Tt (x, y) ---- T (x, y) nu T ( x , - y ) ;

t
o ~ ( x , O) = 2 [~,~ (x, O) - - ~yy (x, 0)].

It follows from here that Fig. id (curves i) and Fig. 5b show the graphs of the stresses
2O~x (x, 0) if the regions D are semicircular with centers at the point with the coordinates
x' = y' = 0, and Figs. Id, 2d, 3d, and 4d (curves 2) give the graphs of the stresses O~x (x,
0) if the regions D are circular or also delineated by the ellipses with the centers at the
point with the coordinates x' = O, y' = 6. The results obtained for the semiinfinite plate
can be used if the welded joint is positioned in the vicinity of its boundary.
LITERATURE CITED
i. V . N . Maksimovich, E. N. Novosad, and G. V. Plyatsko, "Determination of the optimum heat-
ing regions in hardening longitudinal welded joints in cylindrical shells," Probl.
Prochn., No. 4, 79"83 (1982)
2. Ya. S. Podstrigach and Yu. M. Kolyano, Nonstationary Temperature Fields and Stresses in
Thin Plates [in Russian], Naukova Dumka, Kiev (1972).
3. V . N . Maksimovich and G. V. Plyatsko, "Boundary problems of thermoelasticity for piece-
wise homogeneous flat shells," in: Thermomechanical Processes in Piecewise Homogeneous
Structural Members [in Russian], Naukova Dumka, Kiev (1978), pp. 50-56.
4. B . G . Kornev, Problems of the Theory of Heat Conductivity and Thermoelasticity [in Rus-
sian], Nauka, Moscow (1980).
5. I . N . Sneddon, Fourier Series, Routledge & Kegan (1973).

SOLVING A CONTACT PROBLEM OF ELASTICITY THEORY

BY THE FINITE-ELEMENT METHOD

D. M. Barlam UDC 539.3.01

Examination of the force interaction between machine components forms a special group of
the problems of elasticity theory, i.e., the so-called contact problems, e.g., the interaction
of the sleeve of a working cylinder with a rim, the interaction of a bearing shell with compo-
nents in which the bearing is installed, various types of die problems, etc. In all problems
of this type, the pressure between the contacting bodies, displacements in the contact zone,
and in a number of cases, the contact zone itself, are unknown. Because of the highly compli-
cated nature of these problems, analytical solutions have been obtained only for bodies and
contacting surfaces of comparatively simple form.
The extensive development of the numerical methods of solving the problems of elasticity
theory, especially the finite-element method (FEM), has made it possible to obtain solutions
of contact problems for regions of quite complicated geometrical form. A detailed review of
the investigations concerned with the application of FEM to solving the contact problems of
elasticity theory has been presented in [i].
Two approaches have been outlined [2, 3]. In the first case, contact interaction is re-
alized by means of a layer of intermediate "contact" finite elements with specific properties,
and in the other case it is realized directly through the nodes of finite elements distributed
at the boundary of the contacting body.
Use of FEM in the classic form (in the form of displacements) for the problems of elas-
ticity theory makes it possible to determine the displacements of the nodes under the effect
of given external forces at given displacements of the boundary points. However, as already
mentioned, the unknown quantities in the contact problem are not only the displacements, but
also the forces in the contact zone and, in many cases, the contact zone itself. This deter-
mines the iteration solution procedure during which the displacements, forces, and the contact
zone are determined more accurately in each iteration.

Special Design Bureau of Diesel Construction, Leningrad. Translated from Problemy Proch-
nosti, No. 4, pp. 39-43, April, 1983. Original article submitted January 19, 1982.

480 0039-2316/83/1504-0480507.50 9 1983 Plenum Publishing Corporation

 Y~

Rn~

R~

Fig. I. Forces in the contact zone.

In the solution method described in this article, it is assumed that the contact inter-
action between the bodies takes place through the coinciding pairs of the nodes of the finite
elements distributed in the possible contact zone.
To simplify considerations, the proposed procedure is used for a planar problem of elas-
ticity theory. No limitations are imposed on the type of finite elements used.
Contact without Taking Friction into Account. We shall examine two elastic bodies which
are in contact under the effect of external forces (Fig. i). In the regions of the boundary
in which contact may occur, the following conditions should be satisfied:
A B
u~ - - u~ ~< 6.; (1)
~<0; (2)
(3)

where u A u B are the projections of the displacements of the boundary points on the external
n ~ n

normal to the boundary of the body A; o~ is the component of the stress along the normal to
the boundary; and ~#, component of the stress along the tangent to the body boundary A.
The inequality in Eq. (I) represents the condition of absence of penetration of one body
into the other. The equality in Eq. (i) occurs in the case of boundary points entering into
contact. Here 6n ~ 0 determines the initial gap between the bodies (for the given initial
tension, e.g., in interference fit, 6n ~ 0). The inequality in Eq. (2) exists in the presence
of contact, and the equality holds in the absence of contact. Condition (3) characterizes the
absence of interaction between the bodies along the tangents to the boundary, i.e., absence
of friction. Let it be that Uin, Rin, RiT are the displacements of the nodes along the normal
and the components of the nodal forces along the normal and the tangent to the boundary, ~in
is the initial gap between the i-th pair of the nodes. In accordance with Eqs. (1)-(3) for
the nodes of the contact zone we have

(4)
A~<7
Rt~ -.~ O; (5)

= = o.
(6)
The displacements and the components of the nodal forces along the normal and the tangent
will be expressed by means of their components along the coordinate axes
Urn ~ U~x COS0~ ~ Utyo~sin oci;
Rin -]- Rix cos czi -[- R~g sin al;
Rt~ ------- Rt= sin at Jr R~,j cos a i.

481
Consequently, the conditions [4, 5] assume the form
ufx cos a i q- u~ sin a l- u B cos el --ut~ sin ct~ ~<8i,~; (7)

R~ cos a~ -t- R~ sin a i ~ O. (8)

The potential energy of the system of the examined bodies will be expressed by means of
the stiffness vector [K], the column vector of the nodal displacements {u}, and the column
vector of the nodal forces {R}:
I
n ( u ) = ~ {u}~[Kl {u} - - {u} ~ {R}, (9)

where {u} T is the transposed vector. Consequently, the problem is reduced to determining the
minimum H(u) for the limitations (7); for this purpose it is necessary to solve the family of
linear equations

[KI {u} = {~} (lO)

at the given limitations.
The following procedure was used to solve this problem.
In subsequent considerations, the limitations in Eq. (7) for which the equality is valid
will be referred to as active [4]. Let us assume that some of the limitations in Eq. (i0) are
active. We shall determine the solution of the family (7) taking into account the active lim-
itations. The method of solving the family of the linear equations taking into account the
limitations in the form of equalities is described below. From the resultant solution we de-
termine Rin and verify:
a) fulfillment of inequality (7) for the nonactive limitations. Violation of this in-
equality causes the transfer of the appropriate limitation to the group of the active limita-
tions;
b) fulfillment of the condition (8) for the active limitations. If this condition is
not satisfied, the contact between the corresponding pair of the nodes is violated and the
limitation (7) for this pair is transferred into the group of the nonactive limitations. The
accuracy of determination of the contact zone, the forces, and the displacements of the con-
tacting nodes is continuously improved until the set of the active limitations is stabilized.
The described process is similar to the method of the reduced gradient for problems of qua-
dratic programming [4] and coincides with this method if the positive determinacy of the ma-
trix [K] is taken into account.
To verify the proposed algorithm, the problem of the penetration of a rigid punch into an
elastic half-plane was solved. The force acting on the punch was I0 kN. In solving the prob-
lem by the finite-element method, the elasticity moduli of the punch and the half-plane were
equal to 2.10 ~ and 2.10 4 ~ a , respectively. The dimensions of the regions and division into
the finite elements are shown in Fig. 2. Three- and four-node finite elements were used in
the calculations. Figure 3 shows the distribution of contact pressure below the punch. For
the sake of comparison, Fig. 3 also gives the analytical solution [5]. The small difference
between the finite-element and analytical solutions is explained by the fact that in the for-
mer case the punch is not absolutely rigid. Two iterations were required to solve the prob-
lem.
Contact Taking Friction into Account. If friction between the contacting surfaces is
taken into account, components of the nodal forces along the tangent to the boundary appear.
Two states may occur for the points of the boundary taking place in contact, i.e., complete
coupling or slipping. In this case, condition (3) is replaced by

f ~ I~< -- V ~ , (ii)

where ~ is the friction coefficient.

The nonequality in Eq. (ii) occurs if the friction force is higher than the shearing
forces. Slipping does not take place in this case. If the inequality is violated, the con-
tacting bodies slip in relation to each other. In this case, the equality exists in Eq. (ii),
i.e., the tangential force is equal to the friction force.

482
 , ,, 50 ,

i,
~[ 150 MPa

0.15

0,15

Q05
o Q025
Jl
t

Qom m

Fig. 2 Fig. 3
Fig. 2. Division into finite elements in the problem of the punch.
Fig. 3. Contact pressure (broken curve shows the analytical solution for
the case of an absolutely rigid punch, the solid curve gives the numeri-
cal solution).

The following conditions are fulfilled for the pairs of the nodes in which complete cou-
pling takes place:
tl A UB O;
Iv-- it: (12)

UAln-Ufn----6in, (13)

where uim = --Uix sin mi + Uiy cos mi is the displacement of the node along the tangent to the
boundary.
For the nodes of the contact zone, the condition (ii) may be written in the form

I R~, I ~< -- ~R~.. (14)

This problem differs from the contact problem without friction in that the direction of
the friction forces is not known in advance and must be determined by the direction of dis-
placement of the body A in relation to the body B or, which is equivalent, by the sign of the
shearing force at the node, Ri~.
The following procedure is proposed for solving the problem. The nodes initially includ-
ed in the set of the active restrictions are assumed to be fully coupled. The conditions
(12) and (13) are fulfilled for these nodes. These conditions are taken into account in search-
ing for the solution for the family of equations (i0). This is followed by determination of
the nodal forces and verification of:
a) fulfillment of inequality (4) for the nonactive limitations. If this inequality is
not fulfilled, the corresponding pair of nodes is included in the number of active limitations
and the given pair is regarded as completely coupled (the conditions (12) and (13) are fulfilled
for the pair in the subsequent iteration);
b) fulfillment of condition (5) for the pairs of the nodes included in the set of the
active limitations. If this condition is not fulfilled, the corresponding pair is excluded
from the number of active limitations. If (5) is fulfilled, the condition (14) is checked.
If this condition is fulfilled in the form of a strict inequality, the given node pair is re-
tained in the coupled state and, conversely, if the above condition is fulfilled in the form
of an equality or is violated, this indicates the occurrence of slipping; for this reason,
the tangential forces at the nodes in the subsequent (k + l)-th iteration are regarded as
equal to

R ~ k+l~ ------P, I R ~ h ) l s i g n (R~,(k));

483
 MPa
5 ~,

Fig. 4 Fig. 5
Fig. 4. Division of the strip and the layer into finite elements.
Fig. 5. Distribution of the contact pressure and tangential forces in the
zone of contact between the elastic layer stretched by the forces Q and
loaded by the uniformly distributed pressure, and the strip placed on the
rigid base.

~nus, in each consecutive iteration the friction forces are determined by means of the
normal forces of the previous iteration. ~ e n changes in the state of the contact nodes have
st~pped~ an additional iteration is carried out which gives the final values of the friction
forces~
As an example of the use of the proposed solution method, we shall examine a problem of
an elastic layer clamped with a uniformly distributed pressure p = i00 kPa to an elastic strip
placed on a rigid base and tensile loaded along the axis with forces Q = 1 kN. The friction
coefficient ~ is assumed to be equal to 0.15. The authors found no analytical solution of the
given problem in the literature.
The elasticity moduli of the strip and the layer are equal and amount to 2.10" MPa. The
dimensions of the zones and the division into finite elements are shown in Fig. 4, and the cal-
culation diagram and the distribution of the contact pressure and the tangential forces between
the layer and the strip are shown in Fig. 5. In the given problem the contact zone consists
of two regions. In region AB the layer is coupled with the strip, whereas in region BC the
contacting surfaces slip in relation to each other. Five iterations were required to solve
the problem by the proposed method.
It should be noted that while in the contact problem in which friction is not taken into
account the existence and singularity of the solution requires only the convexity of the func-
tional of the potential energy or, which is the same, the positive determinacy of the stiff-
ness matrix; in the problem in which friction is taken into account the singularity conditions
are not formulated and the problem of finding these conditions has not as yet been solved [6].
The contact problem in identical formulation was examined in [7]. In contrast to [7], where
the problem was solved for increments of the force and displacement vectors, the full values
of these vectors were examined at each iteration in the present work.
Solution of the Family of Linear Equations with Connections. The family of linear equa-
tions with connections in the form of equalities is solved in most cases by Lagrange multipli-
ers. These multipliers are not suitable for finite-element analysis because their introduc-
tion increases the number of unknown quantities and, consequently, the order of the family of
equations. The direct substitution of the limitations into the family of equations and sub-
sequent reduction of the family by eliminating the columns and lines are unsuitable because
they require renumeration of the unknown quantities. The solution method described below is
-free from these disadvantages.
We shall examine the family of equations

n
~Kijuj=R~, i--l,2,...,n, (15)
./=1

and the limitations in the form of equalities, assuming that only two unknown quantities are
interrelated

484
 Avuv -4:-6r = ur. (16)

To ensure determinacy, it will be assumed that p < r (this condition is not principal). Sub-
stituting Eq. (16) into (15), we obtain the transformed family
ft

Y. K~jul + (Ktv + AvK~) uv + Kt,6, -~ Rt. (17)

"]~!,t;~p,r

This operation is equivalent to the following operation: The column r is multiplied by ~ and
added up with the column p.
The above operation leads to a family of equations with a nonsymmetric matrix. To re-
store the symmetry of the matrix, the column r is multiplied by ~ and added up with the line
p. The family has the form

K, juj + (Kiv q- AvK,~) up -b gi,6, ---Rl,

l=l.l#p ,r
i-~- 1,2 ..... n, i =/~p ;

(Kv~ + AvK~j) uj + (Kvv + 2AvKpr +

t=1.]~P,r
+ A~Krr)up + (Kpr + AvKrr) 6~----Rp + ApR,.

The position of the u n k n o w n q u a n t i t y u r is occupied by the known quantity d r and, to avoid

eliminating the r-th line and the column, we shall use the method proposed by Zienkiewicz to
take into account the given nodal displacements [8].
In the r-th equation, the coefficient Krr is multiplied by a larger number, e.g., 108 ,
and the right-hand part is replaced by Krrdr'10 s. As a result of solving the family, the r-th
unknown will be equal to d r with an accuracy of 10 -8 , and the final value of u r will be deter-
mined in accordance with Eq. (16).
The proposed method of solving the family of linear equations with connections retains
the order of the system and the tape structure of the matrix so that standar@ programs of
solving families of linear equations with a symmetric matrix can be used. It should be men-
tioned that the width of the tape increases in these transformations but the magnitude of this
increase in the appropriate numeration of the nodes is insignificantly small.
LITERATURE CITED
i, F. D. Fischer, "Solving contact problems of elastic bodies with stretched contact sur-
faces by quadratic programming," Computing, i__33,Nos. 3-4, 353-384 (1974).
2. S. Chan and I. S. Tuba, "A finite-element method for contact problems of solid bodies,"
Int. J. Mech. Sci., 13, No. 7, 615-639 (1971).
3. R. Goodman, R. L. Taylor, and T. L. Brekke, "A model for the mechanics of jointed rock,"
Proc. Am. Soc. Civ. Eng., J. Soil Mech. Found. Div., 94, No. 3, 637-659 (1968).
4. D. M. Himmelblau, Applied Nonlinear Programming, McGraw-Hill (1972).
5. N. I. Muskhelishvili, Some Basic Problems of the Mathematical Theory of Elasticity [in
Russian], Nauka, Moscow (1966).
6. G. Duvant and J. L. Lions, Inequalities in Mechanics and Physics [Russian translation],
Nauka, Moscow (1980).
7. B. Fredriksson, "Finite-element solution of surface nonlinearities in structural mechan-
ics with special emphasis to contact and fracture mechanics problems," Comput. Struct~
i, Nos. 4-5, 281-290 (1976).
. O. C. Zienkiewicz, Finite Element Method in Engineering Science, McGraw-Hill, New York
(1971).

485