Engineering Analysis with Boundary Elements 108 (2019) 49–59

Contents lists available at ScienceDirect

Engineering Analysis with Boundary Elements

journal homepage: www.elsevier.com/locate/enganabound

Boundary element method for two-dimensional frictional contact problems

of anisotropic elastic solids
Van Thuong Nguyen, Chyanbin Hwu∗
Department of Aeronautics and Astronautics, National Cheng Kung University, Tainan, Taiwan, ROC

a r t i c l e i n f o a b s t r a c t

Keywords: The conventional boundary element method solving for the problems of two-dimensional anisotropic elastic solids
Boundary element method with prescribed traction and/or prescribed displacement boundary conditions is extended to the frictional contact
Frictional contact problems. A complete system of linear equations is constructed by boundary integral equations and contact
Anisotropic elastic
constraint relations. The contact solutions are obtained by using an efficient, iterative and fully incremental
Contact constraint
loading technique. By using this technique, the nonlinearity raised by unknown contact region and unknown slip
Incremental load
direction of frictional contact can be approximated by the accumulation of linear increments. The incremental
load is determined by using the load extrapolation technique that allows only one or two node pairs come into
contact in each iteration. The slip direction of frictional contact is decided by referring to the relative tangential
slip in the frictionless state. To avoid reassembling the whole system equations in each iteration, a suitable
arrangement of the equation system is made and a fast solver is adopted to get the solution without resolving the
entire system of equations. When the contact bodies contain holes, cracks or inclusions, we use a special boundary
element whose fundamental solution satisfies the boundary condition along the hole/crack/inclusion boundary.
The validation of the proposed method is demonstrated through several numerical examples, which further lead
to the discussion of the effects of friction coefficient, material anisotropy, holes, cracks and inclusions on contact.

1. Introduction
Contact problems of anisotropic elastic solids are challenging prob-
lems since the complexities arise not only from the involving of direc-
tional dependence of materials but also from the non-linearity caused
by the unknown contact area and frictional contact. The problems are
even more difficult if the defects such as holes, cracks, and/or inclu-
sions exist inside the contact bodies. In the last few decades, many re-
searchers have made great efforts on handling contact problems either
by using analytical or numerical treatments. By using the analytical or
semi analytical treatment, several works can be found such as [1–3] for
isotropic elastic solids, [4,5] for transversely isotropic elastic solids, and
[6–10] for anisotropic elastic solids. Even some of these studies worked
on anisotropic elastic solids, their results are applicable for idealized
problems such as contact of rigid punches on a half plane without holes,
cracks, or inclusions due to the limitation of analytical solutions.
Among the possible numerical approaches that can be used to han-
dle the contact problems, finite element method (FEM) and boundary
element method (BEM) are two popular numerical methods. References
of using FEM to solve anisotropic elastic contact problems can be found
in [11–15]. A common issue of these works is that fine mesh in con-
tact regions is usually required in order to obtain accurate results. BEM
has shown its advantages over the other methods since it reduces the
problems by one dimension and the boundaries are the primary con-
cern in contact problems [16–22]. Although BEM was applied to the
contact problems long time ago, most of the studies are limited to the
cases with isotropic elastic solids. Due to the complexity of the associ-
ated fundamental solutions, relatively few studies focus on the contact
of anisotropic elastic solids [23–25].
In this work, the conventional BEM solving for the problems of
anisotropic elastic solids with prescribed-traction and/or prescribed-
displacement boundary conditions is extended to the problems with fric-
tional contact conditions. The contact solutions are obtained by using
an efficient, iterative and fully load-incremental technique proposed in
[18]. This method has the advantage that the contact constraints can be
directly and exactly enforced at the contact node pairs. In addition, by
suitable arrangement of the structural matrix, we only need to update
the parts of the final system matrix related to the contact conditions
instead of reassembling the whole system equations during the calcula-
tion process. Furthermore, the fast solver based on Sherman-Morrison
formula [26] is used to update the solution from one iteration to the
next without resolving the final system of equations.
To demonstrate the correctness and applicability of the present ap-
proach, several examples such as a flat-ended or parabolic punch on an

∗
Corresponding author.
E-mail address: chwu@mail.ncku.edu.tw (C. Hwu).

https://doi.org/10.1016/j.enganabound.2019.08.010
Received 11 June 2019; Received in revised form 14 July 2019; Accepted 4 August 2019
0955-7997/© 2019 Elsevier Ltd. All rights reserved.
V.T. Nguyen and C. Hwu Engineering Analysis with Boundary Elements 108 (2019) 49–59

anisotropic elastic foundation without or with holes/cracks/inclusions,

and contact between two dissimilar elastic cylinders, were implemented.
The influences of friction coefficient, normal or tangential force, ma-
terial anisotropy, and holes/cracks/inclusions on the contact solutions
are all discussed. Here, the problems with holes/cracks/inclusions are
solved using special BEM whose fundamental solutions already satis-
fied the boundary conditions along holes, cracks, or inclusions. Thus, no
meshes are needed along the hole/crack/inclusion boundaries, and the
meshes are only required on the outer boundaries and contact regions
of the anisotropic elastic solids, which saves us a lot of computational
time.

2. Boundary element method for contact problems

If all three components in x1 , x2 , and x3 directions of displacements

and tractions are considered for the most general two-dimensional de-
formation of anisotropic elastic bodies and body forces are omitted, the
boundary integral equations for anisotropic elastostatics can be written
in matrix form as [27]

𝐂(𝝃)𝐮(𝝃) + 𝐓∗ (𝝃, 𝐱)𝐮(𝐱)𝑑Γ(𝐱) = 𝐔∗ (𝝃, 𝐱)𝐭(𝐱)𝑑Γ(𝐱), (2.1a)

∫Γ ∫Γ
where

⎡𝑐11 (𝝃) 𝑐12 (𝝃) 𝑐13 (𝝃)⎤ ⎧𝑢 (𝐱)⎫ ⎧𝑡 (𝐱)⎫

⎪ 1 ⎪ ⎪1 ⎪
𝐂(𝝃) = ⎢𝑐21 (𝝃) 𝑐22 (𝝃) 𝑐23 (𝝃)⎥, 𝐮(𝐱) = ⎨𝑢2 (𝐱)⎬, 𝐭(𝐱) = ⎨𝑡2 (𝐱)⎬,
⎢ ⎥ ⎪𝑢3 (𝐱)⎪ ⎪𝑡3 (𝐱)⎪
⎣𝑐31 (𝝃) 𝑐32 (𝝃) 𝑐33 (𝝃)⎦ ⎩ ⎭ ⎩ ⎭
⎡𝑡11 (𝝃, 𝐱) 𝑡12 (𝝃, 𝐱) 𝑡13 (𝝃, 𝐱)⎤
∗ ∗ ∗

𝐓 (𝝃, 𝐱) = 𝑡21 (𝝃, 𝐱) 𝑡22 (𝝃, 𝐱) 𝑡∗23 (𝝃, 𝐱)⎥,

∗ ⎢ ∗ ∗ Fig. 1. Contact between two anisotropic elastic solids.
⎢∗ ⎥
⎣𝑡31 (𝝃, 𝐱) 𝑡∗32 (𝝃, 𝐱) 𝑡∗33 (𝝃, 𝐱)⎦
⎡𝑢11 (𝝃, 𝐱) 𝑢12 (𝝃, 𝐱) 𝑢13 (𝝃, 𝐱)⎤
∗ ∗ ∗ To simplify the following discussions for contact problems, (2.2) can
𝐔 (𝝃, 𝐱) = 𝑢21 (𝝃, 𝐱) 𝑢22 (𝝃, 𝐱) 𝑢∗23 (𝝃, 𝐱)⎥.
∗ ⎢ ∗ ∗ (2.1b) be simply written as
⎢ ∗ ⎥
⎣𝑢31 (𝝃, 𝐱) 𝑢∗32 (𝝃, 𝐱) 𝑢∗33 (𝝃, 𝐱)⎦ 𝐘𝐮 = 𝐆𝐭 , (2.3)

In (2.1a,b), Γdenotes the boundary of the elastic solid; uj (x) and tj (x), in which Y, G, u, and t are the global matrices and vectors assembled by
j = 1,2,3, are the displacements and surface tractions at the field point x their associated element matrices and vectors Yij , Gij , uj , and tj . If two
of the boundary Γ; 𝑢∗𝑖𝑗 (𝝃, 𝐱) and 𝑡∗𝑖𝑗 (𝝃, 𝐱), 𝑖, 𝑗 = 1, 2, 3,are the fundamental contact bodies A and B are considered (see Fig. 1) and each of them is di-
solutions of displacements and tractions; cij (𝝃),i, j = 1, 2, 3, are the free vided into contact and non-contact parts, then (2.3) can be re-expressed
term coefficient dependent on the location of the source point 𝝃, which as
in general can be determined by setting rigid body motions in three co-
𝐘𝐴 𝐴 𝐴 𝐴 𝐴 𝐴 𝐴 𝐴
𝑐 𝐮𝑐 + 𝐘𝑛𝑐 𝐮𝑛𝑐 − 𝐆𝑐 𝐭𝑐 − 𝐆𝑛𝑐 𝐭𝑛𝑐 = 𝟎,
ordinate directions. To solve the displacements u(x) and tractions t(x)
of (2.1), in boundary element formulation the boundary Γ is generally 𝐘𝐵 𝐵 𝐵 𝐵 𝐵 𝐵 𝐴 𝐵
𝑐 𝐮𝑐 + 𝐘𝑛𝑐 𝐮𝑛𝑐 − 𝐆𝑐 𝐭𝑐 − 𝐆𝑛𝑐 𝐭𝑛𝑐 = 𝟎, (2.4)
discretized into m elements with n nodes, and the field points x, dis-
where the subscripts c and nc denote, respectively, the contact and non-
placements u and tractions t on the boundary are approximated by the
contact parts; the superscripts A and B denote, respectively, the contact
nodal points xn , nodal displacement un and nodal traction tn through
bodies A and B. Since the actual contact zone may not be known a priori,
different interpolation functions. Following the standard procedure of
it is essential that a potential contact zone be chosen to be larger than
BEM, (2.1a) will then lead to a system of algebraic equations as [28]
the likely final contact region. If the non-contact parts are further di-
𝑛
∑ 𝑛
∑ vided into the part with displacement-prescribed (𝐮𝐴 (𝑘) 𝐵(𝑘) 𝐴(𝑢) 𝐵(𝑢)
𝑛𝑐 , 𝐮𝑛𝑐 , 𝐭𝑛𝑐 , 𝐭𝑛𝑐 )
𝐘𝑖𝑗 𝐮𝑗 = 𝐆𝑖𝑗 𝐭𝑗 , 𝑖 = 1, 2, ..., 𝑛, (2.2) 𝐴(𝑢) 𝐵(𝑢) 𝐴(𝑘) 𝐵(𝑘)
and traction-prescribed (𝐮𝑛𝑐 , 𝐮𝑛𝑐 , 𝐭𝑛𝑐 , 𝐭𝑛𝑐 ), where the superscripts
𝑗=1 𝑗=1
(u) and (k) denote, respectively, unknown and known, we may re-write
where Yij and Gij are the matrices of influence coefficients associated (2.4) as
with nodes i and j, whose determination depends on the choice of the
fundamental solutions and the interpolation functions. In our study for 𝐘𝐴 𝐴 𝐴(𝑢) 𝐴(𝑢) 𝐴(𝑘) 𝐴(𝑘) 𝐴 𝐴 𝐴(𝑢) 𝐴(𝑢) 𝐴(𝑘) 𝐴(𝑘)
𝑐 𝐮𝑐 + 𝐘𝑛𝑐 𝐮𝑛𝑐 + 𝐘𝑛𝑐 𝐮𝑛𝑐 − 𝐆𝑐 𝐭𝑐 − 𝐆𝑛𝑐 𝐭𝑛𝑐 − 𝐆𝑛𝑐 𝐭𝑛𝑐 = 𝟎,
simplicity the linear function is selected for the interpolation of x, u and 𝐘𝐵 𝐵 𝐵(𝑢) 𝐵(𝑢) 𝐵(𝑘) 𝐵(𝑘) 𝐵 𝐵 𝐵(𝑢) 𝐵(𝑢) 𝐵(𝑘) 𝐵(𝑘)
𝑐 𝐮𝑐 + 𝐘𝑛𝑐 𝐮𝑛𝑐 + 𝐘𝑛𝑐 𝐮𝑛𝑐 − 𝐆𝑐 𝐭𝑐 − 𝐆𝑛𝑐 𝐭𝑛𝑐 − 𝐆𝑛𝑐 𝐭𝑛𝑐 = 𝟎.
t, whereas the fundamental solutions 𝑢∗𝑖𝑗 (𝝃, 𝐱) and 𝑡∗𝑖𝑗 (𝝃, 𝐱) are obtained (2.5)
from the Green’s functions. In the literature, several useful Green’s func-
tions for the anisotropic elastic solids have been obtained by using com- By letting
plex variable Stroh’s formalism such as the basic function for a perfect
⎧ 𝐴⎫
infinite anisotropic elastic plate, and the special function for an infinite { } { } ⎪𝐮 𝑐 ⎪
plate with holes, cracks, or inclusions [27]. In this study both of the ba- 𝐮𝐴𝑛𝑐
(𝑢) 𝐵(𝑢)
𝐮𝑛𝑐 ⎪𝐭𝑐𝐴 ⎪
𝐱𝐴 = 𝐴(𝑢) , 𝐱𝐵 = 𝐵(𝑢) , 𝐱 𝐶 = ⎨𝐮 𝐵 ⎬,
sic and special fundamental solutions obtained from the existing Green’s 𝐭𝑛𝑐 𝐭𝑛𝑐 ⎪ 𝑐⎪
functions were employed. Using the special fundamental solutions for ⎪𝐭𝑐𝐵 ⎪
⎩ ⎭
holes/cracks/ inclusions possesses some advantages comparing to the [ (𝑢) ] [ ]
fundamental solutions for 2D infinite anisotropic elastic plate such as 𝐘𝐴
𝑛𝑐
(𝑢)
−𝐆𝐴 𝑛𝑐 𝟎 𝟎
𝐊𝐴 = , 𝐊𝐵 = 𝐵(𝑢) 𝐵(𝑢) ,
no meshes are needed along the hole, crack or inclusion boundaries. 𝟎 𝟎 𝐘𝑛𝑐 −𝐆𝑛𝑐

50
V.T. Nguyen and C. Hwu Engineering Analysis with Boundary Elements 108 (2019) 49–59

[ ] { }
𝐘𝐴𝑐 −𝐆𝐴𝑐 𝟎 𝟎 𝐆𝐴 (𝑘) 𝐴(𝑘) 𝐴(𝑘) 𝐴(𝑘)
𝑛𝑐 𝐭𝑛𝑐 − 𝐘𝑛𝑐 𝐮𝑛𝑐
for this node pair can be written as
𝐊𝐶 = , 𝐟 = 𝐵(𝑘) 𝐵(𝑘) ,
𝟎 𝟎 𝐘𝐵𝑐 −𝐆𝐵𝑐 𝐆𝐵( 𝑘) 𝐵(𝑘)
𝑛𝑐 𝐭𝑛𝑐 − 𝐘𝑛𝑐 𝐮𝑛𝑐 ⎧𝑢𝑖 ⎫ ⎡ cos 𝜃 sin 𝜃𝑖 0⎤⎧𝑢𝑖1 ⎫
⎪ 𝑠𝑖 ⎪ ⎢ 𝑖 ⎪ ⎪
(2.6)
⎨𝑢𝑛 ⎬ = ⎢− sin 𝜃𝑖 cos 𝜃𝑖 0⎥⎨𝑢𝑖2 ⎬,
⎪𝑢 𝑖 ⎪ ⎣ 0 ⎥⎪ 𝑖 ⎪
⎩ 3⎭ 0 1⎦⎩𝑢3 ⎭
(2.5) can now be written as
⎧𝑡𝑖 ⎫ ⎡ cos 𝜃 sin 𝜃𝑖 0⎤⎧𝑡𝑖1 ⎫
𝐊𝐴 𝐱𝐴 + 𝐊𝐵 𝐱𝐵 + 𝐊𝐶 𝐱𝐶 = 𝐟 . (2.7) ⎪ 𝑠𝑖 ⎪ ⎢ 𝑖 ⎪ ⎪
⎨𝑡𝑛 ⎬ = ⎢− sin 𝜃𝑖 cos 𝜃𝑖 0⎥⎨𝑡𝑖2 ⎬, 𝑖 = 𝑎 or 𝑏. (3.2)
Let nA and nB be the number of nodes in the non-contact region of ⎪𝑡 𝑖 ⎪ ⎣ 0 ⎥⎪ 𝑖 ⎪
⎩ 3⎭ 0 1⎦⎩𝑡3 ⎭
bodies A and B, respectively; nC denotes the number of node pairs in
the contact region; qA and qB be the number of known displacement For stick/slip contact mode, this node pair is supposed to be bring to-
components of bodies A and B, respectively. The dimensions of each gether, and hence, to describe the relation for (3.1)2 and (3.1)3 the
matrices and vectors in (2.5) and (2.7) can then be expressed as normal-tangential coordinates of nodes a and b should be arranged to
( ) be parallel but in opposite direction. If we define the direction pointing
𝐮𝐴 (𝑘) 𝐴(𝑢)
𝑛𝑐 , 𝐭𝑛𝑐 ∶𝑞𝐴 × 1, 𝐮𝐴 (𝑢) 𝐴(𝑘)
𝑛𝑐 , 𝐭𝑛𝑐 ∶ 3𝑛𝐴 − 𝑞𝐴 × 1,
( ) from node a to node b be the average normal direction and use the out-
𝐮𝐵( 𝑘) 𝐵(𝑢)
𝑛𝑐 , 𝐭𝑛𝑐 ∶𝑞𝐵 × 1, 𝐮𝐵( 𝑢) 𝐵(𝑘)
𝑛𝑐 , 𝐭𝑛𝑐 ∶ 3𝑛𝐵 − 𝑞𝐵 × 1, (2.8a) ward normal for each point, then the average normal 𝐧̄ 𝑎 and 𝐧̄ 𝑏 can be
𝐮𝐴 𝐴 𝐵 𝐵
𝑐 , 𝐭𝑐 , 𝐮𝑐 , 𝐭𝑐 ∶ 3𝑛𝐶 × 1, determined by
( )
and ̄ cos 𝜃,
𝐧̄ 𝑎 = −𝐧̄ 𝑏 = 𝐧̄ = −sin𝜃, ̄ 0 , (3.3a)
( ) ( ) ( )
𝐘𝐴 (𝑢)
𝑛𝑐 ∶ 3𝑛𝐴 + 3𝑛𝐶 × 3𝑛𝐴 − 𝑞𝐴 , 𝐆𝐴 (𝑢)
𝑛𝑐 ∶ 3𝑛𝐴 + 3𝑛𝐶 × 𝑞𝐴 ,
where
( ) ( ) ( ) ( ) ( )
𝐘𝐴 (𝑘)
𝑛𝑐 ∶ 3𝑛𝐴 + 3𝑛𝐶 × 𝑞𝐴 , 𝐆𝐴 (𝑘)
𝑛𝑐 ∶ 3𝑛𝐴 + 3𝑛𝐶 × 3𝑛𝐴 − 𝑞𝐴 , cos 𝜃̄ = 𝑥𝑏2 − 𝑥𝑎2 ∕𝑔0 , sin𝜃=
̄ − 𝑥𝑏 − 𝑥𝑎 ∕𝑔0 ,
( ) ( ) √
1 1
𝐘𝐴
𝑐 ∶ 3𝑛𝐴 + 3𝑛𝐶 × 3𝑛𝐶 , 𝐆𝐴
𝑐 ∶ 3𝑛𝐴 + 3𝑛𝐶 × 3𝑛𝐶 , (2.8b) ( 𝑏 )2 ( )2
𝑔0 = 𝑥1 − 𝑥𝑎1 + 𝑥𝑏2 − 𝑥𝑎2 . (3.3b)

( ) ( ) ( ) and g0 is the normal gap between node a and node b. If g0 is equal

𝐘𝐵( 𝑢)
𝑛𝑐 ∶ 3𝑛𝐵 + 3𝑛𝐶 × 3𝑛𝐵 − 𝑞𝐵 , 𝐆𝐵( 𝑢)
𝑛𝑐 ∶ 3𝑛𝐵 + 3𝑛𝐶 × 𝑞𝐵 ,
( ) ( ) ( )
𝐘𝐵( 𝑘)
𝑛𝑐 ∶ 3𝑛𝐵 + 3𝑛𝐶 × 𝑞𝐵 , 𝐆𝐵( 𝑘)
𝑛𝑐 ∶ 3𝑛𝐵 + 3𝑛𝐶 × 3𝑛𝐵 − 𝑞𝐵 ,
( ) ( )
𝐘𝐵
𝑐 ∶ 3𝑛𝐵 + 3𝑛𝐶 × 3𝑛𝐶 , 𝐆𝐵
𝑐 ∶ 3𝑛𝐵 + 3𝑛𝐶 × 3𝑛𝐶 .
and
𝐊𝐴 ∶ 3𝑛 × 3𝑛𝐴 , 𝐊𝐵 ∶ 3𝑛 × 3𝑛𝐵 , 𝐊𝐶 ∶ 3𝑛 × 12𝑛𝐶 ,
𝐱𝐴 ∶ 3𝑛𝐴 × 1, 𝐱𝐵 ∶3𝑛𝐵 × 1, 𝐱𝐶 ∶12𝑛𝐶 × 1, 𝐟 ∶ 3𝑛 × 1, 𝑛 = 𝑛𝐴 + 𝑛𝐵 + 2𝑛𝐶 .
(2.8d)
From (2.8d) we see that the system of linear Eq. (2.7) contains 3n equa-
tions and 3n+6nC unknowns. To solve all the unknown variables, we
need 6nC additional equations which comes from the contact constraint
relations discussed in the next section.
3. Contact constraint relations
As stated at the end of the previous section, we need 6nC addi-
tional equations to solve the contact problems by using BEM. To ob-
tain these equations, we consider the contact region ΓC whose trac-
tions and displacements are unknown. The contact condition in this re-
gion can be represented by the contact modes of each node pair. In
general, there are three different modes: separation, stick and slip. If
(𝑡𝑎𝑛 , 𝑡𝑎𝑠 , 𝑡𝑎3 ), (𝑡𝑏𝑛 , 𝑡𝑏𝑠 , 𝑡𝑏3 ), (𝑢𝑎𝑛 , 𝑢𝑎𝑠 , 𝑢𝑎3 ) and (𝑢𝑏𝑛 , 𝑢𝑏𝑠 , 𝑢𝑏3 ) are the vectors of the sur-
face tractions and displacements in the local normal-tangential coordi-
nate of a node pair (a, b) of Fig. 1, the mathematical expressions for
these three modes can be stated as follows.
Separation ∶ 𝑡𝑎𝑛 = 𝑡𝑎𝑠 = 𝑡𝑎3 = 0, 𝑡𝑏𝑛 = 𝑡𝑏𝑠 = 𝑡𝑏3 = 0.
Stick ∶ 𝑡𝑎𝑛 = 𝑡𝑏𝑛 , 𝑡𝑎𝑠 = 𝑡𝑏𝑠 , 𝑡𝑎3 = 𝑡𝑏3 , 𝑢𝑎𝑛 + 𝑢𝑏𝑛 = 𝑔0 , 𝑢𝑎𝑠 + 𝑢𝑏𝑠 = 0, 𝑢𝑎3 + 𝑢𝑏3 = 0.
Slip ∶ 𝑡𝑎𝑛 = 𝑡𝑏𝑛 , 𝑡𝑎𝑠 = ±𝜂𝑡𝑎𝑛 , 𝑡𝑎3 = 𝑡𝑏3 , 𝑢𝑎𝑛 + 𝑢𝑏𝑛 = 𝑔0 , 𝑡𝑎𝑠 = 𝑡𝑏𝑠 , 𝑢𝑎3 + 𝑢𝑏3 = 0.
In the above, 𝜂 is the friction coefficient and the sign ahead of 𝜂 depends
on the slip direction. The frictionless contact can be considered by letting
𝜂 = 0. Note that in (3.1) the local coordinates of node a and node b are
different (see Fig. 1). For all node pairs of contact bodies, the tangential
direction s is always chosen such that when one faces the direction of
increasing s the body lies on the right hand side. Counterclockwise 90°
we get the normal direction n, which will always direct outward of the
body. If we use 𝜃 a and 𝜃 b to denote the tangential directions of node pair
(a, b) (see Fig. 1), the transformation relation of local-global coordinates
to zero, we simply choose the average normal to be the normal of the
point a. Under the action of an external force during deformation the
node pair will be brought closer if the load increases and contact oc-
(2.8c)
curs when g0 =0. Note that in the above 𝜃 a ,𝜃 b ,and 𝜃̄ depend on the po-
sition of each node pair. Normally, if an incremental loading technique
is employed, these angles will change at each loading stage since the
associated position of node pairs will change under different loading
condition. However, if only small deformation is considered for the en-
tire contact process, these angles change slightly and can be assumed to
remain unchanged during the iteration process. Combining (3.1), (3.2),
and (3.3), the contact constraint relations can be stated symbolically as
𝐃0𝐶 𝐱𝐶 = 𝐟0 , (3.4)
in which the dimensions of 𝐃0𝐶 and f0 are 6nC × 12nC and 6nC × 1, re-
spectively. As stated in (3.1)-(3.3), the explicit expressions of 𝐃0𝐶 and f0
will depend on the contact modes, the angles 𝜃 a ,𝜃 b , and/or 𝜃̄ of local
coordinates, and/or the normal gap g0 , and/or the friction coefficient 𝜂.
Eq. (3.4) provides 6nC additional equations. Thus, combination of
(2.7) and (3.4) constructs a (3n + 6nC ) × (3n + 6nC ) system of linear
equations, in which the unknown variables can be determined uniquely.
In the case when the contact region is known a priori (e.g. a flat-ended
punch on a flat elastic foundation), and the size of contact region is inde-
pendent of the magnitude of the applied load, only stick and slip contact
modes will be assumed in the known contact region. In this situation al-
though an iteration process is still required to determine a satisfactory
contact state inside the contact region, the total load can be applied di-
rectly to the system of linear equations to solve the contact problems.
On the other hand, if the size of the contact region is unknown, which
generally occurs in contact problems, and is dependent on the applied
load, then an incremental method must be employed to overcome the
(3.1)
nonlinear feature of contact problems.
4. Incremental loading technique
As stated previously in the general contact problem the size of con-
tact region is usually unknown a priori. Different magnitude of applied
load will lead to different contact size and contact mode, and hence con-
tact problem is generally nonlinear. To solve a nonlinear problem by a
system of linear equations, (2.7) and (3.4), a fully incremental loading
technique was suggested so that the nonlinearity can be approximated

51
V.T. Nguyen and C. Hwu Engineering Analysis with Boundary Elements 108 (2019) 49–59

by many linear steps [18]. To implement an incremental load process, ensure that the normal traction at this pair is zero. If two node pairs
in each step the load is updated by are at the same distance to the contact edge, which usually occurs for
the contact under symmetric condition, both node pairs will be brought
𝐩𝑚 = 𝐩𝑚−1 + Δ𝐩𝑚 , 𝑚 = 1, 2, ..., 𝑀, (4.1)
to contact. Under this consideration, the node pair next to the current
and in the last step pM equals to the external load p. The system of linear contact edge is set to be in slip mode. To get zero normal traction at this
Eqs. (2.7) under the external loads pm and pm − 1 can be written as node pair, two trial incremental loads Δ𝐩(1) and Δ𝐩(2) are applied. If
𝑚
𝐊𝐴 𝐱𝐴 𝑚
+ 𝐊𝐵 𝐱𝐵 + 𝐊𝐶 𝐱𝐶𝑚 = 𝐟𝑚 , under 𝐩𝑚 , 𝑡(1) (2)
𝑛 and 𝑡𝑛 are the normal traction associated with these two trial loads
at the selected node pair, the incremental load Δpm which will lead to
𝑚−1 𝑚−1
𝐊𝐴 𝐱𝐴 + 𝐊𝐵 𝐱𝐵 + 𝐊𝐶 𝐱𝐶𝑚−1 = 𝐟𝑚−1 , under 𝐩𝑚−1 . (4.2) 𝑡𝑚
𝑛 = 0 can then be extrapolated by
Subtracting (4.2)1 by (4.2)2 , we get ( (2) )
𝑡(1)
𝑛
𝑚 𝑚 Δ𝐩𝑚 = Δ𝐩(1) + Δ𝐩 − Δ𝐩(1) . (4.8)
𝐊𝐴 Δ𝐱𝐴 + 𝐊𝐵 Δ𝐱𝐵 + 𝐊𝐶 Δ𝐱𝐶𝑚 = Δ𝐟𝑚 , 𝑚 ≥ 1, (4.3) 𝑡(1) − 𝑡(2)
𝑛 𝑛
where Δfm = fm − fm − 1 . Like the total form system equations, to solve Note that the above extrapolation formula is obtained based upon the
the incremental displacements and tractions from the incremental form linear relation between traction and load, which is applicable if the trial
(4.3), we still need 6nC additional equations which are constructed by incremental load are small enough that the contact conditions are unal-
the constraint relations. tered and the material is linear elastic.
To use the constraint relations (3.1) in incremental form, we first For the convenience of computer programming, no matter the con-
update the displacements and tractions by tact is symmetric or unsymmetric, both node pairs nearest to the two
edges of contact region are brought into contact and are set in slip mode.
𝑢𝑚 𝑚−1
𝑛 = 𝑢𝑛 + Δ𝑢𝑚 𝑚 𝑚−1
𝑛 , 𝑢𝑠 = 𝑢𝑠 + Δ𝑢𝑚 𝑚 𝑚−1
𝑠 , 𝑢3 = 𝑢3 + Δ𝑢𝑚
3
,
By using (4.8) for each node pair, we get two incremental loads. For
𝑡𝑚 𝑚−1
𝑛 = 𝑡𝑛 + Δ𝑡𝑚 𝑚 𝑚−1
𝑛 , 𝑡𝑠 = 𝑡𝑠 + Δ𝑡𝑚 𝑚 𝑚−1
𝑠 , 𝑡3 = 𝑡3 + Δ𝑡𝑚
3
, (4.4) unsymmetric contact, the smaller incremental load is selected and its
where Δ𝑢𝑚 𝑚 corresponding node pair is the new edge of contact region. For symmet-
𝑗 and Δ𝑡𝑗 , j = n, s, 3, are, respectively, the incremental changes
in displacements and tractions due to the incremental load Δpm . Sub- ric contact, these two incremental loads will be equal, and both node
stituting (4.4) into (3.1), the constraints for different contact modes can pairs become the new edge of contact region. Since the slip direction is
be expressed as follows: unknown at the beginning of applying load and the sign ahead of the
Separation mode: friction coefficient 𝜂 is undetermined, the above concept can only be
( 𝑎 )𝑚 ( )𝑚−1 ( 𝑎 )𝑚 ( )𝑚−1 ( 𝑎 )𝑚 ( )𝑚−1 executed by assuming 𝜂 = 0 which is irrelevant to the sign. With this un-
Δ 𝑡𝑛 = − 𝑡𝑎𝑛 , Δ 𝑡𝑠 = − 𝑡𝑎𝑠 , Δ 𝑡3 = − 𝑡𝑎3 , derstanding, to solve the general frictional contact problems we suggest
( 𝑏 )𝑚 ( 𝑏 )𝑚−1 ( 𝑏 )𝑚 ( 𝑏 )𝑚−1 ( 𝑏 )𝑚 ( 𝑏 )𝑚−1 the following process to determine the slip direction and incremental
Δ 𝑡𝑛 = − 𝑡𝑛 , Δ 𝑡𝑠 = − 𝑡𝑠 , Δ 𝑡3 = − 𝑡3 . (4.5)
load.
Stick/slip mode:
(1) Assume frictionless movement in tangential direction, i.e., 𝜂 = 0.
( )𝑚 ( )𝑚 [( ) ( )𝑚−1 ]
Δ 𝑡𝑎𝑛 − Δ 𝑡𝑏𝑛 = − 𝑡𝑎𝑛
𝑚−1
− 𝑡𝑏𝑛 , stick∕slip; (2) Apply two trial incremental loads Δp(1) and Δp(2) and determine
[( ) ] their corresponding values of Δf(1) and Δf(2) by (2.6)7 .
( 𝑎 )𝑚 ( 𝑏 )𝑚 𝑚−1 ( ) 𝑚−1
Δ 𝑡𝑠 − Δ 𝑡𝑠 = − 𝑡𝑎𝑠 − 𝑡𝑏𝑠 , stick∕slip; (3) Both node pairs nearest to the two ends of current contact region,
( 𝑎 )𝑚 ( 𝑏 )𝑚 [( 𝑎 )𝑚−1 ( 𝑏 )𝑚−1 ] (m-1)th step, are assumed to be brought into contact, i.e., node
Δ 𝑡3 − Δ 𝑡3 = − 𝑡3 − 𝑡3 , stick∕slip; pairs (a, b) and (a′, b′) of Fig. 1 are assumed in slip mode. Thus,
( )𝑚 ( )𝑚 [ ( )𝑚−1 ( 𝑏 )𝑚−1 ] 𝐃𝑚 and 𝐟𝐶𝑚−1 of (4.7) can be determined.
𝐶
Δ 𝑢𝑎𝑛 + Δ 𝑢𝑏𝑛 = 𝑔0 − 𝑢𝑎𝑛 + 𝑢𝑛 ≡ 𝑔0𝑚−1 , stick∕slip; (4) Solve ΔxA ,ΔxB ,ΔxC corresponding to Δf(1) and Δf(2) from (4.10)
( )𝑚 ( )𝑚 [( ) ( )𝑚−1 ] by fast solver (4.12), which will be stated later in this Section.
𝑚−1
Δ 𝑢𝑎𝑠 + Δ 𝑢𝑏𝑠 = − 𝑢𝑎𝑠 + 𝑢𝑏𝑠 , stick;
From these solutions we can calculate the normal traction 𝑡(1) 𝑛
( 𝑎 )𝑚 ( 𝑎 )𝑚 [( ) ( ) ]
Δ 𝑡𝑠 ∓ 𝜂Δ 𝑡𝑛 = − 𝑡𝑎𝑠
𝑚 −1
∓ 𝜂 𝑡𝑎𝑛
𝑚 −1
, slip; and 𝑡(2)𝑛 at the node pair (a, b), and the normal traction 𝑡𝑛
′(1)
and
′(2)
[ 𝑡𝑛 at the node pair (a′, b′).
( 𝑎 )𝑚 ( 𝑏 )𝑚 ( 𝑎 )𝑚−1 ( 𝑏 )𝑚−1 ]
Δ 𝑢3 + Δ 𝑢3 = − 𝑢3 + 𝑢3 , stick∕slip. (4.6) (5) The incremental loads calculated by (4.8) using (𝑡(1) (2)
𝑛 , 𝑡𝑛 ) for the
′(1) ′(2)
left node pair and (𝑡𝑛 , 𝑡𝑛 ) for the right node pair can now be
Same as the construction of (3.4), by using (4.5)–(4.6) and (3.2)–(3.3) set to be Δ𝐩𝓁𝑚 and Δ𝐩𝑟𝑚 , respectively.
the additional contact constraint relations can be written symbolically (6) If Δ𝐩𝓁𝑚 < Δ𝐩𝑟𝑚 , the node pair (a, b) in slip mode is confirmed and
as the node pair (a′, b′)is released to be separated. If Δ𝐩𝓁𝑚 > Δ𝐩𝑟𝑚 ,
(a′, b′) slips and (a, b) separates. If Δ𝐩𝓁𝑚 = Δ𝐩𝑟𝑚 , both (a, b) and
𝐃𝑚 𝑚 𝑚−1
𝐶 Δ𝐱𝐶 = 𝐟𝐶 , 𝑚 ≥ 1. (4.7)
(a′, b′) remain in slip mode.
Combination of (4.3) and (4.7) can then solve the incremental displace- (7) After updating the contact status through the comparison of Δ𝐩𝓁𝑚
𝑚 , Δ𝐱 𝑚 , Δ𝐱 𝑚 .
ments and tractions of the mth step Δ𝐱𝐴 and Δ𝐩𝑟𝑚 , repeat (3)-(6) with only one node pair brought into
𝐵 𝐶
contact for Δ𝐩𝓁𝑚 ≠ Δ𝐩𝑟𝑚 , and get the new Δpm . Otherwise, Δ𝐩𝑚 =
4.1. Determination of slip direction and incremental load Δ𝐩𝓁𝑚 = Δ𝐩𝑟𝑚 , and no new updating is required at this stage.
(8) Apply Δpm obtained in stage (7) and determine their corre-
To solve the incremental displacements and tractions from (4.3) and sponding values of Δfm by (2.6)7 . With 𝐃𝑚 𝐶
and 𝐟𝐶𝑚−1 obtained
(4.7), the vectors on their right hand sides, i.e., Δfm and 𝐟𝐶𝑚−1 , should in stage (3), solve ΔxA ,ΔxB ,ΔxC corresponding to Δfm from
be known before calculation. According to (2.6)7 , we see that to obtain (4.10) by fast solver (4.12). Calculate the relative tangential dis-
Δfm we need to provide a proper value for the incremental load Δpm . placement (𝑢𝑎𝑠 )𝑚 + (𝑢𝑏𝑠 )𝑚 for each node pair in contact region. If
Whereas the relations shown in (4.5) and (4.6) tell us that 𝐟𝐶𝑚−1 is de- (𝑢𝑎𝑠 )𝑚 + (𝑢𝑏𝑠 )𝑚 > 0, choose the positive sign for 𝜂, otherwise, nega-
termined from the displacements and tractions obtained in the previous tive sign for 𝜂. This selection will make the direction of tangential
step. Moreover, the sign ahead of the friction coefficient 𝜂, which de- traction be opposite to the slip direction.
pends on the slip direction, is required to be determined prior to the (9) At each incremental load step m, store the sign determined in
evaluation of 𝐟𝐶𝑚−1 . As suggested by [18,20], the incremental load Δpm stage (8) for the friction coefficient 𝜂 and the contact status de-
can determined by the load that is required to bring a node pair, which termined in stage (6) for the nearest node pair. With these in-
is nearest to the contact edge of current contact region, into contact and formation, all the cases with non-zero friction coefficient can be

52
V.T. Nguyen and C. Hwu Engineering Analysis with Boundary Elements 108 (2019) 49–59

performed directly from (3) to (8) by bringing the correct node 4.4. Iteration procedure
pairs into contact with the correct sign for the friction coefficient.
Based upon the boundary element method and the contact constraint
4.2. Criteria for contact status relations described in Sections 2 and 3, and the incremental loading
technique presented in this section, a flowchart for computer program-
Because the solutions obtained at each step are solved under certain ing is shown in Fig. 2. In order to clearly explain this flowchart, an
assumptions of contact modes, after getting the solutions we need to iteration procedure is stated as follows.
check the contact status of each node pair. Followings are the criteria Step 1: Input material properties, friction coefficient and make boundary
for checking the contact status. meshes
Separation: Like the usual BEM, input the material properties for two contact
( 𝑎 )𝑚 ( 𝑏 )𝑚 bodies and make boundary meshes for each of them. Note that indepen-
if Δ𝑢𝑛 + Δ𝑢𝑛 < 𝑔0𝑚−1 remains separate, otherwise change to contact .
dent nodes are set for two bodies. The friction coefficient 𝜂 is input at
(4.9a) this step.
Contact (slip/stick): Step 2: Set a potential contact region and input boundary conditions for
( )𝑚−1 ( 𝑎 )𝑚 each node
if 𝑡𝑎𝑛 + Δ𝑡𝑛 ≤ 0 remains contact, otherwise change to separate. If the contact region is known, the potential contact region is the ac-
(4.9b) tual contact region. If the contact region is unknown, a potential contact
region is suggested to be larger than the likely final contact region. Oth-
Stick:
erwise, it is possible that after completing the iteration process the final
( 𝑚−1 )
if |𝑡𝑚
𝑠
−1
+ Δ𝑡𝑚
𝑠 | < |𝜂 𝑡𝑛 + Δ𝑡𝑚
𝑛 | remains stick, otherwise change to slip.
solution is corresponding to a load pm which is less than the actual total
load p, and new iteration process should start again. The potential con-
(4.9c)
tact region (ΓC ) is generally divided into two parts: stick/slip (Γs ) and
Slip: separate (Γp ). Put ns and np node pairs for Γs and Γp , and nC =ns +np . For
(( ) ( )𝑚 )(( 𝑎 )𝑚 ( 𝑏 )𝑚 )
𝑚−1 the cases whose contact region is known, we set np =0. Initially, if the
if 𝑡𝑎𝑠 + Δ𝑡𝑎𝑠 𝑢𝑠 + 𝑢𝑠 ≤ 0 remains slip,
contact region is unknown all the boundary values are input based upon
otherwise change to stick . (4.9d) the assumption that only one node pair contact, i.e., ns =1 and np =nC -1.
An initial normal gap g0 and a local coordinate for each node pair are
4.3. Fast solver for iteration also calculated at this step.
Step 3: Calculate KA , KB , and KC , for the entire contact bodies
Note that the complete system of equations, (4.3) and (4.7), for solv- According to the definition given in (2.6) we can calculate KA , KB ,
ing the incremental displacements and tractions can be written in matrix and KC , which are related to the matrices of influence coefficients Y and
form as G, and will not change in the entire iteration process.
[ ]⎧Δ𝐱𝑚 ⎫ { } Step 4: Calculate 𝐃0𝐶 , 𝐟 , 𝐟0 or 𝐃𝑚
𝐶
, Δ𝐟𝑚 , 𝐟𝑐𝑚−1 for updating contact status
𝐊𝐴 𝐊𝐶 𝐊𝐵 ⎪ 𝐴 ⎪ Δ𝐟𝑚 To calculate 𝐃𝐶 , 𝐟 , 𝐟0 by (2.6), (3.1)-(3.4), an initial load p0 should
0
⎨Δ𝐱𝐶𝑚 ⎬ = , 𝑚 ≥ 1. (4.10)
𝟎 𝐃𝑚
𝐶
𝟎 ⎪ 𝑚⎪ 𝑚
𝐟𝐶 −1
be applied. We suggest p0 = p for the problems whose contact region
Δ𝐱
⎩ 𝐵⎭ is known. Whereas for the problems whose contact region is unknown,
From (4.10) we see that KA , KB , and KC keep unchanged in the entire an initial contact node pair is assumed and the load p0 is evaluated by
iteration process, only 𝐃𝑚 𝐶
, Δfm , and 𝐟𝐶𝑚−1 need to be updated before (4.8) with the two trial incremental loads. In the following examples,
𝑚 𝑚 𝑚
solving Δ𝐱𝐴 , Δ𝐱𝐵 , and Δ𝐱𝐶 in each step. Here, Δfm is due to the change these values are suggested to be Δ𝐩(1) = 0.003𝐩 and Δ𝐩(2) = 0.005𝐩 for
of incremental load, and 𝐃𝑚 𝐶
and 𝐟𝐶𝑚−1 are influenced by the changes of all updating steps m including the initial one. When m ≥ 1, the values
contact conditions and the solution obtained by the previous step. Com- of 𝐃𝑚 𝐶
, Δ𝐟𝑚 , 𝐟𝑐𝑚−1 can be calculated by (4.3)-(4.7) with the incremental
paring to the unchanged portion, these changes are usually only a small loads evaluated by (4.8). As stated in the nine stages of Section 4.1,
fraction of the total. To avoid reformulation of the entire system ma- the new edge of contact region and the sign of friction coefficient at
trix, all the changes in the left hand side of (4.10) have been purposely each iteration process are determined based upon the assumption of
arranged in the bottom part of the matrix. With this arrangement, a frictionless contact. This information can then be employed for the cases
fast solver suggested by [29] can be employed, which can be stated as with non-zero friction coefficient.
follows. Consider a system of linear equations which is written as Step 5: Solve xA ,xB ,xC or ΔxA ,ΔxB ,ΔxC for the displacements and trac-
𝐊𝑠 𝐱 = 𝐟 , 𝐊𝑠 = 𝐊 + 𝐞1 𝐯𝑇1 + ⋯ + 𝐞𝑠 𝐯𝑇𝑠 , (4.11) tions
When m = 0, solve xA ,xB ,xC by (2.7) and (3.4). Whereas when m ≥ 1,
where ei , i = 1,2,…, s, is a column vector which contains only one com- solve ΔxA ,ΔxB ,ΔxC by (4.10) using the fast solver introduced in (4.12).
ponent that is unity at the (di )th row and all the others are zero, and 𝐯𝑇𝑖 Step 6: Check the criteria for contact status and update contact status
is a row vector which contains the difference between the components If all the node pairs pass the criteria set in (4.9), output the solutions
of K and Ks in the (di )th row. Here, di denotes the row number of the of tractions and displacements for this iteration and go to the next step.
ith different row between K and Ks . Without repeating the evaluation of Otherwise, update contact status by (4.9) and go to step 4. Since the
K−1 in each iteration, which is usually the most time consuming part in contact solutions associated with the new contact status will be changed,
the entire calculation, the solution of x can be calculated as [29] the normal tractions associated with the two trial loads in (4.8) will also
𝐱 = 𝐊−1
𝑠 𝐟 = 𝐐𝑠 𝐊 𝐟 ,
−1
(4.12a) be changed. Thus, the incremental load Δpm needs to be updated by
where (4.8) before going back to step 4.
𝑠 Step 7: Check the total load and total contact area
∏ ( ) ( )( ) ( )
𝐐𝑠 = 𝐈 − 𝐂𝑠−𝑖 = 𝐈 − 𝐂𝑠−1 𝐈 − 𝐂𝑠−2 .... 𝐈 − 𝐂0 , (4.12b) If the magnitude of the total incremental load |pm | equals to the
𝑖=1 actual total load |p|, stop the iteration process and output all the infor-
and mation required in contact analysis. If |pm | < |p| and the contact area
𝐲𝑘𝑘−1 𝐯𝑇𝑘 is over the one set in step 2, which can be indicated by np =0, stop the
( )
𝐂𝑘−1 = , 𝐲𝑘𝑗−1 = 𝐈 − 𝐂𝑗−2 𝐲𝑘𝑗−2 , 𝐲𝑘0 = 𝐊−1 𝐞𝑘 , iteration and send a warning message to suggest the user to increase the
1 + 𝐯𝑇𝑘 𝐲𝑘𝑘−1 size of potential contact region and repeat running the program again.
𝑘 = 1, 2, ..., 𝑠, 𝑗 = 2, ...., 𝑘. (4.12c) In the case the potential contact region reaches its maximum value such

53
V.T. Nguyen and C. Hwu Engineering Analysis with Boundary Elements 108 (2019) 49–59

Fig. 2. Flowchart for computer programming of contact prob-

lems.

as complete indention of parabolic punch, the remaining load p − pm will

be applied on the system in the next step. If |pm | < |p| and the contact
area is less than the one set in step 2, go to the next step.
Step 8: Apply the incremental load
To apply the incremental load Δpm , we first set the node pairs (one
or two pairs depending on the status described in Section 4.1) nearest
to the edge of current contact region to slip mode. In order to obtain a
proper value of Δpm which can make the normal traction at the edge
of updated contact region vanish, the updated formula (4.8) is used and
two trial loads are required. In our program, an indicator q is set to
indicate the status of load pm- 1 . If the load pm- 1 is applied at the first
time, q = 0 and a small trial load Δ𝐩(1) is applied to get the data of 𝑡(1)
𝑛
and switch the indicator q to 1. When q = 1, another trial load Δ𝐩(2) is
applied to get the data of 𝑡(2)
𝑛 and switch the indicator q to 2. When q = 2,
apply the incremental load Δpm which is evaluated by (4.8), reset q = 0,
and update m to m + 1. No matter q = 0, 1 or 2, after getting the value
of Δ𝐩(1) , Δ𝐩(2) or Δpm , go to step 4 to update the contact status.

5. Numerical examples Fig. 3. A flat-ended punch in contact with an elastic foundation.

To show the versatility of the proposed contact BEM, several repre-

sentative examples are implemented such as a flat-ended or parabolic
punch on an elastic foundation; contact between two dissimilar elastic
cylinders; and a punch on an elastic foundation with a hole, crack or
inclusion. To verify the correctness of our proposed method, compari-
son is made by the results obtained from the commercial finite element
software ANSYS. Because the fundamental solutions used in our BEM is
for the generally anisotropic elastic materials, there will be one exam-
ple discussing the effects of anisotropic properties on contact behavior.
Moreover, we also consider the use of special fundamental solutions
dealing with the existence of holes, cracks, or inclusions, and discuss
their effects on contact.
Example 1. A flat-ended punch on an elastic foundation
Consider a flat-ended punch in contact with an elastic foundation.
The geometry, loading, and boundary meshes are shown in Fig. 3. The
vertical uniform load p = 1N/mm2 is applied on the top of punch. Both

54
V.T. Nguyen and C. Hwu Engineering Analysis with Boundary Elements 108 (2019) 49–59

Fig. 4. Tractions under the flat-ended elastic punch. Fig. 6. Traction evolution within contact region (loaded by vertical force only).

Fig. 5. A parabolic punch in contact with an elastic foundation.

Fig. 7. Influence of horizontal force on contact tractions.
of the punch and elastic foundation are made by an orthotropic material
whose properties expressed by the elastic stiffness Cij are

𝐶11 = 147.34 GPa, 𝐶22 = 𝐶33 = 10.78 GPa, 𝐶23 = 3.32 GPa,
𝐶12 = 𝐶13 = 4.23 GPa, 𝐶44 = 𝐶55 = 𝐶66 = 4.1 GPa. (5.1)

Since a flat-ended punch is considered, the contact region is known

and the total load can be applied directly at the initial stage without any
updating process for the incremental load. After the convergence test on
element meshes, in this example 141 nodes and 133 elements are used
in BEM, whereas 37,080 nodes and 36,794 elements are used in ANSYS.

Fig. 4 shows the results of tractions under the punch. From this figure
we see that the results for 𝜂 = 0.2 obtained from BEM are well matched
with those obtained by ANSYS. To know the contact status at different
nodes within the contact region, in this figure we also compare the nor-
malized tangential traction ts /𝜂p and normalized normal traction tn /p
for different frictional conditions. From this comparison we observe that
the contact region is separated into two parts. In the middle part of con-
tact region, |ts /𝜂p| < |tn /p| which is in stick status. Whereas outside
the stick zone, |ts /𝜂p| = |tn /p| which is in slip status. Moreover, when
the friction coefficient increases, the stick zone increases and slip zone
decreases.

Example 2. A parabolic punch on an elastic foundation

In this example, we replace the flat-ended punch considered in Fig. 8. Contact between two dissimilar elastic cylinders (R1 =R2 =5 cm).
Example 1 by a parabolic punch whose head profile is represented by

55
V.T. Nguyen and C. Hwu
Fig. 9. Influence of fiber orientation on contact pressure 𝜎 22 .
y = x2 /2R where R = 10 cm (see Fig. 5), and keep the elastic foundation
unchanged. The material properties of these two contact bodies are the
same as those shown in (5.1). At the beginning, only the vertical uni-
form load p = 172 N/mm2 applied on the top of the punch is considered.
The friction coefficient is considered to be 𝜂 = 0.2. Due to the change of
the punch head, unlike the flat-ended punch the contact region under
the parabolic punch cannot be known at the initial stage of iteration
process. In general, the contact region will increase gradually as the ap-
plied load increases, and the incremental load is applied step by step
using the formula proposed in (4.8). After a standard convergent test
for BEM, the total numbers of nodes and elements used for the present
problem are 166 and 158, respectively, in which 41 contact node pairs
are used to represent the contact region. Due to the symmetry of the
problem, only 20 incremental load steps (starting from p0 to p19 ) are
used to get the final solutions.
For the present example, the contact status needs to be updated in
each incremental load. Thus, it is interested to see the changes of contact
status through the traction evolution in each iteration. Fig. 6 shows the
56
Engineering Analysis with Boundary Elements 108 (2019) 49–59
results of traction evolution. The left-hand side of this figure shows that
the results obtained by BEM and ANSYS are in good agreement. The
evolution of the slip and stick zones in contact region can be observed
from the right-hand side of Fig. 6, in which only the results obtained
from BEM are presented. There is no slip zone inside the contact region
at the 5th load step (m = 4), while at the 10th and 15th load steps the
stick and slip zones are both increasing. At the 20th load step the slip
zone increases, whereas the stick zone decreases and is even smaller
than that of the 15th step.
To see the effects of frictional force, in addition to the vertical load
p we apply a horizontal force f on the left hand side of elastic punch
(see Fig. 5). Fig. 7 shows the distribution of contact tractions in the
presence of the horizontal force. Two different situation are considered:
one is f = 3 N/mm2 , and the other is f = 10 N/mm2 . The results without
horizontal force, i.e., f = 0, are also plotted in this figure as a reference.
From this figure we see that |tn | = |ts /𝜂| on the outer portion of contact
region, whereas |tn | > |ts /𝜂| at the inner portion. In other words, slip
mode occurs on the outer portion, and the stick mode occurs at the
inner portion. Moreover, the slip zone shifts from right to left when the
horizontal force is increased.
Example 3. Contact between two dissimilar elastic cylinders
Consider the frictionless contact between two dissimilar elas-
tic cylinders (see Fig. 8 with radii of cylinder R1 =R2 =5 cm, load
p = 6.786MN/m). The upper cylinder is made by an isotropic material
whose elastic properties are
𝐸 = 74 GPa, 𝜈= 0.3. (5.2)
As to the lower cylinder, it is made by a unidirectional fiber-
reinforced composite whose elastic properties are
𝐸11 = 134 GPa, 𝐸22 = 𝐸33 = 11 GPa, 𝐺12 = 𝐺13 = 5.84 GPa,
(5.3)
𝐺23 = 2.98 GPa, 𝜈12 = 𝜈13 = 0.3, 𝜈23 = 0.49.
In (5.2) and (5.3), E, G, and 𝜈 are, respectively, Young’s modulus,
shear modulus and Poisson’s ratio. The subscripts 1, 2 and 3 denote, re-
spectively, the direction of fiber orientation, transverse direction and the
direction normal to the cylinder plane. To study the effects of anisotropy
on contact behavior, the angle 𝛽 denoting the fiber orientation varies in
Fig. 10. Contact surface deformation u2 of the
lower cylinder whose 𝛽 = 45°.
V.T. Nguyen and C. Hwu
Fig. 11. A parabolic punch on an elastic foundation with an elliptical hole or
inclusion.
Fig. 12. Traction evolution within contact region for elastic foundation with
and without hole.
our example. To get a convergent result, 248 nodes and 244 elements are
used in BEM, and 11,293 nodes and 3765 elements are used in ANSYS.
Fig. 9 shows the distribution of contact pressure for several different
fiber orientation. From this figure we see that the results obtained by the
present method are in good agreement with those obtained by ANSYS
for the case with 𝛽 = 0°. Moreover, we observe that the contact area
decreases and the maximum contact pressure increases when the fiber
orientation angle 𝛽 increases. This is understandable since the material
we consider in (5.3) is softer in the transverse direction than in fiber
direction. In addition, we see that when the fiber orientation is 0° and
90°, the contact pressure distribution is symmetric with respect to x = 0.
Whereas, slightly unsymmetry occurs when 𝛽 = 45°, which can be seen
clearly by the contact surface deformation u2 shown in Fig. 10.
Example 4. A parabolic punch on an elastic foundation with holes, cracks
or inclusions
To study the effects of holes, cracks, or inclusions, we now consider
Example 2 by adding an elliptical inclusion into the elastic foundation
(see Fig. 11). In addition to the generally elastic inclusion, to have the
solutions of upper and lower bounds, the results for a hole and a rigid
inclusion are also presented to simulate the situation of very soft and
very hard inclusions. A crack can then be simulated by letting the minor
57
Engineering Analysis with Boundary Elements 108 (2019) 49–59
Fig 13. Influence of hole shape on contact traction (2a = 1 cm).
Fig. 14. Influence of hole orientation on contact surface deformation.
axis of the elliptical hole to zero, i.e., b = 0. If the inclusion is elastic, it is
considered to be isotropic as shown in (5.2). Note that in this example
the special fundamental solutions mentioned in Section 2 have been
used in BEM. With the special fundamental solution, which satisfies the
traction-free boundary conditions of holes or cracks, or the continuity
conditions of inclusions, no meshes are required on the boundary of
holes, cracks, or inclusions. And hence, the meshes used in this example
are exactly the same as those of Example 2.
To see the shape effects of holes, both of inclined angle 𝛼 and
minor axis b for the problems with holes can be varied. Table 1 shows
the load history for the case of 𝛼 = 0°. From this Table we see that to
reach the same contact region at each step, the load required for the
contact without hole is greater than that with hole, and the sequence
of magnitude is pm (without hole)> pm (with crack, b/a = 0) > pm (with
hole, b/a = 0.5) > pm (with hole, b/a = 1). Moreover, when m = 19
the load applied on the punch for the case without hole reaches the
value set for the problem, whereas for the case with holes or cracks
it is still lower than p = 172 N/mm2 . Under this situation, in general
we need to increase the potential contact region. However, in this
case the contact region cannot be increased since it has reached
the maximum set for the parabolic punch. Thus, the additional step
(m = 20) is done by adding the remainder, i.e., p-p19 . Fig. 12 shows
the traction convolution for the cases with (b/a = 0.5) and without
hole. To see the effects of hole shape, the contact tractions obtained
at the final step (m = 19 for the case without hole, and m = 20 for the
cases with hole) are presented in Fig. 13. From Fig. 12 we observe
V.T. Nguyen and C. Hwu Engineering Analysis with Boundary Elements 108 (2019) 49–59

Table 1
Load history vs. contact region.

Step m contact region (x /𝓁)∗ without hole with hole

2 2
pm (N/mm ) pm (N/mm ) pm (N/mm2 ) pm (N/mm2 )
(b/a = 0) (b/a = 0.5) (b/a = 1)

0 0.05 0.224 0.224 0.224 0.224

1 0.10 1.056 1.055 1.054 1.051
2 0.15 2.491 2.485 2.480 2.465
3 0.20 4.534 4.514 4.497 4.448
4 0.25 7.197 7.146 7.105 6.987
5 0.30 10.494 10.387 10.303 10.063
6 0.35 14.447 14.246 14.094 13.674
7 0.40 19.079 18.736 18.481 17.801
8 0.45 24.424 23.874 23.476 22.448
9 0.50 30.519 29.682 29.091 27.620
10 0.55 37.415 36.191 35.351 33.334
11 0.60 45.172 43.442 42.287 39.618
12 0.65 53.870 51.489 49.945 46.515
13 0.70 63.609 60.405 58.391 54.085
14 0.75 74.530 70.295 67.717 62.413
15 0.80 86.831 81.312 78.061 71.628
16 0.85 100.833 93.706 89.651 81.933
17 0.90 117.145 107.970 102.936 93.724
18 0.95 137.429 125.460 119.149 108.081
19 1.00 172.000 154.772 146.091 131.735
20 1.00 172.000 172.000 172.000
∗
Contact region: [-x, x] denoted by x/𝓁.

accuracy, to show the efficiency in this case the computational time is

147 s by our program coded by matlab, whereas we take 8901 s by AN-
SYS to get this comparison on Intel Core i7-7700 CPU 3.6 GHz computer
processor, which means that our method is around 60 times faster than
the commercial code.
To see the effects of inclusion properties, we now fix the size and
orientation of inclusion by letting b/a = 0.5, a = 0.5 cm, and 𝛼 = 0°.
Fig. 15 shows the contact surface deformation of the elastic foundation
for different inclusions (hole, elastic, or rigid). From this figure we see
that the deformations for rigid inclusion and hole become, respectively,
the lower and upper bounds for that of elastic inclusion.

6. Conclusions

By following the standard procedure of BEM, a system of linear equa-

Fig. 15. Influence of inclusion properties on contact surface deformation.
that the difference between the cases with and without hole becomes
apparent after m = 14, and the stick zone is smaller and slip zone is
larger for the cases with hole compared to those without hole. Fig. 13
shows that when the minor axis of hole decreases, the stick zone
increases and the slip zone decreases, and the extreme case (the hole re-
duces to a crack, b = 0) leads to the results closest to those without hole.
In other words, the crack does have the smallest influence on contact,
which is reasonable for a crack under pressure instead of tensile force.
When we keep the hole size and shape unchanged with b/a = 0.5
and a = 0.5 cm, and only vary the inclined angle 𝛼, the variation of con-
tact surface deformation of the elastic foundation on the contact region
is shown in Fig. 14. From this figure, we see that the indentation is
greater with the presence of hole and it decreases when the inclined
angle 𝛼 increases, which is consistent with our intuition. Moreover, for
the cases of slanted holes, 𝛼 = 30° and 𝛼 = 60°, the normal displacement
u2 is unsymmetric with respective to the central axis (x = 0), the abso-
lute value of u2 on the right hand side (x>0) is smaller than that of the
corresponding left hand side (x<0). Comparison is made by the results
of 𝛼 = 30°obtained by ANSYS, which is in good agreement. Besides the
tions for two anisotropic elastic contact bodies is shown in Section 2.
Since the number of equations in this system is less than the number
of unknown variables, more equations are needed which are provided
by the contact constraint relations shown in Section 3. Since the size of
contact region is usually unknown a priori, the entire contact problem is
generally nonlinear. To solve a nonlinear problem by a system of linear
equations, an incremental loading technique together with its iteration
procedure is proposed in Section 4. During the iteration procedure, we
need to determine the incremental load, predict the slip direction to
justify the sign ahead of friction coefficient, update the contact status,
and then solve the system of equations. Since contact region is usually a
small part of the entire problem, to avoid reassembly the whole system
of equations during the iterative process we employ fast solver for each
iteration, which save us a lot of computational time. Moreover, if the
contact bodies contain holes, cracks or inclusions we use a special BEM
whose fundamental solution satisfies the traction-free condition along
the hole/crack boundary, or satisfies the continuity condition along the
inclusion interface. With this special BEM, no meshes are required along
holes/cracks/inclusions boundaries and the solutions satisfy the bound-
ary conditions exactly, which makes the proposed BEM for contact prob-
lems much more accurate and efficient than the conventional one.
To demonstrate the accuracy and efficiency of the proposed method,
in Section 5 several representative numerical examples are implemented
and these results are compared with the finite element solutions ob-
tained from ANSYS. Through these examples we also study the effects

58
V.T. Nguyen and C. Hwu
of friction coefficient, horizontal force, fiber orientation, holes, cracks
or inclusions on contact. We observe: (1) when the friction coefficient
increases, the stick zone increases and slip zone decreases; (2) when
the horizontal force applied from left to right increases, the slip zone
shifts from right to left; (3) when the fiber orientation angle 𝛽 increases,
the contact area decreases and the maximum contact pressure increases,
also slightly unsymmetry occurs for 𝛽 ≠ 0°, 90°; (4) when a hole exists
in one of the contact bodies, the load required to reach the same con-
tact region is less than that without hole; (5) when the minor axis of
hole decreases (the extreme case is a crack normal to the applied load),
the stick zone increases and the slip zone decreases; (6) when the in-
clined angle 𝛼 of hole increases, the indentation decreases, also slightly
unsymmetry occurs for 𝛼 ≠ 0°, 90°; (7) contact surface deformation for
rigid inclusion and hole are, respectively, the lower and upper bounds
for that of elastic inclusion; etc.
Acknowledgment
The authors would like to thank Ministry of Science and Technology,
Taiwan, Republic of China, for support through Grant MOST 106–2221-
E-006–127-MY3.
References
[1] Muskhelishvili NI. Some basic problems of the mathematical theory of elasticity:
fundamental equations plane theory of elasticity torsion and bending. Netherlands:
Springer; 1977.
[2] Antipov YA, Arutyunyan NK. Contact problems of the theory of elasticity with fric-
tion and adhesion. J Appl Math Mech 1991;55(6):889–901.
[3] Vázquez J, Navarro C, Domínguez J. Analytical solution for a cylindrical contact
with reverse slip. J Strain Anal 2012;48(3):189–97.
[4] Liu H, Pan E. Indentation of a flat-ended cylinder over a transversely isotropic and
layered half-space with imperfect interfaces. Mech Mat 2018;118:62–73.
[5] Pan E, Liu H, Zhang Z. Vertical and torsional vibrations of a rigid circular disc on
a transversely isotropic and layered half-space with imperfect interfaces. Soil Dyn
Earthq Eng 2018;113:442–53.
[6] Willis JR. Hertzian contact of anisotropic bodies. J Mech Phys Solids
1966;14:163–76.
[7] Fan CW, Hwu C. Punch problems for an anisotropic elastic half-plane. ASME J Appl
Mech 1996;73:69–76.
[8] Hwu C, Fan CW. Contact problems of two dissimilar anisotropic elastic bodies. ASME
J Appl Mech 1998;65:580–7.
[9] Batra RC, Jiang W. Analytical solution of the contact problem of a rigid indenter
and an anisotropic linear elastic layer. Int J Solid Struct 2008;45:5814–30.
59
Engineering Analysis with Boundary Elements 108 (2019) 49–59
[10] Bagault C, Nélias D, Baietto MC, Ovaert TC. Contact analyses for anisotropic half-s-
pace coated with an anisotropic layer: effect of the anisotropy on the pressure dis-
tribution and contact area. Int J Solid Struct 2013;50:743–54.
[11] Alart P, Lebon F. Numerical study of a stratified composite coupling homogenization
and frictional contact. Math Comput Model 1998;28:273–86.
[12] Lovell M. Analysis of contact between transversely isotropic coated sur-
faces: development of stress and displacement relationships using fem. Wear
1998;214(2):165–74.
[13] Arakere NK, Knudsen E, Swanson GR, Duke G, Battista GH. Subsurface stress fields
in face-centered-cubic single-crystal anisotropic contacts. J Eng Gas Turbines Power
2005;128(4):879–88.
[14] Brockaerta H, Mazerainb PE, Rachikb M, Ho Ba Tho MC. Nanoindentation on
isotropic and anisotropic materials confrontation with fem simulations. Comput
Methods Biomech Biomed Eng 2009;12:63–4.
[15] Karin I, Niehuesbernd J, Bruder E, Lipp K, Hanselka H, Muller C. Finite-element
analysis of a rolling contact model with anisotropic elastic material properties. Mater
Werks 2013;44(4):298–303.
[16] Andersson T. Boundary elements in two-dimensional contact and friction. Linkoping
University; 1982.
[17] Paris F, Garrido JA. An incremental procedure for friction contact problems with the
boundary element method. Eng Anal Bound Elem 1989;6(4):202–13.
[18] Man KW, Aliabadi MH, Rooke DP. BEM frictional contact analysis: load increment
technique. Comp Struct 1993;47(6):893–905.
[19] Huesmann A, Kuhn G. Automatic load incrementation technique for plane elasto-
plastic frictional contact problems using boundary element method. Comp Struct
1995;56(5):733–44.
[20] Aliabadi MH, Martín D. Boundary element analysis of two-dimensional elastoplastic
contact problems. Eng Anal Bound Elem 1998;21(4):349–60.
[21] Gun H. Elasto-plastic static stress analysis of 3D contact problems with friction by
using the boundary element method. Eng Anal Bound Elem 2004;28:779–90.
[22] Sfantos GK, Aliabadi MH. A boundary element sensitivity formulation for con-
tact problems using the implicit differentiation method. Eng Anal Bound Elem
2006;30(1):22–30.
[23] Blázquez A, Mantic V, París F. Application of bem to generalized plane prob-
lems for anisotropic elastic materials in presence of contact. Eng Anal Bound Elem
2006;30(6):489–502.
[24] Tembleque LR, Buroni FC, Abascal R, Sáez A. 3D frictional contact of anisotropic
solids using BEM. Eur J Mech A-Solid 2011;30(2):95–104.
[25] Nguyen VT, Hwu C. A boundary element approach for indentation by rigid punches
on two-dimensional anisotropic elastic or viscoelastic solids. In: Proceeding of the
6th ACMFMS; 2018. p. 71–4.
[26] Sherman J, Morrison WJ. Adjustment of an inverse matrix corresponding to changes
in the elements of a given column or a given row of the original matrix. Ann Math
Stat 1949;20:621.
[27] Hwu C. Anisotropic elastic plates. New York: Springer; 2010.
[28] Brebbia CA, Telles JCF, Wrobel LC. Boundary element techniques: theory and appli-
cations in engineering. Berlin: Springer; 1984.
[29] Aliabadi MH. The boundary element method: applications in solids and structures.
Chichester: Wiley; 2002.