 62 compute nodes, each with dual core AMD Opteron 265’s
and 4GBytes of main memory.

III. Computational Domain and Boundary Conditions

A simplified representation of the scramjet geometry, coordinate system, and constant pressure boundary condi-
tions is shown in Figure 1. As shown, the origin, and reference location of the Cartesian system is placed at the
beginning of the rectangular isolator, which is located approximately 0.297 m forward of the start of the nozzle. The
isolator then extends to a total length of 0.69 m, or 0.987 m from the start of the nozzle.
The inlet and outlet boundaries were specified as constant pressure boundaries corresponding to 49.86 psia and
20.94 psia respectively. The inlet turbulence intensity (TI), unless otherwise stated, was assumed 1%. The Reynolds
number based on the streamwise velocity and isolator hydraulic diameter is approximately 1.4E6.

Figure 1. Three dimensional scramjet geometry showing reference origin.

IV. Results
A. Two-Dimensional
Comparisons between adaptive versus non-adaptive meshes, inlet turbulence intensity and turbulence models are eval-
uated. Figure 2 shows typical contour results of Mach number, static pressure, temperature, and modified turbulent
viscosity, simulated using the Spallart-Allmaras turbulence model. As indicated, the pressure gradient in conjunction
with the convergent-divergent nozzle accelerates the flow to Mach 1 at the nozzle throat and approximately Mach 2
thereafter, through most of the isolator. At approximately x = 0.75 m the first set of oblique shocks appear, resulting
in dramatic increases in turbulent viscosity, static pressure and temperature.
Figures 2 (b) and 4 show contours of Mach number, and reveal the distinctive shock diamonds or Mach discs,
which are characteristic of supersonic nozzle flows. In particular, the overexpanded flow resembles the wave structure

2 of 9

American Institute of Aeronautics and Astronautics

as shown in Figure 3.

(a) Pressure (Pa) (b) Mach Number

(c) Temperature (K) (d) Modified Turbulent Viscosity (m2/s)

Figure 2. 2D contours of number density, Mach number, static pressure and temperature simulated using Spallart-Allmaras

Figure 3. The Isolator shock train with shock diamonds: characteristic of an overexpanded flow

The effect of adaptive grid refinement in accordance with flow-field density gradients at the isolator shock location
was first evaluated. As shown in Figure 4 the level of mesh refinement increases from no adaption involving only
10,000 cells to the level 2 adaption involving 61,000 cells. As indicated, the level 2 grid adaption clearly shows the
appearance of a series of oblique shock waves occurring aft of the initial disturbance. The corresponding Mach number
contours are shown in adjacent views, and further illustrate the phenomenon. As shown, the advantage of using the
grid-adaption approach appears in the qualitative comparisons between the resolved levels. In particular, the level
2 adaption clearly results in a much improved resolution of shock detail. The corresponding Mach contours reveal
distinctive features well aft of the initial few shocks.
Figure 5 (a) shows the static pressure along the isolator baseline, and the corresponding leading edge shock train.
The results shown, compare the results from progressive grid adaptations. As indicated, the disparity between adapta-
tions is fairly minimal, with levels 1 and 2 showing the closest agreement. The greatest level of disparity occurs well
aft of x = 0.8m, and reveals again, the non-adaptive case showing the least agreement. Figure 5 (b) shows the effect of
varying the inlet turbulence intensity (TI) from a minimum of 1% through 5% and a maximum of 10%. As indicated,
the effect of increasing the inlet TI results in a slight increase in pressure forward of approximately 0.8 m, and a small

3 of 9

American Institute of Aeronautics and Astronautics

decrease in pressure thereafter. On the whole however, the observed changes are minimal. Figure 5 (c) shows the
comparison between the Spallart-Allmaras and k-epsilon turbulence models, with all other simulation parameters (as
stated in Section II) equivalent. As indicated, the k-epsilon model predicts the initial shock train location forward of
the Spallart-Allmaras predicted location by approximately 6.5%.

Figure 4. 2D, Grid-adaption comparisons of Mach number using Spallart-Allmaras turbulence model.

It should be noted, that the two-dimensional results presented herein were not included in order to ascertain the
precise location of the shock train leading edge. Indeed the 3D geometry (as shown in Section III) precludes such a
simplifying assumption. Rather, its inclusion was incorporated in order to facilitate the test cases, already mentioned,
and arrive at possible simplifying assumptions for the follow-on, 3D simulations.
It is clear from these two-dimensional results, that the effects on the shock train leading edge location, as quantified
by the static pressure along the isolator baseline, are not greatly affected by either the inlet turbulence intensity or the
use of excessive grid adaptation. However, the choice of turbulence model does play a significant role. In light of
these results, the follow-on, three-dimensional simulations will focus primarily on comparisons between different
turbulence models and not on the inlet turbulence conditions, or the use of adaptive grids. The models that will
be compared include: the Spallart-Allmaras, the k-epsilon model, and the Large Eddy Simulation (LES) turbulence
models.

B. Three-Dimensional
1. RANS Models
Both the Spallart-Allmaras and the k-epsilon, models were employed in the RANS study. A grid resolution of
50x50x200 cells was employed, and resulted in a steady-state solution after approximately 4 hours running time
using 16 processors. The third order MUSCL discretization approach was utilized along with the Roe flux differenc-
ing scheme. Figure 6 shows contours of Mach number, and temperature using the Spallart-Allmaras turbulence model.
Unlike the 2D results, the initial oblique shock location has moved forward to approximately x = 4.0m, rendering the
Mach 2, isolator “test” section much smaller than its 2D counterpart. In similarity to the 2D simulations however, the
distinctive series of oblique shocks is clearly visible, and distinguished by the successively smaller regions of circular
profiles as indicated. Figure 7 shows quantitative comparisons between the two models via plots of the static pressure
along the isolator baseline (0 ≤ x ≤ 0.69). As indicated, the k-epsilon model resulted in a much steeper pressure
profile than did the Spallart-Allmaras case, achieving an approximate magnitude of 1.2E5 Pa at x = 0.42m. The
k-epsilon model also shows the initial shock location (x = 0.4m), slightly aft of the Spallart-Allmaras model.

4 of 9

American Institute of Aeronautics and Astronautics

2. LES Model
The results from the LES simulation are shown in Figures 8 and 9. A 50x50x200 grid resolution was employed, along
with the Non- Iterative Time Advancement (NITA) scheme employing a second order fractional step method,11 and
second order up-wind spatial discretization. The parallel simulation was run continuously for approximately 3 weeks
using a 16 processor domain decomposition. This resulted in approximately 800 flow through times, based on the
isolator length and average flow-field velocity. Figure 8 shows contours of density, Mach number, pressure and Subgrid
Turbulent viscosity at the initial shock location. As indicated, unlike the 3D, RANS solutions, the distinctive series
of oblique shock waves is not as readily visible, instead, the Mach contours show a significant degree of dissipation.
Figure 9 shows the corresponding time-averaged, static pressure along the isolator baseline. The shock location
(x = 0.7m) is well aft from the previous RANS simulation results.
Figure 10 shows the comparisons between the experiment and numerical simulations with respect to initial shock
location along the isolator baseline. As indicated, the best agreement with the experimental results occurs for the
RANS cases, showing an initial shock location at approximately x = 0.4 m. The maximum experimental, static
pressure of approximately 1.5E5, however, is not obtained by either of the RANS models. In stark contrast, the LES
model shows the greatest degree of disparity with the experimental results, indicating the initial shock disturbance
occurring at approximately x = 0.7 m, or approximately 3 m aft of the experimental and RANS results.

V. Conclusions
In this study, we have investigated the compressible, turbulent flow through a scramjet nozzle/isolator configura-
tion without combustion. The primary objective, was to better identify the shock train leading edge location using both
RANS and LES approaches, and comparing these to experimental observations. The 2D results revealed that the pre-
cise value of inlet turbulence intensity was not critical in the determination of the leading edge shock location, nor was
the implementation of adaptive grid refinement. The 2D results did however, show that the isolator shock was signif-
icantly influenced by the selection of turbulence model, showing a distinctive contrast between the Spallart-Allmaras
and the k-epsilon results. The 3D simulations therefore, were conducted used a variety of different turbulence models
including the Spallart-Allmaras, k-epsilon, and LES models. Comparison results showed that the RANS approaches
showed the least disparity with experimental values, while the LES simulations showed the greatest departure from
experiment. Further LES simulations, however, may be warranted, particularly since these were performed prior to the
inception of a new FLUENT release, (version 6.3) which contains several improvements over previous releases with
respect to compressible, turbulent flows.

References
1 Giacomazzi, E., Rossi, A. D., and Bruno, C., “Numerical Simulation of a Scramjet Combustion Chamber,” Proceedings of the International

Academy of Astronautics Space Debris and Space Traffic Management, San Diego, CA, 2003.
2 Lin, K.-C., Tam, C.-J., Jackson, K., Eklund, D., and Jackson, T., “Characterization of Shock Train Structures inside Constant-Area Isolators

of Model Scramjet Combustors,” 44th AIAA Aerospace Sciences Meeting and Exhibit, No. AIAA Paper 2006-0816, 2006.
3 Lee, D., Goyne, C., Krauss, R., and McDaniel, J., “Shock Train Leading Edge Detection in a Dual-Mode Scramjet,” 44th AIAA Aerospace

Sciences Meeting and Exhibit, AIAA, Reno, NV, January 2006.

4 Hamed, A. and Shang, J., “Survey of Validation Data Base for Shockwave Boundary-layer Interactions in Supersonic Inlets,” Journal of

Propulsion and Power, Vol. 7, No. 4, 1991, pp. 617–625.

5 Fluent 6.3 Users Guide, Lebanon, NH, 2006.
6 Spallart, P. and Allmaras, S., “A One Equation Turbulence Model For Aerodynamic Flows,” Tech. rep., American Institue of Aeronautics

and Astronautics, 1992.

7 Launder, B. and Spalding, D., Lectures in Mathematical Models in Turbulence, Academic Press, London, England, 1972.
8 Roe, P., “Characteristic based schemes for the Euler equations,” Annual Review of Fluid Mechanics, Vol. 18, 1986, pp. 337–365.
9 Germano, M., Piomelli, U., Moin, P., and Cabot, W., “Dynamic Subgrid-Scale Eddy Viscosity Model,” Tech. rep., Stanford, CA, 1996.
10 Lilly, D., “A Proposed Modification of the Germano Subgrid-Scale Closure Model,” The Physics of Fluids, Vol. 4, 1992, pp. 633–635.
11 Armsfield, S. and Street, R., “The Fractional-Step Method for the Navier Stokes Equations on Staggard Grids: Accuracy of Three Variations,”

Journal of Computational Physics, Vol. 153, 1999, pp. 660–665.

5 of 9

American Institute of Aeronautics and Astronautics

 1.5e+05

No Adaption; NCells=10,000
Level 2 Adaption: NCells=24,307
Level 3 Adaption: NCells=61,003

Static Pressure (Pa)

1e+05

50000

0.6 0.8 1
x (m)

(a) Effect of Grid Refinement

1.5e+05

Inlet Turb. Intensity=1%

Inlet Turb. Intensity=5%
Inlet Turb. Intensity=10%
Static Pressure (Pa)

1e+05

50000

0.6 0.8 1
x (m)

(b) Effect of Inlet Turbulence Intensity

1.5e+05

Spallart-Allmaras
k-epsilon
Static Pressure (Pa)

1e+05

50000

0.6 0.8 1
x (m)

(c) Effect of Turbulence Model

Figure 5. 2D comparisons results of static pressure along isolator baseline and initial shock location. Subfigure (a) shows the effect of
progressive grid adaption. Subfigure (b) shows the effect of changing the inlet turbulence intensity, and subfigure (c) shows a change in
turbulence model.

6 of 9

American Institute of Aeronautics and Astronautics

Figure 6. 3D, Spallart-Allmaras contours of Mach number, and temperature with close up views showing the distinctive series of oblique
shock waves.

Spallart-Allmaras
1.5e+05 k-epsilon
Static Pressure (Pa)

1e+05

50000

0.2 0.4 0.6 0.8 1

x (m)

Figure 7. 3D RANS comparisons of Spallart-Allmaras and k-epsilon, showing the static pressure along the isolator baseline (0 ≤ x ≤ 0.69).

7 of 9

American Institute of Aeronautics and Astronautics

 Figure 8. LES contours of density, Mach number, subgrid turbulent viscosity, and pressure.

1.5e+05

Time Average
6th deg. Poly Curve Fit
Static Pressure (Pa)

1e+05

50000

0.6 0.8 1
x (m)

Figure 9. Static pressure along isolator baseline using LES turbulence model.

8 of 9

American Institute of Aeronautics and Astronautics

 2e+05

k-epsilon
Experiment
Spallart-Allmaras
1.5e+05 LES
Static Pressure (Pa)

1e+05

50000

0
0.2 0.4 0.6 0.8 1
x (m)

Figure 10. Comparison plot of initial shock location along the isolator baseline (0 ≤ x ≤ 0.69) using RANS and LES turbulence models

9 of 9

American Institute of Aeronautics and Astronautics