SOLUTIONS

TO

THREE-DIMENSIONAL IN ROTATING THROUGH

THIN-LAYER COORDINATES

NAVIER-STOKES FOR FLOW

EQUATIONS

TURBOMACHINERY

By Amrit Raj Ghosh

A Thesis Submitted to the Faculty of Mississippi State University in Partial Fulfillment of the Requirements for the Degree of Master of Science in Aerospace Engineering in the Department of Aerospace Engineering

Mississippi

State,

Mississippi 1996

December

SOLUTIONS TO THREE-DIMENSIONAL

THIN-LAYER NAVIER-STOKES

EQUATIONS IN ROTATING COORDINATES FOR FLOW THROUGH TURBOMACHINERY

By Amrit Raj Ghosh

Approved:

David_'_S_L. _Jhitfiel-d/' Professor of Aerospace Engineering (Director of Thesis) Jo__ C. McWhorter Graduate Coordinator of the Department of Aerospace Engineering

,/9 _. Mark

./ / Janus

,. Jen-Ping Chen Research Engineer, ERC (Committee Member)

Assistant Engineering (Committee

Professor Member)

of Aerospace

Richard Dean

D. Koshel of the Graduate

t--School

A. Wdyne Dean

Bennett College

/ of Engineering

of the

Name: Amrit Raj Ghosh Date of Degree: December 13, 1996 Institution: Mississippi State University Major Field: Aerospace Engineering Major Professor: David L. Whitfield Title of Study: SOLUTIONS TO THREE-DIMENSIONAL THIN-LAYER NAVIER-STOKES EQUATIONS IN ROTATING COORDINATES FOR FLOW THROUGH TURBOMACHINERY

Pages in Study: 73 Candidate for Degree of Master of Science

The viscous, Navier-Stokes solver for turbomachinery

applications,

MSUTC has been modified to include the rotating frame formulation. The threedimensional thin-layer Navier-Stokes equations have been cast in a rotating Cartesian frame enabling the freezing of grid motion. This also allows the flowfield associated with an isolated rotor to be viewed as a steady-state problem. Consequently, local time stepping can be used to accelerate convergence. The formulation is validated by running NASA's Rotor 67 as the test case. Results are compared between the rotating frame codeand the absolute frame code.The use of the rotating frame approach greatly enhances the performance of the code with respect to savings in computing time, without degradation of the solution.

DEDICATION

This thesis is dedicated to my parents, Nirmalendu and Krishna Ghosh, and to the everlasting memory of my grandmother, Shanti Lata Bose.

ii

ACKNOWLEDGEMENTS

I would like to take this opportunity to express my sincere appreciation and deep senseof gratitude to Dr. David Whitfield for introducing me to the field of CFD, his teaching, and his patience, understanding, and guidance throughout the course of this work. Thanks to Dr. Mark Janus for being on my committee. A special thank you to Dr. Jen Ping Chen for the guidance he provided throughout this work, for painstakingly answering all my questions and queries and for being on my committee. I am also grateful to Dr. Lafe Taylor and Dr. Abi Arabshahi for getting me over the initial hang-up of understanding and running a 3-D CFD code.I also thank Dr. Murali Beddhu for reading my thesis and making valuable suggestions and clearing a lot of my doubts about the confusion of reference frames. I am grateful to Dr. Z.U.A. Warsi for introducing me to the field of tensors and removing my fear of superscripts and subscripts. I would like to extend my warmest thanks to all my friends for making my stay here, one of the most memorable phases in my life. Sreenivas, 'the answer guy', for helping everytime I went and cried to him about anything. Shyam, Rajeev,and Rambabu for their support, encouragement and advise. Laxman, for being a good friend and helping me out in my time of crisis. Pradeep 'Bhaisaab' and Bhabhiji for providing me with food and shelter and for always boosting my spirits when I was going through trying times. Ramesh, for his jibes about my graduation. Gayatri, Vikram and Minnie for all those nice parties. And to Rao,
oo

111

Goyalia, Dilnaz,

Krishnaraj, Sudeep This and

Sanjay, Nivu was Dr. Eric Engineering

Gill,

Aditya, good by

Edward, times NASA Contract Center

ND,

Rahul,

Lokesh,

Rizwan,

for all the supported McFarland,

we shared Lewis

together. Center under time was

research

Research

Grant provided

NAG-3-1712, by NSF

Monitor. at MSU.

Computing

Research

iv

TABLE

OF

CONTENTS

Page DEDICATION ...................................................... ........................................... .................................................. ................................................. ii iv vii viii

ACKNOWLEDGEMENTS LIST LIST OF TABLES

OF FIGURES

CHAPTER

I.

INTRODUCTION

..........................................

II.

GOVERNING 2.1 Continuity 2.2 Momentum 2.3 Energy

EQUATIONS Equation Equation Equation Law

................................. .................................. ..................................

4 8 8 9

...................................... Vector form of the of the Equations ................. ...................... .......... Equations ...............

2.4 Conservation 2.6 Curvilinear 2.6.1

9 13 14 19

2.5 Non-Dimensionalization Coordinate Thin-Layer

Transformation

Approximation

III.

NUMERICAL 3.1 Finite 3.2 Implicit 3.3 Newton's 3.3.1 3.3.2 3.4 Flux

SOLUTION Volume

METHOD

.......................

22 22 23 24 26 27 28

Discretization

...........................

Formulation Formulation Flux Viscous Vector Flux

.................................. ................................. Splitting Jacobian ........................... ..........................

Formulation Convective

..................................... Flux Computation ...................

3.4.1

28

CHAPTER 3.4.2 3.4.3 3.5 The Diffusive Turbulence Scheme Conditions Subsonic Flux Computation ................................... .................................. .............................. .....................

Page 29 30 31 32 32

Modeling

...........................

N-Pass

3.6 Boundary 3.6.1

Outflow

IVo

RESULTS 4.1 Rotor 67

................................................ ..............................................

37 37

V.

SUMMARY

AND

CONCLUSIONS

..........................

55

REFERENCES

....................................................

57

APPENDIX

Ao

DERIVATION OF THE VELOCITY OF COORDINATES .......................................

THE

MOVING 59

So

CURVILINEAR

COORDINATE

TRANSFORMATION ........................

OF

THE 62

NAVIER-STOKES

EQUATIONS

C.

CYLINDRICAL

COORDINATE

FORMULATION

............

68

vi

LIST OF TABLES

TABLE 4.1 Performance Enhancement .................................

Page 40

vii

LIST OF FIGURES FIGURE 2.1 3.1 3.2 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 4.10 4.11 4.12 4.13 4.14 Rotating Cartesian Coordinate Frame ....................... Unit Volume Computational Cell ........................... Codirectional Outflow .....................................
= 0.75) back ......................... pr. = 0.75) ......................... pr. = 0.85) ......................... pr. = 0.90) ............... ............... ...............

Page 6 22 33
43 43 44 44 45 45 46 .......... 46 47 .......... 47 48 .......... Peak 49 Span for Near Peak 50 48

Mass Flow History (back pr.

Mass Mass Mass Mass Mass Mass Mass Mass Mass Mass Mass Relative Flow Flow Flow Flow Flow Flow Flow Flow Flow Flow Flow History History History History History Ratio Ratio Ratio Ratio Ratio Ratio (log-linear, (back

pr. = 0.85) back

(log-linear, (back

pr. = 0.90) back

(log-linear, History History History History History History (back

pr. = 0.75) back

................... pr. = 0.75) ................... pr. = 0.85) ................... pr. = 0.90)

(log-linear, (back

pr. = 0.85) back

(log-linear, (back

pr. = 0.90) back Span

(log-linear, at 10%

Mach

No. Contours

for Near

Efficiency Relative Mach

.............................................. No. Contours at 30%

Efficiency

..............................................

ooo

Vlll

FIGURE 4.15 Relative Mach No. Contours at 70% Efficiency .............................................. Relative Relative Relative Plane Mach Mach Mach of Rotation No. No. No. Contours Contours Contours at 10% at 30% at 70% Span for Near Peak

Page

51 Span Span Span for Near for Near for Near Stall Stall Stall ...... ...... ...... 52 53 54 60

4.16 4.17 4.18 A.1

..........................................

ix

CHAPTER INTRODUCTION

The plex. length to model that This scales

flow flow

field field

associated is highly goal

with

turbomachinery involving for simulating reasonably feasible

applications a wide such well. for the range flows

is very of time

comand

unsteady, code

[1]. The

of a CFD

is to be able time

all aspects

of the

flow phenomena be practically

The CFD

computing code

is required tool

should

also

to be useful

as a design The is an attempt NSF ERC

for turbomachinery code, which this

applications. is a turbomachinery This code has been flow analysis research tool, at the

MSUTC

to achieve for Computational flow solver

goal. Field

under

development State

Simulation

at Mississippi of simulating even

University.

It is a viscous count,

which

is capable

or uneven-bladeor nonnoted code, that they

single-rotating, multistage, converged expensive, time. Newton

counter-rotating at angle-of solutions the the use

or rotor-stator, attack geometries. using

axisymmetric Chen the [2] above requires can be enhanced

symmetric, though are dous can very

unsteady because Thus,

can be obtained of Newton

sub-iterations of the code

tremenif one

computing avoid the

performance

sub-iterations. equations in the system. in the which vector For flows are the governing form equations so that machinery, Adamcyzk they of fluid are me-

Navier-Stokes chanics dent to cast can of the the be expressed coordinate equations

invariant in rotating

indepen-

it is convenient et al. [3] cast

rotating

coordinate

system.

2 the equations in a rotating cylindrical coordinate system to simulate viscous flow through turbines. Although unsteady computations are very important in turbomachinery simulation, many of the problems can beviewed as steady-state problems in the rotating frame, e.g. an isolated rotor. Consequently, local time stepping and multigrid methods can be utilized to accelerate convergence. Due to this motivation, at the start of this study an effort was made to develop a cylindrical coordinate system Navier-Stokes code for simulating the steady flow through a single blade row machine. During the course of the investigation it was realized that only a few modifications were necessary to incorporate the rotating frame formulation into the existing unsteady code if one retained the Cartesian coordinate system but solved the problem in a rotating frame similar to the approach in [4] and [5]. This is the approach that was adopted for the work presented in this study. The existing viscous solver is modified to include the rotating frame formulation using the absolute velocity vector. This thesis is divided into five chapters. Chapter II develops the governing equations cast in the rotating Cartesian coordinate frame. The velocity used in the equations, however is the absolute velocity represented in the rotating frame. The equations are then non-dimensionalized and transformed from Cartesian coordinates to steady curvilinear coordinates to enable the use of a body conforming grid. The thin layer approximation is also explained in this chapter. Chapter III describes the aspects of the flow solver with all the changes that are made due to the change in the formulation. The implicit formulation, Newton's method, flux formulation, flux Jacobians, the N-Pass solution scheme and the boundary conditions are all explained and discussed in this chapter. Chapter IV

3 presents the results obtained from the modified code. The test case used is NASA's Rotor 67. The results obtained from the rotating frame approach are compared with those obtained form the absolute frame approach. Chapter V is devoted to summary and conclusions.

CHAPTER II GOVERNING EQUATIONS The development of the Navier-Stokes equations in the rotating coordinates are presented in this chapter. The primary objective in this study is to make use of the quantities in the fixed frame formulation so that a minimum amount of changes need to be made to the existing code.Toachieve this goal, the formulation presented in this chapter develops the Navier-Stokes equations in rotating coordinates using the absolute velocity, i.e. velocity with respect to the fixed inertial frame. The differential forms of the equations of conservation of mass, momentum and energy can be collectively stated in a general form. The differential form of the general conservation principle can be stated as [6]:
0.4 ot + div f = C (2.1)

where

= Au

+ B and

A,

B,

C, are

tensor

quantities

such

that

and

have A and

the

same

tensorial (2.1)

order, is written velocity.

and

if B with

# 0 then respect

it is an

order frame from [6]:

higher so that the

than the

C. Equation u is the

to an inertial

vector frame

absolute

Consider frame i = which 1,2,3

a transformation is of the and form v

inertial

to a moving
Xi =

non-inertial t),

xi(Xl,X2,X3,

= t

5 where (Xl,
how, inertial and x 2, X 3) are frame. a
--i

the

rectangular

Cartesian the general vectors is the the

coordinates coordinates of the moving

defining defining frame.

the the If the

somemovrelato the

(x 1, x 2, x 3) are the fluid covariant particle, is denoted absolute [6]:

ing frame tive velocity

are a

base which by

of

velocity velocity

with of the

respect moving can

nonsteady nates by the

coordinates,

v and

coordi-

is denoted following u = v -

by w; the equation w

velocity

of a fluid

particle

be expressed

(2.2)

where

w _ The the

Oxi = -Ot

a -i components in the moving frame of w are frame with velocity notice. and the partial time derivatives

(2.3)

contravariant coordinates of the

of each

of

it can to the absolute of the

be proven inertial frame, derivation

to be negative frame. which are Thus, the

of the vector in

velocity w can the

moving

respect of the Details

be thought frame

of as the would

an observer presented

rotating A. to the time

in Ap-

pendix Due tial

above derivatives

transformation, in the two

the frames

following

relation [6]:

exists

between

the

par-

of reference

0( )
Ot Thus, the

0( )
Or

w _

grad(

(2.4)

unified

conservation

law,

equation

(2.1),

transforms

to [6]"

0.4

(gradA)

div

(2.5)

The
lOW.

rotating

Cartesian

coordinate

frame

utilized

in this

study

is illustrated

be-

/\
t_

/
Z

>

Figure It is evident sian in nature, due from

2.1 Rotating Figure

Cartesian the

Coordinate rotating frame between vary

Frame is rectangular covariant in space, Carteand so that: con-

2.1 that there the

to which Also

is no distinction base vectors do not

travariant
Oa

components.

-j
dx i

0 derivatives Ow c)xi of the base vectors are given by [6]:

(2.6)

The

temporal Oa -i Ov

(2.7)

Choosing

a rotating

frame

illustrated

in Figure

2.1 results

in:

w =

_2 x r

where

Y2 = -

YJa
-1

. Therefore,

w =
-

Qx3a
-2

+ Qx2a
-3

. Consequently,

c_w

div

w
_

= _
c)X i

ai
_

OwJ
c_Xi

a
-j

ai
_

Owi
Ox i

(2.8)

Using

the

following

tensor

identity

div

(Aw)

= (grad

A)

(div

w)A

(grad

A)

(2.9)

and

equation

(2.8),

one

can

rewrite

equation

(2.5)

as:

0,4
0r

+ div

(Aw)

div

= C

(2.10)

As explained cast the Navier-Stokes The

at the

beginning in the

of this rotating and absolute are

chapter, frame

the

primary the that

objective absolute are derived

is to velocbe-

equation

using

ity vector. low in this

continuity, have

momentum used the (2.10) can

energy

equations vector using

chapter,

velocity represented

u. However, the This base

all the vectors of

vectors the

appearing frame.

in equation This

rotating

concept

appear

confusing. related to any

confusion vector

is rein the is repre-

moved above sentation

by addressing equations. of the The The vector.

two important first being

aspects definition

velocity the

of the

vector,

and

second

keywords reference absolute flowfield, who 2.1, flow

to defining frame velocity with selects but one as

a velocity

vector to when as the

are:

with

respect the

to. This velocity particle

clarivector. in the

fies Thus,

which the

one is referring u, is defined respect a right which to the handed is not

defining

velocity

of a fluid

concerned ary one observer

fixed

inertial

frame.

Therefore, system velocity moving

a stationlike the

rectangular moving, other hand,

coordinate observes the

in Figure in the

of a fluid with the

particle

field

u. On the

an observer

8 axes described in Figure 2.1, observes the velocity ofthe fluid particle in the flow field as
ticle with Once the tor vector v. Thus, respect the arises. with vector vector the relative velocity v, is defined as the velocity frame. of, the of each issue other of representing and the once a vecof a fluid par-

to the definition These respect using can

rotating of the two issues

non-inertial vector are is taken

reference care

independent there system is no need in the

is defined same velocity rotating in this

to a frame, a coordinate be expressed coordinate the

to express same frame. with

components the to the the absobasis exerframe, derivits base rotating

of the lute of the cise

Thus,

in component system Navier-Stokes arising out in the in the shown

form

respect

Cartesian chapter

in Figure

2.1. Since rotating viz.,

is to cast operations div, are

equations equations frame

in the below, and

all the atives, vectors. frame

mathematical curl, grad,

time

carried and

rotating

involve the

To be consistent has been chosen

to simplify

mathematical and

operations, tensors.

to represent

all vectors

_.1 The law, the equation continuity 0Q 0-_ + continuity (2.10), equation equation

Continuity is obtained

Equation from f the = Qu, generalized B = 0, as: conservation C = 0. Thus,

by substitutingA in the rotating

= _, frame

can

be written

div

(Qw)_ +

div

(Qu)_

= 0

(2.11)

2.2 Momentum The momentum equation in the

Equation non-inertial frame is obtained by substi-

tutingA

= Qu,

= -

T = p I

a,

C = O in equation

(2.10).

9 The resulting equation is:

O(Qu ) or + div (_uw) + div (Quu) = div (T) (2.12)

where identity viscous

T is the tensor shear

stress and stress.

tensor, a is the It can

p is the deviatoric

thermodynamic part of the by: stress

static

pressure, containing

I is the the

tensor

be represented

= 2(div

u)I

+ tt[(grad

u)

(grad

u) T]

(2.13)

wherett ing can u

and = u_a

2 are and

the

first

and the

second result

coefficients in equation

of viscosity, (2.7), the

respectively. momentum

Writequation

utilizing as:

be rewritten

v,e i_ ,z:'u o_a

_g

div

(Ouw)
....

div

(Ouu)

div

_ (T)

. Ow OU_ox i

(2.14)

2.3 Energy The equation generalized by conservation A

Equation law, equation B = T" (2.10), u + q, yields the energy where

substituting

= et,

C = 0,

P et _ __ _-1
m

+ lp

I u I 2 is the in the

total

energy

and frame,

q is the therefore,

conductive is:

heat

flux

vector.

The

energy
lge t

equation

rotating

0--r" +

div

(etw)_

div

(etu)_

= div

(T-u)_

div(q)

(2.15)

2.4 Conservation Equations (2.11), (2.14)

Law and

Vector (2.15)

form can

of the

Equations and written in the

be expanded

10
conservation gence the and law form by carrying on the the various base and out the mathematical and tensors the operations involved. results axes are of diverWhile that the doing base

dot products involving vary in x 2 =y, result

vectors one the

operations do not

vectors that since

utilizes coordinate

vectors Cartesian x 1 = x, tions

forIn:

in space nature;

rectangular let five equaof the

a i " aj

= _ij"

For

simplicity, u 3 = w. vector The

x 3 =z, can then

v = t and

u 1 = u, into

u 2 = v, one single

that

be combined

equation

Oq Of O---i +-g-_

Og +g-y

Oh Oz

OF Ox

Og v 03,

Oh _ + 07

(2.16)

where vectors

the fv,

dependent

variable the

vector source

q, the term

flux

vectors s are

f, defined

g,

h, the

viscous

flux

h v, gV, and

vector

as follows:

(2.17) Ow
et

Qu

Ov Ovu
g

_Qz Ou_z
-QvQz
h

Ow + QQy
Owu , Owv
I I _ w2

Ou 2 +p
f

+ QuQy + Qv_y
4-

Quv

QV 2 +

(2.18)

_uw u(e t + p)

Qvw

OwQz et.Qz

4- p

Ow_y + et.Qy

v(e t + p)

w(et

+ p)

Z'xx Vxy Txz -IVTxy

0 ]
qW_'xz q-qx

11 0 ryx
gU

Tyy

(2.19)
WVyz + qy

ryz
UTyx + Vryy +

"gzx hV

T'zy
Tzz

VZ'zy

0 ]
+ WTzz +

qz

0 0
S

Ow_ OvQ 0

(2.20)

The

viscous

stress

terms

appearing

in the

equation

(2.19)

are

defined

below 2 _p.

un-

der

the

assumption

that

bulk

viscosity

is negligible,

so that

2 = -

OU

OV q- -ff'_)

Vxx = 2it(20u -_
"gyy =

Ov 03, Ou Ox Ou ox

Ow) Oz' Ow )' _ Ov -_) '

"gxy

"gyx =

_('-_

Ov _p(2 Ov 03,
T.yz = Tzy = _U("_

Ow
"1- "-_')

(2.21)

Vzz = _/_(20w oz

r_x = r:,z = _(--ff + -_)

appear vector in the viscous as: flux q vectors = are k(grad deT)

Ow

Ou

The fined

conductive below. The

heat

flux

terms

which heat flux

conductive

is defined

where of the

k is the heat flux

thermal vectors

conductivity, along each

and of the

T is the coordinate

temperature. directions,

The

components are:

therefore,

12
q.
qy qz

= k OT
Ox
=

k o7'

Oy

(2.22)

= kOT Oz components in terms frame vectors of the of the tensors base and vectors of the the For the that rotating components purpose the appeared frame. in the The involved of validation flow fields between needs equa-

All the tion code terms rotating ated (2.16) uses are

vectors so that frame.

existing are in

a fixed base

formulation of the fixed

of the frame as the

of the genercomto be obwith for with

formulation from

one needs the two and

to be able

to compare Thus, in the

solutions rotating

formulations. components components in the of the

a relation fixed frame absolute and, frame relations

ponents tained. respect

in the Letting

frame

Ua, Va, and

Wa be the

of the fixed

velocity assuming matches are

to a Cartesian that the

coordinate initial (say

system

frame,

convenience, the stationary

Ua =

configuration

rotating

frame
U

at instant

t = 0); the

following

obtained:

Va Wa

v cos(tgt)

w sin(_2t) w cos(t2t)

(2.23)

= v sin(t2t)

where, velocity equation ponents nents

it is emphasized with that of the in the Now, respect

that

u,

v,

and

w are

the

components It is apparent full are

of the from rotation, to the

absolute the the above com-

to a rotating t = 0 and velocity frame.

Cartesian after every rotating

basis.

at instant absolute stationary the

subsequent frame

in the

identical

compo-

Navier-Stokes there is no need

equations to move

have the grid

already anymore.

been

cast The

in the field

moving that is

frame,

so that

flow

13 generated in terms of the components, by solving the Navier-Stokes equations in the form presented in equation (2.16), for purposes of comparison, can be changed according to equation (2.23) to match the flow field which results from solving Navier-Stokes equations derived in the inertial frame. However, the distinct advantage of the former over the latter is that local time stepping can be used for steady flows in the rotating frame. Another advantage is the freezing of grid motion becausein the fixed frame formulation the grid motion is imperative.
2.5 Non-Dimensionalization All the ables. In order variables to obtain used in the formulation equation variables of the Equations are with dimensional non-dimensional by ^ ) is carried vari-

till now (2.16)

a form

of the

quantities, out using

a scaling the following

of all the relations:

dimensional

(denoted

x=--z, D

y =-z D

z=--z, D

u =-;--, ao

v =-z-, ao

w =-;--, ao

t2 =-7-ao

P =-^

_ ^---_, Q = 7--, Qoao Qo

at t =--;-, D

e =-^

^---5' _oao

h ---_, ao

It

= -^ t_o

V iJ ^

TiJ ^2'

V wall

Twall , _ ,_o \

qxi

?Ix, = ^ ^3 Ooao

'

T = -7To

'

^2 ^^ ao = yRT 0

Qa

( D )'--r--

The noted which the

scaling

has

been

accomplished are the total

using conditions blade tip.

the

reference and the detailed

quantities reference length/)

(de-

by subscript is the

0) which

maximum of the

diameter can

of the be found

The

development results in a form

of of

scaling

equations

in [2]. The

scaling

14 the vector equation which is exactly similar to equation (2.16) but with non-dimensional variables. The form of the dependent variable, flux, viscous flux and the source term vectors is also exactly same as shown in equations (2.17) - (2.20), but with all variables non-dimensional. The only difference arising out of the scaling of equations is the introduction of two new non-dimensional quantities, viz., Reynolds and Prandtl numbers (Re, Pr), which appear in the non-dimensional form of the viscous shear stress and the conductive heat flux terms. The non-dimensional viscous shear stress and the conductive heat flux terms are: 1 2
Vxx l_e_tt(2 1 2 au _-_ av av 03, au Ow _-)' aw 1 ,au Vxy = ryx = _-_w,.._ 1 ,av av + -_) aw

_yY- _-_-_t'(2 Oy
1 Vzz _ Re 2 p(2 aw az

Ox
Ou ax

_)'
av _)'

_yz = _zy = _-_t-_

1 .,aw rzx = rxz = _eett_--_

+-y-y-)
au + -_)

(2.24)

aT
qx

Re

(_ -

1) Pr

_x

qy

Re

1[

aT 1)Pr

(_,-

oy
aT

(2.25)

qz

where

Re

0aD
^

and

Pr

_ Cp I_

2.6 Curvilinear In order on a configuration conforming grid to carry out

Coordinate

Transformation of the governing equations to use a bodyto

numerical

computation geometry, in the

comprising in which the

of a complex solid

it is simpler geometry

surfaces

correspond

15 constant coordinate surfaces. To this end, a curvilinear coordinate transformation needs to be introduced in the rectangular Cartesian spacebeing used in the formulation thus far. The following general, nonorthogonal, steady, curvilinear coordinate system is introduced in the (x, y,
= _(x,y, r
T=t

z) space:

z) (2.26)

= r/(x,y,z) = (x,y,z)

Only tailed metric

the

final

results

due

to the

transformation involved in Appendix and

are the

stated how

in this the chain

section.

A defor the

presentation terms 0 at Ox _ are 0 Or

of all the obtained,

steps

relations rule

is made

B. Using

yields:

(2.27)
c9 oy -,y-_ -F O r/yT + Cy__

o Oz
The Jacobian _

+r/zT +z
J of the IO(x,y,z)l (2.28) z_y_) y_(x_z_z_x_) + z_(xrty_y_x) inverse transformation is:

J = detlo- , ,
= x_(y_z_-

The

metric

terms

in equation

(2.27)

are:

_x
_y

=
=

_(yrtz-z_)
_(Zr/X_Xr/Z_)

r/x
r/y

=
=

_(z_y_-y_z)
_(X_-Z_Z_X_)

Cx
_y

= _(y_z_= X(z_r/-

z_y_)
X_-Zr/)(2.29)

l(x_y-yrtx) 3

r/z

1 _(y_x

x_y_)

_z

--_(x_y_

- y_x_)

16

Substituting sults nates. in the Only oQ form the OF

relations of the final

obtained

in equation equations here.

(2.27)

into

equation curvilinear form is:

(2.16)

re-

governing

in stationary The OGv + resulting OH v +

coordi-

equation OG OH

is stated _ OF v

(2.30)

where

ol
)U

Q=J

(2.31)

ee]
oK' QuK' K=J QvK' QwK' etK' + kxp + kyp + kzp + pK where K' = kxu kxu + kyv + kzw + kt (2.32)

kt = K=

kyQz + kyv

+ kz_Jy + kzw

i;_

= J

(2.33)

T,z
Qk
0 0 S=J Ow_2 OvQ 0 (2.34)

In

the

above V, W;

equations i_

(2.32)

and

(2.33), for k

K = F, = _,

G,

H;

K'

= U', The

V',

W';

K = U, ance

= F v, G v, HV;

t], _ respectively. terms in

appear(2.11),

of the

term

k t is to account

for the

divergence

equations

17 (2.14), and (2.15) in which the vector

explained the other 00 0-7 which by considering equations. The the continuity w explicitly appears. This can be further

equation, equation can

the

treatment

being as:

same

for

continuity

be rewritten

div

[_(u_

+w)_

0 becomes

(2.35)

in curvilinear

coordinates

o(Jo)
Ot + 0.._0.[Jo(u Ox k that: + w) _
a k ]

(2.36)

It can

be verified

kt

w"

a k

(2.37)

and K' = (u + w) ak (2.38)

In the appears nent grid is just sions

formulation

of the

continuity similar vector.

equation and present not

in the denotes

absolute the

frame

the

term compo-

kt

in a strikingly of the grid velocity and

manner In the

contravariant however, any Q x grid there

formulation, with w = terms are

is no It

motion the

involved contravariant

k t should component

be associated vector, flux

velocity. expres-

of the and heat

r. The

for transformed

shear

stress

as follows:

Tkx
Tky

= kxvxx
= kxvxy

+
+

kyvyx
kyvyy

+
+

kzzzx kzvzy kzvzz wT_ + kxqx + kyqy + kzqz (2.39)

Tkz Qk

= kxvxz = uTkx

+ +

kyvyz VTky

+ +

where

18

=12[

Ou

Ou

OU

(_y_

q- _y_

+ _y_)

(_(+

_z_-_ ;z_-;) + 1

"_y y

8u Vzz 1 = _-#[ 2 Re3 [ Ow 2(_-ff + Ow ,7_-_ + _ -_) _, (_x

Ou

Ou

-_ + ,7_ + _x-_) (2.40)

(_Y_o _Y-_ + _yo_)] + ov

Ou

Ou

Ou

Ov

Ov

) + (_x

-_

v 1
]

1 .[(_ ow

ow l;,,_) + (,_u

ou

_e[ 1 [ qY - -- Re[(y

# #

Prl]_e oT + ___aT + ],e OT aT + + r/y--_

aT (2.41) aT _,y-_)

_-- " 1)Prl

_sY _a

19

2.6.1 Since very study. cid and the numerical

Thin-Layer simulation the because interactions of the fact normal very coarse

Approximation of the thin-layer it can handle between that most full Navier-Stokes approximation flows the grids which two [2]. equations is used have The both are in this invis-

computationally This approach

intensive, is useful and use

viscous

regions makes in the they are this

thin-layer equastreamcalculaand _ the in the the normal

approximation tions wise tions are packed

for Navier-Stokes whereas along the

direction usually study, Thus, are

to the and

walls, good

direction only [2].

enough 77 the heat flux

for inviscid spanwise, derivatives only are

In

_ is the all the

streamwise, viscous and

pitchwise streamwise second This

directions. direction derivatives are

dropped.

In the while

other the

two mixed

directions, derivatives and lead

considered are

neglected. calthe fol-

is done

because

they the

insignificant approximation equations: OGv 0rl

in corners

to expensive gives

culations lowing

[2]. Applying final oQ form oF + of the oG -_-

above

to equation

(2.30)

governing OH 0_

0-7
where

OHv 0_

(2.42)

Q = j

(2.43)

J
QU' mU' F=J _vU' owU' etU' + _p + _yp + _zP + pU G=J ouV' ovV' owV' etU' oV' + r/xp + r/yp + r/zP + pV ,H=J oW' ou W' ovW' owW' etW' + _xP + _yp + _zP + pW 2.44)

2O

01
T_x GV=J
HV =j

"0"

(2.45)

Q_
0 0 S=J Owt'2 OvQ 0

Q_

(2.46)

The

terms

U', for the in the

V',

W',

and

U, shear

V,

W are stress are:

defined and the

in equation transformed

(2.32). heat flux

The

ex-

pressions appearing

transformed viscous flux

terms

vectors

Tkx
Tky Tkz

=
= =

kxrxx
kxvxy kxvxz

+ + +

kyvyx
kyvyy kyvyz

+
+ +

kzvzx kzrzy (2.47)

kzvzz

Qk

= uT_

VTky

wTkz

kxqx

kyqy

kzqz

In

the

equation the

(2.47) thin-layer heat flux

above,

= _], _ for in the become:

flux

vectors

G v, H v respectively. the viscous stress and

Applying the

approximation terms will

k direction

conductive

rxx

_ _ee-_/u[_x"_-_- _,_, 1 2 .[,,_ -

ky'_-Ov

Owl ]_z"_
J

_ 1 2.[,,,_ vv oU_kz_g _ _yy- g_gF,[_.,_y_- kxg

rzz _ Re-3P 12 [, "_kz Ow Ok kx-_Ou ky Ov]

(2.48)

21

_'xy

_-

"gyx

,uLy

[k ou k
Ok -.Ix Ok J

1. [k Ov k owl Vyz = Vzy = -_eettL z-_ + y ok J

(2.49)

rzx = _xz =

tt kx--_

+ kz

qx =

(7-

1)Pr

-_-)

qY

----

(_'-

1)Pr

_)

(2.5o)

qz =

(7 -

1) Pr

The ing was proach entails this equations via the

approach was

described not initially

in this

chapter

for the

development The initial

of the route that

governtaken the ap-

employed

for this

study. later less

cylindrical above

coordinates. is much amount more

However, elegant

it was and

realized

described the particular For

time code

consuming.

It also for

minimum

of changes

to the

existing

for it to work

formulation. the development frame of the governing equations using C. The from equations cyfinal

completeness, coordinates equations

lindrical form

in a rotating derived

is presented C are then match the

in Appendix transformed governing

of the

in Appendix to see if they

cylindriderived

cal to Cartesian here in this

coordinates chapter.

CHAPTER III NUMERICAL SOLUTION METHOD The finite volume approach is utilized to obtain the numerical solution of the thin-layer Navier-Stokes equations which were derived in the previous chapter and are presented in their final form in equation (2.42).
3,1 Finite The main unit tained which curvilinear coordinate domain A_ Volume Discretization maps the into volume volume entire collection formulation computational physical of cells doof

transformation which is discretized finite a unit

to a computational volume. by Thus, integrating in the

= A_/ = A_ equation figure

= 1. The over

is obcell,

(2.42) below.

is shown

f
i I I I (i, j, k) I F
/ / l J /

/
_____>

Figure

3.1 Unit

Volume

Computational

Cell

22

23 In the cell-centered finite volume approach the dependent variable vector

assumed the cell. to be the The fluxes cell face. can then flow F, quantities G, The H, at the G v, H v are form cell center, and is uniform Q, is

throughout over each cor-

assumed

to be uniform (2.42) after

responding zation,

integral as:

of equation

spatial

discreti-

be written

-_r A_AyA_

+
+

(F i +-_ 1[(Vj+__ -

Fi_)ArIA_
V

Vj___)

(Vj+

V vs__I)]A_A_

[(H k +_1 - Hk_

) -

(H_k +_1 -

HVk_)]A_Ari

Defining ing that

the the as:

central finite

difference volume cell

operator has sides

_ as _l ( of unit

)l+ -

)/ _ and can

notbe re-

length,

equation

(3.1)

written

oQ
O-r- + hi(F) + 6j(GG v) + 6h(HHv) = S (3.2)

In the

above

definition

of 6,

l + 1 represents

the

face

adjacent

to the

cell center

in the center

positive in the

direction negative

of l and direction

l of/,

1 represents where l = i, j,

the k

face fort,

adjacent r/,

to the

cell re-

_ directions

spectively.

3.2 Implicit An implicit cause cous due these schemes scheme can very tolerable is utilized handle fine grid large

Formulation to numerically CFL numbers. is chosen step integrate equation (3.2) beviseffects, code

In problems to resolve the

involving viscous

computations, to which the

spacing time

minimum

cannot

be too small

for the

24 to have any practical value [2]. A

numerically equation: /iQn where /iQn Rn+l = Qn + l _ Qn + _ 6j (G) n+l 6k (Hv)n+l + 6k _
(H) n+l

first-order

time-accurate by the

implicit following

scheme algebraic

to

solve

equation

(3.2)

can be implemented

/iv

(R n+l)

(3.3)

= 6 i (F) n+l 6j (GV) n +1

(3.4)

sn+l

As can and the

be observed source term

from are

the

above

equation, implicitly.

the

convective

and

diffusive

fluxes

all treated

3.3 Newton's The (3.4). tem Newton's iteration method of the the =

Formulation is used in this work to solve equation a sys-

A detailed of nonlinear

explanation equations _ (3.3) x k)

method

is presented method can

in [2] and be stated as

[7]. For [7]:

Newton's _ F(x k) result

F,(x k) (x k+ l Dividing equation ,4q n where = d__[v J method = 0 =

(3.5) in (3.6)

by J, will

A--_ (R n + 1)

/iv

(3.7)

The

Newton's L(qn+l)

can

now

be applied

to the

following

vector

equation: (3.8)

where

25
L(qn+l) = qn+l _ qn + ,_ (Rn+l) (3.9)

The

resulting

matrix

equation

is:

[ I -

_-_ (__qS)*,k-

1 + _-_ (M)*,k-1

"l Aq k-1 (3.10) _ [qk-1 _ qn + Av (R *'k-l) ]

The of the served ual

operator unknown from

indicates vector

that

the

difference

operator = qk

M acts _ term to the qk-1

on the

delta

form be obresidof

Aq k- 1, where (3.10), that the help This and the

Aq k- 1 time

[2]. It can in the

equation arises The by

derivative method

is present nonlinear

which

applying

Newton's

system

equations. solution lems. The

sub-iterations time step.

in obtaining

a better better

approximation for unsteady (3.10) are

of the probdefined

at each various

method, terms

is therefore, appearing

operators

in equation

as follows:

M*,k-1.

= 5i (A)*,k-1.
(_j

+ Sj (B) *'k-l" 1 . _ 5k (CV)*,k1

+ Sk (C)*,k-1.

(BV)*,kS

where

OF =-b-q'

OG = 0--q-'

C-

oH 0-_'

Bv

OG v = 0--q- '

C v_

OH v Oq

(3.11)

R*,k-1

6i

(F)*,k-1

+ _

Oj (G)*,k-1 (_k (HV)*'k-1

(_k (H)*'k-1 _ s*,k-1

(_j (GV)*,k-1

The for computing mass and first flow a stator, iteration.

superscript the without the metric

* is used terms.

in the Janus when

above

equation that the n the and

to specify in order interface q0 should terms

the

time

level the a rotor for the

[13] found crossing level

to maintain between be used at the

discontinuities metric terms

at old time

For

the

subsequent

iterations,

metric

present

26 level
n + 1 and with qk - 1 are real to the of the used. A, B, and The C are the flux flux Jacobians, vectors which are 5

x 5 matrices ter The

eigenvalues. flux vectors

form

of the in the

derived

in chap-

2 is identical derivation

obtained

absolute their

frame

formulation. is given

transformed thesis [8].

flux Jacobians

and

eigensystem

in detail

in Janus's

3.3.1 The agating tates porate vector flux Jacobians in each have

Flux real

Vector eigenvalues direction. for the are

Splitting which The sign correspond of the of the the to wave eigenvalues fluxes. propdic-

speeds what this

curvilinear should

information approach method G-

be used vectors

evaluation using flux

To incorflux

the

flux

split

Steger-Warming are split

splitting F=F++F G

described

in [9]. The

vectors

as follows:

= G + +

(3.12)

H=H++H

where positive from

the

superscript

of plus so that direction. using set of flux this

denotes flux

the

flux

sub-vector is computed sub-vector the positive

corresponding using with the

to the

eigenvalue the negative is evaluated a new

sub-vector the flux from can

information superscript Utilizing

Similarly, information Jacobians

of minus this

direction. as follows:

concept

be formulated

A + A-

= OF + Oq ' OFOq '

B + B-

= 0__QG + Oq ' = O---G--GOq '

C + C-

= -

OH + Oq OH@q

(3.13)

The

detailed [10].

derivation

of these

new

flux

Jacobians

is available

in Belk's

dis-

sertation

27 3.3.2 Viscous Flux Jacobian The viscous flux Jacobians

computed elements numerically of the Jacobian in this study. B v and The C v, defined in equation used (3.11) to obtain are the

numerical

derivative

matrix

is [11]:

F l (x + hem)
alto ---h

Fl

(3.14)

where

em is the chapter, variable which

m th unit

vector

and fluxes

v/machine are

t. As will be explained of the derivative right will and

latleft

er in this dependent of two

the

viscous

at a face viscous

functions

vector. can

Thus,

the

numerical as:

consist

parts

be written

separately

(Blm)

G_ =
G_

(qj

+ hem,

qj+l) h

G_

(qj,

qj+l)

(qj,

qj+l

+ hem)

G_

(qj,

qj+l)

(B[m) -

= H[ (qk + hem,

h (3.15) qk+l h H_ (qk, qk + 1 + hem) h H_ (qk, qk + 1)

) --

(cL)+
( CV lm) --

=
=

H_

(qk,

qk+l)

Equation

(3.10)

can

be reformulated

by using

the

new

flux

Jacobians

as:

[ I -

A-_ (_qS)*,k-

1 +

"_/(M+)*,k-i _

. + (M-)*,k-1
[ qh1 _ qn

. }] +/I-'-z

Aqk-1 (R *,k -1)

(3.16)
]

where

28 (M +)*'k 1

6 i (A+)

*'h-1

+ (_j (B+)

*'k-l"

+ 5 k (C+) *'k-1 "

*,h-1

(M-)*,k1 .

dj [(BY)+] d i (A-)

*'k-1 *'k-l-

5 k [(CV)+] *'k-1

(3.17) = + dj (B-)
" -- (_k

+ d k (C-)
"

*'k-l"

-- (_j [(BV)-l*'k-1

[(CV)-]*'k-1

The curate ety

difference in time.

expressions A general time

that form

have

been

mentioned (3.16) which

till now

are

first

order

acvari-

of equation schemes

incorporates [2]:

a wide

of different

accurate

is written

below 1

1 OAr _ I +

"-_" (c)S,) *,k-1

I"

+ (M+
_ qn

"

M-

)*,k- lJ Aqk-

1 = (3.18)

1 + _0[

o-J;

qk-1

I_

1 1+_

{A_ [ (0 -

1)R n'n -

0 R *,k-1

+ _

Aqn-1}]

A few

important

time

accurate (0 = 1, order

schemes _ = 1/2),

are and

Euler

implicit

(0 = 1, (0 = 1/2,

_0 = 0), _ = 0). order work

three-point The first

backward scheme in time. out in this is first The

trapezoidal the other has two been

accurate point

whereas

are used

second for the

accurate carried

three

backward

scheme

study.

3.4 Flux The computation vective and computation of the diffusive fluxes. flux of the The terms, RHS term

Formulation of equation R defined of which (3.18) requires (3.4), a method consists for the of con-

in equation is computed

each

differently.

3.4.1 The convective fluxes

Convective are

Flux

Computation in nature which implies that the ei-

hyperbolic

29 genvalues associated with the flux Jacobians are real. Thus, there is a preferred direction of information propagation in such systems. Thus, to be consistent with the physics of the problem one needs to use the method of upwinding to evaluate the convective fluxes. One such method is Roe'sflux formulation which is a flux difference split scheme [12]. The method entails solving an approximate Riemann problem exactly at each cell face. However, Roe'smethod is first order accurate in space which is too dissipative for simulation of realistic problems. Higher order accuracy is obtained by adding corrective flux terms with limiters [2]. The limiters act as a switching mechanism to control the addition of the corrective flux terms such that the formulation is higher order in smooth regions. In regions containing shocks or contact discontinuities the formulation is first order accurate becausein such regions higher order schemesare dispersive. The limiters used in the corrective flux terms are Roe's Superebee, van Leer's and minmod limiters. The detailed development of the above formulation is presented in [13], [14], and [15]. 3.4.2 Diffusive Flux Computation The evaluation of the diffusive fluxes defined in (2.45) requires the numerical representation of derivatives needed for the stress calculations which can be noted in equations (2.47) and (2.48). The metric terms such as _x, _y,...., are also needed instead of
areas fluxes tion in the are Cartesian J_x, J_y, and so that Thus, .... , which are there the the are the projections code. direction scheme The of the face

directions in nature

computed

in the

diffusive

parabolic

is no preferred differencing

of informais the most

propagation and

for systems. easiest way

central

popular

to compute

derivatives

appearing

in the

diffusive

30 fluxes [2]. The implementation of the above approach is illustrated by considering an example stress term,
rxx _ _ee'_/x 1 2 2kx-_Ou_ rxx:

ky Ov-kz_

The appear

following in the

numerical above

formulae

are

used

to calculate

the

various

terms

that

equation: ( Jky ( ky )'

OV

( kx

)k

( Jk, 1 = )k+ +_ ( Jk + Jk+a

= )k+_ ( Jk

1 +_ .... + Jk+l )'

)k

(3.19)

OU

('_ The central

)k

+_ i

uk+l

uk

('_

)k+ (3.20)

vk+l

--

Vk , .... order

(3.20) accuracy.

difference

scheme

in equation

is of second

3.4.3 Equation Reynolds-averaged aged Navier-Stokes /_ Similarly k

or

Turbulence is the thin-layer

Modeling Navier-Stokes equation flows, when which equation, is the becomes time as: (3.21) aver-

(2.42),

which

thin-layer equation

Navier-Stokes for turbulent

_ is written

= /xz + _t the = thermal kl + k t conductivity is written as:

(3.22)

Pr

/_ (y 0.72

_ 1) and and Pr/ Pr t =

/Q (_ 0.90

+ 1) for air

/_t Pr t (y [2]. model of the

(3.23) 1)

where

Pr/ The

Baldwin

Lomax

turbulence description

[16] model

is used is given

in this

study

for mod-

the

calculation

of/x t. A detailed

in [2]. This

31 el is a two-layer algebraic eddy viscosity model in which _tt

ently the then for the inner the and inner outer and outer layers. _t t at each [2]. Since there is no clear is calculated is evaluated differbetween methods and

demarcation by both

layers, value

cell face

smaller

is chosen

3.5 The Equation (L where upper ing ation the + D (3.18) + U)X lower can be written = b block The

N-Pass in the

Scheme general form as: (3.24)

L is strictly block

triangular,

D is block defined

diagonal above can

and can

U is strictly be solved us-

triangular.

system The

of equations "N-Pass"

"N-Pass" scheme

algorithm. based on the

algorithm

be viewed [17]:

as a relax-

symmetric

Gauss-Siedel

algorithm

(L LX

+ 1 +D)X1 (D

UX 2 + + U)X

= b = b

i st pass

(3.25)

(L x2N-X D )+X_-I(D + +U)X + _UX _-2__ L

=b b

N_

pass

The

nature

of L, D, and needed to be equivalent been illustrated more

U can be exploited to be carried to one pass out

to reduce [7]. The

the modified

number

of matrix scheme

vector can above. steniterathe scope

multiplications be shown This has

two-pass

of the

"N-Pass"

scheme

described

by considering than two iterations solution, solver.

a representative are the This needed "N-Pass" is so, because

one-dimensional for the Gauss-Siedel allows each

cil in [2]. Since, tive for method a more

to yield accurate

a converged and stable

scheme with

additional

32 pass one can add two additional Gauss-Siedel iterations, which gives a better solution. In this study, only one pass has been used to obtain the solution. This schemehas also been vectorized using the diagonal plane processing developed in [13] and [10].
3.6 Boundary_ The conservative developed walls, plied. reservoir boundary cause direction boundary proach conditions. coupled ple radial [21. with the The in characteristic vector [18] form is used wall, boundary for does variable of the boundary transformed study. boundary that flows, between at the For Conditions conditions Euler the as applied with to the source nonterms at the [14] is apthe

equations viscous

in this no-slip condition internal not change

computations, developed in [2] which in

adiabatic inflow conditions condition

condition is developed called the the

specifies

total-condition-preserved This only is so bein the x-

two formulations. entrance The only the and

inflow

is assumed is normal that

to be uniform to the is made plane when

grid

which

of rotation. one arises changes in the conditions equation produced

difference fixed frame

in the ap-

conditions to the rotating The the

from subsonic with described

frame

approach

outflow source

boundary terms are sim-

subsonic simple

outflow radial

boundary

equilibrium for the swirl

in [19]. The of the

equilibrium

accounts

downstream

blades

3.6.1 Subsonic outflow condition

Subsonic

Outflow by the figure below.

is illustrated

33

+k

Figure In the come from above the situation,

3.2 Codirectional depicted by Figure domain and that the are outflow evaluated and

Outflow 3.2 four one from pieces outside. of information If the _ direction, the computational outflow then

inside

computational _ plane

boundary the

is a constant

is in the from

positive inside

characteristic are [18]:

variables

domain

V-_y

w]b=
[_x- (Q P--) + _z
V

- _yw / + J in

(3.26) F1

"1
[_y

(Q --P---') c_

U -_z + _x W]

.,b

]
Cy (O C2

(3.27) +

P--) - _z

_xW

I in

r2

L)
[_z (Q

+
_y U

-_xv]
b

=
(3.28)
U m

+_ _z (q - P-P-)

_x

Vii

+r3

34

_yV

+ zw]
b

=
(3.29) + _y v

plV l

_x u

where

Fi

= AT

I V_ I F_ j , is considered S is defined

= P- 1 M_,o to be locally in (2.46)

1 S. The

subscript

"0" denotes

that

the

matrix

P_-1

constant. one can

The write

P_-1 the

and

M-1

are off

derived as:

in [8] and

so that

elements

F 1 1" 2 ['3 /'4

= = = =

AT A'_

[-[_x

_z v w

W $"_] Q] w

--

_y

V Q]

(3.30)
_ q_z v .(2]

A Z" [_x A Z" [-_y

By constructing the metrics

_y

the and

grid

such

that be

the zero

outflow [19].

boundary Utilizing this to:

is normal result

to the in

x axis,

_z will

equations

(3.26)-(3.30),

the

characteristic

variables

simplify

(3.31)

Vb

---- Vin

--

AT

Wav

.(-2

(3.32) (3.33)

W b

---- Win

AT

Vav

_)

+u)
b

+u)
in

where phantom

Oo and points

c o are which

reference are

quantities, later

ray

and

Way section.

are

calculated The calculation

using of

explained

in this

35 pressure at the outflow boundary is carried out by using the result from the first three equations above in the radial equilibrium equation [19]:
(3.35) Or
b

A first right ing

order hand

finite side

difference

scheme as the will

is used average yield the

to evaluate of the quantity

the

derivative at the two

and points

the be-

is calculated This procedure

considered.

following

algebraic

equation:

P2-Pl r2 rl

1 \ 7_ + al 2 ]

rc , )
[
rl the

v21 +

a2 r2 ]

1
(3.36) the points along the

where

0-

P] _00 in and

subscripts

1 and

2 denote

spanwise

direction.

Letting

v2 c2 r

and

a v2 r one

gets

the

following

ex-

pression:

= Pl
P2

(r2-rl)( 2 1

Plfl
( r2-rl 2

+ gl )-f2

+ g2

(3.37)

The sure the tions

calculation is taken boundary (3.31)

proceeds to be the

from specified

the

casing

(maximum to the u can

value hub.

of r), where Once the from

the pressure the

presat equa-

back the

pressure, and

is determined, and (3.34) the

density

be calculated

respectively. boundary domain conditions is used. The a layer conditions of phantom at these points points just are outdeter-

To implement side the computational

36 mined by the following relations: _p

where point. can = 2_b Win (3.38)

_ is any Once be the

of the primitive at

primitive variables the

variables are

and

subscript the Using (3.33)

p denotes conservative the can above be written

a phantom variables definition, as:

determined, points.

calculated
Wav

phantom

Vav and

appearing
Vin 4Vp

in equations

(3.32)

and

Va v

-Win

2
4Wp

(3.39) 2

Wav

--

It should solution

be noted at the

that

Vp and time

Wp in equation step.

(3.39)

are

calculated

using

the

previous

CHAPTER IV RESULTS The aim of the work carried out in this study is to show that, for turbomachinery applications, solving the Navier-Stokes equation cast in the rotating frame has a significant advantage over solving the same in fixed coordinates. However, one has be sure that the change in formulation doesnot introduce any additional inaccuracy which will degrade the solution. The test case chosen in this study to validate the formulation is NASA's Rotor 67, which has one row of rotating blades. All the computations were performed on the SGI Challenge 10000 XL computer. The calculations have been carried out with third order spatial and second order temporal accuracy. The higher order computations and viscous computations are delayed for a few cycles.
4,1 Rotor The has The It has inlet been test first used case stage to validate rotor the of a NASA code Lewis 67 designed two-stage geometry transonic and 1.38 fan, Rotor 67, flow. rotor. The

for a single-rotating damperless, design rotor tip was pressure

in internal fan vanes.

is a low-aspect-ratio, of multiple-circular-arc number rpm). at the The

axial-flow guide designed rotor

22 blades relative m]sec flow

no inlet at the ratio

Mach (16,043 rate

tip speed is 1.63 at

of 428.9 a mass

designed [2].

for the

of 33.25

Kg/sec

37

38 The grid used for the computations is very coarse. An H-type grid for the whole computational domain was generated by TIGER, an interactive turbomachinery grid generation code developed by Shih [20]. The grid has dimensions of 55x31x25 with twenty one points on the blade axially and twenty six points in the spanwise direction. There is a gap clearance between the blade tip and the casing. Since this case is at zero angle of attack, only one blade passage needs to be simulated by taking advantage of the symmetry. Total-condition-preserved B.C. is used to apply constant uniform conditions at the upstream boundary. Axisymmetric B.C.s are used for re-entry

boundaries. No-slip B.C.s are used on the blades, hub and casing. At the exit plane, the radial equilibrium boundary conditions with the sourceterms are applied with the back pressure specified at the radial location corresponding to the casing. The Reynolds number based on the tip diameter at the designed speed and standard day conditions was about 8,000,000. The solution is considered converged when the following are true: 1. 2. The inlet mass flow rate changedless than 0.2% of the average of the maximum and minimum values in 1000 cycles. Mass flow ratio (outlet
! inlet) varied between 1 + 0.005 in 1000 cycles.

For sional peak viewed with code back

the

above

test

case

three 0.85,

conditions and 0.90

were which

run

with

the

non-dimennear can be

pressures and

of 0.75, near state stall flow

correspond Since the relative the

to choke, problem frame code

efficiency

conditions in the and rotating minimum time

respectively. frame, time

as a steady both, local

is run frame

time

stepping with

stepping. Figures

The

absolute and

can only be run

minimum

stepping.

4.1, 4.3,

4.5 show

39 a comparison of mass flow histories between the relative frame and the absolute frame codes,for the three conditions, respectively. The comparison of the behavior of the mass flow ratio is presented in Figures 4.7, 4.9, and 4.11. As can be observed from the mass flow histories, the absolute frame code requires a much longer time to yield a satisfactorily converged solution. The absolute code was run with a CFL=70 which corresponds to 2600 time steps per revolution, and with 2 Newton sub-iterations per time step. When using local time stepping the rotating frame codewas run with a CFL=50; no Newton subiterations. While using minimum time stepping the relative code was run with a CFL=10000, corresponding to 18 time steps per revolution. However, the initial 400 cycles were run with a CFL=500. Thereafter, the CFL was increased to 10000. The absolute coderequires 10400 cycles (4 revolutions) to produce a converged cycles When Thus, with can solution, to give local there the whereas a converged stepping the rotating frame when run code requires a minimum time were of 1500 stepping. needed. method CFL that be

solution was used,

using

minimum cycles time very

time

a minimum of using the

of 1500 local

is no significant frame running plots are code.

advantage This the can

stepping high

rotating when the

be attributed time

to the stepping

be used

with that

minimum back to obtain

method.

It can

observed more let

from

as the

pressure

is increased solution. by less which

progressively, The than has final in-

number

of cycles

required the

a converged varies plot

mass

flow rates can 4.2,

between

two formulations on a semi-log mass flow

+ 1%. The been and shown 4.12

advantage in Figures

be better 4.4, and

appreciated 4.6 for inlet

and

Figures

4.8,

4.10,

40 for mass flow ratio. Another observation that can be made from the plots is that the mass flow ratio settles down much faster than the inlet mass flow for all the three cases. The speedup that one observes due to the above description is huge. The absolute frame coderequires 38 hours of running time to complete one revolution, whereas the relative frame code requires 10 hours running time for 2000 cycles on an SGI Challenge 10000 XL. The CPU times required by the two formulations for the three different conditions are illustrated in the table below. Since there was no difference in the number of cycles required with the relative frame formulation when using local time stepping or minimum time stepping, the CPU times corresponding to the relative method using both approaches is the same. Table 4.1 Performance Enhancement

METHOD

TIME (hrs.) back pr. =

TIME 0.75 back (hrs.) pr. = 0.85 back

TIME (hrs.) pr. = 0.90

RELATIVE

7.5

12.5

15

ABSOLUTE

152

171

190

In order tween the relative

to ensure Mach

the

accuracy

of the

solution,

a comparison from by solving

is made the

be-

number

contours

obtained

two for-

41 mulations. The comparisons are made at two operating points, one near peak efficiency and the other near stall condition, and at 10%, 30%, and 70% span locations from the blade tip. The operating point near peak efficiency occurs at a non-dimensional back pressure of 0.85 which results in an inlet mass flow of 33.89 kg/sec, which is 97.86% of the computed choked flow, which occurs at a non-dimensional back pressure of 0.75. For the same back pressure the absolute frame codehas a mass flow rate which is 98.39% of the choked flow rate. Figures 4.13-4.15 illustrate the comparisons for this operating point and for 10%, 30%, and 70% spanwise locations respectively. The near stall condition occurs at a non-dimensionless pressure of 0.90. The relative frame codeyields an inlet mass flow rate of 32.49 kg/sec which is 93.82% of the computed choked flow rate, for this condition. The absolute frame yields an inlet mass flow rate which is 94.91% of the choked flow rate. The relative Mach number contours for this condition at the various spanwise locations are shown in Figures 4.16-4.18. It can be observed from the figures that there is a good agreement between the relative Mach number contours. Thus, it can be concluded that the formulation presented in this study does not introduce any additional errors. The discrepancies between the computed relative Mach number contour plots and the experimental plots are the same for the two formulations and are not discussed in this work. A detailed description of the flow field and the agreement and discrepancies between computed and experimental results is given in [2]. Another observation that was made during the course of this study is that the relative

42 frame codedoesproduce a converged solution for a non-dimensional back pressure of 0.95 whereas the absolute frame code exhibits a stall condition beyond a non-dimensional back pressure 0.90.

43
35.5

35.0

/ \, , 1 p I _ i_ / /. I

34.5

!
:

i
p J

i
1 _ _ ! j'

34.0 i p

i _ t 33.5 Loclll Mlnm. Ablll. _ f

lj' r"

'_,/

.... --

- --

33.0 0 2600 5200 No. of 7800 Time Steps 10400 33000

15600

Figure

4.1 Mass

Flow

History

(back

pr.

= 0.75)

35.5

.......

........

........

........

........

35.0

34.5

i i'
34.0 ,j

33.5 /

/f

"_, ,

Locl_l Mlnm. Abel.

33.0

.......

-i'o

.....
No.

loo
of

.....
Time

-_ _5'oo ....
Steps

-_ _,',hoo-

" :I"0_,ooo

Figure

4.2 Mass

Flow

History

(log-linear,

back

pr. = 0.75)

44

35.5

35.0

34.5 i 34.0

33.5

:i

,'

33.0

i i' i

32.5 I' _E 32.0 i 31 .5 I 31 .0 I 30.5 i 30.0 I I

i I

.... --

--

LOGS! MInm. Ab81.

29.5 0 2600 5200 No. of 7800 Time Steps 10400 33000 35800

Figure

4.3 Mass

Flow

History

(back

pr. = 0.85)

35.5

35.0

34.5

34.0

33.5

33.0

32.5

32.0

31

.5

31

.0

30.5

Local Mlnm. Abel.

30.0

29.5 10 NO. 1 O0 of Time 1000 Steps 10000 100000

Figure

4.4 Mass

Flow History

(log-linear,

back

pr. = 0.85)

45

35.5

34.5

'

[.

'
33.5 x,

/,J

.\

32.5

I 31 .5

!
30.5 I .... -- --

Local Mlnm. Abel.

29.5 f

i
28.5

!
/

27.5

I
J

26.5

2600

5200 NO. of

78'00 Time

" Steps

10400

"

13()00

15600

Figure

4.5

Mass

Flow

History

(back

pr.

= 0.90)

35.5 ," ', 34.5

' L

33.5

32.5 .j 31 .5

30.5 I t 29.5

i 28.5 i Local Mlnm. Abel.

27.5

t J

26.5

1o
No.

1 oo
of Time

1 ooo
Steps

10000"

1"ooooo

S_

Figure

4.6 Mass

Flow

History

(log-linear,

back

pr. = 0.90)

46
1.1

1.O

O.9

O.8

O.7

O.8
.... ---

Local Mlnm. Absl.

O.5

2800

5200 No. of

7800 Time Steps

10400

13000

15600

Figure

4.7 Mass

Flow

Ratio

History

(back

pr. = 0.75)

1.1

1.O

O.9

0.8

0.7

0.6

Local Mlnm. Abel.

0.5 10 No. 1 O0 of Time 1000 Steps 10000 100000

Figure

4.8 Mass

Flow

Ratio

History

(log-linear,

back

pr. = 0.75)

47
1.2

1.1 ! 1.0 '1

!
!\ ,!

0.6 0.7

0.6 Local Mlnm. Abel.

0.5

0.4 0

2600

5200 NO. of

7800 Time Steps

10400

13000

15600

Figure

4.9 Mass

Flow

Ratio

History

(back

pr. = 0.85)

1.2

1.1

,," ',.
1.0

0.9

/-/
,' "

U
/" _ / r'

0.8

ir

0.7

0.6
/ _ ' t' j --Local

0.5

0.4 1

...............

10 No.

4 oo
of Time

1 ooo
Steps

1 oooo

1 ooooo

Figure

4.10

Mass

Flow

Ratio

History

(log-linear,

back

pr.

= 0.85)

48
1.3

i ._. 1.2 s r 1.1 j

1.0 . \ /.

o.g'

i
I

o 0.6

0.7

:=E

0.6

0.5 Local MInm. Absl. 0.4

0.3 0 2600 5200 NO. of 7800 Time Steps 10400

13000

= 15600

Figure

4.11

Mass

Flow

Ratio

History

(back

pr. = 0.90)

1.3

1.2

1.1

1.0

0.9

0.6

0.7

0.6

0.5 Local MInm. Abel 0.4

0.3

................. 10
No.

-i oo
of

............... 1000
Time Steps

4o_oo"

" " :_00ooo

Figure

4.12

Mass

Flow

Ratio

History

(log-linear,

back

pr.

= 0.90)

49

\
1.05 1.26

a) Absolute

1 40

b) Relative Figure 4.13 Relative Mach No. Contours at 10% Span for Near Peak Efficiency

5O

I.i0

1.25

a) Absolute

1.02

C
1.10

\ 1.22 Figure 4.14 Relative Mach b) Relative at 30% Span for Near Peak Efficiency

No. Contours

51

\
0.66 0.77

\
1.07

0.93

a) Absolute

\
0.65 0.77

0.91

1.08

k,__"

b) Relative

Figure

4.15

Relative

Mach

No. Contours

at 70%

Span

for Near

Peak

Efficiency

52

a) Absolute

1,43

?
\
1.23 0.97

b) Relative Figure 4.16 Relative Mach No. Contours at 10% Span for Near Stall

53

0.82

1.30

0.90

1,12

a) Absolute

0.80

1.30

0.89

1.17

b) Relative Figure 4.17 Relative Mach No. Contours at 30% Span for Near Stall

54

\
0.62 0.72

\
1.07

0,87

a) Absolute

\
0.61 0_70 0.87 1.05

b) Relative Figure 4.18 Relative Mach No. Contours at 70% Span for Near Stall

CHAPTER V SUMMARY AND CONCLUSIONS The results presented in Chapter IV amply demonstrated that solving the Navier-Stokes equations in rotating coordinates is a worthwhile exercise and provides a huge advantage in terms of savings in computer time. The savings are due to two main advantages of the rotating frame approach over the inertial frame approach. Since the flow can be considered to be steady in the rotating frame, no Newton sub-iterations are required so that less computing time is required for each time step. This is augmented by the improved stability of the rotating frame codewhich allows it to handle a far larger CFL than the absolute frame code. This reduces the total number of cycles that are needed to obtain a converged solution. Also the need to move the grid is obviated due to which the metric terms do not need recomputing after eachtime step. All the above factors contribute to the huge savings in computing time. The speedup in CPU time obtained for the test casein this study was a factor of 16. Moreover, the approach used in this study did not utilize multigrid method or Jacobian freezing. These two factors can contribute to further savings in computing time. The enhancement in the performance of the codeis achieved by making a few minor modifications to the codeto incorporate the rotating frame formulation. To summarize, they are the following: 1. 2. Freezing of the grid motion. Calculating the term
kt, analytically as described in Chapter II.

55

56 3. 4. Addition of the source term implicitly in the flux balance. Addition of the source term in the subsonic outflow CVBC which is coupled with the radial equilibrium equation. The results presented for the test caseshow that the solution obtained by the relative frame formulation is in very good agreement with the results obtained with the relative frame formulation. The significant improvement in stability is useful for problems with different time scales in turbomachinery computation. Since the rotating frame formulation freezes the frequency associated with the grid rotation, the time scalesassociated with other frequency (e.g.blade fluttering ) can be used to investigate the unsteady behavior related to that particular time scale.Thus, it can be concluded that this approach is the right direction to take for simulation of flows through turbomachinery, which can be treated as steady state problems in the rotating frame. Future work will entail implementing the relative frame formulation for a multi-blade row configuration to simulate the flow through a stage. However, this would involve transformation of the dependant variable vector between the fixed and rotating frames before exchanging information between blade rows. Addition of the multigrid method and parallelization of the code are two other areas of future work.

REFERENCES [1] [21

Adamcyzk, Turbomachinery," Chen, Solutions Mississippi J.P., J.J., "Model ASME "Unsteady State J.J., Journal Equation Paper for Simulating Flows November Thin-Layer Flow," December T.A., Vol. and Ph.D. 1991. Barnett, a Multistage No. 3, July M., "SimulaTurbine," 1990. pp. in Multistage 1984. Navier-Stokes Dissertation,

No. 85-GT-226,

Three-Dimensional in Transonic Mississippi, M.L., Viscous Beach, Flow

for Turbomachinery University, Celestina,

[3]

Adamczyk, tion ASME 370-376.

of Three-Dimensional

Within 112,

of Turbomachinery,

[4]

Agarwal, licopter MDRL

R.K., Rotor Report R.V.,

and Deese, in Hover," June 86-10,

J.E., 1986.

"Euler

Calculations Douglas

for Flowfield Research

of a He-

McDonnell

Laboratories,

[51

Chima, al Viscous 798-806.

and

Yokota, Flows,"

J.W.,

"Numerical Journal,

Analysis Vol.

of Three-Dimension5, May 1990. pp.

Internal

AIAA

28, No.

[61

Warsi, proaches,"

Z.U.A., CRC D.L.,

"Fluid Press,

Dynamics: Inc., Boca

Theoretical Raton, 1993.

and

Computational

Ap-

[71

Whitfield,

"Newton-Relaxation

Schemes

for Nonlinear

Hyperbolic

Systems," Engineering and EIRS-ASE-90-3, Mississippi

Industrial Research Station State University, Mississippi,

Report, MSSUOctober 1990. Flux Vector Mississippi

[81

Janus, Euler State

J.M., Solver

"The with

Development of a Three-Dimensional Split Dynamic Grid Applications," M. S. Thesis, Mississippi, "Implicit August Upwind 1984. pp. 14-27. Scheme and Mississippi Industrial State

University, D.L., Report,

[91

Whitfield, Dimensional Station Mississippi,

Finite Engineering

Volume

for the

Three-

Euler September

Equations," 1985.

Research University,

MSSU-EIRS-ASE-85-1,

[10]

Belk, Blocked sippi,

D.M., August

"Three-Dimensional Ph.D. 1986. Dissertation,

Euler

Equations

Solutions State

on Dynamic Missis-

Grids,"

Mississippi

University,

57

58 [11] Whitfield, D.L., and Taylor, L.K., "Discretized Newton-Relaxation Solution of High Resolution Flux-Difference Split Schemes,"AIAA Paper No. 91-1539, June 1991. Roe,P.L., "Approximate Riemann Solvers, Parameter Vectors, and Difference Schemes," Journal of Physics, Vol. 43, 1981. pp. 357-372. Janus, J.M., "Advanced 3-D CFD Algorithm for Turbomachinery," Ph. D. Dissertation, Mississippi State University, Mississippi, May 1989. Simpson, L.B., "Unsteady Three-Dimensional Thin-Layer Navier-Stokes Solutions on Dynamic Blocked Grids," Ph.D. Dissertation, Mississippi State University, Mississippi, December 1988. Whitfield, D.L., Janus, J.M., and Simpson, L.B., "Implicit Finite Volume High Resolution Wave-Split Schemefor Solving the Unsteady Three-Dimensional Euler and Navier-Stokes Equations on Stationary or Dynamic Grids," Engineering and Industrial Research Station Report, MSSUEIRS-ASE-88-2, Mississippi State University, Mississippi, February, 1988. Baldwin B.S., and Lomax, H., "Thin Layer Approximation and Algebraic Model for Separated Turbulent Flows," AIAA-78-257, January 1978. Sreenivas, K., "High Resolution Numerical Simulation of the Linearized Euler Equations in Conservation Law Form," M. S. Thesis, Mississippi State University, Mississippi, July 1993. pp. 30-31. Kisielewski, K.M., "A Numerical Investigation of Rain Effects on Lift Using a Three-Dimensional Split Flux Vector Form of the Euler Equations," M. S. Thesis, Mississippi State University, Mississippi, May 1985. pp. 29-36. Whitfield, D.L., Swafford, T. W., Janus, J.M., Mulac, R.A., and Belk, D. M., "Three-Dimensional Unsteady Euler Solutions for Propfans and Counter-Rotating Propfans in Transonic Flow," AIAA-87-1197, June 1987. Shih, M.H., "TIGER: Turbomachinery Interactive Grid genERation," M.S. Thesis, Mississippi State University, Mississippi, December 1989.

[12] [13] [14]

[15]

[16] [17]

[18]

[19]

[20]

APPENDIX DERIVATION OF THE VELOCITY

A MOVING COORDINATES

OF THE

59

6O It was of the deavor velocity in this stated of the in Chapter moving II that frame with the the vector w can be proven to the statement. exercise. The stationary Only figure to be negative frame. the The encase

respect above

exercise in this study

is to prove is treated the

specific

encountered the situation

in this plane Z

below

describes

by considering

of rotation.

Y
f

Figure In the ary inertial with paper; Figure frame A. 1 above, whereas

A. 1 Plane Y, Z represent

of Rotation the axes coordinate of the the make point axes moving x axis, of the frame which stationwhich points is out the

y, z represent angular time velocity t the

the

rotating of the

a constant so that after axes.

t2 about axes a fixed coordinates moving following

moving

an angle in space inertial

of t2t with denoted frame,

respective dot in the

stationary above figure,

If one have

considers fixed in the by the

by the viz.,

it will

in the viz.,

(X, Y, Z). However, over time x y z and =X = Ycos(t2t) = Z cos(t2t) can

it's coordinates be expressed

frame, relations.

(x, y, z) will

change

+ -

Zsin(sgt) Ysin(_2t)

(A.1)

Therefore,

the

partial

time

derivatives

of the

moving

coordinates

are

given

by:

61
OX

Ot _J Ot
OZ

(A.2)

= f2 [Zcos(Qt)

Ysin(f2t)]

= Qz

(A.3)

dt The vector

t2 [Ycos(_t)

Zsin(f2t)]

f2y

(A.4)

w is defined

in equation

(2.3)

as:

w _ According
X 1 = X,

Oxi Ot to

a -i the coordinate system let i, described j, and in Figure the

(A.5)

A.1, Car-

X 2 = y,

x 3 = z and

for simplicity

k represent

tesian

base _ -

vectors Ox -_-fi

of the Oy. + _-fj

moving Oz _-fk

frame.

Thus,

f2zj

Qyk

(A.6)

The

velocity

of the

moving ijk f200

xyz

frame

with

respect

to the

inertial

frame

is given

by:

t2zj

f2yk

(A.7)

It can

be observed w= -f2X

from r

equations

(A.6)

and

(A.7)

that: (A.8)

APPENDIX CURVILINEAR COORDINATE STOKES

B OF THE NAVIER-

TRANSFORMATION EQUATIONS

62

63

The mensional

governing variables

equations in rotating

in conservation Cartesian

law

vector

form by:

using

nondi-

coordinates

is given

Oq

o-7 + o-_ + _
the vectors q,

Ot

Og

+ oz
f, g, h,

Oh

fv,

Ofv

ox
gV,

-_

OgV

Ohv

oz

(B.1)

where

h v, and steady,

s are curvilinear

all

defined coordinate

in equations system is

(2.17)-(2.20). introduced

A general in the = _(x,y,z)

nonorthogonal,

(x, y, z) space

as follows:

rl
z=t

= =

rl(x,y,z) _(x,y,z)

(B.2)

It should nates are

be noted transformed

that

through to rotating place

equations curvilinear two the

(B.2)

the

rotating

Cartesian In other within form

coordiwords, the

coordinates. systems in matrix

transformation frame. Using

takes chain

between writing "0" Ov 0 0

coordinate result

the

rotating

rule

and

yields:

"0" Ot 0 Ox 0

'1 0

000

(B.3)

Oy
0 Oz

0 0

Cy T]y _y

_z _z _z

07 0

The

inverse

transformation

in matrix

form

is:

64
000" x y z xv 27 z_ x_ y_ z_ "0" at 0 Ox 0

a-_ 0 -0

"1 0 =

a_ o_

(B.4)

o
0

oy
0 Oz

0 _

This

implies

that: "1 0 0 0 0 0 0"

-1

"10

0"

0 _x _x _x 0 _y _y _y

x y z (B.5) x_ yn z_ x_ y_ z_

Thus,

J, which 1 0
J

is the 0 0

Jacobian 0

of the

inverse

transformation,

can

be written

as:

x$ y$ z_ x_ y_ z$ = 0 0 = x_(y_zx, y_ z_ x y z x_ y_ z_ z_y) y_(x_zz_x_) + z_(x_yy_x) x_ 27 z_ (B.6)

The tions:

metric

terms

in equation

(B.3)

can

now

be expressed

by the

following

rela-

_x
_y

=
=

l(y_z_(ZtIX_-

z_y)
Xr/Z_)

_/x
r]y

=
---

l(z_y-yCz)
_(X_-Z_Z_X_)

_x
Cy

= l(ycz--_(Z_,Xt_-

z_y)
X_Z_/) (B.7)

!/(x_y_

65

The governing equations in curvilinear coordinates, therefore are:

0-7 + _x (f-if) + rlxo_(f-f _) + _x (f-if) (B.8)

O(g

Multiplying

throughout

by J yields

the

following:

(B.9)

= Js

Using

the

identity: (B.10)

J_x_
the equation (B.9) becomes:

-_v(Jq) + _[J(_xf

q-_

[J(_xf

+ _yg (f-

+ _zh)] ff)[_(J_x)

[J(t/xf

+ tlyg

+ rlzh)] _(J_x)]

+ _yg+_zh)]+ "_ (jtly) 0

+ O-_(Jrlx)+

_ (g _ gV)[__(j_y) (h hV)[-_-_ (J,z)

+ _(j_y)] (B.11)

+ O_-_(J_lz) + _(J_z)] 0

= 0__ [j(_j_ + _yg_ + _zh_)] + -_ [j(_j_ + _yg_ + _zh_)] + _ [j(_j_ + _ygV+ _zhV)l+ Js
Since the curvilinear transformation is steady:

66
OJ 0v Using write: the 0 (B.12)

expressions

for the

metric

terms

defined

in equation

(B.7)

one

can

= -_(xny)_

-_-_(xCy,7)+
c_X O2Y

_(xCy$) Oy 02X

_(x_y)

+ _(x_y_) Oy 02X

_(x_y$) (B.13)

Oy 02X

c_X 02Y

OX O2Y

00_07 + o70_0_
Oy O2x Ox 02Y

o70_0
+

o_0_07 + ot 070 + 0 070t

+

Oy 02x

Ox O2Y

Oy 02x

OX

02Y

aCaTa_
=0

ata7a_

07a0_

a_aCa7

ataCa7

07aCa_

Similarly

O"_-(J_x) at
0 -_(Jty)

O-_(JTx)

+
+

-_(J_x)
-_(J_y)

=0 (B.14)
= 0

0 + -_ (JTy)

Thus,

the oQ

governing OF

equations OG OH

in curvilinear _ OF v

coordinates OGv + OH v

are: + S

(B.15)

where

Q =Jq

=J

(B.16)

67
m

K=

J(

kx.f + kyg QK' _uK'

+ kzh

+ kxp +kyp + kzp + pK

where

K'

= kxu -

kyv

kzw

+ k t (B.17)

= j

9vK' _wK' etK'

kt =

kyQz + kyv

+ kz_y + kzw

K = kxu

_v

= j(

kxf v + kyg v + kzh v ) "0]

=J

Tky Tkzl
.QkJ
0 0

(B.18)

S =Js

=J

Ow_2 QvQ 0

(B.19)

In K=

the U,

above V,

equations W; _ stress

(B.17)

and

(B.18), for

K = F, r/, _

G,

H;

K'

= U', The

V',

W';

= F v, G v, HV; and heat with flux

k = _,

respectively. (B.18) can

trans-

formed in terms the terms stress chain

viscous of the rule

terms

in equation to curvilinear in the

be expressed by applying and heat flux

derivatives to the

respect appearing

coordinates viscous The stress

derivatives flux terms vectors are

of the and

viscous flux

in equation defined in detail

(2.19).

transformed (2.39)-(2.41).

viscous

heat

in equations

APPENDIX C CYLINDRICAL COORDINATE FORMULATION

68

69 The Navier-Stokes equations in cylindrical coordinates, in an absolute frame with absolute velocity components using non-dimensional variables can be written as:

oq
where

o-7 + _

of

+-_7

08

+ oO -

oK

of

Oz

+ o_
-_-

o_

oO

(C.1)

q=

rQ rQUz rQUr
"20U _ re t

(C.2)

r_Uz

rQUr rQUzUr , g----

'

QU 0

r(Qu 2 + p) rOU rU z
f

OUzU 0
M

.(eU2r+ p)
r2oUoUr

h=

OUrU o r(ou 2 + p)

(C.3)

r2OUoUz roHuz

rQHu

QHu e

[rTOz]

[rTOr]

[TzO1

(C.4)
F ---- /rr_r'Zoz/ ' g-V ---- /rr_;r I' h-V -['/'rO lrvO0

[ rqz j

[ rqr J

qo

0 0 k=

(C.5) 2u_ + 0 - _ I p
0
2

7O

In the solute

above velocity

equations

Uz,

Ur,

and

u 0 are

the et

axial, P

radial,

and

tangential

ab-

components

respectively,

1 2) + -_O(Uz 2 + u 2 + Uo and

et + p

"

Also,

20Uz Oz 2 OUr Or 2/_ 3 Re "20Uo r O0

OUr

IOUo

Ur

Or 10Uo r O0 OUz Oz

r O0 OUz Oz
OUr

r Ur r

TOO--

Or

Ur + 2--y(C.6)

[OUz OUr]
_zr -_ Tr-z = _ VrO = "gOr

Re [ Or
_ [I_OUr

Oz J
OUo

Re [ r O0

Or

. [ouo lOU 1
VOz = VzO = R'e[ Oz + r OOJ /u qz = VzzUz +
7JzrUr -4- _'zOUO + _ee

aT az aT Or (C.7)

(y-

1)Pr ju

qr

VrzUz

VrrUr

VrOU 0 +

_ee

(_, -

1)Pr

qo = %zUz

VorUr

vOOUo

R_e

(?, - /_1)Pr

]lOT O0 ?"

For (rotating) Oab s where into

flows frame

in a rotating by the _t

frame,

equation [3]:

(C.1)

can

be cast

in the

relative

transformation

---- Ore 1 -

(C.8) speed (negative with 0). Introducing equation (C.8)

Q is the equation

rotational (C.1) yields:

71
Oq Of Og O(h+ ..Qq) O/W 0_' + oh-' + _ (C.9)

o-7 + az +-_7
equation velocity the

o0

az

0--7-

aO

The ever, lute

above the

is in cylindrical components only difference If one (z, r,

coordinates (C.9) that the

but are

in the the

rotating

frame. of the now relative Y,

Howabso-

in equation being

components are in the (X,

velocity, relative the

components

expressed frame, Z) coordi-

in the from nates

frame.

introduces 0) coordinates

a transformation to the Cartesian

cylindrical is of the
z

which
X -_

form:

Y Z

= rcos0 = rsin0

(C.10)

equation

(2.16)

can

be derived rule and

by repeating writing 0' Ot 0 the result

the

procedure in matrix

discussed form yields:

in Appen-

dix A. Using "0" Ot 0 Oz 0 Or 0 oO

.

chain

"1000 0100
0 0 YrZr

oX
0 OY 0 OZ

(C.11)

YoZo

The

inverse "0 Ot 0 0 aY 0 O2

transformation

in matrix "0" Ot 0 Oz 0 Or 0 O0
.

form

is:

'1 0 0 0

0 100 0

(C.12)

ry

rZ z

00yO

From

equation

(C. 10) the

following

results

can

be derived:

72
Yr ry = cos 0, Y0 rZ = = r sin 0, Oy
Zr =

sin 0, Sinr0,

Z 0 = r cos 0 (C.13) Oz _ COSr0

= cos0,

sin0,

Thus,

the

Jacobian

of the

inverse 1 -- -_ hold

transformation

is given

by: (C.14)

J Also, the

= ryO z following in the

-'-- U

rzO Y relations

between systems

the

components

of the

absolute

veloc-

ity vector
UX

two
-_ Uz

coordinate

Uy uz

= UrCOSO = UrsinO

u6sin0 uecos0

(C.15)

= w

Although, the

in the

present of the

context, absolute

the velocity,

above

relations

give

the

relation that they

between are valid the fol-

components components rule

it must two

be mentioned coordinate derived

for the chain lowing

of any (C.9) equations: Og -_

vector and

in the using the

systems. above,

Applying yields the

to equation of the Of -_

relations

form Oq 0-_ +

Oh OZ

Of v OX

Og v + --OY

Oh v OZ

(C.16)

where

(C.17)

73
Qu Qu 2 -I- p
f

Qv Ovu
g

QI2z Out2z OvI2z

Ow + O_2y Owu Owv

?W 2 + w(e t +

+ Out2y + ovi2y
p p) + +

Ouv Ouw

Qv 2 + p Ovw
v(e t +

(C.18)

p)

Owt2z
--

Owi2y efl2y

u(e t + p)

et.Qz

I l
Txx Vxy Txz

rxx

Vrxy

Wrxz

+ qx

0 ry_
gU __

ryy
Tyz

(C.19) WVyz + qy

Vryy

Tzx
h v

Tzy Tzz 4I)'t'zy

0 1
4WVzz 4-qz

0 0
S

pw_ Ova2 0

(C.20)

It can Cartesian equation

be observed coordinates (2.16).

from

the

above via

equations the

(C.17)-(C.20)

that

the are

equations identical

in to

derived

cylindrical

coordinates