CALCULATION OF ,THRE,ETDIMENSIONAL TURBULEN T SUBSONIC FLOWS IN TRANSITION DUCTS

Ralph Levy, Henry McDonald and W. Roger Briley

Scientific Research Associates, inc.

Glastonbury, Connecticut 06033, U.S.A.

ABSTRACT

A method for computing three-dimensional turbulent subsonic flow in curved

ducts is being developed. A set of tube-like surface oriented coordinates is
employed for a general class of geometries applicable to subsonic diffusers with
offset bends. The geometric formulation is complex and no previous treatment of
this class of viscous flow problems is known to the authors. The duct centerline
is a space curve specified by piecewise polynomials. A Frenet frame is located on
the centeriine at each axial location. The cross sections are described by super-
ellipses imbedded in the Frenetframe. Duct surfaces are also coordinate surfaces,
which greatly simplifies the boundary conditions. The resulting coordinates are
nonorthogonal.

An approximate set of governing equations is employed for viscous flows

having strong flow in a primary flow direction. The derivation is coordinate
invariant and the resulting equations are expressed in tensor form. These equations
are solved by an efficient alternating direction implicit (ADI) method. This
numerical method is generally stable and permits solution in difficult geometries
using the general tensor formulation.

ANALYSIS

Subsonic flow is known, to be governed by equations which are elliptic; that is,
by equations which require downstream boundary conditions. Solution by forward
marching integration is not appropriate, at least not without some sort of iterative
procedure to satisfy the downstream boundary conditions. To circumvent this problem
for subsonic flows, so-called "parabolic flow assumptions" are used and it is
thus assumed that the pressure field appropriate for irrotational inviscid flow
through the passage represents a given, reasonable first approximation to the actual
pressure field. The parabolic flow assumptions also presume a primary flow in a
predominantly streamwise flow direction and a Reynolds number sufficiently high to
permit the assumption of negligible streamwise diffusion. The inviseid axial
pressure gradients computed with appropriate downstream boundary conditions are
"imposed" upon the flow, m u c h a s in conventional boundary layer theory, so as to
permit solution by forward marching integration for subsonic flows (of., Ref. I).
The forward marching integration requires there be no flow reversal in the primary
flow direction. For internal flows, the inviscid pressure gradients are corrected
for internal flow losses associated with the well-known viscous pressure drop and
blockage effects by a process which is consistent with forward marching integration.
The imposition of inviscid pressure gradients incorporates a ri~_r~_~the elliptic
effects associated with a subsonic pressure field without the necessity of solving
elliptic equations other than for an inviscid flow.

This work was sponsored by NASA Lewis Research Center under Contracts NAS3-17522
and NAS3-19856.
 362

In conservation law form and in general curvilinear coordinates, the time

average Navier-Stokes equations from Ref. 2 are:

.... o

(1)

o I (pViVJ + gUg+~ij)
OyJ .. Oyi
Oxs ~ J =0
i, -i,
where the y s are the curvilinear coordinat'es, the v s are the curvillnear
velocity components, ~ is the density, ~ij is the viscous portion of the stress
tensor, ~ is the pressure, gij is the inverse of the metric tensor and g = Det (glj)
and density fluctuations are neglected. In the present application, the stress
tensor was modified to neglect streamwise diffusion (Ref. 3) and the pressure
was modified by parabolic flow assumptions (Ref. 2).

At present a mixing length model is being used which employs the following eddy-
viscosity formulation for the Reynolds stresses:

= Re OX,I
The mathematical form of the expression for the turbulent viscosity follows that of
Beer and Chigier, Ref~ 4:

#_Z:pL2(2~:~)v2
Re
O)

where ~ is the mean flow rate of strain tensor

(4)

The mixing length % is determined from the empirical relationship of McDonald &
Camarata (Ref. 5) for equilibrium turbulent boundary layers which can be written

(s)
(y) = o.o9 Bb ,o,,h

where 6b is the local boundary layer thickness, K is the yon Karman constant, taken
as 0.43, y is distance from the wall, and ~ is a sublayer damping factor defined
by

,.~ = pl,."@( y _ ' ~ ) / ~ . (6)

where P is the normal probability function, y+ = ~(~/p)i/2/(p/p), T is local shear

stress, ~+ = 23, and o I = 8.

The numerical method presently under development follows the general approach
developed by McDonald & Briley (Ref. 6) for laminar supersonic flow in rectangular
jets. A detailed discussion of the calculation procedure is not included here, as
such a discussion would be lengthy, and discussions of the general approach being
followed are given elsewhere (Refs. 6 and 7). The method is based on implicit
schemes which are potentially stable for large step sizes. Thus, as a practical
matter, stability restrictions which limit the axial step size relative to the
transverse mesh spacing and which become prohibitive for locally refined meshes are
not a factor in making the calculations. The general approach is to employ an
implicit difference formulation and to linearize the implicit equations by expansion
about the solution at the most recent axial location. Terms in the difference
equations are then grouped by coordinate direction and one of the available
 363

alternating direction implicit (ADI) or splitting techniques is used to reduce

the multidimensional difference equations to a sequence of one-dimensional
equations (Refs. 8 and 9). These linear one-dimensional difference equations can
be written in block-tridiagonal or a closely related matrix form and solved
efficiently and without iteration by standard block elimination techniques. The
general solution procedure is flexible in m a t t e r s of detail such as the type and
order accuracy of the difference approximations and the particular scheme for
splitting m u l t i d i m e n s i o n a l difference approximations. By solving the equations
fully coupled and by employing%a formal linearizatlon procedure the application of
coupled implicit boundary conditions is straightforward. The use of coupled
implicit boundary equations played a significant role in the development and
implementation of this analysis.

This analysis in its general form has been tested by application to the flows
in ducts whose centerlines are specified by piecewise polynomial functions. Cross
sections of the ducts in planes normal to this centerline are described by the
equation of a superellipse (in local cartesian coordinates):

(7)
blxl .lyi = rn a

where a is the superelliptic exponent, b is the ratio of major to minor axes

(sometimes called the shape factor), and m is the length of the minor axis.

COMPUTATIONS

A series of test calculations has been performed to assist in the development

of the analysis and computer program. Flow in the entrance region of a circular
straight pipe was computed at a Reynolds number of 500. At an inlet Mach number
of 0.01 the results, Fig. i, compared very favorably with incompressible experimental
data, Ref. i0. For a pipe of noncircular cross section at a Reynolds number of
50 O, the inlet Math number was progressively increased from 0.01 to 0.i, 0.2 and
0.3 and the expected effects of compressibility were observed. For the case with an
entrance Math number of 0.3 the calculation proceeded until the flow choked, Fig. i.
This behavior corresponds to that expected from solution of the initial value problem
posed. Additional calculations using a variety of cross sections have also been
performed including elliptical cross sections of shape factors, b, from 1.0 to 2.0
and superelliptical cross sections of superelliptic exponent (a in Eq. 7) from
2.0 to i0.0, shown in Fig. 2. Three of these are shown in Fig. 2.

A series of calculations has been performed for flow through a duct which
undergoes an S-shaped bend, as in Fig. 3. These cases were run at a Reynolds number
of 500 and an inlet Mach number of 0.I. Four different duct cross sections were
used: circular, elliptic with shape factor of 1.5, elliptic with shape factor
of 0.667, and superelliptic of exponent I0.0. Although the required inviscid
pressure field can be generated by numerous methods, in the S-shaped bend cases
considered here a suitable three-dimensional inviscid pressure field was not yet
available. The inviscid pressures were generated at each axial station by a highly
approximate method based on a free vortex pressure distribution using the local
radius of curvature of the centerline. Some sample cross flows near the start of
the first bend are presented in Fig. 4. Turbulent flow through the tighter
S-shaped bend of Fig. 3 was computed at a Reynolds number of 105 . Several laminar
and turbulent calculations were also performed for transitions from a nearly square
to round cross section as shown in Fig. 5,
 3~

CONCLUSIONS

A method for computing three-dimensional turbulent subsonic flow in curved

ducts has been outlined. A surface-oriented coordinate system is used for a general
class of geometries applicable to subsonic diffusers with offset bends. The
approximate set of governing equations are applicable to viscous flows having
strong flow in a primary flow direction. The formulation is coordinate invariant
and the resulting equations are expressed in tensor form. These equations are
solved by an efficient alternating direction implicit (ADI) method. This numerical
method is generally stable and permits solution in difficult geometries using the
general tensor formulation. Further development of the analysis and numerical
method appears warranted by the results described here.

REFERENCES

i. Briley, W R.~ Numerical Method for Predicting Three-Dimensional Steady

Viscous Flow in Ducts. Journal of Computational Physics, Vol. 14, 1974, p. 8.

2. Eiseman, P. R., R. Levy, H. McDonald and W. R. Briley, Development of a Three-

Dimensional Turbulent Duct Flow Analysis. NASA CR-3029, 1978.

3. Eiseman~ P. R., A Unification of Unidirectional Flow Approximations. Sixth

International Conference on Numerical Methods in Fluid Dynamics, 1978.

4. Beer, J. M. and N. A. Chigier, Combustion Aerodynamics. Wiley, New York, 1972.

5. McDonald, H. and F. J. Camarata: An Extended Mixing Length Approach for

Computing the Turbulent Boundary-Layer Development. In Proceedings,
Stanford Conference on Computation of Turbulent Boundary Layers, Vol. I,
Published by Stanford University, 1969, pp. 83-98.

6. McDonald, H. and W. R. Briley: Three-Dimensional Supersonic Flow of a Viscous

or Inviscid Gas. Journal of Computational Physics, Vol. 19, No. 2, 1975,
p. 150.

7. Briley, W. R. and H. McDonald, Computation of Three-Dimensional Turbulent

Subsonic Flow in Curved Passages. United Technologies Research Center Report
R75-911596-8, 1975.

8. Douglas, J. and J. E. Gunn, Numer. Math, Vol. 6, 1964.

9. Yanenko, N. N., The Method of Fractional Steps, Springer-Verlag, New York, 1971.

i0. Reshotko, E., Experimental Study of the Stability of Pipe Flow, Pt. i
Establishment of an Axially Symmetric Poiseuille Flow. Jet Propulsion
Laboratory Progress Report No. 20-364, October 1958.

ACKNOWLEDGEMENT

The authors are indebted to Mr. B. Anderson of NASA Lewis Research Center
for kindly constructing the plots presented in Figs. 2, 3, and 5.
 365

FIGURE I
O I"1 0 ' ~ DATA OF RESHOTKO
= PREDICTION

RADIUS, R

0 0 0 0 O 0.4 0.8 1.2 1.8 2.0

A X I A L VELOCITY RATIO, u/G

V e l o c i t y Profiles in t h e Entrance Region o f a Circular Pipe

SUPERELLIPTIC EXPONENT C = 3

NOTE: FLOW --"~'~ M - O 3

4.0 CHOKED / I i

&5

&C
uE
Mi - 0.2

2.6

Mj - 0.1
2.0 M i " 0.01

t.5

I I I, I I I I
1.0 0 0.1 0.2 0.3 0.4 0.6 0.6 0.7 0,8
X/R
Re
E f f e c t o f Mach N u m b e r on Centerline V e l o c i t y R a t i o
 FIGURE 2

Entrance Region to an Elliptic Straight Pipe

CO
03
03

Computational
planes
1.1 1.5
2.0
Shape factor
 FIGURE :5

CO
03
.,,j

S-SHAPED BENDS WITH CIRCULAR CROSS SECTION

(FLOW LEFT TO RIGHT)
 Ol = 1N~NOclX':] '1::)no OIJ.cll773~3clFIS CI3A~IFK)
....... i":i', * ~vo ".'i:" a i ........
f , . . . I ! |
' I
I t , . '
I " " "
I I
. . . . . . . //
S'I = 3dVHS '1017(11 011d1773 C]3A~I17C)
_. :..~ ...... .. _ - . .
"~, _-- ---~'-~'~~-- /
,m ,.e ,p~ ,ha ~* . 'lo . 1.4, *

~,,,, .'.... .. : . . , / ~
10170 ~IV717D~IIC) CI3A~II7O
,,.'" .".'"'..--"'.'-. "".. I
0 IJ P ~ p
\' ,,' ,' , ' . :. . ... "z/
\\-., : : - ; ; //
~"~- "-- - - - i //
s10ncl Cl3A~ln0 NI $31110073A M07.-I SSO~IC)
3~Ii'19 I:I
89~
 369

FIGURE 5

I
i

Transition Duct Superellipse Exponent = 10 to Circular