AIAA JOURNAL

Vol. 39, No. 8, August 2001

Technical Notes
TECHNICAL NOTES are short manuscripts describing new developments or important results of a preliminary nature. These Notes cannot exceed six manuscript
pages and three Ž gures; a page of text may be substituted for a Ž gure and vice versa. After informal review by the editors, they may be published within a few
months of the date of receipt. Style requirements are the same as for regular contributions (see inside back cover).

Geometry-Based Hyperbolic Grid .xm ; ym / D i

xb lab ¡ yb C ya i
lab i
; yb lab C x b ¡ xa i
lab (1)
Generation for Computational i i i i
.xn ; yn / D xb lbc C yb ¡ yc lbc ; yb lbc ¡ xb C xc lbc (2)
Fluid Dynamics
.x o ; yo / D [.x m C x n /=2; .ym C yn /=2] (3)
¤ † ‡
S. Anil Lal, B. V. S. S. S. Prasad, and N. Sitaram i
where lab i
and l bc are lengths of the edges ab and bc on 0i , respec-
Indian Institute of Technology, tively.
Madras, Chennai 600 036, India
Downloaded by MONASH UNIVERSITY on February 5, 2015 | http://arc.aiaa.org | DOI: 10.2514/2.1491

Cell Area
Consider that a cell should be constructed about an edge ab. The
Introduction i
cell area is a functionof the length of the speciŽ c edge l ab on the base

I
1
N computational  uid dynamics hyperbolic grid system is best curve 0i , length of the correspondingedge lab on the boundarycurve
suited for obtaining orthogonal grids near solid walls for accu- .01 /, and a distanceparameter ±i , i D 1 to N . The distance parameter
rately estimating the  uxes and gradients of  ow properties. The ± determines the normal dimension of the cell. It is formulated with
available hyperbolic grid-generation techniques are based on two the help of an exponential stretching function:
coupled nonlinear, hyperbolic partial differential equations: one
for the cell area (Jacobian) and the other for the orthogonality.1 ±i exp[s.N ¡ i /=.N ¡ 1/] ¡ 1
D1¡ (4)
These equations typically require a linearization procedure and 1 es ¡ 1
a block tridiagonal matrix solver for their numerical solution.2
In some cases artiŽ cial dissipation terms are added for dampen- where 1 D ±N , the value of 1 is chosen as a function of the charac-
ing the grid oscillations.3 Such numerical methods, though mathe- teristic dimension in question, e.g., chord length C h for an airfoil.
matically elegant, are complex and time consuming. In this Note The area of cell formed over the edge ab on 0i is obtained by
simple algebraic relations based on analytical geometry are de- giving chosen weightages to the lengths of edges on the boundary
rived for ensuring the orthogonality and providing required cell curve 01 and the base curve 0i . This area A abed is given by
area.
i 1
Aabed D lab " C lab .1 ¡ "/ h i (5)
Methodology
The boundary curve chosen for grid generation is deŽ ned by a set where " is called the cell size control factor and h i .D ±i ¡ ±i ¡ 1 / is
of discrete data points. The grid generation is carried out by starting the cell height. A suitable value for " (between 0 and 1) is selected
from the boundary curve and marching in successive layers. The to get an appropriate cell size distribution within the domain.
boundary curve is regarded as the base curve for the generation of
the Ž rst grid layer. Quadrilateral cells are constructed on outward Cell Construction
normals at all grid points of the base curve. The outer surface of The cell above the edge ab of Fig. 1 is formed by knowing the
these cells is treated as the base curve for the generation of the next coordinatesof the points a, b, d, o and the prescribedcell area A abed .
grid layer. This marching process is continued until a prescribed Here e is the grid point in the new layer, and it corresponds to the
number of (say, N ) layers are formed. The procedure for obtaining point b on the base curve. Whereas the coordinatesof the point o are
the outward normals and the cell area is explained in the following calculated from Eqs. (1– 3), the coordinates of point d are obtained
sections. during the cell formation on the preceding edge of ab. From the

Construction of Outward Normal

Consider a base curve 0i in Fig. 1, which has discretized points
such as a, b, and c. At point b, bm, and bn are unit vectors along the
outward normals of the edges ab and bc, respectively. The outward
normal at the point b of the curve 0i is along the ray bo, which is
the bisector of angle 6 mbn. The coordinates of the points m, n, and
o are derived as

Received 28 November 2000; revision received 20 March 2001; accepted

for publication 4 April 2001. Copyright °c 2001 by the American Institute
of Aeronautics and Astronautics, Inc. All rights reserved.
¤ Research Scholar, Thermal Turbomachines Laboratory, Department of

Mechanical Engineering.
† Professor, Department of Mechanical Engineering, Thermal Turboma-

chines Laboratory; prasad@acer.iitm.ernet.in.

‡ Associate Professor, Thermal Turbomachines Laboratory, Department Fig. 1 Schematic diagram illustrating the hyperbolic grid-generation
of Mechanical Engineering. method.

1631
 1632 AIAA JOURNAL, VOL. 39, NO. 8: TECHNICAL NOTES

properties of triangles dbe and dbo in Fig. 1, the ratio of areas A dbe view of the grid near the trailing edge when " takes a value of unity.
and A dbo is equal to the ratio of the lengths lbe and lbo . Thus, the As " increases, the grid lines near the rear stagnation point tend
coordinates of the point e are estimated as to diverge as the cell areas increase from inner layers to the outer.
Figure 3 shows the grid generated around a 90-deg sharp corner
.xe ; ye / D [ f xo C .1 ¡ f /x b ; f yo C .1 ¡ f /yb ] (6) with 1 D 1, N D 20, s D ¡1, and " D 0:5, which demonstrates the
capability of the method to discretize open geometries having sharp
where corners. The case of possible grid line intersection is handled with
controlledcell area variationsfor the corner(cavity) problem,shown
f D lbe = l bo D Ad be = Ad bo D . Aabed ¡ A abd /= A bod in Fig. 4.

The initial cells of every layer are constructed as rectangular cells.

Method of Damping Grid Oscillations

Dampening grid oscillations in the present method is done by
averaging the normal distances. For instance, the point e in Fig. 1
is relocated to e 0 , such that the distance between b and e0 is the
arithmetic mean of the distances lad , lbe , and l c f . In this manner
all of the points in the new grid layer are relocated repeatedly for
a few iterations, typically Ž ve times. For situations such as sharp
corners where there is a danger of grid line intersection, the cell
height .h i / in Eq. (5) is varied linearly to facilitate gradual turning
Downloaded by MONASH UNIVERSITY on February 5, 2015 | http://arc.aiaa.org | DOI: 10.2514/2.1491

of grid lines. Thus while smoothing the grid, small variations in cell
area are allowed.

Results and Discussion

Figure 2a shows the hyperbolic grid generated over an ellipse
having a major axis of 2 units and a minor axis of 0.2 units with
1 D 1, N D 20, s D ¡5, and " D 0:25. Figure 2b shows the enlarged

Fig. 3 Grid exterior to a corner.

a) " = 0:25

Fig. 4 Grid inside the corner of a cavity.

b) " = 1
Fig. 2 Hyperbolic grid generated around an ellipse with grid smooth-
ing (D = 1, N = 20, s = ¡ 5). Fig. 5 Hybrid grid system around NACA-0012 airfoil.
 AIAA JOURNAL, VOL. 39, NO. 8: TECHNICAL NOTES 1633

Mixing Enhancement in Compressible

Base Flows via Generation
of Streamwise Vorticity

C. J. Bourdon¤ and J. C. Dutton†

a) Streamline plot in laminar  ow
University of Illinois at Urbana– Champaign,
Urbana, Illinois 61801

Introduction

P REVIOUS studiesemploying ow visualizationtechniquesand

pitot pressure measurements1¡3 have shown that asymmetries
in the pressure Ž eld of the jets issuing from ideally expanded con-
verging and ideally or overexpanded converging– diverging nozzles
are caused by stationarystreamwise vorticespresentin the  owŽ eld.
The origins of these vortices have been traced to imperfections in
the nozzle surface. Krothapalli et al.3 assert that imperfections as
small as 121 th of the boundary-layer velocity deŽ cit thickness are
b) Mach contours in turbulent  ow
Downloaded by MONASH UNIVERSITY on February 5, 2015 | http://arc.aiaa.org | DOI: 10.2514/2.1491

sufŽ cient to trigger such behavior.

Fig. 6 Flow over an isolated NACA-0012 airfoil. Stationary streamwise vortices such as these were shown to im-
prove the mixing characteristics of axisymmetric jet  ows4 by in-
creasing the stream interface area and overall shear layer thickness.
A C-type boundary-layer grid around a NACA-0012 airfoil Therefore, these researchers found it beneŽ cial to promote stream-
has been generated up to a normal distance corresponding to wise vorticity generation by inserting surface disturbances onto the
1 = C h D 0:06. The values of N and s are 20 and ¡4, respectively. nozzle lip. King et al.4 found that the most effective shape for gener-
This boundary-layer mesh is coupled and with unstructured and al- ating streamwisevorticesin a supersonicjet is an isoscelestriangular
gebraic meshes as shown in Fig. 5. Figure 6a shows the streamlines tab, placed  ush on the surface, with an apex angle of 25– 30 deg.
in laminar  ow with the conditions for freestream Mach number This study also found that increasing the tab thickness increased
M1 D 0:5, Reynolds number based on chord Rech D 5 £ 103 and the shear layer thickness, although the beneŽ t was relatively small
incidence ® D 0 deg, same as those used by Liu.4 The present re- when compared to that of the thinnest signiŽ cant tab disturbance.
sults, including a small separation zone near the trailing edge, are Extension of such a technique to base  ows seems quite natural.
in good agreement with Liu.4 Figure 6b shows the isentropic Mach- If the streamwise vorticity generated from the tabs survives the base
number contoursfor the turbulentcompressible ow at M1 D 0:754, corner expansion fan, which has been shown5 to damp turbulencein
Rech D 3:76 £ 106 , and ® D 3:02 deg. With the help of hyperbolic the developing shear layer, it could signiŽ cantly in uence the tur-
grid, the Baldwin– Lomax turbulence model has been implemented bulence structuralorganizationin the near-wake region. In uencing
without any complex changes, such as those suggested by Turner the turbulent structure organization (and thus mixing between the
and Jennions.5 The standard k – " model is used in the unstructured freestream and core  uid) may substantially alter the base pressure
and algebraic grid zones. This result is in good agreement with the and drag characteristics of a bluff object. This is the subject of the
predictions of Kallinderis.6 current Note.

Flow Facility and Equipment

Conclusions The axisymmetric, supersonic  ow facility in the University of
A hyperbolic grid-generationtechnique is proposed based on an- Illinois Gas Dynamics Laboratory was employed in this study.
alytical geometric considerations. The main feature of the method The base model is supported by a 63.5-mm-diam sting, which
is generation of orthogonalquadrilateralcells with controllable cell extends through the supersonic converging– diverging nozzle. The
area. The grid generated by the proposed method is found to be con- freestream  ow before separation from the base model is at a Mach
venient for accurate prediction of laminar and turbulent  ow past number of 2.46, with a unit Reynolds number of 52 £ 106 m¡1 ,
arbitrary shaped bodies. and typical stagnation conditions of P0 D 368 kPa and T0 D 300 K.
The turbulent boundary-layer thickness on the sting/afterbody just
References before separation has been measured to be 3.2 mm (Ref. 5). A
1 Thompson, J. F., “Grid Generation Techniques in Computational Fluid schematic of the main features of the base region are given in Fig. 1.
Dynamics,” AIAA Journal, Vol. 22, No. 11, 1984, pp. 1505– 1523. The surface disturbanceswere formed in this study by application
2 Hoffman, K. A., and Chiang, S. T., “Grid Generation-Structured Grids,”
of pieces of adhesive shipping label to the surface of a blunt-based
Computational Fluid Dynamics for Engineers, Vol. 1, Engineering Educa- 1
afterbody. As stated earlier, disturbances as small as 12 th of the ve-
tion Systems, Wichita, KS, 1993, pp. 344– 411. locity deŽ cit thickness have been found sufŽ cient to produce asym-
3
Tai, C. H., Yin, S. L., and Sorg, C. Y., “A Novel Hyperbolic Grid Gen- metries in overexpandedand ideally expanded jets.3 In our facility,
eration Procedure with Inherent Adaptive Dissipation,” Journal of Compu-
tational Physics, Vol. 116, No. 1, 1995, pp. 173– 179.
this translates to a disturbance thickness of approximately 0.1 mm
4 Liu, F., “Numerical Calculation of Turbomachinery Cascade Flows,” (Ref. 5), the approximate thickness of the shipping label material.
Ph.D. Dissertation, Dept. of Mechanical and Aerospace Engineering, The disturbance thickness was altered by applying multiple layers
Princeton Univ., Princeton, NJ, June 1991. of the labeling material.
5 Turner, M. G., and Jennions, I. K., “An Investigation of Turbulence Mod-

elling in Transonic Fans Including a Novel Implementation of an Implicit

k – " Turbulence Model,” Journal of Turbomachinery, Vol. 115, No. 2, 1993, Received 11 March 2001; revision received 16 April 2001; accepted for
pp. 249– 307. c 2001 by C. J. Bourdon and J. C. Dut-
publication 18 April 2001. Copyright °
6 Kallinderis, Y., “Algebraic Turbulence Modelling for Adaptive Unstruc-
ton. Published by the American Institute of Aeronautics and Astronautics,
tured Grids,” AIAA Journal, Vol. 30, No. 3, 1992, pp. 631– 639. Inc., with permission.
¤ Graduate Research Assistant, Department of Mechanical and Industrial

Engineering. Student Member AIAA.

J. Kallinderis † W. Grafton and Lillian B. Wilkins Professor, Department of Mechanical
Associate Editor and Industrial Engineering. Associate Fellow AIAA.