Professional Documents
Culture Documents
V.B.WATWOOD Jr.
Battelle Memorial Institute, Pacific Northwest Laboratory,
Richland, Washington, USA
A direct method of computation of the stress intensity factor, K, of linear fracture mechanics is discussed. Poor
accuracy of the method, unless extremely small elements are used near the crack tip, leads to its abandonment in
favor of direct computation of the strain energy release rate which is equivalent to K. This procedure consists of com-
puting the strain energy for two slightly different crack lengths and employing numerical differentiation to determine
the strain energy release rate. The accuracy is quite good for a relatively coarse finite element mesh for the example
problems presented. Finally the question of bounds to the true solution is considered and it is suggested, but not
proved, that the method will always bound the solution in a manner depending on whether the stress intensity
factor increases or decreases with crack length growth.
able if not necessary to briefly describe the ideas of where a = crack length, N = number of cycles of load-
what is referred to here as "classical elastic fracture ing, C = constant depending on environment, ',3K =
mechanics and crack propagation" since that field is change in stress intensity factor K during a ]oad
also evolving rapidly. cycle.
First of all, both subjects make use of the fact that Obviously, this explanation has been oversimpli-
the limiting solution for the elastic state of stress fied as attempts to make the hypothesis fit real life
around a sharp crack as the distance from the crack have demanded some compromises, e.g. thickness
tip becomes small is of the form: corrections. Nevertheless it can be seen that if K can
be determined for a given structural configuration
K and crack length, the behavior of the crack can be
o =- f(o)
r½ predicted using previously catalogued material data.
No attempt is made here to defend or condemn
where o = one of the stress components, e.g. Or; tile procedure. The facts of the case are that the real
r, 0 = polar coordinates with origin at the crack tip; world demands best estimates from engineers of
f = a trigonometric function of 0 only which depends structural behavior. Since the above procedure seems
on the stress component considered and symmetry of to be the best available means of estimating that
the problem. behavior short of full scale tests, an engineer is tbrced
K is a constant dependent on the geometry such as to use it.
crack length and loading. Since this solution is based Before proceeding, one final statement on fracture
on linear elasticity K is linearly dependent on the mechanics theory is warranted. Another quantity
magnitude of the loading. It is important to note that generally referred to as the strain energy release rate,
this general form of the solution holds for any arbi- G, is directly related to K through an equation involv-
trary structure in which the crack is loaded. The K is ing only the elastic constants of tile material. This
referred to as the stress intensity factor since it is equation is (ret'. [51):
some kind of measure of the intensity of the local
crack tip stress field. K2
G = ~ - ( 1 - v 2) (plane strain) ,
The hypothesis is made that for a given material
the behavior of the crack, whether it remains stable ha
length, grows without bound or grows at some defi- K2
G = ~:- (plane stress) ,
nite rate, depends only on the magnitude of this
parameter. Thus for a given material it is possible to
test a single specimen for which K can be analytically where E = Young's modulus, u = Poisson's ratio.
determined in terms of the load and crack length, G conveys more physical meaning than K because in
thereby resolving its behavior as a function of K. This addition to the interpretation implied by its name,
same behavior as a function of K is assumed to hold strain energy release rate, it can be viewed as the por-
for any structure. For example, if unstable crack tion of the work performed by the external loads
growth occurs at a value of loading and crack length during crack extension that goes into failure of the
for which K is 100 ksi X x / ~ then Kcr for that mate- material per unit area of crack growth (or one half of
rial is 100 ksi × X/~. Then for any other structural the total new surface area).
configuration of the same material in the same
environment, if K reaches the value of 1O0 ksi × ,v/in
unstable crack growth is said to occur. In the same 3. Application of the finite element methods to elas-
manner data for crack growth is empirically related to tic fracture mechanics and crack propagation
K as e.g., from ref. [18] : theory
strain energy release rate, G. In the past use was made b .......... ~
of a few closed form solutions, ref. [3], or a colloca-
tion technique, ref. [10], to determine these param-
eters. These methods were, however, not general
enough or convenient enough for most applications
although they have performed quite well in analyzing
specimens.
Finite element methods, on the other hand, are
well known for their generality particularly with
respect to geometric irregularities. In addition, they
are becoming increasingly easier and more efficient to
use. Therefore the decision was made to attempt to
use existing finite element codes to determine the
appropriate value of the stress intensity factor. For
the sake of completeness the finite element code used
here is described in Appendix A.
i ]11
J[]]ll]l]JllilZlt
~ l l l l ] l [ : l l : l i11111111]1 ']] Ii I--I
i L I
ox - (27rr)
KI cos ~ 0 { 1 - sin 1 3
½ ] 0 sin ~ 0} (1)
Fig. 1. Finite element idealization for center slotted panel
KII and side slotted panel. L = 85, b = 40, a is variable. N u m b e r
- - sin 71 0 { 2 + c o s ~-
1 0 cos~0} o f elements = 470, n u m b e r o f nodal points = 478.
(27rr) ½
plus similar equations for Oy and Yxy" The origin here geometrical boundary conditions are shown in fig. 2.
is at the crack tip and the x-direction is parallel to the Each element furnishes an estimate of the three
crack axis and into the uncracked material, r and 0 stress components. Making use of the appropriate
are the usual polar coordinates. Since a finite element values o f r and 0 one can use eq. (1) to solve for K I
solution provides a stress distribution it is a simple for each stress component. This procedure is dis-
matter to solve eq. (1) and the other component cussed in more detail in Appendix A. Therefore it
equations for K I and KII utilizing the finite element can be seen that each element offers three possible
solution at discrete points. One would expect, of values for K I (actually more since similar computa-
course, to utilize only those elements near the tip of tions can be made for the predicted displacements).
the crack because of the local nature of the theoreti- A blowup of the elements near the tip is shown in
cal solution. fig. 3 and the results are tabulated in table 1. Since
To verify this process a problem was solved for these results range from a low of 1.5 to a high of
which the solution is known. This was the center 18.5 (correct value K I = 5.82) it becomes apparent
slotted panel under tension shown in fig. 1. The ele- that this procedure lacks consistency. By choosing
ment idealization is indicated in the figure and con- only those values derived from Oy where 0 remains
sists of 478 nodes and 470 elements. The stress dis- less that 90 ° much better results are obtained. This is
tribution is symmetrical about the axis through the perhaps a logical choice for Mode I and perhaps
crack and therefore KII is zero. It is also symmetrical those associated with 7-xy would be appropriate for
about the vertical plate centerline so that only one Mode II. There still remains an annoying degree of
fourth of the plate need be analyzed. The appropriate arbitrariness and considering the apparent conse-
326 V . B . W A T W O O D Jr.
o-
ttttttt 'ltttTf
1~4 Panel Considered • ,1/2 Panel Considered
Below a~ Below
1)
Table 1
Predicted values of KI for center slotted panel based on finite element stress solution.
Element idealization shown in fig. 1, 2, and 3.
tedly improve the estimate but at considerable in- displacements. For the finite element method this
crease in data preparation and computer time. If one means a multiplication of the generalized forces at the
has the automatic element ref'ming scheme built into nodes by one-half the nodal displacements. Such a
his finite element code such as in ref. [9] then only procedure is simple, both in concept and in imple-
computer time would be lengthened. However it mentation. For those problems with a small number
would be most desirable for most engineers to be able of external loads, it can be done by hand from the
to use their existing codes without such extensive existing output of most finite element codes. Only a
modification. few additional FORTRAN statements were necessary
to modify an existing finite element code to accom-
3.2. Energy or compliance m e t h o d plish this objective. Alternatively, the strain energy in
Since the above procedure lacks sufficient accu- each element may be calculated directly from the
racy without resorting to very small element nets it nodal displacements and these summed. This proce-
was considered desirable to seek a better procedure. dure is a little more complicated but is sometimes
The approach that finally evolved was basically an more convenient since the summation can proceed as
analog to the so called compliance method of experi- stresses are being evaluated. Furthermore, no special
mentally determining the K or G value. The method is manipulations are required when displacement
quite simple and consists of computing the strain boundary conditions are applied.
energy stored for two or more slightly different crack About the only word of caution necessary here is
lengths and making use of the definition of G, i.e., that the basic element mesh and nodal point coor-
dinates must remain fixed from one crack length to
dU another. This is required because the quantity which
G = ( ± ) * d--A' (2) is desired is the change in stored strain energy as the
crack changes in length. Therefore the only changes
*( + if constant load, that should occur in the strain energy estimate
- if constant displacement) should be related to change in crack length and not
change in the element mesh. In order to eliminate
where U = strain energy stored, A = crack opening possible confusion in this matter the recommended
area i.e., the crack length for a plate of unit thickness. procedure for extending the crack is simply to add
The easiest way to calculate the strain energy is to another node at the crack tip (with the same coor-
make use of Clapeyron's theorem i.e., that the strain dinates as the original one) and renumber the ele-
energy stored for an elastic body is equal to one-half ments in such a way as to separate the two elements
the work that would be done by the applied forces just inside of the crack tip. Thus the crack is extended
(of the equilibrium state) acting through their total in length by an amount equal to the element length.
328 V.B.WATWOOD Jr.
Table 2 Table 3
Values of K I by finite element energy method for center Values Of Kl/o by finite element energy method for side
cracked panel with evenly distributed end loads a. a as shown; cracked panel with evenly distributed end loads o. a as shown;
b = 40;L = 85; unit thickness. b = 40; L = 85; unit thickness.
7 3160.04 9 3271.28
21.50 4.86 7.5 4.96 2.0 55.10 7.78 9.5 8.00 2.8
8 3181.54 10 3326.38
24.76 5.22 8.5 5.32 1.9 67.10 8.58 1 0 . 5 8.76 2.!
9 3206.30 11 3393.54
28.14 5.56 9.5 5.66 1.8
10 3234.44
31.66 5.90 1 0 . 5 6.00 1.7
11 3266.10 fig. 2. Again the results are excellent when compared
35.38 6.23 11.5 6.34 1.8 to another numerical scheme.
12 3301.48 Note that for both cases the finite element proce-
39.26 6.57 12.5 6.67 1.5 dure consistently underestimates the stress intensity
13 3340.74
43.36 6.90 13.5 7.01 1.6 factor.
14 3384.10
U1 = U1 + el '
under tension are examples where G increases with
where U 1 = strain energy estimate of configuration 1, crack length and indeed as pointed out earlier the
U1 = exact strain energy for configuration 1, e I = predictions herein are less than the values given in the
error associated with the strain energy estimate for other references. Also note that since the other meth-
configuration 1. ods used may well err in the opposite direction, the
A similar relation holds for the case where the error o f the finite element procedure herein with
crack length has grown slightly and is referred to as respect to the true solution may even be smaller than
configuration 2. Substituting such equations into eq. that indicated.
(2) for a plate of unit thickness gives:
4.2. Upper bound on G
U2 U1 + e2 -- el If G is decreasing with crack extension then e 1
a = , (4)
a2 al exceeds e 2 in absolute value though both are still
negative and e G is positive thus producing an upper
which reduces in the limit to: bound on G.
To check for the other bound the center slotted
G=G+e G , infinite length panel with a centered wedge load was
worked using the finite element scheme. The element
where G = estimated value o f the strain energy release layout and displacement boundary conditions were
rate, G = exact value o f strain energy release rate, similar to fig. 1 except that the quarter-plate was
e G = (e 2 - e l ) / A a = error associated with estimate. taken longer (210 versus 85) to better simulate the
Since U is underestimated using compatible ele- infinite length situation. Also the loading was changed
ments then e I and e 2 are negative. Then as before if from an end tension to a vertical concentrated load
G is increasing with crack extensions, e 2 should be applied at the left edge (which puts it at the center of
greater in magnitude than e I so that e G will be nega- the full simulation, of course) of the plate. The
tive. The accuracy of this argument rests heavily on results are given in table 4 with a comparison to an
the assumption that the element net changes asso- approximate closed form solution as given in ref. [3].
ciated with the crack extension only affect the strain Note that the finite element values are greater than
energy estimate because of the crack length change the reference values. It can be seen then that the
and do not produce significant side effects. limited results presented herein bear out the bounds
The center slotted panel and the side slotted panel concept outlined before. However, the author wishes
330 V.B.WATWOOD Jr.
to make clear that no claim is being made that the example, the mode 11 contribution could be suppressed
statements on bounds have been proven. Clearly this by forcing the nodal points on opposite side of the
is not so and a much more precise argument involving crack opening to move together in a direction parallel
numerical differentiation as well as a more definitive to the crack surface. In a similar manner the mode I
relationship between K and the quality of the finite contribution could be suppressed by forcing these
element estimate of the strain energy is required. same modes to move together in a direction normal
The converse argument applies to an element to the crack surface.
model that satisfies equilibrium throughout the struc- It can be seen that the proposed finite element
ture. For a discussion of equilibrium models see ref. method above offers a convenient, economical and
[15]. The combination of upper and lower bounds suitably accurate method of determining stress inten-
would provide a much needed assurance of accuracy sity factors. Probably the most glaring weakness is
without resorting to finer gridworks. one of knowing when the element mesh is refined
enough to provide suitable accuracy. The problem is
best met by utilizing more than one mesh size and
5. Generalization of the method observing the sensitivity of the results or by providing
a dual equilibrium model analysis and examining the
The procedures outlined above may be applied size of the "gap" between the two solutions. The
directly to any geometry desired which enables one severity of this problem is diminished however when
to apply the method to real engineering problems as one considers the overall accuracy of classical fracture
well as to specimen analysis. The finite element code mechanics.
used herein was for two dimensional problems only,
either axisymmetric or plane but the method applies
equally well to the three dimensional case. However
a problem which rears its head immediately is that in /
21_
the general case the distinction between modes I and
II is lost. This at first appears disturbing as almost all I
material fracture data is classified by mode. However
this apparent problem becomes less significant when
one considers that almost all such data is for mode I.
The writer knows of only two apparently successful Fig. 4. Centre cracked panel with center wedge loads.
attempts to measure Knc, refs. [11] and [12]. In
ref. [11] Erdogan and Sih find that Kic andKit c are
about equal in value for plexiglass (Kic = 460 psi V ~ 6. Summary and conclusions
and KII c = 410 psi x/~). However, in ref. [12] Corten
states that GII c >> Glc and gives some experimental The application of classical crack propagation
data where KII c is about four times Kic for a glass analysis is made much easier and more general by the
fiber in epoxy resin matrix material. The presence of use of finite element methods to provide the neces-
fibers in the material may well influence the relation- sary estimates of the stress intensity factor K. By a
ship between K I and KII. straight forward procedure utilizing existing codes,
Mode II data is completely lacking for most mate- accurate estimates of this factor can be made. This
rials of interest. Hence for most practical purposes procedure consists simply of calculating the strain
one is justified and conservative to say that Guc = energy at several crack lengths and numerically dif-
Glc and to simply ignore the mode concept. The ferentiating the resulting strain energy-crack length
same conclusion can be reached for crack propaga- curve. Indications have been given of the direction
tion. It should be pointed out however that with the of the error involved and it has been suggested that
above finite element procedure the modes could be use of both equilibrium and displacement finite ele-
separated with acceptable accuracy by applying the ment models with the same element idealization will
proper constraints in the vicinity of the crack. For result in estimates on both sides of the exact result.
FINITE ELEMENT METHOD FOR PREDICTION OF CRACK BEHAVIOR 331
Appendix A
1
4
KI 5 F(Oi)
Description of finite element code used for the i=1 a i - 4 x / ~ i=1 X/Tii
numerical calculation
or
The code used is fairly typical of the codes cur-
rently in widespread use to solve two dimensional 4x/~a
problems in plane and axisymmetric elasticity. For a KI- 4 F(Oi ) '
complete description of the theory involved see refs.
[1] and [2]. i=1 N/~i
For both the axisymmetric and plane cases the
basic element is triangular and within that element where r i and 0 i are the coordinates of the geometric
the displacement field is constrained to be a linear center of the ith element. The second alternative was
function of the coordinates. Thus, the resulting strain considered the most accurate and therefore was the
field is constant within each element. Complete con- one used to generate table 1.
tinuity of displacements is achieved at the element
boundaries by demanding continuity at the element
vertices. Notation
The quadrilateral element in the code is made up
of an assembly of four triangles with an additional A = crack opening area
a = crack length
node introduced at the geometrical center of the
E --- Young's modulus
quadrilateral. This node is reduced out by considering
G = strain energy release rate, general
the subassembly of four elements as a substructure.
GI, GII = strain energy release rate, mode I
Element stresses are computed directly from the
or II
displacement field for each triangle. For the quadri- g stress intensity factor, general
lateral these stresses are merely averaged and the
resulting state o f stress is said to exist at the geomet- KI, Kn = stress intensity factor, critical
ric center of the quadrilateral. Kcr = stress intensity factor, critical
N = number of load cycles
Some care must be exercised in regard to the
U = strain energy
interpretation of stresses for use in computing K I
x,y = Cartesian coordinates,
from eq. (1). Two choices exist. First one may simply
r, 0 = polar coordinates
use the values of r and 0 corresponding to the geo-
o, Or, Ox, etc. = stress component, subscript indicates
metric center of the element and compute K I from
direction,
(KII is assumed zero here):
= Poisson's ratio
x/ gc
KI = F - - ~ c ) 0 •
References
[5] G.R.Irwin, Encylopedia of physics, elasticity and plas- [12] R.T.Schwartz and H.S.Schwartz, eds., Fundamental
ticity, vol. 6 (1958). Aspects of Fiber Reinforced Plastic Composites, Chap-
[6] G.R.Irwin, Structural Mechanics, Proceedings of the ter 6, Influence of fracture toughness and flaws on the
First Symposium of Naval Structural Mechanics (Perga- interlaminar shear strength of fibrous composites by
mon Press, 1960). H.T.Corten (Interscience, New York).
[7] W.F.Brown and J.E.Srawley, Plane strain crack tough- [ 13] O.C.Zienkiewicz and G.S.Holister, eds., Stress Analysis
ness of high strength metallic materials, ASTM Special (Wiley, New York, 1965).
Techn. Pub. No. 410 (1966). [ 14] M.J.Turner, R.W.Clough, H.C.Martin and L.J.Topp,
[8] B.Gross and J.E.Srawley, Stress intensity factors for Stiffness and deflection analysis of complex structures,
single-edge-notch specimens in bending or combined J. Aeronaut. Sci. 23, No. 9 (1956).
bending and tension by boundary collocation of a stress [ 15] V.B.Watwood and B.J.Hartz, An equilibrium stress field
function, NASA TND-2603, January 1965. model for finite element solutions of two-dimensional
[9] S.K.Chan, I.S.Tuba and W.K.Wilson, On the finite ele- elastostatic problems, Int. J. Solids Structures 4 (1968)
ment method in linear fracture mechanics, presented at 857.
the 2nd National Fracture Mechanics Symposium, [16] J.R.Rice, A path independent integral and the approxi-
Lehigh U., 17-19 June 1968. mate analysis of strain by notches and cracks, Brown
[ 10] B.Gross, J.E.Srawley and W.F.Brown, Jr., Stress-inten- Engineering Report No. E39, May 1967.
sity factors for a single-edge-notch tension specimen by [ 17] O.C.Zienkiewicz and Y.K.Cheung, The Finite Element
boundary collocation of a stress function, NASA TN Method in Structural and Continuum Mechanics (Mc-
D-2395, August 1964. Graw-ltill, New York, 1967).
[11] F.Erdogan and G.C.Sih, On the crack extension in [18] G.R.Irwin, J.M.Krafft, P.C.Paris and A.A.Wells, Basic
plates under plane loading and transverse shear, Journal aspects of crack growth and fracture, Naval Research
of Basic Engineering, ASME, December 1963. Lab. Report 6598, November 1967.