See

discussions, stats, and author profiles for this publication at:

https://www.researchgate.net/publication/23577730

The line spring model for surface flaws

Article February 1972
Source: NTRS

CITATIONS

READS

32

55

1 author:
James R. Rice
Harvard University
361 PUBLICATIONS 43,821 CITATIONS
SEE PROFILE

All in-text references underlined in blue are linked to publications on ResearchGate,

letting you access and read them immediately.

Available from: James R. Rice

Retrieved on: 04 September 2016

Published in "THE SURFACE CRACK: Physical Problems

and Computational Solutions" (edited by J. L. Swedlow),
ASME, New York, 1972.

THE LINE SPRING MODEL FOR SURFACE FLAWS

James R. Rice
Division of Engineering
Brown University
Providence, Rhode Island

ABSTRACT

A model is discussed for the analysis of long part-through surface cracks

in the walls of plate or shell structures. Such problems are formulated within
the context of two dimensional plate and shell theory with the part-cracked
section represented as a line-spring in the middle surface. The spring allows
relative separations and rotations of the middle surface, and constitutive laws
relating these discontinuities to the prevailing force and moment per unit length
at any point are taken from the plane strain solution for a strip in combined
tension and bending, which contains an edge crack of a corresponding depth.
Prior work is reviewed and further line spring constitutive laws are discussed
as appropriate to elastic analysis with thermal or residual stresses and to
elastic-plastic analysis, with yielding in the ligament between the crack front
and far wall in the latter case.
INTRODUCTION

This paper is concerned with a model for the analysis of part-through surface cracks in the walls of plate or shell structures. It does not give exact
answers, although the approximations involved become increasingly more accurate
the greater the surface length of the flaw in comparison to plate or shell
thickness.
The virtue of the model is in its simplicity. It reduces prohibitively
complex three-dimensional stress analysis problems to two-dimensional problems
in plate and shell theory. Further, the model is readily incorporated within
any existing finite element program for plate or shell analysis. To date the
model has been employed only for the elastic analysis of surface flaws under
external loading, but perhaps its greatest potential lies with the ease by which
it may be extended to more complex cases. Indeed, the groundwork is laid here
for its extension to thermal stress analysis and to the elastic-plastic range,
with yield occurring over all or part of the ligament between the crack front
and the far surface of the plate or shell in the latter case.

171

Description of the Model

1
Rice and Levy (1) introduced the model whereby a surface flawed plate is
analyzed within the context of the two-dimensional generalized plane stress and
Kirchhoff-Poisson plate bending theory. The part cracked section is represented
as a line spring in the two-dimensional model, and the middle surface of the
plate on one side of the line spring may both displace and rotate relative to
'the middle surface on the other side. Of course, the magnitudes of the separation displacement o and rotation discontinuity 9 at any point along the line
spring are considered to be functions of the force N and moment M per unit
length transmitted across the spring at that point.
The functional relation between 6,9 and N,M at any given point of the
line spring, corresponding to a location along the surface flawed section at
which the crack depth is t, is taken from the plane strain solution for an
infinite strip of thickness h (corresponding to the plate thickness) which contains an edge crack of the same depth t, and which is subjected to an axial
force Nand bending moment M, both per unit length in the direction of plane
strain constraint. o and e are then identified as the ad~onat extension and
rotation of one end of the strip relative to the other due to the presence of
the crack. In this way the line spring represents in lumped form the additional
localized ~ompliance of the surface cracked plate. It is clear that the surface crack problem would reduce to the plane strain problem for the edge cracked
strip in the limit of a very long surface flaw of essentially constant depth,
and it is because of this feature that the model may be considered increasingly
more accurate the greater the surface length of the crack in comparison to
plate thickness and the more gradual the changes in depth t along the profile
of the crack.
Once the relation between o,e and N,M is established at each point along
the line spring, the stress analysis problem is solved within the context of
the two dimensional theory of stretching and bending of a plate subject to
boundary conditions appropriate to the mode of external loading and to additional boundary conditions along thv spring consisting of this relation between o,
e and N,M. It is, in fact, this latter boundary condition which couples the
stretching and bending fields in the plate because a force N per unit length
induces a rotation as well as a separation of the line spring, just as a
moment M induces a separation as well as a rotation. The net result of the
analysis is that o ,e and N,M are determined at each point along the line spring.
From these the severity of conditions as regards crack extension at any point
of the spring corresponding to crack depth t is judged to be the same as for
the strip in plane strain, containing a crack of the same depth and subjected
to the same loads N and M. Hence in the linear elastic case the crack tip
stress intensity factor at each point along the surface flaw front is estimated
from the known solution for the intensity factor of the edge cracked strip.
The description of the model has been developed in a sufficiently general
form to encompass its proposed extension here to cases of thermal and residual
stress states and to the elastic-plastic range. However, work to date has
dealt exclusively with the elastic version of the model for which eq. (7), to
follow, gives the required relation between 6,9 and N,M. The description may
be summarized with reference to Fig. 1.
Fig. la shows the profile of a surface crack in a plate, and Fig. lb shows
the corresponding two-dimensional plate model, here supposed to be subjected to
boundary loads N*, M* and to contain the line spring of length 2a at the part
cracked section. If u (x,y), u (x,y), u (x,y) denote displacements of the plate
middle surface referre~ to righ* handed ~.y,z coordinate system with the z axis
perpendicular to the middle surface, then
1underlined numbers in parentheses denote references listed at the end of the
text.
172

,._~--2.a
(a)

Fig. 1 Description of Line Spring Model

( 1)

These are related by the line spring constitutive law to

(2)

M(x) = Myy(x,O)

where

+h/2

NaS(x,y) =

+h/2

MaS(x,y)

'as(x,y,z)dz

-h/2

z-raS(x,y ,z)dz

(3)

-h/2

Here 'ij(x,y,z) denotes the stress state in the three dimensional cracked body
and a,S = x,y. This constitutive law is dependent on the crack depth t{x) at
the point under consideration and it is determined from the plane strain solution for the edge cracked strip of Fig. lc. Hence, if the plane strain solution for the additional extension o and rotation e of one end of the strip
relative to the other, due to the presence of the crack, is supposed to take
the general functional form
o

e = G{N,M;.t)

F{N,M;.t)

(4)

then the discontinuity relations imposed as boundary conditions along the line
spring are
uy{x,O+)-uy(x,O-)

F[NYY(x,O), MYY(x,O); .e.(x)]

auz{x,O-)/ ay- auz(x,O+)/ ay

(S)

G[NYY(x,O), MYY(x,O); .t(x)]

The mathematical problem then involves solving the equations of plate extension
and bending subject to these and to the boundary conditions of external loading,
so that the distribution of N(x), M(x) along the part-cracked section is determined.
Prior Work
A primary aim in the present work is to examine forms for the line spring

173

constitutive laws of eqs . (4). In the linear elastic case Rice and Levy (l)
have shown that they may be determined directly from an expr2ssion for thecrack tip stress intensity factor K (defined so that (l- v2 )K /E is the energy
release rate). K may be written in the form
K = hl/2[ cr gt(;) + m gb(;)]
where

; =

t/h ,

= N/h

(6)

m=

The dimensionless functions g (;), g ( ; ) have been determined by Gross and

Srawley (2) through a boundarj collo~ation solution thought to be accurate up
to ; : 0.7, and their results have been adopted for the particular formulae
derived by Rice and Levy . The resulting elastic constitutive laws are

Eo
Ee

2(l- v2 ) h[crtt(;)cr + crtb( ; )m]

l2(l- v2 )

(7)

[crbt(;)cr + abb(;)m]

and the compliance coefficients aA~ are given by

;
g (;") g (;")d ; "
A,~
=
aA~(~)
A
~

t,b

(8)

The problem depicted in Fig. lb was reduced in (l) to two coupled integral
equations for the nominal tensile and bending stresses cr(x) and m(x) transmitted
across the line spring, under the assumption of an effectively infinite plate.
These were solved numerically for crack profiles of semi-elliptical shape,
and stress intensity factors at the crack mid-section were determined for both
remote tension and bending loads. The results demonstrate quite substantial
reductions of K from the values for plane strain edge cracks of the same depth
as at the mid-section of the semi-elliptical surface cracks. For example, when
the mid-section depth is h/2, the surface length 2a must be greater than approximately 35h forK to be within 20% of the plane strain value. Somewhat
surprisingly, the results for remote tensile loading were within a few percent
of the presumably more accurate three dimensional analysis of Smith and Alavi
(l) for 2a/h in the neighborhood of unity. However, Prof. C. W. Smith of
Virginia Polytechnic Institute has suggested in a private communication (Jan.
1972) that substantially larger values of 2a/h may be necessary for accurate
predictions with remote bending loads. Bending is further complicated by the
comparitively larger stress concentrations expected in the regions near to
where the crack front meets the plate surface, and the model cannot be expected
to give accurate results in these regions.
The elastic line spring model has been further extended to part-through
surface cracks in shells by Levy and Rice (4). The formulation is identical
to that illustrated in Fig. 1 except that now Fig. lb is replaced by the appropriate shell middle surface. Again, the line spring constitutive laws are
taken from the edge cracked strip in plane strain, as given in eqs. (7).
Solutions were given in closed form for complete circumferential cracks and
long axial cracks of constant depth extending part-through the wall of a
pressurized and axially loaded cylindrical shell. In the case of the axially
loaded shell with a complete circumferential part-through crack, the restraint
against rotation about the line spring afforded by shell curvature is found to
substantially reduce K from the corresponding plane strain value. When the
crack depth is h/2, the shell radius R must exceed approximately 50h to be
within 20 % of the plane strain K. In addition, a finite element formulation
has been presented for the line spring model in (4), in a form which can be
merged with any existing finite element program for plate or shell analysis.
This involves representing the spring as a collection of line finite elements
taking displacements and rotations on either side in accord with those of the
bordering plate or shell elements. Contributions of the line spring to the
overall stiffness matrix are then obtained by adding a term of the form

174

+a

f-a

[N

ao / a~

M ae;a~ n ] dx

to the virtual work statement of equilibrium under an arbitrary variation of

any nodal parameter~ . Here o and e are linear functions of the nodal
parameters describingnthe bordering elements, and N, Mare determined as linear
functions of o,e and hence of the nodal parameters by inverting eqs. (7).
One difficulty with the finite element formulation is that for sufficiently
deep cracks the matrix of coefficients in eq. (7) is nearly singular. Physically, this means that the spring deforms almost as if it was rigidly hinged about
a point in the uncracked ligament, unless very large forces are applied with a
line of action through the hinge point. Hence small inaccuracies in the numerical predictions of o,e can cause quite large inaccuracies in the values of
N, M computed from these displacements. In this case it is best to estimate
N, Mfrom the stress states created in the bordering elements. This finite
element formulation has been employed in conjunction with the MARC program to
study the influence of surface crack shape (rectangular, semi-elliptical,
segment of circular arc) on K for plates and also to estimate K for cylindrical
shells containing part-through cracks of limited surface length.
The following portions of the paper will deal with the formulation of
line spring constitutive laws for cases beyond those covered by eqs. (7) and,
in much less detail, with methods of stress analysis with these constitutive
laws.
ELASTIC ANALYSIS WITH THERMAL OR RESIDUAL STRESSES
Here we continue to suppose that the material of the plate or shell is
elastic but that it is subjected to thermal or residual stresses which would
cause deformation when a crack is introduced even if no net force or moment
acted across the part-cracked section. We consider stresses of a kind which,
as they exist in an uncracked body, would transmit only a normal stress (varying
with distance through the thickness) across the prospective crack surface.
Thus with reference to the strip in plane strain, the stress state existing
before a crack is introduced may be divided (Fig. 2a) into a nominal tensile

mh-

--- h ---

~M
(b)

ca>

Fig. 2 Thermal or Residual Stresses

175

stress a equilibrating N = ah, a nominal bending stress distribution of surface

magnitude m equilibrating M = mh 2/6, and a remaining stress distribution p(~')
of thermal or residual origin, where ~'h is the depth from the surface.
We wish to compute the additional extension o and rotation 8 of one end of
the strip relative to the other when the crack is introduced and it is perhaps
clear that the result will be given by eqs. (7) with the modification that p(~')
will cause a Tq~tribution to o and 8 which remains even when a and m vanish.
Let g(~' .~)h- I be the stress intensity factor induced by unit point loads
wedging open the crack at c' when the crack length is ~;;. Then since the stress
intensity factor due to introducing the crack can evidently be computed from
the load system acting in Fig. 2b,
I;
(9)
K = hl/2[ agt( c) + m gb(l;) +
p(!; ' ) g(E;',I;) dE;']

The weight function g(l;' ,1;) is not known, but many of its properties are known:
since K due to a and m could have been obtained by superposing an opening stress
distribution a+ (l-2 ~ ')m on the crack surface
I;

gt( c)

f g( E; ' , 1; ) dE; '

I;

gb( c) =

J g( c' .c) (l-2c') de'

(10)

Further, if the load point is very near the crack tip the resulting K cannot
depend on overall specimen dimensions but rather is the same as for a semiinfinite crack in an infinite body and hence g becomes infinite as
(ll)

These would allow an approximation to g to be constructed in the form

g(E;' , 1; ) : (2/ n) 112 (c- c' f 1/ 2 [l + A(i;)(E;-1;') + 8( 1; )(1;-1;' ) 2]
(12)
with A(i;), 8(1;) determined from eqs. (10) in terms of g ( 1; ), g (1;). Actually,
however, the work of determining g has already been don~ even 9f the information is not readily available. Rice (5) has shown recently that g can be
constructed from the expressions for crac k surface displacements and stress
intensity factor for any one load system acting on the same cracked body , a
result which is closely related to 8ueckner's (6) theory of the weight function.
In particular, if the total opening displacement between the upper and lower
crack surfaces at 1; ' due to the tensile or the bending loads is written as
f'.Uy = 2[(1-\) 2)/E] ah Ut( l; ' ,1;) , or f'.Uy = 2[(1- \)2)/E] mh Ub(l;' , 1; )

(13)

then the weight function is given by

( 14)

That is, it can be computed either from the tension or bending solution and
both must, of course, give the same result . The functions U and U were
determined by Gross and Srawley in their collocation solutio~ but nBt directly
given since it is only recently that their real significance has become clear .
Thus we may assume that the weight function will become known and procede to
compute the contribution of p( i; ') too and 8 in terms of it.
Energy Release Rates and Compliance Changes
It is simplest to develop the relation between the expression for K and
compliance changes in general terms. Thus assume that the strip of Fig. 2b
is subjected to various generalized forces Q., each representing some load
system symmetrical about the crack line, and 1 let q. be the corresponding work
1

176

absorbing displacements.
strip,
{Qi}

={ a ,

Specifically, for the load systems acting on the

( 15)

m, p( t; ') dt; '}

the last entry indicating an infinite number of infinitesimal forces acting

along the crack surface, and from the requirement that Qidqi be a work increment,
( 16)
Here each qi is measured from zero when no crack is present and 6uy(t;') is the
total opening between upper and lower crack surfaces at t;'. Eq. (9) for the
stress intensity factor appropriate to a crack of length t=t;h may be put in
the form
1 2
K = Qiki(t) , {ki(i)} = h 1 {gt( t;) , gb(t;), g(t;',t;)}
(17)
The load-displacement relations take the form
qi = cij(t) Qj

'

where

(18)

cij(t) = cji (t)

Hence the Irwin relation between energy release rates and compliance
changes,
(19)

becomes, by virtue of the two prior equations,

[(l-v 2 )/E] QiQj ki(t) kj(t) = (l/2) QiQj dCij(t)/dt

(20)

From these it is clear that the compliance coefficients are given by

i
C. . (i) = 2[(1-v 2 )/E]
k.(t") k.(t") dl"

(21)

1J

Hence, from the identifications of {Q.}, {q.}, and {ki(t)} given in eqs.
(15,16,17) it is clear that eq. (18) gives 1 for t~e line spring constitutive
relations
t;

Eo

2(1-v )h [att(t;)o+atb(t;)m +

at(t;',t;)p( t; ') dt; ']

(22)
t;

Ee = 12(1-v 2 )

[abt(t;)o+abb(t;)m +

ab(t;',t;)p(t;') dt;']

where att(t;), atb(t;), abt(t;), abb(t;) are given by eqs. (8) and where
t;

aA (t;' ,t;)

J gA (t;") g(t;', t;") dt;"

A = t, b

(23)

t;'
(A similar equation may also be given for 6u (t;'), but this is of no interest
for our present purposes.) In view of eqs. {13) for the weight function, we
may now make th~ simplification
A

= t,b

(24)

a result which may alternately have been deduced from the elastic reciprocal
theorem.
177

Formulation of Stress Analysis Problems

Eqs. (22-24) and (8) complete the formulation. Further direct progress
in applications cannot be made until the weight function g( ~ ' .~) is determined
or, as is the same, until prior work of the type in (2) is reviewed for a
determination of at least one of the functions Ut( ~ ' .~) or Ub(~' .~) giving the
. separation displacements of the crack surface due to tensile or bending loads
on the edge cracked strip.
However, the general procedure is now clear: When thermal or residual
stresses act we must first determine their part p( ~ ') which remains after the
net tension and bending parts have been subtracted away. From p(~') we determine the last terms in eqs. (22) which give a contribution to o and e analogous
to a thermal strain which must exist even when a and m vanish. We then solve
the plate stretching and bending equations subject to the boundary conditions of
eqs. (22) along the line spring, with the understanding that the plate stretching
and bending constitutive laws employed in the equations are to include any
thermal straining contributions to the middle surface stretch and curvature.
Once the values of a and mare determined along the line spring, the elastic
stress intensity factor along the part-through crack is determined from eq. (9).
This procedure would, for example, be applicable to determining the stress
intensity factor induced along a surface crack by the thermal shock in sudden
cooling of a hot pressure vessel wall.
ELASTIC-PLASTIC LINE SPRING
The constructional materials employed for pressure vessels and piping are
often so ductile that it must be assumed that all or part of the ligament between the part-through crack front and far plate or shell wall has passed into
the plastic range when fracture propagation conditions are met. Hence, it is
important to have an elastic-plastic line spring model. A suitable general
procedure follows the description of the line spring model in the introduction
section. Eqs. (4) are viewed as functional relations giving o and e in terms
of the history of N and M, as determined from an elastic-plastic stress analysis
of the edge cracked strip in plane strain. These serve as boundary conditions
on the plate or shell stretching and bending problem, to be formulated in terms
of the appropriate elastic-plastic plate or shell equations.
However, this general procedure lacks adequate simplicity and we examine
here a simpler formulation suitable for cases in which the material away from
the part-cracked section remains elastic and in which the line-spring may be
considered perfectly plastic at least when the uncracked ligament has entered
into the fully plastic range.
Elastic-Perfectly Plastic Model
Suppose that the result of a plastic limit analysis for the edge cracked
strip in plane strain (Fig. lc) is available in the form
cp(N, M; .e.)

=0

(25)

For the simplest formulation, the strip is idealized as elastic when Nand M
lie within the yield surface thus defined (cp < 0) . That is, the line spring
constitutive laws are taken to be of the elastic-perfectly plastic type, so
that eqs. (7) relate do, de to dN, dM for all load variations within the elastic
range. ~owevpr, when the yield condition is met there will be plastic increments do , de as well and these will be related to the yield function by the
normality rule
(26)

with the multiplier determined in terms of do , de in the following manner.

178

Again adopting the language of generalized forces and displacements, with

{Q . } = {N, M} and {q.} = {6,6}, the invented form of eqs. (7) may be written in
th~ stiffness form (~ee (i) for the actual expressions):
(27)

Hence the elastic-plastic constitutive law is

(28)

We solve for dA by requiring that when it is not zero the resulting load increments satisfy d~ = 0. The resulting incremental constitutive law is

- [

2..L (2..L
aQ

dQi - Eij dqj - aQ.

J

lt.._)-1 2..L
aQ

El~ aQ

l.l

Ea~dq~]

(29)

Of course the term involving ~ is deleted in the elastic range or when ~ 0

but the imposed {dqi} is such that dA would be predicted negative by the formula.
The combined tension-bending yield function ~must be known to give the
specific version of this law for the line spring. It is not known as of yet,
at least for the N/M ranges of greatest interest, although it is straightforward
to derive an outer appro xi mati on to it by the upper bound theorem of 1i mit
analysis. This is done in the last part of this section.
Some Further Refinements
There will be stress concentrations at the ends of the line spring and
hence it is possible in many cases that yield zones will exist there even
though stresses far from the part cracked section are still in the elastic
range. It is important that account be taken of the existence of these yield
zones and a suitably simple way of doing so is through a Dugdale-Barenblatt type
yield model. Hence, the line spring is supposed to extend beyond the actual
surface length of the flaw so that the distance 2a in Fig. lb is taken to include both the part cracked section and the unknown prolongation of the DugdaleBarenblatt line plastic zone.
The plastic zones are described by constitutive laws of the line spring
kind corresponding to crack depth t = 0. Hence there is no elastic contribution to the discontinuities 6, 6 across the plastic zones, and 6,6 are there
given by rigid-plastic equations of the form (26) based on the special form of
the yield function ~ for the uncracked strip. The extent of these plastic zones
is determined from the condition that the yield condition not be violated by
stresses existing directly ahead of their tips, and in particular that the net
stress singularity there due to the applied loads and to the restraining load
distributions N(x), M(x) along the line spring must vanish. As of yet there
seems to be no experience with the Dugdale-Barenblatt model under combined
tension and bending. Analysis is considerably more difficult for this case,
except for very special circumstances which will lead to constant values of
Nand Mat all points within the line plastic zones. These cannot be expected
to prevail in the application of the model to plastic zones at the ends of a
part-cracked section.
The factor most limiting to the accuracy of the model (given a sufficient
2a/h ratio for applicability of the line spring concept) will evidently be
the neglect of deviations from linearity due to contained yielding prior to the
fully plastic range. Yield begins at a natural crack tip at essentially negligible load levels, although substantial deviations from linearity occur only as
the limit load is approached. It is not yet clear as to how this can be incorporated into the model. One possibility to examine would be the formulation
of isotropic hardening constitutive relations for the line spring based of the
attainment of values of ~(fl, M; t) less than the limit load value of zero.

179

These will however not be accurate for significant deviations from proportional
loading. Another possibility is to introduce the effective crack depth correction whereby the crack is considered deeper by an amount proportional to (K/To) 2 ,
where T 0 is the yield stress in shear and K is the stress intensity factor.
This too has obvious drawbacks with highly non-proportional loading, and also
for deep cracks it may be that the effective crack depth exceeds the wall thickness. However, for small scale yielding this approach may enable a rational
definition of ~ for use in the early stages of an isotropic hardening formulation.
Limit Analysis of Edge Cracked Strip
McClintock (7) has recently summarized limit load solutions for cracked
bodies in plane strain. Two of the solutions, namely the Green and Hundy solution for pure bending and the solution for pure extension, are summarized in
Figs. 3a and b respectively. For tensile loads on surface flawed elements, 1t
can generally be assumed that both 6 and N will be positive along the line spring.
Hence the most important cases will lie between these two extremes.

~12.
, ... -

..

'----:-r.:-M-:--'

(a)

N:O

(d)

Fig. 3 Limit Analysis of Strip

The pure bending (N=O) limit load is
M = 0.63 To (h-!)2

(30)

for a sufficiently deep crack, and the pure extension (6=0) limit load is
N = 2T 0 (h-i)

( 31)

with the line of action of N passing through the center of the uncracked
ligament. The exact limit load for all cases of positive M and negative N may

180

be computed from the slip line field of Fig. 3a, with the circular hinge radius
progressively reduced in size and the extent of the 45 slip lines correspondingly increased until the pure compression analog of Fig. 3b is attained. This
solution will be useful for remote bending loads on part cracked elements.
Likewise, a solution which is possibly exact and certainly an outer bound
is obtained for small positive values of N by enlarging the hinge circle radius
up to the point at which the 45 slip lines disappear and the hinge circle
touches the far boundary. It is easy to compute the limit loads resulting from
these alterations of the Green and Hundy solution because the 45 slip line
regions contain a compressive stress of 2To Hence reduction or enlargement of
the hinge circle radius is equivalent to adding or subtracting material from the
wall surface, with this material bearing the compressive stress 2To.
Fig. 3c shows an upper bound flow field appropriate to the range N > 0,
e ~ 0, and by suitably varying the parameters this may be made to cover all
cases between the above limiting alteration of the Green and Hundy solution and
the pure extension solution of Fig. 3b. Flow occurs by rotation along the
circular arc of radius R with center at a distance L measured positive to the
right of the wall surface. It is most convenient to give results with reference
to the moment M' taken about a point on the wall surface as shown, where

M'

= M+

N h/2

(32)

For a unit rotation rate on the circle, the upper bound inequality reads

M'

+ NL S ToR 2 (a- S)

(33)

Further, for a given choice of L, only one of the parameters R, a, a is independent and this is, of course, to be chosen to optimize the bound. The two geometrical constraints are
L=Rsin S

L + ( h-.e) = R sin

(34)

When the right side of (33) is minimized subject to the above two constraints,
there results the additional equation

2( a- S)
so that R, a, and

'

tan a - tan

( 35)

a are now determined in terms of L.

Hence for each choice of L the upper bound inequality results in a straight
line limit toW + NL and these are entered on Fig . 4. The heavy line envelope
of all such straight lines is the resulting outer approximation to the yield
surface, and this is constructed graphically in Fig. 4. The parameter on each
straight line has been entered as the value of s rather than L, and this covers
the full range from -45 to +45. The point on the envelope corresponding to
a = 45 is the exact pure extension limit load of Fig. 3b. Further, it is easy
to show that the yield surface contains a pointed vertex at this point. The
normal to the yield surface at each point has components proportional to
( h-.e) deP
in theM' and N directions respectively. The construction of the envelope is
completed for small values of N by drawing the yield surface tangent through
the point on theM' axis representing pure bending from the known ratio of
components of the normal at that point, namely
dop - deP h/2 = -0.63 (h-.t) deP
from the Green and Hundy solution.

181

(36)

N
n~th-0

30.

1.0

.6

.'L

.2.

.6

M'
2."t 0 (h-i)'l

Fig. 4 Combined Tension-Bending Yield Surface

The dashed line in Fig. 4 represents an elliptical approximation to the
portion of the yield surface shown, and its equation is ~ = 0 where ~ is given
by
4>

= j [N/2 To(h-l )] - 0. 3l 2 + 9 j
M'
_
N
l
1
0. 7
~
12To(h-l} 2
4To (h-l} f

(37}

This coincides exactly with the pure extension point (e=45 ), reproducing the
vertical normal there, and coincides almost exactly in the range near pure
bending. The disparity for intermediate values of s may be somewhat compensated
for by the knowledge that the heavy line envelope is certainly an outer bound
in this range. Hence it is suggested that this elliptical approximation be
used for simplicity in the line spring elastic-plastic constitutive relation
of eq. ( 29).
This result is for cases of sufficient crack depth so that yield may be
considered to be confined to the remaining ligament, but will also represent an
upper bound for the very small depths at which this may not be so. When the
crack depth is zero as for the Ougdale-Barenblatt type yield zone model the
yield surface is, from Fig. 3d,
(38}

This is also an upper bound for any non-zero crack depth, and an approximate
indication of the requisite crack depth for yield to be confined to the remaining ligament is obtained by determining the smallest value of l for which
this expression for 4> results in a lower limit load than the previous expression.
This completes an at least
elastic-plastic range, although
value problems has not yet been
desirable to include deviations

preliminary extension of the model into the

application to part-through crack boundary
made and further refinement of the model is
from linearity prior to limit load.

182

SOLUTION METHODS
The last two sections have presented generalizations of the elastic line
spring constitutive equations (7) to the case of thermal or residual stresses
(22) and to the elastic-plastic range (29) and this has been the primary aim of
the paper. However, it is appropriate briefly to review here the methods of
formulation and solution of boundary value problems. The constitutive laws
relate o(x), e(x) to N{x), M(x) at each point of the line spring. These must
be complemented by a pair of relations giving o(x), e(x) in terms of the unknown
distributions of N{x), M(x) along the spring and in terms of external loadings
such as N*, M* in Fig. lb. This last pair of relations may come from the solution of the boundary value problem for a plate or shell with a through cut,
subjected to the given external loadings and to the unknown N(x), M(x) distributions along the cut.
Surface Crack in a Large Plate
For example, when the dimensions of the plate in Fig. la, b are sufficient
to be considered infinite, it may be shown {l) from the solution of the elastic
plate equations that
+l
Eo{X)/4a = a* /(l-X 2 ) -

f G(X,X') o{X') dX'

-1

{39)

+l
(3+v) E h e(X)/ 8(l+v)a = m* /{l-X 2) -

G(X,X') m(X') dX'

-1

where the nominal stress equivalents of N, M have been adopted, where X= x/a,
and where the influence function is
G{X,X') = {l/11) log [1-XX' + /(l-X 2) /{l-X'2)]
IX-X' I

(40)

It is also possible to treat more general loadings. For

virtue of inhomogeneous thermal stressing in the plane of the
other than a uniform set of boundary stresses, a distribution
nominal stresses would act along the prospective part-cracked
crack was actually present, then eqs. (39) are modified to
+l
Eo(X)/4a =
G(X,X') [a*( X') - o(X' )] dX'

example if by
plate or of loads
o*(X) and m*(X) of
section if no

(41)

-1

and a like equation relating e tom*, m. These equations may also be inverted
to give o,m in terms of the applied loads and the distribution of o and e.
The results are

+l

o(X) = o*(X) - [E/(4Tia)]

J
-1

~)

X-X'

(42)
+l
m(X) = m*(X) -

{[(3+v)Eh] I [811 (l+v)a]}

J
-1

~)

X-X'

Equations of a kind similar to (39-42) will be available in some other cases of

plates and shells, and the final governing integral equations are obtained by
eliminating either o,m or o, e by use of the line spring constitutive law.

183

These must be solved numerically; Rice and Levy (1) have employed such a method
in connection with the coupled integral equations-obtained after elimination of
6,6 in (39). Their technique is also applicable to elastic analysis with thermal
or residual stresses. However, the most suitable numerical technique is not yet
clear for elastic-plastic problems, which are to be formulated in terms of the
incremental versions of (39-42) and will, of course, involve further complications if the Dugdale-Barenblatt yield zones are to be appended at the ends of
the part cracked section.
Finite Element Analysis
Finite element and finite difference methods may of course be employed to
obtain equations analogous to (39-42). For example, the cut surface is divided
into a number of segments coincident with bordering finite element boundaries or
finite difference mesh spacings and a.m are given piecewise linear variations
taking on unknown values a. ,m. at the nodal points. Also a ~. m~ denote the
nominal stresses at the sa~ ~odal points from the solution 1 for 1 the uncracked
plate or shell. Then by superposition the solution for 6., 6. is determined by
applying reverse loads a~-a., m~-m. to the cut surface with hOmogeneous boundary
conditions elsewhere. I~ a~alogy to (41), the solutions may be put in the form
of linear relations relating 6 ., e. at any given node to the collection of values
{aj-a., m~-m.} at all the nodeS, w~th the final set of discrete equations to be
obtai~ed ~Y ~limination through the constitutive laws.
This approach requires that the influence functions be determined for all
the node pairs. A more direct approach is that followed by Levy and Rice (4)
and sketched out in the introduction, although this does require that special
line segment finite elements be introduced to simulate the line spring. Their
contribution to the overall stiffness matrix or to the incremental stiffness
matrix in the elastic-plastic case is obtained by including the integrated
No+ Me terms in the virtual work statement of equilibrium. This method also
has the attractive feature that non-linearities in the plate or shell equations
are readily included, whether due to large rotation effects or to constitutive
non-linearities.
ACKNOWLEDGMENT
These studies were supported in part by the National Aeronautics and
Space Administration under Grant NGL-40-002-080 to Brown University and in part
by the National Science Foundation through a Senior Postdoctoral Fellowship
at the University of Cambridge.
REFERENCES
Rice, J. R., and Levy, N., "The Part-Through Surface Crack in an Elastic
Plate", JouJtrtal. o6 Applied Mec.hruuCL>, Vol. 39, No. 1, March 1972, pp. 185-194.
2 Gross, B., and Srawley, J. E., "Stress Intensity Factors for Single Edge
Notch Specimens in Bending or Combined Bending and Tension by Boundary
Collocation of a Stress Function", NASA Technical Note D-2603, 1965.
3 Smith, F. W., and Alavi, M. J., "Stress Intensity Factors for a Part Circular
Surface Flaw", PJtoc.eecU.ng.o o6 :the F)N.,t IrtteMat.<.onal. Con6eJtenc.e on Plte!.>'->Wte
Vu.,.oet Te~tno!ogy, ASME, New York, 1969, pp. 793-800.
4 Levy, N., and Rice, J. R., "Surface Cracks in Elastic Plates and Shells",
JoU!tVlai. o6 Applied Mec.hanic..o, publication pending.
5 Rice, J. R., "Some Remarks on Elastic Crack-Tip Stress Fields", IrtteJtrta..tiortal.
JouJtrtal. o6 So~d6 and StJtuc.tuJtu.,, Vol. 8, No. 6, June 1972, pp. 751-758.
6 Bueckner, H. F., "A Novel Principle for the Computation of Stress Intensity
Factors", ZeA.JAc.h!U.6t 6ii!t angewandte Ma.-themat.<.k urtd Me~tarU.k, Vol. 50, 1970,
pp. 529-546.
184

7 McClintock. F. A "Plasticity Aspects of Fracture". Flulctwz.e.: An Adva.n.c.e.d

T~e.~e., Volume. 111 (edited by H. Liebowitz). Academic Press. New York. 1971.
pp. 47-225.

185

reprinted from

The Surface Crack Physical Problems and Computational Solutions

published by

THE AMERICAN SOCIETY OF MECHANICAL ENGINEERS

345 East 47th Street, New York. N.Y. 10017
Printed i n U.S .A.