Académique Documents
Professionnel Documents
Culture Documents
2073-2099, 1998
Pergamon
PII
I~IIIUglII-7~3(9"/)001
- - - - - - ' - - - - - ' - - - 7 - - -1-0
1998ElsevierScienceLtd
Allrightsreserved.Printedin GreatBritain
0020-7683/'98 $19.00+ .00
DECOMPOSITION
OF THE MIXED, MODE
J-INTEGRAL--REVISITED
R. H. RIGBY
British Aerospace Airbus Ltd., Bristol, U.K.
M. H. ALIABADI*
Department of Engineering, University of London, Queen Mary College, Mile End,
London El 4NS, U.K.
E.mail: M.H.ALIABADI@QMW.AC.UK
(Received 29 September 1996; in revised form 6 June 1997)
Abstract--In this paper the correct decomposition of three-dimensional stress and displacement
fields for decoupling the mixed mode J-integral into mode I, II and III is presented. It is believed
that this has not been achieved before as certain components for the correct derivation have been
missing. It is shown that by using the correct decomposition method, a different area integral for
the mode II and III J-integrals is obtained. Several test examples are presented to demonstrate the
accuracy of the method. 1998 Elsevier Science Ltd.
1. INTRODUCTION
2074
contour to yield two parts of the J-integral : one comprising of symmetric elastic fields (js)
and the other comprising of anti-symmetric elastic fields (jAS). The integral j s is equal to
the mode I J-integral, whereas the integral jAS contains both the mode II, III J-integrals. The
decomposition method further decouples the mode II, III stresses, strains and derivatives of
strain. The stress intensity factors are then obtained directly from the mode I, II, III Jintegrals.
In this paper, the proper derivation of the decomposition method for mixed mode Jintegral in the three-dimensional boundary element method is presented. It is believed that
this has not been obtained before, as some of the components for a correct derivation have
been missing. The first of these is the proper decomposition of stress and strain into their
mode I, II and III constituents. The equation used in previous papers [see e.g. Nikishkov
and Atluri (1987b) ; Shivakumar and Raju (1990) ; Rigby and Aliabadi (1993) ; Huber et
al. (1993)] is shown to be incorrect. Another component is a different area integral for the
mode II and III J-integrals to that quoted in Rigby and Aliabadi (1993) and Huber et al.
(1993). These two components are significant for the mode II and III J-integrals derived
by the decomposition method.
2. THE J-INTEGRAL
Here the energy momentum tensor of Eshelby (1970) is obtained for elastic or nonlinear
elastic materials, and from this the three-dimensional J-integral is derived. The strain energy
density W in linear elastostatics is defined as :
W=
W(eij) =
i , j = 1,2,3
aijdeij
(1)
where % is the stress tensor and % is the strain tensor with components
e u = ~ (l u , , j - u j , i).
(2)
Here u,,j denotes the derivatives of displacements ui with respect to crack coordinates xj
shown in Fig. 1.
Differentiating the strain energy density W with respect to Xk gives
Xa
X!
2075
0W
Oeij
Oxk - Oxk
1 ?,,<,+o,,;~
= ~ trq It aXk
0 {
OooOu~
(3)
Ou~
Ou,,j
(4)
From equilibrium Oaq/Oxj = 0 ; so substituting eqn (4) into eqn (3) yields
aw
11-a /
a
au,'~
a [
au,'~-I
au,\
(5)
as oi~ = *ii by equilibrium of moments. Equation (5) can be rewritten in a more compact
form as
dPei
Oxj
Ox~
Wfikj- a,.j ~
-- 0
(6)
where 6ki is the Kronecker delta function and [ Wrkj-trq(Oui/axk)] is Eshelby's momentum
tensor and is denoted by Pkj [see Amestoy et ill (1981)]. All parameters in this tensor are
in terms of the crack coordinate system.
Consider a cross-section of a crack shown in Fig. 2. Integrating Pkjj over any area
in the plane x3 = 0, whilst excluding the crack singularity, gives
.II
C~
~(c-c,)
2076
R. H. Rigby and M. H.
Aliabadi
~(W~kj--aiJ~xk)d~=O
aui
(7)
fn~c- c~)Oxj
where f~(C-Ce) = Q(C)-~(C~) denotes the area delimited by the contours C, C~ and
crack surface V. According to Green's theorem
(8)
Since dx~ = - n 2 d F and dx2 = n~ dF (where n is the normal to the contour F) one obtains
from eqn (7)
fr ( Wn,
tu~ "~
c~ [
Ou~'~
(9)
where F is the contour around area t ~ ( C - Ce) and is, therefore, given by
r = c+c~+~.
(lO)
0 /
=-
Oui\
(11)
c3ui
J~(s)= fv~(Wnk-",J~xS nj) dF
=f +,(wnk
an,,
a.,,
(12)
jl(s)=fv.(Wn
' -a~Jff~x~nj)dF
Ou, \
Jck
-ru--nj|dr-
~x,
--|a,-3
Jo,o ~x3 k
~x, )
(13)
Consider contour F~ held constant. Then the fight-hand side o f e q n (13) is constant for any
contour C, i.e. the right-hand side is path-area independent. The path-area independency
of J1 (s) is limited to small regions around position s on the crack front. If parts of the path
are distant from the crack front, then J~(s) is influenced by the singular fields of the points
neighbouring s on the crack front. The J-integral is path-area independent in a global sense,
2077
i.e. the total strength of the singularity of the whole crack front is independent of the surface
enclosing it [see Shivakumar and Raju (1990)].
2.1. M i x e d mode J-integral
For each of the three modes of fracture there is a corresponding J-integral, J', jll and
j m and these are related to Jk as follows [see Cherepanov (1979)] :
Jl = J' + j n +jill
(14)
12 = -- 2 ~ / J IJn.
(15)
-4~n;)dr,
M= LII, III
(16)
GIII=~F,(i,3n
'
0/,/3
"~.~
(17)
a3j d~3j.
(18)
For linear elasticity, the relationship between Jt and the mode I, II and III stress
intensity factors can be obtained by substituting the three dimensional stress fields in the
Appendix into eqn (13). This yields:
Jl = j i .jr.j i I + jin
1
2
= ~--~(KI
+ K I2I ) + 1K21, '
(19)
where E* equals Young's modulus E for plane stress, E* = El(1 - v 2) for plane strain and
# is the shear modulus. Note that a plane stress or plane strain assumption is required to
obtain stress intensity factors from the J-integral.
Equations (14)-(19) can be used to obtain the mode I, II and III stress intensity factors
from J~, J2 and G m as follows :
KI =1
( , / J, -
_ : i +,/g, +
KI 1 = 1 / ~ ( ~ j I _ J 2 _ G I I I _ N / j
Kill = V/2gG m.
-6"')
I + J 2 - - G Ill)
(20)
(21)
(22)
However, the use of J2 leads to numerical difficulties as it will involve integration of singular
2078
elastic fields over the crack surface. Also Herrmann and Herrmann (1981) found that J: is
only path independent, in the same way as J~, i f the integral of Wn2 over the crack faces
vanishes (e.g. if the field stresses o"n and a22 are equal).
There is another approach which avoids the use of J2. The first step is to split the J~
integral into two parts :
J1 = j s + jAs
(23)
where j s is found from the symmetric elastic fields about the crack plane whereas jAs utilizes
the anti-symmetric elastic fields. This is demonstrated in the next section. The integral js is
equal to J~ as the mode I elastic fields are symmetric about the crack plane. This leaves
j~s = j ~ + jm, from which J" and jm are decoupled by the decomposition method (Section
2.3). Equation (19) is then used to calculate the mode I, II and III stress intensity factors.
2.2. Derivation of symmetric and anti-symmetric components
In this section the symmetric and anti-symmetric components of the J-integral in eqn
(23) are derived. These components are obtained from the symmetric and anti-symmetric
stresses and strains. Hence, these fields are presented first.
Consider points P(a, b, c) and P'(a, - b , c) which are symmetric about the crack plane
x2 = 0 (see Fig. 3). For any arbitrary deformation, the stresses at points P(a, b, c) and
P'(a,-b, c) can be expressed in terms of symmetric and anti-symmetric components as
follows :
"'l
and
(24)
0"22S
~
//UI 2S
%3~
G22AS
%5
~tl S
(~12As
G11AS
%#s #'#~
C
/
%
~2 S
--
uS
- XI
(723.~
A:
--~
%3~GIIS
/
~,3 s
~3 s
Symmetric components
(TI2AS
(711AS
{:733
AS
a,~s
Anfisyrnmetric components
olt~
OSl, r
~ 7S 21r
ol2rl
'l 3e" I
OSl3P,
o22," I
oS2v
o23e" I
2079
vl2P'
(25)
~231
~3r
0"33P" .)
where S and AS denote symmetric and anti-symmetric components. Therefore, the symmetric components of stress are given as :
"os~
as2
as3
oh
os3
0"$33
(26)
2 / 022e+O22r
23P~O23P"
1L 0O'33P
"Jr-O'33P'
and similarly
e0"i Ip-- 0"i iP
0"12p-~ Ol 2P
O'l 3 p - - O'l 3P
"=~
(27)
0 2 2 P - - 0"22P .
0"23P -~- 0"23P,
o~3s"
O'33P - - 0"33P
for anti-symmetric components. The strains are related to stresses by Hookes' law :
8iJ = ~
where 6q is the Kronecker delta and # the shear modulus. Thus the strains can be written
as
- S - - ~AS
611P--611P"
gl2e+gl2P"
~12P ~/~I2P" !
_lq",3P--",3e"
813p-~-813 P
22P "3U~22P
21
" t
e22P-e22r
~23pl~23P '
/ 823P-~-823 P'
~33P--33P"
The stresses and strains at point P' are now examined. Consider a function
x3) = ou(x,,
Then differentiating with respect to x2 at P', one obtains
x,).
(28)
2080
= &r,j(x,, -x2,
xO
= _ Ocr:j(x~, x : , x O
ko,
Oo',j
oo:, o<,]
_ [Oo';j
OXj (a,--b,c)
OX 2 ~-
OXl
OX 3 Jta, b, c)
= O.
0,~ s
c3xj - t3xj - 0
(29)
i.e. the symmetric and anti-symmetric stress fields exhibit equilibrium of forces.
The displacement derivatives OuS/Oxj and O~S/Oxj are now obtained from the strains
%. An example of this procedure is presented for O@S/Ox2. Starting with
AS I
I
~23 = ~(~23P"~ e23P") = 2 [e23 (Xl, X2, X3) "}-~23(Xl, -- X2, Xs)](a,b,c)
= ~(~3 +~3)
then
~f23(XI,X2,X3) ~-- F,23(Xl,X2,X3)l(a,_b.c)
=
Ox3
8x2
~ [_On2(Xl'--X2' X3)
j,o,-b,,,
OUs(XI,--X2,X3)" ]
Ox3
O(-x2)
J~o.b,,.>
_ 1 VOu'2(~,x:,x3) _ Ou's(x,,xz,xO1
2L
Ox3
Ox2
d(.,b,~)
and so
AS
1 VOu2
o,,;]
au3
Ox3
Ox2_k.,b,,.)"
The symmetric and anti-symmetric strains are related to the displacement derivatives by
and so
1 (0u~2S _1 0~3S~
= 2 \ 0x~
0x~ }
(3o)
2081
au,
aus
a~ s
Ox~ - O~ + dxj
au,
1
Ou,+
= ~" Ox,
du3
Oul
a.+,
du'+
ax2
au~
aul
axj
au'~
Ou3
Ou;
axj
axj
"Ou~
Oul
+~1
,.
(31)
Returning to the J-integral, eqn (13) can be written using the symmetric and antisymmetric components as
Jl
=fc[(f~+(es+~s)d(4+~S))n,
(32)
using the definition of W in eqn (1) and including the (zero) area integral for f~(C~). By
inspecting Fig. 3, the symmetric and anti-symmetric stresses a~j at P' are related to the
stresses a u at P as follows :
(33)
+/
_+S+l
+++s
.o.;%J
,,s+ j
.,,+s.
,
-~;+
-~,~ Ip
A~
--8~9
[
an
-,++3i
--
A~
~ ":t "-I I
(34)
2082
(35)
Consider contour C is symmetric about the crack plane x2 = 0. Then the normal n' at F is
related to the normal n at P as follows :
,M , Ou;N
(36)
OuN
(37)
(38)
where M, N = S or AS and a u, ~o, Ou~/~x~are values at point P and a;j, e;~, Ou~/dxn are
values at point P'. In eqns (36)-(38), the positive sign denotes the case of M = N while the
negative sign denotes the case M ~ N. Thus in the case of M # N, the integrands in eqn
(32) cancel each other at symmetric points about the crack plane. As the contour C is
symmetric about the crack plane, then eqn (32) reduces to
Jr= ~ f c ( Wun'
~Ou~ \
'
O z' u Ou~\
= j s + jAs
(39)
ny)dF
J, =f,.(Wn, -aU-~xOu,\
~ n,-o,~~n~)~.
C40~
(41)
where F = C + C~+7. It is now shown that this holds for each N. Applying Green's theorem
[eqn (8)] to eqn (41) one obtains, for N = S or AS,
;o~-q~ Lr ax,w,,
2083
0(
Oxj ~ Ox,)J
(42)
where area f ~ ( C - C,) contains no singularities and noting n3 = 0. From eqn (3)
(43)
as C = tr~j,..Also
a2~
a2~
(45)
fo~c-q~ Lr.Ox~
Oxj
fc(W~nNOU~V~
Ou~
as the contour integrand is zero along y as tractions ~ = a~un/ = 0 and n~ ffi 0. If the contour
C, is kept constant, then the left-hand side o f e q n (47) is constant for any symmetric contour
C. It is, therefore, path-area independent. It has already been noted that the integral over
area f~(C~) tends to zero as e --* 0. Therefore, one can obtain from eqn (47) the path-area
independent integral :
j~=fr,(W~n, -aouOu~/
\
~-~xnQdF
for N = S or AS. Here the contour F. is identical to the contour C. except that it pr~:eeds
2084
in an anti-clockwise direction. The mode I J-integral is given by js, whereas jAs is related
to both the mode II and III J-integrals :
jAS = j l l q_ j i l l .
The decomposition method for decoupling the mode II and III terms from the jAS integral
is now presented.
2.3. Decomposition method
In this method the stresses, strains and derivatives of strain are decoupled into their
mode I, II and III components. This enables the mode I, II and III J-integrals to be
obtained. The mode I stresses are the same as the symmetric stresses [eqn (26)], whereas
the anti-symmetric stresses in eqn (27) are split into mode II and III.
2.3.1. Re-evaluation of stress decomposition. The decomposition of stress has been
given by various authors [e.g. Nikishkov and Atluri (1987a) ; Shivakumar and Raju (1990) ;
Rigby and Aliabadi (1993) ; Huber et al. (1993)] as :
1
I1 - 111
0,ij = 0,ij -~- 0,ij ~ 0,ij
"0,1,+0,;,
0,12 --O"12
1
0,12 -'~-O"12
0,22 + 0,~2
1
+~-
0 ,
0"22 --0,22
1 GI3 - - 0,tl 3
-~-2"
0,23 --0,~3
0,23 +a~3
733 --0,33
In this section this equation is re-evaluated in the light of the three-dimensional stress
equations near the crack front given in the Appendix. It is found that the expressions
quoted for a~ and ojj
_,i are incorrect.
In the Appendix the singular stresses and the constant stresses in the vicinity of the
crack front are listed. It can be seen that the 0"33stress is zero for mode III in the vicinity of
the crack front and that the singular stresses are functions of sin(a0) and cos(b0) where a
and b are constants. Therefore, because of the properties of sin(a0) and cos(b0), one can
relate the stresses tr~j at P' to the stresses trij at P for each mode I, II or III. Thus, Table 1
is obtained from the Appendix.
Referring to Fig. 3, it can be seen how the stresses in Table 1 relate to the symmetric
and anti-symmetric stresses. For example, from Table 1 the mode I a'~2 at P' is equal to
minus the a,2 at P, i.e. mode I 0,'~2 = -0,12. From Fig. 3 the symmetric 0,s2 behaves in the
same fashion, i.e. ~rs2r = -aS2e. Therefore, try2 = as2 and similarly
Table 1. Relationship between stresses a,j at P and o~j at F for mode I, II and III
Mode I
Mode II
o~2 = --o~2
o~2 =a12
o~2 =o22
a~a = --o22
O33 = G33
G33 = --O33
Mode III
Non singular
~3 = 023
0~3 = -025 = 0
2085
I + ~}I + ~}IX
0-,j = 0-,j
":, , "
"~, "
0-~2
0-L
-"
.0"33
o "
0
~
"+"
0
(49)
""
~
0
The non-singular stress al3 behaves in the same way as as3, i.e. non-singular a~ 3 = al 3 and
as3r = as3e. The singular stress o13,
_in ~i] are both anti-symmetric and the non-singular
stresses 0"13,0"23are both symmetric. As the mode I stresses are considered to be symmetric,
then the non-singular a~3, a23 are included in mode I.
Therefore, from eqns (26), (27) and (49), the resulting stress decomposition is :
I
II
--I11
0-ij = a ij - ~ 0-ij + O ij
"all +a'~l
"~ll--U~l
0-12 ~ 0-'12
1 0"13"~ 0-'13
+~"
'' -o 30
(50)
~22--~22
0"23 -- 0-23
0'33 "~"0-~ 3
~33--~33
a23+a[3
0
It can be seen that the term a33-a33' has swapped from the e-lll expression to the a~jll
expression compared with the decomposition equation used in Nikishkov and Atluri
(1987b). An example of the effect of the incorrect decomposition equation can be seen in
e.g. Rigby and Aliabadi (1993), where the mode II and in particular mode III results for a
penny-shaped crack have been adversely affected (see e.g. decomposition results in Fig. 7
of that reference). Equation (50) enables the proper decomposition of the j^s integral into
its mode II and III components, and this reflected in the accurate penny-shaped crack
results presented in Section 4.
2.3.2. Strain and displacement components. The strains are related to stresses by
Hookes' Law:
(51)
where ~j is the Kronecker delta and # the shear modulus. The decomposition of the strains
can be achieved by simply inserting eqn (50) into eqn (51), for example
= ~(~,-~,~).
(52)
2086
/l/j .~- -1 ~ - 1 1 t
^ili
"~,, +~'~
0
0
~12 - - 8 / 1 2
~,3+eq3
~ +ei~
.+~.
t
823 - - 8 2 3
~33
813 --8113
(53)
0
t
[,. ~33 - - ~ 3 3 .
"3L 833 .
Ouf~
1 {Ou~
11
^111
?u,
(Ou, ou;)
2 \ax, + Ox,}
Ou~
Ouz OU'l
+ ~ + Ox2
OU I
OU 3
Ou'2
Ox,
OU"1
OX 1
Ou'~ + Oul
Ox2
Ou, + -Ou2
Ox,
(~Ut3
~+~+~+~
2 ~OX I
m q
=4
2 \Ox2
Ou2
3x2,]
Ou3
~ Ox~
2 \Ox2
cgul Ou;
Ox3 ~ ~x~
Ox,,]
0
(ou ou%
\ Ox 3 -~xd
2 \Ox3 + Ox3]
0
0
OU 1
On 3
1 -~x3 + Oxt
Ou'~
Ox3
Ou;
Ox,
Oul
Ou;
Ox3
Ox2
(54)
0
Ou2
Ou3
0
From eqn (54) the mode I, II, Ill derivatives of displacement are given as
~+~
Oui
1 Ou2
Ou'2
2 Ox~
Ou3
Oxj
Ou'3
~+~
Ou, Ou',
3x~
+2
Ox~
+-~x,
0
0
0
+~"
Ou3
Ox~
Ou;
Ox~
(55)
Ou'~
Ou2
gu'2
c9X2
-4
gu~
8u~
~X2
~X 2
c3u2
au'~
gu3 . gu'3
~u~
~Ui
OU2
OUt).
c~x3 - 2 "
dx3
dx3
c~u3
c9u'3
jl+
1tI1
+~
+~
au'
Ox3
jli_l - jill.
+~"
cgu'~
2087
c3u3 8u's
gx2
dx2
~U 1
~U/I
(56)
c3x3 c~x3
~u~ ~.
8u2
Ox3
(57)
Jl = j s + jAS
- ,j njJdr
+fr~I(f~qo~i2Sd(~iyS))n, -o~,.'sdU~i
] ~i j SnjjdF.
(58)
The symmetric fields are mode I, whereas the anti-symmetric fields are mode II plus mode
III. Therefore,
jAs
(59)
However, there are no cross-products between stress and strain for mode II and mode III
from eqns (50), (53) and (55), i.e.
offe~=0,
u au~
aiJ ~X~-~nj
M#N
= O, M ~ N
(60)
IlI
E
M=II
eil
--O'~'ii~_'_ n j
dE
(61)
2088
R . H . R i g b y a n d M. H. A l i a b a d i
0W
0e;j
Oxl - a Ox---~=
0.
(62)
fn (c-c~) ( ~xl
3 W O- eaiJff~Xl)
'
df~ =
0"
(63)
Wn t d F -
a;j ~x~ d ~ = 0.
(64)
(C - C0
This can be written in terms of symmetric (i.e. mode I) and anti-symmetric (i.e. mode
II + l i d components as
~(~S --!--gAS]
; c0l .O_ . _=
(65)
However, it has already been shown that products a~ d(e~) cancel out for M :# N and a
symmetric contour r [see eqn (36)]. Thus, eqn (65) reduces to
N= 1
WUnl d F - -
_U ~ge;j~,-,
(66)
WNnl d F - -
(C--Ce)
oij--u~-~ =
OXl
(C--C,)
k, 8Xl
i~xl]
= 0
(67)
as
_~ &~
axt - o;j ax-~'
OW ~
For N = AS, eqn (67) can be further decomposed into mode II and III components
as there are no cross-products between stress and strain for mode II and III. Therefore,
fvWMnl
dF--[
dn(C--C~) O/j
0.~71
= 0
M = I, II, III
(68)
holds for each mode M. Here F = C + C~+y (see Fig. 2). Concentrating on the area integral
in eqn (68) :
2089
~'"
02~ \
=
( c - c~)
a~ Ox, Ox----;
(69)
o [ ,~o,~.~ o a g o ~ + a ~ 0 ~
OX, t ail ~ X 1) = ~X, XO~
o [Mau?'~ Oa~O~
ox~V'~)-ox, Ox, + a ~
-
(7o)
OXl OX 1
O2u~
(71)
Ox20x~ "
Now from eqns (55) and (56) the following relationship is obtained:
o (o~)
0 t'0~)
(72)
0 (o.r)
~(o.,%
M = II, III.
(73)
Substituting eqns (70) and (71) into eqn (69) and using relationship (72), gives:
,c_~.,'j~x~"~' =
,~_c., ~
~a" ~ ) + ~
+
a,~
(c-c.) a~3xl~gx3
dn
kOx~ + Ox2]OxtJ
" (74)
~ - - ~ - +-&-U +-~7~3=o
(75)
= oX;
It" 3 1
)/121(
(,.,..III
[v32)
(~1313J
Ox,
Ox2
- -
Ox3
~o~, 0o~
3.,.
vv23
0x,
0x3
+ . . . .
0x2
(76)
(77)
2090
'~ Ill
00-32
Ox~
Ox2
00"113
(78)
8x3 "
O0"12
(79)
Ox~ + - ~ x 2 = 0
8
111
0"21
"~ 111
00"22 - -
(8o)
aa~1, 8a~I:
8x-T
=
(81)
au
ny dF +
fr (WMn,
M SuM \
ax, ]
~( 80-ff o0-M'~SuM
M a2uiM -] "~
-a,j _----n,/dF+ /
~ J a .
= O.
(83)
The contour F can be separated into contours C, C~ and 7. The contour integral is zero on
7 as n~ = 0 and tractions t M = 0-U
Mnj are assumed to be zero. The area integrand is of the
type
~0.i3 8Ui
IA --
8x3 ~x~
82Ui
-~-0.i3 - -
8x~ 8x3
noting eqns (76)-(81). The area integral over f~ will tend to zero as ~ ~ 0 as IA is similar
t o 8/~X3[ai3(SUi/SXl) ]. Replacing C~. by its anti-clockwise equivalent F~. and utilizing the
definition of j~t in eqn (16) yields
JM ~_fF,:(WMnl MSuM ~
=
Wgnl--a,~--nj/dF+
ax, /
t"
--JO(C)
M O2uY
jo(~kax, ~
M ~- I, II, III.
i2 oui
~dO
(84)
The area integral in eqn (84) is different from that quoted in previous papers [see e.g. Rigby
and Aliabadi (1993) ; Huber et al. (1993)]. The area integral has been quoted as
,o
t, " Ox, M
2091
"
c~u, ,,, 0 ~
0,30x---~~ ,,,=, ~,3 c~x---[
which would result in J~ # j l + j l i + jm.
It is noted from eqn (73) that 02U~i/OXI OX3 # 02U~i/OX30X 1 for M = II and III. The
term d2u~/OXlOx3 is obtained by differentiating Ou~]Ox3with respect to x~. Hence, from eqn
(57):
a (ou,)_ 2u,
Ox~\ax3] Ox, tx3
02ul + - 02u]
OXl OX3
OX10X 3
02U2
02U~
OXI~X 3
OXIOX 3
02U3
O2U~
1
~+~.
0
0
02u3
02u'3
Ox~Ox3 8x~Ox3
Ox~Ox~ OXlOX3
O2u~
1
+~"
02u]
Ox~Ox3 Ox~Ox3
02u2
02u~
Ox~Ox3 Ox~Ox3
0
..
(85)
The mode I, II and III J-integrals can be found from eqn (84). The stress intensity factors
can be related to the J-integral by using eqn (19)"
j = j i + j n + jill
= ~---(K~ + K~I) + ~1 K,,12
where E* equals Young's modulus E for equivalent plane stress and E* = El(1-v 2) for
equivalent plane strain and # is the shear modulus.
2.3.3. Path-area independence of JM. Path-area independence is investigated by integrating eqn (6) (with k = 1) over an area containing no singularities for each mode. This
integral is equal to zero for symmetric and anti-symmetric stress fields. Using Green's
theorem :
ox~ /
Jntc-c,)
(86)
2092
for M = I, II, III where F = C + C~+V (see Fig. 2) and f ~ ( C - C,) is the area bounded by F
and free from singularities. The area integral on the left-hand side becomes
02uf 1
(c-c~) ~r~20x~
Oxj Ox~
\cgx~Oxj o ~ , j
Ox~ Ox, l
(87)
I j = 0 and
using eqn (3) and a~ = a~. This integral is zero for mode I as au.
a2u~/Oxl Oxj = 02u~/Oxj Oxl [see eqns (50), (55)-(57)]. Thus, reordering the right-hand side
of eqn (86) :
i Ou~
)oo
(88)
as the contour integrand is zero on y for a traction free crack and n~ = 0. For constant C,,
the left-hand side is constant for any C. Hence the left-hand side is path-area independent.
Also the integrand over area fl(C~,) will tend to zero as noted before. Thus, one obtains
(89)
where J~ is path-area independent and the contour F,, is identical to contour C~ except that
it proceeds in the anti-clockwise direction.
For mode II and III, the area integral in eqn (87) becomes
Io
[ ( o2ui o2u
(C-C~) ~g \Ox, Ox3 ~x3 Ox,]
oOx,, Ou,
lda e o
Ox, j
(90)
as OZuy/Ox~ Oxj = OZuiM/OxjOxl f o r j = 1, 2, but not f o r j = 3 [see eqns (55)-(57)] and also
M ~ 0 for M = II, lII, as can be seen in eqns (50). This area integral is equal to the leftaijj
hand side of eqn (86) and so
~ / Mc?u~'~
kO~lTX3
Therefore
ax3ax,)
Oxs Oxl j
"
(91)
fF(
~x, )
2093
M i " O U i 2~ "u
~ f f OM~ i l O~ri~2 ' O U
Jn~c-q)
(92)
which is eqn (83). Splitting the contour integral into two parts
f l ~
~c~ \
Fl&r~
oai~ko~
O~uV q
;c,(
e ]ur
ox, I
jo,q,
Lt, Ox,
axe) a~,
.o ,ut, 1
(93)
as the contour integrand is zero on y for a traction free crack and n~ = 0. Consider C, held
constant. Then the left-hand side of eqn (93) is path-area independent as it has the same
value for any C. Also the area integrand is of the type
IA --
Oai3 ~Ui
02Ui
- +% - Ox3 -Ox~
Ox~Ox3
noting eqns (76)-(81). The area integral over fie will tend to zero as e --+ 0 as 1A is similar
to d/Ox3[ai3(Ou~/dxO].Therefore
n?dr
uXX') = fr Uu(X"x)tj(x)dF(x)-fr
Tu(X"x)uj(x) dr(x)"
(94)
The terms Uu(X', x) and Ti/(X', x) are the fundamental displacement and traction solutions,
respectively, for a point force in an infinite domain (i.e. Kelvin's solution), us and tj are the
known values of displacement and traction on the boundary F. The strain field throughout
the body may be obtained by differentiation of the above equation to yield (assuming zero
body forces) :
2094
fr U'J'~(X"x)tAx)dr'(x)- fr T,j,~(X',x)u~(x)dV(x)
(95)
where i, j range from 1 to 3 and x is a point on the surface F. The stresses at interior point
X' are given by
o-o(X" )
(96)
The area integrals defined in eqns (48) and (84) require two more boundary integral
equations to give O2uj/Ox~Ox3 and Oa3j/Ox3.As 02u~/Ox~Ox3 = O~uj/Ox30x~then this kernel is
obtained by differentiating eqn (95) (with k set to 1) with respect to x3 and Oa3j/Ox3is
obtained by differentiating eqn (96) (with i set to 3) with respect to x3. Hence :
(97)
(98)
u,3,(x')
and
4. TEST EXAMPLES
The test examples establish the accuracy and reliability of the proposed J-integral
technique for the embedded penny-shaped crack in an infinite domain, for which the exact
solution is available. Path-area independence of the J-integral is investigated.
4.1.
Inclinedpenny-shaped crack
A crack at the centre of a large cylinder of radius d and height 2h (see Fig. 4) was
modelled. The stress intensity factors for this configuration should be very close to those
of an inclined penny-shaped crack under remote tensile loading in an infinite elastic isotropic
body. The crack has a radius a and is inclined at an angle of ct to the xt axis of the cylinder.
The loading is a tensile stress of magnitude tr applied on the ends of the cylinder. The crack
length to cylinder radius ratio was chosen as a/d= 0.1 and the height to radius ratio as
hid = 1.5. The Poisson's ratio was chosen as v = 0.3.
(7
T~
1
cx
2h
(7
D e c o m p o s i t i o n o f the m i x e d - m o d e J - i n t e g r a l
2095
T a b l e 2. R e s u l t s f o r p e n n y - s h a p e d c r a c k , ~t = 0
Mesh density
r/a
K~I
/~1% error
K*
/q~m
Coarse
Coarse
Coarse
0.25
0.5
0.75
1,0126
0,9907
0,9812
1.3
-0.9
- 1.9
7.0 x 10 - s
1,4 x 10 -7
2.1 10 -7
4.7 x 10 - s
9.3 x 10 - s
1.5 10 -7
Fine
Fine
Fine
0.25
0.5
0.75
0.9927
0.9969
0.9922
-0.7
-0.3
-0.8
1,5 x 10 -6
1.6 10 -6
1.6 x 10 -6
2.7 x 10 -7
3.7 x 10 -7
5.1 x 10 -7
4.1.1,. Numerical results--non-inclined crack. For ~ = 0 both Kn and Km are zero and
K~I= 1.0 Where K~I= KI/Ko [see Tada et al. (1985)]. Here K~I has been normalized by the
stress intensity factor K0 for a penny-shaped crack in an infinite body :
Ko = 2 a v / ~ .
7Z
Two boundary element models were set up: the first model has two regions, 120
elements per region and 760 nodes and the second is a finer mesh with two regions, 168
elements per region and 1144 nodes. The J contour to crack radius ratios of r/a = 0.25, 0.5
and 0.75 were tried. The minimum element side length on the crack faces was 0.5a for the
coarse mesh and 0.25a for the fine mesh. The results for 0 = 30 are shown in Table 2.
The ~ results are very accurate for r/a = 0.25 and 0.5, and not much advantage is
gained by using a fine mesh rather than a coarse mesh. However, for r/a = 0.75 the accuracy
of the J-integral results are more sensitive to mesh density. The effect of the coarse mesh
on the accuracy of the results (which is most pronounced for r/a = 0.75) demonstrates that
a certain degree of mesh density is required to reasonably approximate the displacement
and traction fields around the crack front.
It can be seen that the path-area independency of the J-integral is maintained at various
radii. If the ratio r/a <~0.5 then good accuracy is obtained regardless o f fineness of mesh
employed. This indicates the robustness of the J-integral method. It is also noted that the
ratio of the area part of the J-integral to the contour part was approximately 0.067 for
r/a = 0.25 and 0.176 for r/a = 0.5.
4.1.2. Numerical results--inclined crack. Next the crack is inclined at an angle of 30
then 45 . For each configuration two mesh densities were employed o f the same density as
those used for the non-inclined crack. The J contour to crack radius ratios of r/a = 0.3, 0.5
and 0.75 were used for each r/a ratio. The maximum percentage error for the stress intensity
factors compared to the theoretical solution from Tada et al. (1985) is presented in Table
3 for the 45 configuration. The percentage errors for the 30 configuration are slightly
smaller.
T a b l e 3. P e r c e n t a g e e r r o r s f o r p e n n y - s h a p e d c r a c k , ct = 45
Decomposition % error
Mesh
Max K*
Max ~
Max K*I
density
r/a
% error
0 = 0
Coarse
Coarse
Coarse
0.3
0.5
0.75
1.2
-2.1
- 1.7
0.2
-2.2
-0.5
2.5
-1.3
2.6
Fine
Fine
Fine
0.3
0.5
0.75
- 1.3
-0.3
0.9
-0.7
-0.2
0.7
-0.2
2.1
9.5
0 --- 90
2096
1.2
Theory
o o o o o r z a = 0.3
ooooo r/et
= 0.5
zx~xLxzx r i o t
= 0.76
~0. 8
K,
bl
J
Eo
0.
0.
20
40
60
80
"
Angle along crack f r o n t ( 0 )
Fig. 5. Penny-shapedcrack at 30 to tensile stress, fine mesh.
0.8
Theory
00o00 r ~ a = 0.3
ooooo
r~a
= 0.5
0.2
0.~ r
2o
40
60
80
The normalized stress intensity factors K*, K*, K*~ are plotted against angle along
crack front 0 for the crack inclined at angles 30 and 45 in Figs 5 and 6, respectively. This
is for r/a = 0.3, 0.5 and 0.75 using a fine mesh. The K*, K*, K~z results for both the coarse
and fine meshes and r/a = 0.5 are plotted against 0 for ~ = 30 and 45 in Figs 7 and 8,
respectively.
By inspection of Table 3 and Figs 5 and 6 it is clear that the ratio r/a = 0.75 is too
large for accurate g l l I values. The contour is close to the centre of the penny-shaped crack
and so the J-integral will be influenced by other parts of the crack front as well as the region
of interest.
G o o d results were obtained for r/a = 0.3, 0.5 for both the coarse and fine meshes as
can be seen from Table 3 and Figs 5-8. This shows that the implementation of the J-integral
is robust for r/a <~0.5.
2097
1.2
Theory
~0.8
0.,4
0.~
.8 f
L
"~0.
'
~,~
Tho*'ll
Coown mesh
OOO00Fine
mesh
I .
i===|l
O. ~
20
40
60
80
An#re a ~ n # crack .~ont (e)
Fig. 8. Penny-shapedcrack at 45 to tensile stress, r/a = 0.5.
"
The ratio o f the area part of the J-integral to the contour part (Rc~/r) varies according
to the radius of the J contour, and generally increases as r/a increases. F o r r/a = 0.5 the
ratio Ra/r was 0.15, 0.11 and 0.285 for ji, j . and j m , respectively, and there was no
variation with 0 or with ~. However, Rn/r did vary with 0 for jAs with ranges
AS
Rn/r
= 0.11-0.285 for 0 = 0 to 90 , but did not vary with =. The total J-integral Ra/r varied
with both 0 and ~ and was
R~/tr = 0.14-0.19 for ~ = 30 and R~/tr = 0.13-0.225 for = 45
[
(0 = 0-90).
In conclusion the decomposition method gives good results for mode I, II and III
stress intensity factors. The graphs in Figs 5 and 6 show that the J-integrals follow closely
the theoretical stress intensity factors for the entire range 0 = 0--90 for both a = 30 and
45 . Better results are obtained with a finer mesh as would be expected (see Fig. 8), however,
the gain in accuracy in the stress intensity factors is not large.
2098
In this paper mode I, II and III stress intensity factors have been obtained using the
J-integral technique applied to three-dimensional boundary elements. Mode II and III
stress intensity factors have proved problematic in the past and so a rigorous derivation of
the decomposition of the J-integral into its constituent modes has been carried out. During
this derivation it was found that the decomposition equation used for stress by some authors
[see e.g. Nikishkov and Atluri (1987a) ; Shivakumar and Raju (1990) ; Rigby and Aliabadi
(1993) ; Huber et al. (1993)] was erroneous and that the area integral quoted previously for
mode II and III J-integrals was incorrect [see e.g. Rigby and Aliabadi (1993) ; Huber et al.
(1993)]. The path-area independency of mode I, II and III J-integrals was demonstrated.
That the decomposition method has been successfully implemented was demonstrated by
the penny-shaped crack example where the mode I, II and III stress intensity factors were
all obtained to a good order of accuracy.
It should be noted that the best results are obtained if the contour radius is kept as
small as possible (r/a <~ 0.5). This correlates with the observation that the J-integral should
be used only in the vicinity of the crack front in three dimensions. If the remote contour C,
used for the J-integral, is too large then it is influenced by other parts of the crack front
and not just the position of interest [see Shivakumar and Raju (1990)].
REFERENCES
Aliabadi, M. H. (1990) Evaluation of mixed-mode stress intensity factors using the path independent J-integrals.
In Boundary Elements XII, Vol. 1, Applications in Stress Analysis, Potential and Diffusion (eds) M. Tanaka et
al. CM Publications, pp. 281-292.
Aliabadi, M. H. and Rooke, D. P. (1991) Numerical Fracture Mechanics. Kluwer, Dordrecht and Computational
Mechanics Publications, Southampton.
Amestoy, M., Bui, H. D. and Labbens, R. (1981) On the definition of local path independent integrals in threedimensional crack problems. Mechanics Research Communications 8, 231-236.
Cherepanov, G. P. (1967) The propagation of cracks in a continuous medium. Journal of Applied Mathematics
and Mechanics 31, 503-512.
Cherepanov, G. P. (1979) Mechanics of Brittle Fracture, translated and edited by R. W. De Wit and W. C. Coody.
McGraw-Hill, New York.
Dodds, R. H. Jr and Read, D. T. (1990) Experimental and numerical studies of the J-integral for a surface flaw.
International Journal of Fracture 43, 47-67.
Eshelby, J. D. (1970) Inelastic Behaviour of Solids (eds) M. F. Kanninen et al. McGraw-Hill, New York, pp. 77115.
Hartranft, R. J. and Sih, G. C. (1969) The use of eigenfunction expansions in the general solution of threedimensional crack problems. Journal of Mathematics and Mechanics 19, 123-138.
Herrmann, A. G. and Herrmann, G. (1981) On energy release rates for a plane crack. Journal of Applied Mechanics,
A S M E 411, 525-528.
Huber, O., Nickel, J. and Kuhn, G. (1993) On the decomposition of the J-integral for 3D crack problems.
International Journal of Fracture 64, 339-348.
Ishikawa, H., Kitagawa, H. and Okamura, H. (1979) J-integral of a mixed mode crack and its application. In
Proceedings of the 3rd International Conference on Meehanieal Behaviour of Materials, 3, 447--455. Pergamon,
Oxford.
de Lorenzi, H. G. (1982) On energy release rate and the J-integral for 3-D crack configuration. International
Journal of Fracture 19, 183-192.
Murakami, T. and Sato, T. (1983) Three-dimensional J-integral calculations of part-through surface crack
problems. Computers and Structures 17, 731-736.
Nikishkov, G. P. and Atluri, S. N. (1987a) An equivalent domain integral method for computing crack-tip integral
parameters in non-elastic, thermomechanical fracture. Engineering Fracture Mechanics 26, 851-867.
Nikishkov, G. P. and Atluri, S. N. (1987b) Calculation of fracture mechanics parameters for an arbitrary threedimensional crack, by the 'equivalent domain integral' method. International Journal of Numerical Methods of
Engineering 24, 1801-1821.
Rice, J. R. (1968) A path independent integral and the approximate analysis of strain concentration by notches
and cracks. Journal of Applied Mechanics Transactions of the A S M E 35, 379-386.
Rigby, R. H. and Aliabadi, M. H. (1993) Mixed-mode J-integral method for analysis of 3D fracture problems
using BEM. Engineering Analysis with Boundary Elements 11,239-256.
Shivakumar, K. N. and Raju, I. S. (1990) An equivalent domain integral for three-dimensional mixed mode
fracture problems, NASA CR- 182021.
Tada, H., Paris, P. C. and Irwin, G. C. (1985) The Stress Analysis of Cracks Handbook, Second edn. Paris
Production Incorporation and Del Research Corporation.
APPENDIX
Hartranft and Sih (1969) used an asymptotic series expansion in three dimensions to obtain the elastic fields
along planes centred on points located on the crack front, The singular stresses at a point P Sl~eified by dimension
l (see Fig. 9) in a plane normal to the crack front are given as :
2099
front
""
KI(I)
0 I0
30-1 Ku(I) . 0 [0
30]
= - - c o s - / 1 - sin ~ sin ~ - ! - ~
sm ~ L2 + cos ~ cos ~ - J
v'~
2L
J 42,,r
gl(l )
0 [0
30q Kn(I )
0 0
a22 = ~ C O S ~ / 1
+ sin~sin-~-/+ ~sin~cos~cos~-
x/2nr
_1 x/2nr
VK,(I)
= zv - - - c o s ~
a'2
30
g,(t)
0
0
= ~sin~cos~cos-~-
3O
x/2~r
K,,(I) . 0-]
-- - - sln~-
Kil(t)
01-
~2~r
(A2)
(A3)
. 0.
+ ~cos~/1-sln~sln~-j
Kin(l). 0
~sm~
x/ 2nr
or,3
(A 1)
30q
(A4)
(A5)
and
Kin(l)
cr23 = ~ r n r cos ~.
(A6)
The next term in the expansion yields stresses which are constant with respect to r. In the crack coordinate
system these stresses are :
(711)2 = 2C
(a12)2 = 0
(~,3)2 = (A+B)
(a22)2 = 0
(a~3)2 = 0
(a33)2 = 2vC+ (1 + V)D
(AT)
where ( a h represents the second term in the series. A, B, C, D are functions of x3 and are determined from the
loading conditions of the problem under consideration.