AIAA JOURNAL

VOL.

10,NO.2

1590

TINSLEY ODEN* CARTER WELLFORD JR.-J

ANDL.

J.

Analysis of Flow of Viscous Fluids by the Finite-Element Method

University of Alabama, Huntsville, Ala.

has

are

we

in

of

are

of

toa

the

but

ina

We

(or

and

isnot

are

The

ofthe

the

the

is

of

in

the

as

in

of

of

and

the

We

The

of

The

Eof

of

(i

andthe

ofa

bythe

ear

of

and

of

on

PinRare

ofthe

ofthe

of

the

we

the

To fix ideas, consider the motion of a continuous medium

through some closed region R of three-dimensional euclidean
space. We establish in -R a fixed inertial frame of reference defined
by orthonormal basis vectors it = 1,2,3). spatial coordinates place denoted xt components
velocity of the medium at P at time t are denoted v^x^, x 2 , x3, t) =
vt(x, t). The density and the absolute temperature at place P at
time t are denoted p(x, t) and 6(x, t\ respectively. If T0 denotes a
uniform temperature at some reference time t0, we may use as an
alternate temperature measure the temperature change T(x, t) =
0(x,t)-TQ.
To construct a finite element model of the fluid, we replace R
by domain ^consisting finite number subdomains re,
so that R R = uf = t re. The subregions re are the finite
elements; they are generally of simple geometric shapes and are
designed so as to represent a good approximation of R when
appropriately connected together. geometry each finite
element re is characterized by a finite number Ne of nodal places
(the number may vary from element to element) and the nodes of
typical element defined spatial coordinates xfj e) ;
N = 1,2,..., JV_e; i = 1,2, 3; e = 1,2, ...,. The global finite
element model R is obtained by connecting the E discrete elements
at appropriate nodal points by means of simple incidence
ofa

the

tohe

tobe

of

et

and

The

we

and

to

is

and

its

so

the

on

by

in

of

and

Inthe

Received December 1,1971; revision received May 24,1972. Support

of this work by the U.S. Air Force Office of Scientific Research under
Contract F44620-69-C-0124 is gratefully acknowledged.
Index categories: Viscous Nonboundary-Layer Flows; Boundary
Layers and Convective Heat Transfer-Laminar.
* Professor and Chairman, Department of Engineering Mechanics.
Member AIAA.
t Graduate Student, and Engineer, Teledyne Brown Engineering.

Finite Element Models of Fluid Flow

ofthe

and

the

of

at

ofthe

ofthe

by

is

in

has

ofthe

in

ofthe

are

inthe

are

by

is

tohe

is

of

HE

present paper concerned with application

concept finite elements formulation solution
of a wide range of problems in fluid dynamics. The method is
sufficiently general to treat a variety of unsteady and nonlinear
flow phenomena in irregular domains. An intrinsic feature of
finite-element approximations that mathematical model
generated patching together number purely "local"
approximations of the phenomena under consideration. This
aspect of the method effectively frees the analyst from traditional
difficulties associated with irregular geometries, multi-connected
domains, and mixed boundary conditions. Moreover, applications
firmly rooted physics problem hand
preliminary studies indicate that, for a given order of accuracy,
the resulting equations better conditioned than those obtained
by, say, finite difference approximations of the governing differential equations.1
Certain underlying ideas finite element method
were discussed 1943 Courant.2 However, formal
presentation method generally attributed 1956
paper of Turner et al.3 While the method has found wide application in solid and structural mechanics,4 its application to flow
problems come only rather recent times. Early uses
method were always associated with variational statements
problem under consideration, that natural that steady,
potential flow problems were first solved using finite
elements. We mention, in this regard, the works of Zienkiewicz,
Mayer, and Cheung5 on seepage through porous media and
Martin6 potential flow problems. Finite element models
unsteady compressible and incompressible flow problems were
obtained Oden.7"10 Applications finite element methods
to a number of important problems in fluid mechanics have been
reported recent years; among them, mention work
Thompson, Mack, and Lin11 on steady incompressible flow and
Tong,12 Fujino,13 Argyris al.,14~16 Reddi,17 Baker,18
Herting, Joseph, Kuusinen, MacNeal19 various special
incompressible flow problems. recent book Zienkiewicz4
can be consulted for additional references.
present investigation, extend finite element
method to general three-dimensional problems of heat conduction flow compressible incompressible fluids, wherein
no restriction is placed on the constitution or equation of state

of fluid under consideration. Effectively, develop finite

element analogues of the equations of continuity, linear momentum, energy arbitrary fluids. models obtained from
local approximations of the density, velocity, and temperature
fields in each element and represent generalizations of those
proposed earlier.7'9
In addition, we treat the problem of fluids characterized by
equations state which thermodynamic pressure
given explicitly function density, temperature,
velocity gradients. There we develop mixed finite-element models
by approximating locally mean stress thermodynamic
pressure) each element obtain general model which
equation state satisfied average sense over each
element. then consider special important case
viscous incompressible fluids, with emphasis on isotropic Newtonian fluids with constant viscosities. There we address ourselves
certain problems connected with imposing continuity
equation (incompressibility condition) in the discrete model and
to special boundary conditions. A notion of solenoidal finite
element fields is introduced. We then describe computational
schemes for the solutions of the equations governing the model
for uniform steady flow, nonuniform steady flow, and unsteady
flow of viscous fluids. Numerical results obtained from applications number representative example problems
presented.

the

Introduction

the

and

Itis

are

of

General finite-element models compressible incompressible fluid flow derived. These involve local
approximations of the velocity field, the density, and the temperature for compressible fluids and the velocity,
temperature, and pressure for incompressible fluids. Theories of local solenoidal approximations and mixed finiteelement models for compressible flow are derived. A number of computational schemes are developed for the
numerical solution of both transient and steady nonuniform flow problems involving incompressible fluids. Numerical
results obtained from several test problems given. shown that finite element method great potential
for use in flow problems, and represents a powerful new tool for the analysis of viscous flows.

DECEMBER 1972

FLOW OF THE VISCOUS FLUIDS BY THE FINITE ELEMENT METHOD

ina

is

Jv

r q/ijdA

JA

the

DK/Dt =

and introducing Eq. (5) and (la), we obtain

(D/Dt)K = LcMNLpMvLk +
where
C

(9)
Jv

MNL

Ninthe

as

at

canbe

forthe

the

DUJDt

(12)

Mechanics of a Finite Element

where Qe = Ve(qkk + ph)dv is the heat of the element. For an

element of fluid of volume ve and surface area Ae, the power of
external forces
is

[L

Kinematics

where

the

+ CRNkpR

(14)

Ae

of

and

inthe

asthe

the

then generalized forces develop same amount mechanical power external forces continuum element,
the

&e PNkVk
=

the

e.

Finally, introducing Eqs. (9), (12), and Qe into (7), we obtain an

energy equation typical finite element Then, making
fora

of

the

Ifwe

dueto

inthe

at

andthe

and

The

PNk =

isthe

local vorticity field co, co, = ^Nk^ where s

permutation symbol. Local approximations various other
kinematical quantities can be calculated in a similar manner.
is

(13)

first second quantities parenthesis represent

force at node N in the /cth direction due to the surface stress
distribution force node /cth direction
body force, respectively. define generalized force
pNk according to

the

a, = DvJDt = dvjdt + vimvm = ^ 4- $ *,&*$*% (5)

where v? dv*(t)/dt, commas denote partial differentiation
with respect to the spatial coordinates (i.e. \I/N m = d\l/N(\)/dxm\
the repeated indices are, again, summed over their admissible
ranges (i,m = 1,2,3; M, N = 1,2, . . . ,JVJ, and the element
identification label (e) has been dropped for simplicity. Likewise,
the models of the rate-of-deformation tensor dtj and the spin
tensor w/7 are given by

andthe

vNk

ofthe

al

the

byEq.

the

With local velocity field given (la), relevant

kinematical quantities associated with motion element
are determined by the nodal velocities \N(t). For example, the
components of local acceleration are

CRNkpR

Ninthe

The

also obey Eq. (3).

(10)

and M, N, L, F = 1,..., Ne. The first quantity in parenthesis

represents local inertial force node /cth direction,
and the second term in the parenthesis represents the convective
inertial force at node N in the /cth direction.
Since the local form of the energy balance is
p(Ds/Dt) = ps + pvfj = Tikvki + qkk + ph
(11)
wherein Tik is the Cauchy stress tensor, the material derivative
of internal energy element written

(4)

P(e) =

interpolation functions ju

(8)

areth

tohe

isthe

InEq.(8)
Nin

by

or

Notice that Eq. (1) implies that different forms of the interpolation
functions may be used to approximate different local fields over
the same finite element. In certain cases (some of which are to be
discussed later), this may require that certain of the functions
vanish certain nodal points that indices each member
of Eq. (1) may have different ranges. Note also that "higher order"
local representations obtained also specifying values
of derivatives of v, p, and T at the nodes.
We must also remark that, in the case of incompressible fluids,
our formulation requires that, instead of the density p(e)(x, t) we
approximate the pressure field p(x, t) over R. Thus, if p(e)(x, t) is
the restriction of p(x, t) to re, we assume

canbe

Q= \ Phdv +

e internal energy density, Fj components

of the body force vector per unit mass, Sj are components of the
surface tractions, h is the heat per unit mass supplied from
internal sources, q} are the components of heat flux, and rc, is the
normal boundary surface. Noting that material derivative of the kinetic energy for element e is

the

(3)

at

Jv

JA

Jv

f
psdv

at

are

and

etc. functions ^(x), <p(ff(\\ ffi(x) are local interpolation

functions defined so as to have the properties
The

Q = f pFjVjdv + f SjVjdA

areofth

Nof

at

the

and

and

T(e) = $\x)T}(t)
(Ic)
Here henceforth repeated nodal indices summed from
1 to Ne; v(^(t), p"e}(t\ and T"e)(t) are the values of the velocity,
density, relative temperature node element re
time r; i.e.

in

is

f pVjVjdv
Jv

and

e)t

P(e)

ofthe

The

of

of

on

the

is

as

of

ofa

the

by

The

= ^1

U=

the

equations governing motion typical finite element

obtained constructing Galerkin integral Cauchy's
first law of motion over the element and by using the velocity
interpolation functions \l/N(x) weight functions this integral.
If this approach taken, linear momentum balanced
average sense over element. arbitrariness choice
of \l/(x) as weight functions, however, is removed if an alternate
but equivalent approach based energy balances employed.9
The behavior of the medium must be consistent with the
principle conservation energy:
D/Dt(K + U) = Q + Q
(7)
where K is the kinetic energy, U is the internal energy, Q is the
mechanical power, and Q is the heat:

t)

as

Momentum Equations for a Finite Element

canbe

are

inthe

by

ofits

is

t),and

are

andlet

t),

atxsR

Let

and

asour

The

and

and

mappings which merely identify the desired correspondence

between local global nodal labels. These mappings
described elsewhere20 and need not be discussed here. The
important feature of the model is its local character; that is, the
behavior of the medium can be idealized locally in a typical
element independent of the behavior in other elements in the
model independent ultimate location model.
final global model then obtained routinely connecting
elements together through mappings which depend only upon
the topology of the model.
Following guidelines provided by the notion of determinism,22
we shall take fundamental dependent variables, velocity, mass density, temperature change obvious measures
of these primitive characteristics. Therefore, consider a typical
finite element re with Ne nodes isolated from the global
model R. v(x, p(x, T(x, denote global approximations of the velocity, density, and temperature change to be
determined time r, v(e), p(e), T(e) denote their
restrictions to element re. Then finite-element approximations of
these restrictions constructed which form
ejW
(la)

1591

AIAA JOURNAL

JR.

ODEN WELLFORD

Constitutive Equations, Equations of State,

and Mixed Models

in

ofan

(or

and

orim-

anditmybe

ofthe

ofthe

as

byan

is

nis

=0

p,

to)

(ordnt

is

we

ofnt

we

we

the

the

bythe

we

Eq.(1)and

ofthe

the

As

the

ina

its

for

ofthe

and

inthe

(22)

of

Jve

Jv

[orEq.

and

<PRWM<IVPR +

Jve

and

Eq.

the

Equation (28) represents finite-element analogue

equation state, (26), insure that satisfied
weighted average sense over each finite element. Introduction
of (27) into (25) incorporation result into Eqs. (15),
(18), (22) (23)] (along with constitutive equations
qk, e, and /? as functions of p, T, and vt j) yields a complete system
of equations nodal values velocity, density, temperature,
and thermodynamic pressure.
Eq.

ofthe

Introduction of Eq. (20) into Eq. (21) gives the general finite
element analogue energy equation:

(28)

ofthe

-i

(21)

dv = Q

) dv = 0

of

tohe

be

in

we

Eq.

Eq.

By introducing (la) into local form first law,

(11), obtain residual
re = <PRPR + (PRpR^N^Sti - Tik\ljNivNk - qkk - q>RpRh
(20)
Here it is understood that e, Tik, and qk are functions of the local
fields defined Eqs. (1). before, require that residual
orthogonal subspace spanned #M(x) functions:

Eq.

in

Eq.

byuseofEq.

(19)

<PM<PNdv,

Energy Equation for an Element

of

fora

of

the

andT

=0,and

the

and

and

Tu = dun(p, drs, T} + 7^(p, drs. T)

(25)
Here n is the so called thermodynamic pressure and ffj is the
dissipative stress. Generally T^ given explicitly function
of drs, (e.g., class Stokesian fluids, fu 2^dip
H being viscosity); however, defined implicitly
equation state :
F(TC, dip T)
(26)
Assuming that cannot choose eliminate
from (25) (26) propose mixed finite-element
model which restriction ;re(x, f) element re assumed
to be of the form
ne = ftv(x)7rA'(r)
(27)
The interpolation functions /?N(x) have properties similar to
(3). Introducing (27) into (26), obtain residual
rc*. Then, the condition <r*,/? N > = 0 leads to the Ne local
equations

p,

and

ina

tohe

the

</,

the

tohe

can

where = dp/dt. Introducing Eqs. (la) (Ib) into (16) yields

at point x the residual
re(x) = (pN(x)pN + ((pN(x)i//R (x))kpNv%
(17)
We guarantee that residual vanishes average sense
over the element by requiring that it be orthogonal (with respect
inner product #> Vefgdv) subspace spanned
by functions <pM(x). Then <r e , (pMy = Vere(pMdv
we obtain the finite element model of the continuity equation:
aMNpN + bkMNRpv* = 0
(18)
Here M,N,R 1, 2,..., Ne aMN y^NR denote local
arrays

the

of

the

The local form of the continuity equation is

inthe

The Continuity Equation for an Element

isthe

ina

of

to

be

and

The constitutive equation for stress in terms of the approximate

velocity pressure expressions must introduced complete
this equation.

and

v ye

of

rikilsNidv - pNk = Q (15)

the

Equations (15), (18), (22) (23)] describe general

equations motion, continuity, energy heat conduction)
of typical element finite-element model arbitrary
fluid. To apply these equations to a specific fluid, it is necessary to
eliminate a, Tip.qk, and possibly Y\ by introducing appropriate
constitutive equations which uniquely define these functions in
terms v (e) , p(e) T(e). This customary procedure
finite-element formulations.
In the case of compressible fluids, however, the mean stress or
thermodynamic pressure often appears implicitly
equation state fluid, impossible
practical to obtain Ttj explicitly as a function of p, r, p and T.
In such cases, we propose that a "mixed" finite-element formulation be used, the basis of which is now to be described.
Consider a class of fluids described by constitutive equations
for stress tensor form
and

argument that this result must hold for arbitrary values of the
nodal velocity vk, we obtain as the general equations of motion
(momentum) for a compressible fluid element:

[orEq.

ANDL.C

J.T

1592

Ae

(24)

aN =

are

and

for

is

and2)the

=0

forthe

we

in

v=

ofp.

= PMk

and

where m and n^ are the mass and convected mass "matrices"

respectively, hMm array pressure coefficients, pm
are the components of the generalized force defined by Eq. (14)
of

to

is

canbe

Eq.

to

Again, procedure used derive (23) equivalent

Galerkin's method specific forms obtained when
constitutive equations Y\ qt furnished.

and

+ hMNkpN

^k +

Jve

isan

jAe

the

to

q^^dA

Jve

andthe

are

pisnowa

al

be

h%Ndv

at

= qN + aN
(23)
Here qN and GN are the generalized normal heat flux and the
generalized internal dissipation node

Eq.

Jve

We shall now consider purely mechanical behavior of incompressible fluids. Here two principal considerations are involved:
1) motions volume-preserving stress tensor
not completely determined by the motion. The first condition
reveals that known constant continuity equation reduces incompressibility condition
div dn = t'i i
(29)
The second consideration suggests that n = p, p being the
hydrodynamic pressure, and that the local pressure of Eq. (4)
should selected unknown place
Considering now p to be known and following essentially the
same procedure used obtain (15), obtain general
equation of motion of incompressible elements
tohe

t)=

ofthe

in

Eq.

V+

<
Jv e

Incompressible Fluids

asn

of

ph.For

t)andhe

by

forthe

be

h.

In order to obtain a useable expression for the energy equation

constitutive equations must be introduced for stress Tik, internal
energy density e, heat flux qk, and heat produced by internal
sources
We remark that an alternate equation of heat conduction can
obtained element rewriting (11) terms
entropy density rj(x, internal dissipation cr(x,
pOq qkk these choices variables, procedure
similar to that used to obtain Eq. (15) yields the general equation
of heat conduction for an element,

DECEMBER 1972

FLOW OF VISCOUS FLUIDS BY THE FINITE ELEMENT METHOD

1593

-i
(31)

be

can

Vvis

for

no

can

But

The

derivation of. the incompressibility constraint best

cast in physical terms: Observe that the work done by the pressure
p due to a change in volume dv is clearly peV \dv = pev\ej dv.
pe perform work since incompressible flow
zero. It follows that for a finite element of an incompressible fluid,
= 0

the

+ hMNkPN

(33)

the

athe

viathe

9MNkPN fMk ~ "iMNVk

=

(36)

(35)

eby

for

the

canbe

(at

and

so

isofthe

on

isnow

Eq.

ofEq.

the

to

pnk\l/NdA
jAe

the

pNk = -

it

or
JAe

of

ofthe

(-pdik

JAe

= PNk + I Tikniil/NdAi

onthe
The

The pressure term h^ eliminated from discrete

momentum equation Eq. (33) by constructing solenoidal finiteelement approximations of the local velocity. We accomplish this
by introducing called "bubble functions" af(x) which vanish
boundaries each element which satisfy least
approximately) condition
a M = -*NtM
(37)
Any particular integral (37) vanishing dRe form
a;(x) = a/N(jc)^
(38)
incompressibility condition, (29) satisfied locally
by local velocity approximations form
vfa t) = ^Njt(x)v^,
$NJt = cVWx) + a/N(x)
(39)
The remaining terms in Eq. (33) are altered accordingly.
local solenoidal velocity fields makes possible
eliminate hydrodynamic pressure term from discrete
the

Tikn^NdA =

JAe

Solenoidal Approximations

Theusof

f
PNk =

its

are

inthe

the

ofEq.

the

no

to

the

we

ofan

for

are

ofa

If either the total stress or the velocity is specified at the boundary finite element model, particular difficulties
encountered; the prescribed stress is introduced directly into
Eq. (14) to obtain consistent generalized nodal forces, and the
nodal velocities prescribed satisfy "nonslip" boundary
condition at a fixed or moving wall or the specified velocity
distribution on other surfaces. In this respect, the procedure
differs very little from that employed in the finite element analysis
of solids. However, in fluids, specification of the stress at a
boundary may not uniquely determine the pressure; moreover,
the boundary conditions may represent constraints on the
pressure or its gradient rather than the total stress. In these
situations, it may be necessary to develop special analogues of
the boundary conditions.
Consider, example, generalized force (14)
case incompressible fluid. Ignoring body force term
temporarily, observe that

toa

canbe

Boundary Conditions

dv.

wherein g J \l/^k
In the second method, a discrete version of this type of boundary condition obtained element merely introducing Eq. (4) into the local statement of the boundary condition
and evaluating the result at the coordinates of each of the wall
nodes. While this leads cruder approximation, nevertheless much easier to apply in actual calculations.
=

to

To these equations we add the incompressibility conditions,

Eqs. (32), form determinate set.

pnk\l/MdA

ofthe

Jv e

we

the

(34)

-I

and

and

where fMk = |Ve pFk\l/Mdv is the generalized force at node M due

body forces Fk. Transforming surface integral
Green-Gauss theorem collecting terms, arrive
discrete analogue pressure-gradient boundary condition,

tohe

, _f

wlMN
MN

The

the

the

the

v?=

v hMNkpN ZMNV? w^ pMk

where

tousea

for

itsno

is

is

athe

tohe

For

canbe

ponthe

The

of

forthe

isof

The special case of an incompressible Newtonian flow with

constant viscosity special interest because number
classical solutions to various flow problems are available for
comparisons. Of course, since the viscosity is constant, the energy
equation is uncoupled from the momentum and continuity
equations and they may be solved separately. For the Newtonian
fluid the dissipative stress tensor is simply fik = 2^dik, \JL being the
viscosity. Thus, finite element, fik fj[\l/N kViN ijsN p^
and Eq. (30) reduces to

Incompressible Newtonian Fluids

tohe

andthe

notbe

and

and

are

byits

now

on

dueto

quantity pNk generalized nodal force prescribed

pressures boundary.
Similar procedures used various other special
boundary conditions. example, uncommon
a condition at the impermeable boundary with zero velocity the
specification of the pressure gradient as a function of the body
force acceleration (i.e., n = pFn pan where subscript refers normal direction wall). This boundary
condition can either be applied by retaining the momentum
equation at the nodes in contact with the fixed wall or by applying
discrete version nodes contact with wall. first
method, which followed herein, statisfies boundary condition in an average sense over the boundary elements. The
second method satisfies the boundary condition exactly at the
wall nodes.
first method obtained directly from discrete momentum equation, Eq. (33), by setting vk = 0 (for boundary nodes N)
and fik = 0, since fik depends only on vk there. Then

The

J ve

We note that the energy equation, Eq. (22), is also modified

slightly for incompressible elements due to the fact that (pNpN is
replaced presumedly known value fik depends
pN. These alterations straightforward need
displayed here.

0;

Fig. 1 Finite-element model for the calculation of Couette flow.

(32)

HM\I/N t

MN

in

^" = r,

isthe

for

This relation must hold arbitrary pM-; thus finite element

model for the incompressibility condition is,

J. T. ODEN AND L. C. WELLFORD JR.

1594

AIAA JOURNAL

EXACT SOLUTION

VELOCITY (in/sec.)
.10

0.5

FINITE ELEMENT
SOLUTION

Fig. 2 Velocity profile at various

time points for transient Couette
flow.

t=.OO042sec. t=OOI67sec.

t*OO376sec

t=OO669sec. t=oosec.

of this type. With minor changes, these techniques could be

applied to three-dimensional incompressible flow problems or,
in general, isothermal incompressible viscous fluid.
Consider the solution of the system of equations consisting of
momentum equation, (33), continuity equation,
Eq. (32). Three cases can be analyzed based on the properties of
the flow: steady uniform flow; steady nonuniform flow; and
transient unsteady flow. models developed earlier lead
to systems of linear algebraic equations in the first case, nonlinear algebraic equations in the second, and nonlinear differential equations third.

toany

the

on

of

or

-pan

by

of

inthe

is

The

canbe

the

momentum equations, Eqs. (33). Thus momentum equations

can be solved directly for the local velocity fields. Then the
pressure calculation performed independently based
the computed velocity field.
The pressure must satisfy Poisson's equation.
(40)
Pii = T^x, itf) pFit
where 7}/x, v*) is the dissipative part of the stress tensor. This
can be verified by taking the divergence of the local momentum
equation.22 solution this equation means finite
elements very well documented.4

andthe

t^.OOOIOsec.

Eq.

The

Numerical Techniques

Steady, Uniform Linear Flow

andthe

Forthe

ofthe

hMNkPN ZMNI% vO? pMk

to

are

to

the

steady flow problem, term mMNv* convective term n^NPv^ in Eq. (33) vanish and the linearized
momentum equation takes the form

In order obtain preliminary estimate applicability

of the foregoing theory, we consider a series of classical problems
in two-dimensional incompressible Newtonian flow. Several
numerical techniques developed herein handle problems

(41)

.08T

.07-

.06-

.04-

Fig. 3 Time history of the ;c velocity

components at nodes 7, 8, and 9 for
transient Couette flow.

03-

.02-

NODE 7'

.01-

C/ .obi

.002

.003

.004

.005

.006

TIME (SEC.)

.01-

.007

.008

.009

.010

.012

DECEMBER 1972

FLOW OF THE VISCOUS FLUIDS BY THE FINITE ELEMENT METHOD

.001

1595

.002 .003 .004 .005 .006 .007 .008 .009 .010

.Oil

.012

Fig. 4 Time history of y velocity component at nodes 7 and 9 for transient Couette flow.

dv. Thus, we obtain the system of non-

r
MNF = jvc
linear equations

(43)
" = pMk
+
frk + hMNkPN +
A natural choice of a technique for solving Eqs. (43) is the
incremental loading method used nonlinear structural
mechanics.20 This conclusion is based on the observation that if
density p is assumed to be the loading parameter, when the
density equals zero the set of equations reduces to the stationary
linear system, Eq. (42), which can be solved using methods
described above.
This solution process can be described concisely in vector
notation. Suppose that the collection of momentum equations
and continuity equations are expressed in vector form as
f(v, p) = 0
(44)

in

wecan

the

Equations (41) and (32) represent, for the two-dimensional case,

a set of 3Ne algebraic equations in 3Ne unknowns. Taking note
of fact that r^ h express these equations
in matrix form as

In

canbe

and

setof

is

andthe

the

The

coefficient matrix square symmetric. addition,

upon assembling the elements, the global form is sparse and
banded. Boundary conditions must be applied in accordance with
previous discussion resulting equations
solved for the velocity and pressure variables using standard
Gauss-elimination codes.
of

Steady, Nonuniform FlowThe Method Incremental Densities

the

by

Eq.

The momentum equation for steady nonlinear flow can be

obtained from (33) setting local inertial term mMNvkN
equal to zero and using the fact that n^NP = pr^NP, where

where v is the vector of unknown velocity components and

pressures density parameter. solution
vector corresponding particular value density
v + d\ be a solution vector corresponding to the value of density

p,andlet

(42)

O]

of

Letvba

'NM
r
0

r'NM

[fc_ 2 ]

toa

L
MN2

2
tN

MN

andpisthe

Slider Bearing
_.3000
c

o .2333

of

Fig. Finite element model

steady flow of a lubricant through
plane slider bearing.

06667

Moving Surface

1.333

X
T7////////7//////////
2000
(inches)

1596

J. T. ODEN AND L. C. WELLFORD JR.

AIAA JOURNAL

FINITE ELEMENT
SOLUTION

-05

forthe

6x

Fig. velocity component lubrication

problem.

., 3333 >

VELOCITY(in./sec.xlO')

DISTANCE FROM LEADING EDGE OF THE BEARING (ia)

in

canbe

where

=0;

we

Eq.

By differentiating (32) twice observe that r^Lv^

consequently
X

UN rvLmLMlhMNk>

yu ruLmLlHMk
=

yv>

by

is

the

and mLM * is the inverse of the mass matrix mLM. The first
derivative pressure variable formulated similar
fashion. We differentiate Eq. (46) with respect to time to obtain
ofthe

the

then+

for

=0,the

we

p)=

p+d)

f(v+

andCisNx1

an

ofthe

of

BisanNx

dp.

p 4- Then d\, also vanishes. Taylofs series

expansion can be introduced in the neighborhood of v if f(v, p)
is continuously differentiate at v. Thus
f(v + v, p + 6p) = f(v, p) + B(5v + Cdp + higher order terms
where matrix vector: B,-/v,
dffa p)/dvy9 C(v, p) = dftv, p)/dp ij = 1, ... , N. Thus, within
terms first order, have linear system B<5v Cdp.
Given the initial value of the solution vector v0 and the initial
value density p0 series expansion forms basis
of iterative scheme which 1 TH iterate given

(48)

forthe

the

we

r),

x=

arenowfth

and

by

be

of

ofthe

=1

isthe

Vn = v. - B- Hv,, pn)C(vn, pn)dpn+ ! ; p n+ ! = dpt

(45) If hw inverse of XUN, obtain explicit expression
1
derivative pressure variable
This procedure amounts to a piecewise linearization and the
Pv hwyu
(49)
number iterations should determined accuracy requireEquations
(46)
(49)
form
F(x,
which
ments.
can be directly integrated by standard Runge-Kutta schemes.
Unsteady Flow

to

bythe

we

and

are

and

For

the

4th

We shall present results in the next section in which the solution

Some Numerical Results
of the system of nonlinear differential equations, Eqs. (32) and
We shall now cite representative numerical results obtained
(33), were integrated numerically using self-correcting order
by applying theory methods presented earlier specific
Runge-Kutta technique. To outline the essential features, conproblems in two-dimensional flow of incompressible Newtonian
sider first Eq. (33) rewritten in the form
fluids.
demonstration purposes, shall employ six-node
FMk
(46)
triangular elements of the type shown in Fig. 1, for which the local
where
velocity pressure fields given quadratic polyno(47)
pMk
mials
FINITE ELEMENT SOLUTION

.3333
VELOCITY (in./sec. x IO4 )

33

I
.6666

3 ~ |

I23?

3I2

the

Fig. 7

I.OOOO

3 |
1.6666

1.3333

DISTANCE FROM LEADING EDGE OF BEARING

(in.)

2.0

y velocity component for

lubrication problem.

.004-1
> FINITE ELEMENT
SOLUTION

onthe

.003-

inthe

"I
\
u:

of

m
~ .002-

1.0

onthe

Variation pressure fixed slider bearing wall.

of

Fig.

ofthe

7and9

orthe

at

ofthe

An Incompressible Lubrication Problem

ofthe

the

4)The

of

y=0.2)Thexand

1)The

was

and

ora

The two-dimensional flow of lubricant between a slide block

slider bearing moving surface determined. fifteen
element model, as illustrated in Fig. 5 was constructed. The
following boundary conditions were applied: x-velocity
component was 0.01 in./sec and the y velocity component was
zero along velocity components were
equated to zero along the slider bearing wall. 3) The stress at all
unconstrained boundary nodes was prescribed as a hydrostatic
pressure 0.001 lbf./in.2 gradient pressure perpendicular to the bearing wall was set equal to zero.
In initial calculations, convective inertial terms were
ignored. The finite element solution for the velocity profile in the
x direction is presented in Fig. 6. The transverse velocity profile
is presented in Fig. 7, and the pressure along the inclined bearing
the

was

andthey

tobe

andil

to

tobe

The problem of unsteady Couette flow through the domain

indicated in Fig. 1 is considered. The following boundary
conditions were applied: 1) The x velocity component was
assumed equal 0.1 in./sec. velocity component
was assumed to be equal to zero at y = 0.2 in. 2) The x and y
velocity components were prescribed as zero at y = 0. 3) The
stress on the boundaries x = 0 and x = 2.0 in. was set equal to
zero. 4) The gradient of the pressure in the direction perpendicular
to the wall was zero at y = 0. As initial conditions, we set the
pressure and the velocity vector equal to zero at all interior nodes
at t = 0. The value of the mass density used was 0.00242 Ibf.-sec.2/
in.4 viscosity here subsequent results
assumed 0.00362 lbf.-sec./in.2
In Fig. 2 the tangential velocity profile at x = 0.5 in. is presented
at various times and shows good agreement with the exact
solution.21 In Fig. 3, time histories of the tangential velocity

dueto

in

was

Ascanbe

at

7,8and9re

component nodes included. seen,

the initial starting period negative tangential velocity components occur several nodes. This apparently either
the applied stress boundary condition coarseness
model used. Because of this constraint small transverse velocity
components were computed. They were symmetric with respect
to the lines connecting nodes 11 and 15 and 3 and 23. The time
histories transverse velocity component nodes
are given in Fig. 4.

of

Couette Flow

andthe

2.0

DISTANCE FROM LEADING EDGE OF BEARING (IN.)

at

and

the

of

to

to

is

of

for

to

and

of

ofthe

the

the

and

ofa

anytwo

vathe

visa

.001-

and

of

inEq.

Inthe

isthe

visa

and

and

6)

N=

alofthe

k=

thesix

vk = (aN + bmxl
p (aN bNixl cNijxlxj)pN
(5)
Here independent constants %, bNh cNij (cNij cNji;
ij, 1,2,3; 1,2,..., depend only local coordinates of the six nodes of the element. These local approximations
determine relevant arrays matrices local
momentum, energy, and continuity equations described earlier
for each element.
Concerning convergence accuracy this particular
approximation, we mention that Zlamal24 has obtained the error
estimate
v-v\^ (K/sin 9)h2
(51)
where given continuously different?able field, finiteelement interpolant (i.e., coincides with nodal points),
h is the maximum diameter of all finite elements in a given mesh,
6 smallest angle between sides triangle,
K is a constant independent of 9 and h. Clearly, long flat elements
lead to poorer conditioned systems than networks of isosceles
triangles. case elliptic parabolic problems, estimates
of type (51) lead directly convergence error
estimates also involving I/sin and, energy convergence, K/z4.
While study stability convergence finite-element
approximations hyperbolic problems scarcely beginning,
preliminary results seem indicate that local character
approximation lead inherently better conditioned systems
than conventional difference schemes equal accuracy.
the

1597

BYTHE

FLOW VISCOUS FLUIDS FINITE ELEMENT METHOD

OF

DECEMBER 1972

DISTANCE FROM LEADING EDGE (in.)

1.0
I
LU

1
Fig. 9 x velocity component at
section A as computed by the incremental densities technique.

fe
LU
o

I
Q

-7002

.002

.004

.006 .008 .01

X VELOCITY COMPONENT AT SECTION A (in./sec.)

2.0
I

AIAA JOURNAL

JR.

T.

ANDL.C

J. ODEN WELLFORD

1598

DISTANCES FROM LEADING EDGE (in.) NOT SCALE

"cvi

IN

888

jj

Sod d

9-

w/////////// /?// ///////////777/7 /////7/7T*

a

of

10

Fig. Finite-element model flow over

NODE A

and

ofthex

flat plate time histories velocity

'L-

AandB.

at

components nodes

NODE B

.3

.4

.5

TIME (SEC.)

at

tohe

at

was

of1.

ofthex

at

in

The

240

of240is

at

thex

9.For

were 10. in./sec. and the y velocity components were zero at all
unconstrained nodes at t = 0. The pressure was also zero at all
nodes time zero. mass density lbf.-sec.2/in.4 assumed.
Computed time histories velocity components
nodes A and B are presented as representative variations in
Fig. 10. In Fig. 11, the steady state finite-element solution is
compared Blasius boundary-layer solution specific
points on the plate. Again, excellent agreement is obtained with
rather coarse mesh.

8.The

in

is indicated Fig. transverse velocity profile changes

direction between x = 0.3333 and x = 0.6666 in. resulting in a
back flow in the bearing. The nonlinear convective terms were then
included, and the bearing problem was solved by the incremental
densities method described earlier. resulting velocity profile
in direction Reynolds number included
Fig. Reynolds numbers larger than more significant
changes were observed in the velocity profile; these results are
now being evaluated with a more detailed model.
Boundary-Layer Flow

References

p.

of

of

forthe

ofSOR

23,No.9

of

R.,

and

1)Thex

andthey

to10.

Zienkiewicz, O. C., The Finite Element Method in Engineering

Science, McGraw-Hill, London, 1972.

5
Zienkiewicz, O. C., Mayer, P., and Cheung, Y. K., "Solution of
Anisotropic Seepage Problems by Finite Elements," Proceedings of the
American Society of Civil Engineers, EM1, Vol. 92, 1966, pp. 111-120.
6
Martin, H. C., "Finite Element Analysis of Fluid Flows," AFFDL-

thex

The

I.O-,

K.,"Onthe

G.and

10.

of

Fix, Larsen, Convergence Iterations

for Finite Element Approximations to Elliptic Boundary Value
Problems," SI AM Journal of Numerical Analysis, Vol. 8, No. 3, 1970,
pp.2 536-547.
Courant, "Variational Methods Solution Problems
of Equilibrium and Vibrations," Bulletin, American Mathematical
Society, Vol. 49, Jan. 1943, pp. 1-23.
3
Turner, M. J., Clough, R. W., Martin, H. C., and Topp, L. P.,
"Stiffness Deflection Analysis Complex Structures," Journal
the Aeronautical Sciences, Vol. 1956, 805-823.

ofthe

the

for

wasnot

waset

The

is

was

The problem of transient boundary-layer formation over a

flat plate analyzed using finite element grid Fig.
This model primarily useful study velocity profile
away from the leading edge of the plate. Thus results are presented
for spatial points at least 1 in. from the leading edge where the
model converges to the exact solution. In order to predict flow
patterns at the leading edge, where high velocity gradients are
encountered, a more detailed model of the flow would be required.
following boundary conditions were specified:
velocity component equal in./sec. velocity
component was equated to zero at x = 0.05 in. 2) The x and y
velocity components were zero at y = 0. 3) The gradient of the
pressure in the direction normal to the wall was zero at y = 0.
4) The pressure was zero at all boundary nodes at which the
velocity specified.
initial conditions specified that velocity components

BLASIUS SOLUTION

3 FINITE ELEMENT
SOLUTION

a .7-

UJ

I6"

and

of

Fig. Comparison finite-element

Blasius solutions for the steady-state boundary layer on the front portion of the plate.

i*
Q

.2.13

DISTANCE FROM THE LEADING EDGE OF THE PLATE (ia)

10

FLOW OF VISCOUS FLUIDS BY THE FINITE ELEMENT METHOD

p.

"Onthe

New

I,

and

L.

J.

theUsof

for

"A

ofthe

The

the

D.W,

J.

The

M.,

12,

in

byR.H

in

ofthe

and

in

F.S,

and

L.

of

andJ.T

Y.

and

bythe

of

P.,

G.,

"A

J.

of

Deformation and Irreversible Thermodynamics of Nonlinear Continua," Recent Advances in Matrix Methods of Structural Analysis and
Design, University Alabama Press, Tuscaloosa, Ala., 1970.
10
Oden, T., Finite-Element Analogue Navier Stokes
Equations," Journal of the Engineering Mechanics Division, ASCE,
Vol. 96, No. EM4, 1970, pp. 529-534.
11
Thompson, E. Mack, T., Lin, "Finite-Element
Method for Incompressible Slow Viscous Flow with a Free Surface,"
Developments in Mechanics, Proceedings of the llth Midwestern
Mechanics Conference, Iowa State University Press, Ames Iowa,
Vol. 5, 1969, pp. 93-111.
12 Tong, "The Finite Element Method Fluid Flow Analysis,"
Recent Advances in Matrix Methods of Structural Analysis and Design,
edited by R. H. Gallagher, Y. Yamada, and J. T. Oden, University of
Alabama Press, Tuscaloosa, Ala., 1971, pp. 787-808.
13
Fujino, T., "Analysis Hydrodynamic Plate Structure
Problems Finite Element Method," Recent Advances Matrix
Methods of Structural Analysis Design, Edited Gallagher,
Yamada, Oden, University Alabama Press, Tuscaloosa,
Ala., 1971, pp. 725-786.
14
Argyris, J. H., "Two and Three-Dimensional Potential Flow by

1599

the Method of Singularities," The Aeronautical Journal of the Royal

Aeronautical Society, Vol. 73, Nov. 1969, pp. 959-961.
15
Argyris, H., Scharpf, "The Incompressible Lubrication
Problem," Aeronautical Journal of Royal Aeronautical Society,
Vol. 73, Dec. 1969, pp. 1044-1046.
16
Argyris, J. H., Mareczek, G., and Scharpf, D. W., "Two and
Three-Dimensional Flow Using Finite Elements," Aeronautical
Journal of the Royal Aeronautical Society, Vol. 73, Nov. 1969,* pp.
961-964.
17
Reddi, "Finite-Element Solution Incompressible
Lubrication Problem," Transactions of the ASME, Journal of Lubrication, Vol. 91, July 1969, pp. 524-533.
18
Baker, J., Numerical Solution Technique Two-Dimensional Problems Fluid Dynamics Formulated with
Discrete Elements," TN-TCTN1005, Bell Aerosystems Co., Niagra
Falls, N.Y., 1970.
19
Herting, Joseph, A., Kuusinen, R., MacNeal,
R. H., "Acoustic Analysis of Solid Rocket Motor Cavities by a Finite
Element Method," NASTRAN: User's Experience, Vol. NASA
TMX-2378, 1971.
20 Oden, J. T., Finite Elements of Nonlinear Continua, McGraw-Hill,
New York, 1971.
Schlichting, H., Boundary-Layer Theory, McGraw-Hill,
York, 1968.
22
Eringen, A. C., Mechanics of Continua, Wiley, New York, 1967.
23
Ladyzhenskaya, O. A., The Mathematical Theory of Viscous
Incompressible Flow, Gordon and Breach, New York, 1969.
24
Zlamal, M., Finite Element Method," Numerische
Mathematik, Vol. 1968, 394-409.
A.

in

ofthe

D.,

p.

J.Tand

1,No.3

TR-68-150, Wright-Patterson Air Force Base, Ohio, 1969, pp. 517-535.

7
Oden, J. T., "A General Theory of Finite Elements; II Applications," International Journal for Numerical Methods in Engineering,
Vol. 1969, 247-259.
8
Oden, Somogyi, "Finite-Element Applications
Fluid Dynamics," Journal Engineering Mechanics Division,
ASCE, Vol. 95, No. EM4, 1969, pp. 821-826.
9
Oden, J. T., "Finite Element Formulation of Problems of Finite

D.N,

DECEMBER 1972