Sunteți pe pagina 1din 35

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/316927229

Computational fracture mechanics

Technical Report · January 2015


DOI: 10.13140/RG.2.2.30931.17444

CITATION READS

1 2,465

2 authors:

Silvio Rabbolini Wolfgang Brocks


Exergy ORC Christian-Albrechts-Universität zu Kiel
19 PUBLICATIONS   62 CITATIONS    172 PUBLICATIONS   2,991 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Structural integrity assessment procedures for high temperature components View project

All content following this page was uploaded by Silvio Rabbolini on 15 May 2017.

The user has requested enhancement of the downloaded file.


Computational Fracture Mechanics
Students Project 2015

Final report

Wolfgang Brocks

Silvio Rabbolini

May 2015

Dipartimento di Meccanica

Politecnico di Milano

report_Brocks-Rabbolini, wbrocks, 24 June 2015, - 1 -


Part 1: Determination of a true stress-strain curve as ABAQUS input
from tensile test data
Geometry: flat tensile bar (rectangular cross section)
width W0 = 8.01 mm,
thickness B0 = 2.9 mm
measuring length L0 = 30 mm
Material Al 5083
Young’s modulus E = 70300 MPa
Poisson’s ratio ν = 0.3
0.2% proof stress Rp0.2 = RF(εp=0.002) = 242 MPa

10
F [kN]

Al 5083
8

2 test 2.1.21
test 2.1.16

0
0 1 2 3 4 5
∆L [mm]

SCHEIDER, I.; SCHÖDEL, M.; BROCKS, W.; SCHÖNFELD, W.:


Engng. Fract. Mech. 73 (2006), 252–263

(1) Evaluate the test data provided in “tensile-test_Al5083.xls”

Note that • ABAQUS requires true (equivalent) stresses in dependence on


logarithmic plastic strains as flow curve input, , for an elasto-
plastic analysis,
• A curve beyond uniform elongation of the tensile bar is needed
for the fracture mechanics analysis, which necessitates an extrapolation
of the stress-strain curve up to ≈ 3 by a power law

where , are normalisation values, commonly taken as yield limit


= = = 0 and = ⁄ .
• Number of measured data has to be significantly reduced: no more than
30 points for the curve.

report_Brocks-Rabbolini, wbrocks, 24 June 2015, - 2 -


(2) Check the stress-strain curve by numerical simulation of the tensile test beyond uniform
elongation (maximum force) and compare simulation with test data in terms of force, F,
vs. elongation, ∆L .
Analysis of tensile test • plane stress and 3D model
• account for symmetries,
• displacement controlled,
• geometrically nonlinear (“large deformations”).

(3) Derive a correction factor, , for the evaluation of the equivalent stress, , from
measured force, F, and specimen width, W, beyond the onset of necking according to
the procedure of SCHEIDER et al. [2004]. see below, and plot it in dependence on .

Stress and strain measures


1. Homogeneous uniaxial stress state (uniform elongation)

Linear strain = ≤ ! (εu = uniform strain)
%
Logarithmic (“true”) strain ε " = ln = ln&1 + ε )
%

Nominal (“engineering”) stress σ * =+ =, -


≤ R* = /01
+

(Rm = tensile strength)


4 4
“True” (Cauchy) stress σ2 !3 = 5 = 67

Assumption of isochoric deformation 8 9 = 89 , 2 2: = 3 + ≈


4 4
σ2 !3 =5=5 = * &1 + )

VON MISES equivalent stress = σ2 !3

= ε = = 2 2:

σ<=>?
Yield condition , " " @

is the (uniaxial) “flow curve” required as input for ABAQUS


2. Triaxial principal stress state (beyond uniform elongation = onset of necking)
B
VON MISES equivalent stress = AC D& B − C)
C +& C − E)
C +& E − B)
CF

SCHEIDER, I.; BROCKS, W.; CORNEC, A.: “Procedure for the determination of true
stress–strain curves from tensile tests with rectangular cross section specimens”,
J. Eng. Mater. Techn. 126 (2004)
8G = H C 6
7
nominal area eq. (17)
4
equivalent stress = 5I eq. (18)

report_Brocks-Rabbolini, wbrocks, 24 June 2015, - 3 -


1.1. Evaluation of test data

In a first step, the force elongation curves, F(∆L), measured in the tests are converted into
engineering stress-strain curves, σ * &ε ). A magnification of the initial part σ ≤ 240 MPa
exhibits measuring offsets of the test data, see Fig. 1.1, which result in wrong plastic strains,
= − ⁄ . After offset corrections of ∆ε = +0.0002 for test 2.1.21 and ∆ε = -0.000475 for
test 2.1.16 the elastic parts of the test data coincide,1 see Fig. 1.2 .

250 250

σnom [MPa]
σnom [MPa]

200 200

150 150

100 100
test 2.1.21
test 2.1.21 test 2.1.16
50 E*eps
50 test 2.1.16

E*eps
0
0 0 0,002 0,004 0,006
0 0,002 0,004 0,006
εlin [-] εlin [-]

Fig. 1.1: Engineering stress-strain curves Fig. 1.2: Offset corrected test data
for σ ≤ 240 MPa showing
measuring offseta of the test data
Plotting (nominal) stresses2 vs. (linear) plastic strain, = − ⁄ , Fig. 1.3, provides
• yield strength, = * = 0 = 200 MPa , and
• 0.2% proof stress (or “offset strength”) .C = * = 0.002 = 242 MPa

1
Except the very first strain values below σ = 50 MPa of test 2.1.21
2
The difference between nominal and true stresses is small, σtrue/σnom ≤ 1.0055%, for εp ≤ 0.002.

report_Brocks-Rabbolini, wbrocks, 24 June 2015, - 4 -


250

σnom [MPa]
225

200
test 2.1.21
test 2.1.16
ABAQUS
175

150
Fig. 1.3: Nominal stresses vs. plastic 0 0,0005 0,001 0,0015 0,002
strains: determination of yield p
ε [-]
strength3 lin

True stresses result from


4 4
2 !3 =5=5 = * &1 + ) (1.1)

under the assumption of isochoric deformation since no necking occurred in the tests. The
true stress-strain curves are shown in Fig. 1.4.

500 500
test 2.1.16
stress [MPa]

stress [MPa]

test 2.1.21
400 400

300 300

200 200

100 nom stress 100 nom stress


true stress true stress
0 0
0 0,05 0,1 0,15 0 0,05 0,1 0,15
strain [-] strain [-]

Fig. 1.4: Conversion of engineering stress-strain curves to true stress-strain curves


(a) test 2.1.16 (b) test 2.1.21
Extrapolation beyond the test range, 0.12 ≤ εp ≤ 3.0 is required for the later analyses of a
fracture mechanics specimen, as respective plastic deformations will occur at the crack tip.
This is commonly done by a power law,
P RS
= QR T , (1.2)
P

3
The square dots denote the input data for ABAQUS in the range R0 ≤ σ ≤ Rp0.2, see Fig. 1.7

report_Brocks-Rabbolini, wbrocks, 24 June 2015, - 5 -


which at the same time provides a fit curve of the scattering stress data and allows for data
reduction. The stress-strain curves obviously do not allow for a unique power-law fit over the
whole range of εp. Therefore, only the final part is used to define the power law.4
Normalisation is recommended in order to obtain a dimensionless equation so that α does not
depend on the specific dimension used for the stresses. The normalising values can be
arbitrarily selected, commonly taken as the yield-limit values,
= = 200 MPa and = ⁄ = 0.00284.

2,1 2,1

σtrue/σ0 [-]
y= 1,0274x 0,1755
σtrue/σ0 [-]

y = 1,03x 0,1769
2 2

1,9 1,9

1,8 1,8

1,7 1,7
test 2.1.16 test 2.1.21
1,6 Pot.(test 2.1.16) 1,6 Pot.(test 2.1.21)

1,5 1,5
10 20 30 40 50 10 20 30 40 50
εp/ε0 [-] εp /ε0 [-]

Fig. 1.5: Power law fit of true stresses vs. logarithmic plastic strains
(a) test 2.1.16 (b) test 2.1.21

The extrapolation is finally executed for α = 1.03 and n = 1.76, see Fig. 1.6.5 An extrapolation
over a range of 25 times the last strain value of the tensile test may be considered as improper.
There is no other choice, however, and the analyses of the M(T) specimen will have to
legitimate it.
The data points of the flow curve for ABQUS, Fig. 1.7, are established as
• first point: = =0 ,
• 8 points from the tensile test data up to = 0.06
including .C = = 0.002 as fourth point (see Fig. 1.3, above),
• 13 data points σ ε in the range 0,12 ≤ ε ≤ 3.0 from the extrapolation by power
law fit.

4
The fit parameters will in any case depend on the chosen interval. Varying this interval is
advisable.
5
Extrapolating a flow curve up to 25 times the final test point of plastic deformation seems rather
disputably, but there is actually no other choice, and the simulation results of the M(T) specimen
will have to justify it.

report_Brocks-Rabbolini, wbrocks, 24 June 2015, - 6 -


700 700

RF [MPa]
σ [MPa]

600

600 500

400
500
300
test 2.1.21
extrapol
200
400
ABAQUS
100

300 0
0 0,5 1 1,5 2 2,5 3 0 0,5 1 1,5 2 2,5 3
εp [-] εp [-]

Fig. 1.6: Extrapolation of the true stress vs. Fig. 1.7: Flow curve as input for ABAQUS
logarithmic plastic strain curve.

1.2. Numerical simulation of the tensile test

Due to a two- or threefold symmetry in a 2D or 3D analysis, the tensile bar can be


represented as a quarter or eighth model, respectively, Fig. 1.8. This is realised by applying
boundary conditions along the respective line or surfaces, ZC &[B , 0, [E ) = 0, ZB &0, [C , [E ) =
0, ZE &[B , [C , 0) = 0. The symmetry to the x1-axis guarantees that necking occurs at x2 = 0.
In order to realise a homogeneous stress and strain state over the measuring length, L0, up
to uniform elongation, the fixation of the specimen with a larger width at the upper end has to
be incorporated in the model. Fig. 1.9. In a panel of constant width, necking might occur in
the section where the prescribed displacement is applied instead of the centre section. The
correct location of necking has to be checked, see below. A finer mesh is applied in the centre
segment of the bar.
The specimen is loaded in displacement control, as the axial force will decrease under still
increasing elongation in the simulation due to local necking beyond uniform elongation.6 A
prescribed displacement is applied at the upper cross section of the model by connecting all
respective nodes to some reference point using the coupling constraint option in ABAQUS.
This point moves in the x2-direction with displacement u2. The corresponding external force is
obtained as reaction force at the reference point.

6
This does not occur in the tests as the specimens broke before.

report_Brocks-Rabbolini, wbrocks, 24 June 2015, - 7 -


Fig. 1.8: Quarter model (2D) of tensile bar Fig. 1.9: 3D mesh
esh of tensile bar (1/8
model)

The deformed mesh at ∆L ∆ = 10mm with iso-contours


contours of transverse and longitudinal
displacements, u1, u2, respectively, is shown in Fig. 1.10. Necking has occurred in the centre
section, whereas in some distance, the deformation is homogeneous, again. Plastic
deformation is revealed in Fig. 1.11.
1.1

Fig. 1.10: Deformed 3D mesh showing localisation of deformation (necking) at ∆L = 10mm


(a) transverse displacement,
displace u1 (b) longitudinal displacement,
displace u2
Fig 1.12 finally shows a comparison between the numerical results of a 2D plane stress and
a 3D model with the test data in terms of force, F, versus elongation, ∆L. ∆ There is little
difference between the 2D and the 3D simulation. Due to the symmetry conditions, the t total
external force results from the reaction force at the reference point as F = 2 FRP in the 2D
model and F = 4 FRP in the 3D model, and the elongation is ∆L = 2 u2(L0/2) for both models.
Both simulations coincide perfectly with each other up to uniform elongation and with the test
data. This means that the plane stress assumption is legitimate.

report_Brocks-Rabbolini, wbrocks,
wbrocks 24 June 2015, - 8 -
10

F [kN]
8

test 2.1.21
4 test 2.1.16
FE: 2D
FE: 3D
2

0
0 2 4 6 8 10
∆L [mm]

Fig. 1.11: Deformed mesh with iso-contours Fig. 1.12: Comparison of test data with 2D
strain εeq, at
of equivalent plastic strain, and 3D numerical simulations
simulation
∆L = 10mm

Maximum force, ultimate (linear) stress and uniform elongation7 are:

Fmax (kN) σu (MPa) (∆L)u (mm) εu (-)

2D (plane stress) 8.24 354.8 5.62 0.187

3D 8.26 355.6 5.65 0.188

In a geometrically nonlinear analysis,


analysis the
he external force decreases due to local necking of
the bar in the mid-section. Fig. 1.10.
1.1 . The corresponding value of uniform elongation is
affected by any small imperfection of the bar, e.g. a reduction of diameter by 1% or a mesh
refinement in the mid-section
section as in Fig. 1.9. The plane stress and the 3D simulation start to
differ after uniform elongation.

7
which were tests εf = 0.14
ere not reached in the tests:
report_Brocks-Rabbolini, wbrocks,
wbrocks 24 June 2015, - 9 -
1.3. Relation between true effective stress and applied force beyond uniform elongation

Flow curves which are required as input for finite element analyses can generally not be
directly determined from measured force versus elongation curves beyond the onset of
necking, as the strain field becomes inhomogeneous and the stress state triaxial. An analytical
solution derived by BRIDGMAN8 is used for round tensile bars to determine the effective stress
in the necked section. It requires the continuous measurement of the reduction of diameter
and the necking radius during the test, which necessitates advanced testing techniques.
The BRIDGMAN correction does not work for specimens with rectangular cross sections,
however, as the respective solution assumes an axisymmetric stress and strain field with
constant plastic strain in the smallest necking section. The strain gradients in the cross section
increase with the aspect ratio of rectangular bars. Difficulties also arise with the experimental
determination of the cross section area as the rectangular cross section adopts the shape of a
cushion. ZHANG et al.9 proposed a method using the local thickness reduction in the necking
area to determine the actual cross section area, but the triaxiality of the stress state after
occurrence of local necking is not taken into account.
SCHEIDER et al.10 developed a procedure based on in-situ measurement of the deformation
field on the specimen surface and derived an approximate formula relating true effective
stresses and applied force which is validated by numerical parameter studies. Since the actual
area of the necking section cannot be measured any more after the onset of necking due to its
cushion like shape, they define a ”nominal” area,
8G = H C 6 ,
7
(1.3)

where the width, W, is a measurable quantity. The true effective stress results from the
measured force by means of a correction factor, fcorr,
4
= 5I , (1.4)
which does not only include the conversion of the nominal area, 8 * , to the actual area, A,
but also the effect of triaxiality of the stress state. A finite element parameter study has been
conducted for determining the correction factor for various materials. The quantities F and 8G
result from the finite element calculation and is the employed flow curve. The
equivalent plastic strain, εp, serves as a scalar quantity related to the load step of the
simulation. In testing, it is approximately calculated from the measured deformation field in
the central area of the specimen,
]
≈ ̅ = AE & C
^ + ^ ^^ + ^^ )
C
. (1.5)

The curves fcorr(εp) by SCHEIDER et al. have a typical shape: they start at a value of one
which is kept until maximum force is reached at the respective limit value, εu, of uniform
elongation, and decrease for εp > εp(εu) in the beginning but raise to a value of one again for

8
BRIDGMAN, P.W.: Studies in large plastic flow, McGraw Hill, New York, 1952.
9
ZHANG, Z.L.; HAUGE, M.; ODEGARD, J.; THAULOW, C.: Determining material true stress-strain
curve from tensile specimens with rectangular cross section, Int. J. Solids Struct. 36 (1999),
3497–3516
10
SCHEIDER, I.; BROCKS, W.; CORNEC, A.: “Procedure for the determination of true stress–strain
curves from tensile tests with rectangular cross section specimens”, J. Eng. Mater. Techn. 126
(2004)

report_Brocks-Rabbolini, wbrocks, 24 June 2015, - 10 -


εp ≈ 0.8. For higher values of εp, the curves spread significantly, depending on the hardening
exponent of the flow curve.
Fig 1.13 shows the nominal reduction of the mid-section area as calculated by the
assumption of isochoric deformation, 8⁄8 = 9 ⁄9, and by eq. (1.3), 8G⁄8 = &H ⁄H )C .
Both approaches coincide below uniform elongation, εp ≤ εp(εu)11. Due to the localisation of
plastic deformation, the area reduction in the centre section increases much faster than the
elongation, L0/L, which is better matched by the definition of the nominal area, 8G⁄8 ,
according to eq. (1.3)12.
Fig 1.14 finally presents the factor relating external force and true effective stress, fcorr(εp),
according to eq. (1.4) for the present material, which additionally includes the effect of
triaxiality of the stress state. The initial part corresponds with the results of SCHEIDER et al.,
i.e. fcorr = 1 for εp ≤ εp(εu) and fcorr < 1 for εp > εp(εu), but a value of fcorr = 1 is reached again not
until εp ≈ 1.2.

1 1,2
A / A0 [-]

0,8 fcorr [-]

1,1
0,6

0,4 L0/L
1
(W/W0)^2

0,2

0 0,9
0 0,5 1 1,5 2 0 0,5 1 1,5 2
εp [-] εp [-]

Fig. 1.13: Nominal reduction of the mid- Fig. 1.14: Correction factor, fcorr(εp), for the
section area as calculated by the relation between true effective
assumption of isochoric stress and applied force according
deformation and by eq. (1.3) to eq. (1.4)
red vertical line indicates point of
uniform elongation, εp(εu)

11
εp(εu) = εp(0.188) = 0.171
12
Note, that also eq. (1.3) provides only an approximation of the „real“ cross section area, which
cannot be calculated from B and W due to the cushion like shape of the cross section.

report_Brocks-Rabbolini, wbrocks, 24 June 2015, - 11 -


Part 2: Numerical analysis of a centre cracked panel M(T) with
stationary crack

Geometry: width 2W = 300 mm,


initial crack length 2a = 60 mm,
thickness B = 2.9 mm
length 2L = 1.5·2 W = 450 mm
Material: Al 5083 as above for Part1
YOUNG’s modulus E = 70300 MPa
POISSON’s ratio ν = 0.3
0.2% proof stress Rp0.2 = 242 MPa
yield limit = =0

200
load vs elongation
F [kN]

150
VL = ∆(2L) = “load point” displacement,
elongation
100

50
M(T)150 #1.1.3
M(T)150 #2.1.8
Test results in “MT_150_Al5083_test.xls”
0 SCHEIDER, I.; SCHÖDEL, M.; BROCKS, W.;
0 1 2 3 4 5 SCHÖNFELD, W., Engng. Fract. Mech. 73
VL [mm] (2006), 252–263

(1) Compare FE simulation with test data in terms of force vs. elongation, F(VL), up to
approx. maximum force, i.e. _% ≈ 2``

Note: • Even below maximum load, the test results include significant crack
extension (up to ab ≈ 6``), which is not considered in the present
FE analysis! For simulation of crack extension see Part 3!
• There is also an apparent measuring offset for F → 0.

(2) Evaluate J-integral by domain integral in ABAQUS and plot J(VL); compare with
analytical evaluation of J as “energy release rate” from F(VL) curve, see below.
(3) Investigate path dependence of J.

report_Brocks-Rabbolini, wbrocks, 24 June 2015, - 12 -


2.1. Analysis of test data

The test data show some measuring offset for VL → 0 which becomes visible if the line of
elastic compliance is plotted into the diagramme, Fig. 2.1. The elastic load point displacement
results from a superposition,
_%3 = _% : c
+ _% : c
= d% : c
+ d% : c
F = d% f , (2.1)
of the elongation of an uncracked panel,
C C
d% : c
= = = , (2.2)
@5 C@76 @76
and a contribution which is due to the crack,
g g
d% : c
= _ Q T
@6 C 6
, (2.3)
where the function V2 is given by TADA et al.13.
g g g C g E g ]
_C Q T = −1.071 + 0.250 Q T − 0.357 Q T + 0,121 Q T − 0,047 Q T +
6 6 6 6 6
g j B, kB g
0,008 Q6T − ln Q1 − 6T . (2.4)
:⁄,

According to TADA, eq. (2.4) holds for 9⁄H ≥ 3 but obviously gives a good approximation of
the elastic compliance for 9⁄H = 1.5 as in the present case. The contribution of d% : c is only 5.8%
of the total compliance, CL, however.

200 200
F [kN]
F [kN]

150 150

100 100

M(T)150 #1.1.3 M(T)150 #1.1.3


50 50
M(T)150 #2.1.8 M(T)150 #2.1.8

elastic elastic

0 0
0 0,5 1 1,5 2 0 0,5 1 1,5 2
VL [mm] VL [mm]

Fig. 2.1: Test data exhibiting a measuring Fig. 2.2: Test data after applying an offset
offset for VL → 0 correction

The offset corrections are obtained as


_% = _%23m2 + ∆_% = _%23m2 + d% f&B) − _%&B)
23m2
. (2.5)

13
TADA, M.; PARIS, P.C.; IRWIN, G.R.: The Stress Analysis of Cracks Handbook, Del Research
Corp., Pennsylvania [1973]

report_Brocks-Rabbolini, wbrocks, 24 June 2015, - 13 -


After the respective shifts of the displacements by ∆VL = 0.0485 mm for test #1.1.3 and
∆VL = 0.0930 mm for test #2.1.8, the measured curves coincide in the elastic range, Fig. 2.2.
The offset correction is particularly important for the calculation of plastic deformations, _%
in Part 3.
The (corrected) test data, F(VL), can be used to evaluate the J-integral according to its
interpretation as an energy release rate. To begin with, crack growth, ∆a, is not considered.

Fig. 2.3: Schematic of an F(VL) curve at constant


crack length with definitions
F
n = o fp_% = n 3 + n ,
B B U*
n 3 = C f_%3 = C f _ −_% ,

n = o fp_% ,
Upl
B B
n ∗ = n − C f_% = n − C f_% . Uel

vL

Assuming “deformation theory of plasticity”14, J is split into an elastic and a plastic part,
where the elastic part is calculated from the stress intensity factor, KI, and the plastic part
from the area under the f&_%% ) curve, Fig. 2.3,
stu vw S
r = r3 + r = @
− QC7 vgT , (2.6)
xy
g
where z^ = ∞ √|b } Q6T , (2.7)
4
and = . (2.8)
∞ C76

The geometry function for an M(T) specimen is15


g •g
} Q6T = Asec QC6T , (2.9)

and the plastic J can be calculated from16


w∗
r = . (2.10)
7&6‚g)

The result of this evaluation is shown in Fig. 2.4 for the specimen #1.1.3. Up to VL = 1 mm
only the elastic part, J e, contributes to the total J. As the evaluation does not account for crack
growth, J e decreases beyond maximum load and J p becomes more and more dominant.

14
Or actually hyperelastic behaviour of the material requiring that the strain energy density
represents a potential, ƒ„ = …†⁄… ƒ„ .
15
FEDDERSEN, R.E., ASTM STP 410 (1966), 77-79
16
RICE, J.R.; PARIS, P.C.; MERKLE, J.G., ASTM STP 536 (1973), 231-245. Note that eq. (2.10)
holds for an M(T) only!

report_Brocks-Rabbolini, wbrocks, 24 June 2015, - 14 -


J [N/mm] 200 1400

J [N/mm]
J_total 1200 J_total
J_e J_e
150 J_p J_p
1000

800
100
600

400
50
200

0 0
0 0,5 1 1,5 2 0 1 2 3 4 5
VL [mm] VL [mm]

Fig. 2.4: J-integral for test #1.1.3 evaluated from F(VL) according to eqs. (2.6)-(2.10)
neglecting crack growth
(a) VL ≤ 2 mm, F ≤ Fmax (b) total test range

2.2. FE model

The FE model disregards all technical details of a real M(T) specimen, in particular
• clampings for application of the force17,
• the machined notch used as starter for fatigue cracking
and accounts for the twofold symmetry with respect to x1 and x2. It is thus reduced to a
rectangle of size W×L, Fig. 2.1. The mesh is refined towards the ligament, b ≤ [B ≤ H,
Fig. 2.2, reaching a minimum element size of 0.125×0.125 mm2. The regular arrangement of
square elements in the ligament is specific for simulations of crack extension (see Part 3) but
will yield strongly distorted elements at the crack tip and is hence inappropriate for
calculating reliable stress and strain values at a stationary crack. Since the present analysis is
focused on global values, namely elongation and J-integral, the meshing at the crack tip is
uncritical.
The twofold symmetry is realised by fixing the vertical displacements in the ligament,
ZC &b ≤ [B ≤ H, 0) = 0, and the horizontal displacements of the vertical centre line,
ZB &0, [C ) = 0. The upper edge is kept straight, ZC &[B , 9) = ZC &0, 9). The specimen is loaded
in displacement control mode by applying a vertical displacement u2 to a reference point as in
exercise #1, which is connected to the upper edge of the specimen. The respective force is
picked up as reaction force at the reference point. The elongation of the specimen equals the
displacement of the reference point, a&29) = _% = 2ZC &0, 9)

17
This is different from the tensile bar, Fig. 1.8, where the fixation of the specimen had to be
incorporated into the model in order to prevent necking in the section where the prescribed
displacement is applied. Due to the crack at x2 = 0 this is precluded here.

report_Brocks-Rabbolini, wbrocks, 24 June 2015, - 15 -


Fig. 2.5: FE mesh of the M(T) specimen Fig. 2.6: Refined FE mesh in the ligament
(quarter model) with boundary
conditions

The analysis is performedd as plane stress with the specimen’s thickness B. Fig 1.13 has
shown that the assumption of a plane stress state is admissible. Figs. 2.7 shows
show comparisons
of the FEM results with test data in terms of force, F,, versus load point displacement VL,
where in Fig. 2.7(b) the offset correction of eq. (2.5) has been applied to the test data as in
Fig. 2.2. The elastic compliance of the FE model is obviously lower than in the analytical
solution of TADA, eqs. (2.1)--(2.5), d% ‡ˆ = 0.91d%Š+‹+ , and for the test data.
data The higher
stiffness of FE models is generally due
du to the finite number of degreess of freedom. Fig. 2.7(b)
also contains the FE results of exercise #3 (see below). The cohesive elements obviously
reduce the elastic stiffness of the model.

200 200
F [kN]

F [kN]

150 150

100 100

M(T)150 #1.1.3 M(T)150 #1.1.3


M(T)150 #2.1.8
50 M(T)150 #2.1.8 50
FEM ex#2
FEM ex#2
FEM ex#3
Tada Tada
0 0
0 0,5 1 1,5 2 0 0,5 1 1,5 2
VL [mm] VL [mm]

Fig. 2.7: Comparison of the numerical results with test data up to maximum load
(a) original test data (b) offset correction, eq. (2.5)
Any offset correction is negligible in a plot over the full range of VL, see Fig. 2.8. The
FEM curve starts deviating from the experimental data already below maximum force reached
in the tests,, as no crack extension is considered in the analysis. The force continues to
increase in the simulation whereas it decreases in the tests. Different from the force vs.
elongation curve of the tensile bar, Fig. 1.13,, where the load drop was due to plastic necking,
it is due to crack extension, here.

report_Brocks-Rabbolini, wbrocks,
wbrocks 24 June 2015, - 16 -
200

F [kN]
150

100

M(T)150 #1.1.3
M(T)150 #2.1.8
50
FEM

Fig. 2.8: Comparison of the numerical


results with test data over the 0
0 1 2 3 4 5
whole range of VL
VL [mm]

2.3. J-integral evaluation

The J-integral introduced by RICE18 and CHEREPANOV19 is defined as an integral over a


closed contour, Γ, around the crack tip,

r = ∮• †p[C − ƒ„ •„ Zƒ,B pŽ , (2.11)


where w is the strain energy density, σij the stress tensor and nj the outer normal to the
contour. It is equal to the energy release rate in a plane body of thickness, B, due to a crack
extension, ∆a, in mode I,
‘w
r = − •7‘g’ , (2.12)
xy

which is exploited in the virtual crack extension (vce) method to calculate J numerically as
well as in the experimental determination of J.
Applying the divergence theorem, any contour integral can be converted into a domain
integral over a finite region, i.e. an area in two dimensions or a volume in three dimensions,
surrounding the crack. ABAQUS/Standard uses this domain integral method to evaluate the
J-integral and automatically finds the elements that form each ring from the regions defined
as the crack tip or crack line. Each “contour” provides an evaluation of the respective domain
integral. The number, n, of contours must be specified in the history output request. 20
*CONTOUR INTEGRAL, CONTOURS=n, TYPE=J
By definition, J is a “path-independent” integral, claiming that each contour should yield
the same J-value. However, the derivation of its path-independence requires
(1) time independent processes,

18
RICE, J.R.: “A path independent integral and the approximate analysis of strain concentrations
by notches and cracks”, J. Appl. Mech. 35 (1968), 379-386.
19
CHEREPANOV, C.P.: “Crack propagation in continuous media" Appl. Math. Mech. 31 (1967),
476-488.
20
Note that the largest domain must not touch the outer surface of the specimen, as the energy
release rate is calculated in a ring of elements around this domain.

report_Brocks-Rabbolini, wbrocks, 24 June 2015, - 17 -


(2) absence of volume forces and surface forces on the crack faces,
(3) small strains (geometrically linear),
(4) homogeneous hyper-elastic material (deformation theory of plasticity).
A geometrically nonlinear elasto-plastic analysis applying incremental theory of plasticity
violates the two last-mentioned conditions, so that the values obtained for the various
contours will differ and the J-integral becomes path-dependant, more precisely, J increases
with increasing size of the contour.21 Contour (or domain) dependence thus is physically
significant due to the dissipative character of plastic deformation. What is needed in a fracture
mechanics analysis is a saturation value or “far-field” value of J corresponding to that
evaluated from the mechanical work done by the external force at the load-line displacement,
eq. (2.12). For further details see BROCKS & SCHEIDER22.
The user must define the crack front and the virtual crack extension in ABAQUS to request
contour integral output; this can be done as follows:
• Defining the crack front
The user must specify the crack front, i.e., the region that defines the first contour.
ABAQUS/Standard uses this region and one layer of elements surrounding it to compute the
first contour integral. An additional layer of elements is used to compute each subsequent
contour. The crack front can be equal to the crack tip in two dimensions or it can be a larger
region surrounding the crack tip, in which case it must include the crack tip. By default
ABAQUS defines the crack tip as the node specified for the crack front in the so called
CRACK TIP NODES option of the contour integral command.

*CONTOUR INTEGRAL, CONTOURS=n, CRACK TIP NODES


Specify the crack front node set name and the crack tip node number or node set
name.
Alternatively, a user-defined node set which include the crack tip can be provided to
ABAQUS manually by omitting the CRACK TIP NODES option:
*CONTOUR INTEGRAL, CONTOURS=n
Specify the crack front node set name and the crack tip node number or node set name.
• Specifying the virtual crack extension direction
The direction of virtual crack extension must be specified at each crack tip in two
dimensions or at each node along the crack line in three dimensions by specifying either the
normal to the crack plane, n, or the virtual crack extension direction, q, see Fig. 2.9.

21
BROCKS, W.; YUAN, H.: On the J-integral concept for elastic-plastic crack extension, Nuclear
Engineering and Design 131 (1991), 157-3
22
BROCKS, W.; SCHEIDER, I.: Reliable J-values - Numerical aspects of the path-dependence of the
J-integral in incremental plasticity, Materialprüfung 45 (2003), 264-275.

report_Brocks-Rabbolini, wbrocks, 24 June 2015, - 18 -


Fig. 2.9: Crack front tangent, t, normal to
the crack plane, n, and virtual
crack extension direction, q, at a
curved crack

The normal to the crack plane, n, can be defined in ABAQUS using the so called NORMAL
option23 in the contour integral command:
*CONTOUR INTEGRAL, CONTOURS=n, NORMAL
nx-direction cosine, ny-direction cosine, nz-direction cosine(or blank), crack front node
set name (2-D) or names (3-D)
ABAQUS will calculate a virtual crack extension direction, q, that is orthogonal to the
crack front tangent, t, and the normal, n, see Fig. 2.9. Alternatively, the NORMAL option can
be omitted and the virtual crack extension direction, q, can be defined directly:
*CONTOUR INTEGRAL, CONTOURS=n
crack front node set name, qx-direction cosine, qy-direction cosine, qz-direction cosine
(or blank)
Note, that the symmetry option (SYMM) must be used to indicate that the crack front is
defined on a symmetry plane:
*CONTOUR INTEGRAL, CONTOURS=n, SYMM
The change in potential energy calculated from the virtual crack front advance is doubled
to compute the correct contour integral values.
Different combination of the above mentioned options can be used to request contour
integral within ABAQUS, more details about this could be found in the ABAQUS
documentation.
Two options of defining the contours are used in the present study with a number of
contours24, n = 20, for both:
• The CRACK TIP NODES option with the first contour being the crack tip, hereafter called
“automatic option”, Fig. 2.9;
• A first contour around the crack tip of finite size is defined via a crack tip node set,
hereafter called “manual option”, Fig. 2.10.

23
The normal option can only be used if the crack plane is flat
24
Note that the number of contours is limited by the symmetry line of the specimen.

report_Brocks-Rabbolini, wbrocks, 24 June 2015, - 19 -


Fig. 2.9: Contour (domain) definition25 of J-integral calculation by “automatic option”:
option”
st
(a) 1 domain = crack tip (b) last domain #20 20

Fig. 2.10: Contour (domain) definition for J-integral calculation by “manual option”:
option”
st
(a) 1 domain = „node
node set“
set (b) last domain #2020
Fig. 2.11 shows two comparisons between the domain sizes and the plastic zones. A lower
limit of the equivalent strain εp,=0.002, corresponding to =
quivalent plastic strain, .C
C , has been defined,
so that the grey area approximately represents the elastic region.
region In order to obtain meaningful
far-field values of J, the corresponding contour should imbed the plastic zone emanating from
the crack tip. This would be possible for VL = 1.5 mm, Fig. 2.11(a),, but require a different
definition of the initial node set, accounting for the specific shape of the plastic zone, since
the number of contours is limited to n = 20 by the symmetry line of the specimen,
Fig. 2.10(b). It is not possible any more when the plastic zone extends from the crack tip to
the free surface of the specimen,
specimen Fig. 2.11(b).

lastic strain, εp, and domains for J evaluation (manual option)


Fig. 2.11: Equivalent plastic
(a) VL = 1.5 mm,, domain #1 (b) VL = 4 mm,, domain #20
Test and simulation results of r&_% ) are compared in Fig. 2.12. The J-integral
J has been
evaluated from F(VL) for both, test and FE results according to eq. (2.6) and also as domain

25
The terms “contour” and “domain”
“ are used more or less synonymously, here. More precisely
spoken, the contour is the first
fi ring of elements around the
he respective domain.
report_Brocks-Rabbolini, wbrocks,
wbrocks 24 June 2015, - 20 -
integral, eq. (2.11), for contour #20 (manual option), Fig. 2.10, in ABAQUS. The test data
coincide satisfactorily with the FE results even a bit beyond maximum load, VL = 2 mm,
though the F(VL) curves start differing significantly, Fig. 2.8, since the opposing effects of
decreasing force and increasing ∆a cancel each other. Remember that the evaluation of J p
from the F(VL) curve according to eq. (2.10) does not consider crack extension. The
underlying F(VL) data include crack extension, ∆a, in the tests resulting in the load drop, but
do not in the simulation. Also note that J-results for VL > 2 mm have only numerical
significance. The “far-field” value of the domain-integral definition is supposed to lie close to
the J-definition as energy release rate, J_F(VL).

200 1000
M(T)150 #1.1.3 M(T)150 #1.1.3
J [N/mm]

J [N/mm]
J_20 (manual) J_20 (manual)
J_F(VL) 800 J_F(VL)
150

600
100
400

50
200

0 0
0 0,5 1 1,5 2 0 1 2 3 4 5
VL [mm] VL [mm]

Fig. 2.12: Comparison of FE results with test data #1.1.3


(a) VL ≤ 2 mm, F ≤ Fmax,test (b) total range of VL
The first contours, particularly in the automatic option, lie more or less in a near-field
around the crack tip where large plastic strains with corresponding rearrangement of stresses
occur, Fig. 2.11, which violates the requirements of “deformation theory of plasticity”. This
causes the “path dependence” of J, Fig 2.13. It is still not pronounced for VL ≤ 2 mm
(maximum force in the tests), Fig 2.13(a). Both, automatic and manual option yield a far-field
value for the domain #20 which coincides with the J-value calculated from F(VL), and even
J_01 (node set) of the manual option does so, though the respective domain covers a large
part of the plastic zone, Fig. 2.11(a). In the automatic option, J_01 = Jcrack-tip is significantly
lower, of course.
Due to increasing plastic deformations for VL > 2 mm, significant and growing path
dependence occurs, Figs. 2.13(b) and 2.14. For domain #1 of the automatic option,
J_01 = Jcrack-tip even becomes physically meaningless as J must not decrease with increasing
global deformation. J_20 (manual) differs slightly from J_F(VL) for VL > 3 mm, which calls
for even larger contours, see Fig. 2.12(b).

report_Brocks-Rabbolini, wbrocks, 24 June 2015, - 21 -


J [N/mm] 200 800

J [N/mm]
J_01 (automatic) J_01 (automatic)
J_01 (manual) J_01 (manual)
J_20 (automatic) J_20 (automatic)
150 600
J_20 (manual) J_20 (manual)
J_F(VL) J_F(VL)

100 400

50 200

0 0
0 0,5 1 1,5 2 0 1 2 3 4 5
VL [mm] VL [mm]

Fig. 2.13: Contour dependence of J: 1st contour = node set (Fig. 2.9)
(a) VL ≤ 2 mm, F ≤ Fmax,test (b) total range of VL
J increases with increasing size of the domain, Fig. 2.15, see BROCKS & YUAN 21. The finite
slope at the end of the J(contour) curve for VL = 4.5 mm indicates that no saturation value,
Jfar-field, has yet been reached.

800 800
J [N/mm]
J [N/mm]

J_01
J_02
J_03
600 J_05 600
J_10
J_15
J_20
J_F(VL)
400 400

200 200 VL=3mm


VL=4mm
VL=4.5mm

0 0
0 1 2 3 4 5 0 5 10 15 20
VL [mm] contour

Fig. 2.14: Contour dependence of J for Fig. 2.14: Contour dependence of J (manual
manual option of contour option)
definition, Fig. 2.10

report_Brocks-Rabbolini, wbrocks, 24 June 2015, - 22 -


Part 3: Simulation of ductile crack extension in a centre cracked panel,
M(T)

Geometry: width 2W = 300 mm,


initial crack length 2a = 60 mm,
thickness B = 2.9 mm
length 2L = 1.5·2 W = 450 mm
Material Al 5083 as above for exercises #1, #2
Young’s modulus E = 70300 MPa
Poisson’s ratio ν = 0.3
0.2% proof stress Rp0.2 = 242 MPa
yield limit = =0

200 5
δ5 [mm]
F [kN]

4
150

100

2
M(T)150 #1.1.3
M(T)150 #2.1.8
50
M(T)150 #1.1.3 1

M(T)150 #2.1.8

0 0
0 1 2 3 4 5 0 20 40 60 80
VL [mm] ∆a [mm]

Test results in “MT_150_Al5083_test.xls”

(1) Perform a numerical simulation of crack extension applying cohesive elements (see
SCHEIDER, I.: “The Cohesive Model - Foundations and Implementation“, Geesthacht
[2011])
(2) Compare simulation with test data in terms of force vs. load-point displacement,
F(VL), and CTOD R-curve, δ5 (∆a)
(3) Evaluate J by the domain integral method and compare to the “far-field” J evaluated
from F(VL,∆a) (see formulas given in exercise #2 and below).
(4) Investigate the effect of cohesive parameters σc, Γc

report_Brocks-Rabbolini, wbrocks, 24 June 2015, - 23 -


Cohesive law (traction-separation law)

suggested shape parameters:


1
δ1 = 0.05 δc
0,8
δ2 = 0.5 δc
σ / σc

0,6
“effective” cohesive parameters (plane-stress
0,4 model)
0,5
0,2 σc = 2.3 Rp0.2 = 560 MPa
0 Γc = 10 kJ/m2
0 0,2 0,4 0,6 0,8 1
δ / δc δc = 0.024 mm

Evaluation of J from F(VL, ∆a) for extending crack

“&ƒ) f&ƒ‚B) _%&ƒ) − f&ƒ) _%&ƒ‚B)


r&ƒ) = r&ƒ‚B) +
“&ƒ‚B) 2”“&ƒ‚B)

“ = H − b ; _% = _% − _%3 = _% − d% f
see also
• formulas in exercise #2
• BROCKS, W.; ANUSCHEWSKI, P.; SCHEIDER, I.: “Ductile Tearing Resistance of Metal
Sheets”, Engineering Failure Analysis 17 (2010), 607-616.

report_Brocks-Rabbolini, wbrocks, 24 June 2015, - 24 -


3.1. Analysis of test data including crack extension

Whereas the F(VL) curves of the two tests #1.1.3 and #2.1.8 differ significantly beyond
VL > 3 mm, Fig. 3.1, the δ5(∆a) curves nearly coincide, Fig. 3.2. The reason of this
discrepancy will be discussed and analysed below. In any case, it is another indication, that
local quantities like δ5 and ∆a are better suited for calibrating cohesive parameters by
comparison of test and simulation data than global quantities like F and VL.

200 5

δ5 [mm]
F [kN]

4
150

100
2

M(T)150 #1.1.3
50 elast 1
M(T)150 #2.1.8
M(T)150 #1.1.3
M(T)150 #2.1.8
0
0 0 20 40 60 80
0 2 4 6
VL [mm] ∆a [mm]

Fig. 3.1: Force vs. elongation (offset Fig. 3.2: CTOD R-curves
corrected)

Contrary to the unique CTOD R-curves, δ5(VL) and ∆a(VL) differ between the two tests,
Figs. 3.3 and 3.4. The source of this inconsistency of the test data can evidently be found in
the kinematics between (local) CTOD and the (global) elongation of the specimens.

5 80
∆a [mm]
δ5 [mm]

M(T)150 #1.1.3
M(T)150 #2.1.8
4 M(T)150 #1.1.3
60
M(T)150 #2.1.8

3
40
2

20
1

0 0
0 1 2 3 4 5 0 1 2 3 4 5
VL [mm] VL [mm]

Fig. 3.3: CTOD in dependence on VL Fig. 3.4: Crack extension in dependence


on VL

report_Brocks-Rabbolini, wbrocks, 24 June 2015, - 25 -


The maximum crack extension of more than 60% of the initial ligament, Fig 3.4 is quite
remarkable.
The test data, F(VL, ∆a), can be used to evaluate the J-integral according to its
interpretation as an energy release rate like in section 2.1, eqs. (2.6) to (2.9). The evaluation of
the plastic J now has to accounts for crack extension, ∆a. Since (a/W) is not constant anymore
and F changes with both VL and ∆a, a step-wise procedure is needed,
S S
–&—) 4&—˜™) xy&—) ‚4&—) xy&—˜™)
r& ) = r& ‚B) + , (3.1)
–&—˜™) C7–&—˜™)

where “ = H − b. (3.2)
Eq. (3.1) requires a f _% curve. The total load point displacement is split into an elastic
and a plastic part by
_% = _% − _%3 = _% − d% f , (3.3)
where the compliance, CL(a/W). results from eqs. (2.2) and (2.3). Beyond maximum load, _%3
stays constant whereas _% increases linearly with VL, Fig. 3.5.

5
1200
VLe, VLp [mm]

a=a0
J [N/mm]

Ve=CL*F
a=a0+Da
4 Vp=V-Ve
1000
V_L
800
3

600
2
400
1
200

0
0
0 1 2 3 4 5
0 1 2 3 4 5
VL [mm] VL [mm]

Fig. 3.5: Elastic and plastic parts of the total Fig. 3.6: J as energy release rate calculated
load point displacement according from F(VL, ∆a) neglecting and
to eq. (3.3) for specimen # 1.1.3 considering crack extension for
specimen # 1.1.3

Fig. 3.6 compares the results of the J evaluations for constant crack length, a = a0, and
crack extension, a = a0 + ∆a., where J p is calculated according to eq. (2.10) and eq. (3.1),
respectively. For given VL, crack extension reduces J.
If J is plotted vs. VL, Fig. 3.7, no differences between the test curves are observed since J
has been calculated from the global F(VL) curve which yields consistent results. A crucial
indicator, however, is the relation between the J-integral and CTOD, which is supposed to be
approximately linear,
œ
šj ~ • , (3.4)
S .u

report_Brocks-Rabbolini, wbrocks, 24 June 2015, - 26 -


as plotted in Fig. 3.8.26 It finally answers the question which of the respective two test curves
in Figs.3.1, 3.3 and 3.4 is more reliable. Actually, the test data of #2.1.8 violate the linear
relationship and have hence to be regarded as faulty. The problem with testing M(T)
specimens is generally the existence of two crack tips for which crack extension may be
varying. A local δ5(∆a) curve is not affected, Fig. 3.2, but the global F(VL) is, of course.

1000 1000
J [N/mm]

J [N/mm]
M(T)150 #1.1.3
800 800
M(T)150 #2.1.8

600 600

400 400

M(T)150 #1.1.3
200 200 M(T)150 #2.1.8
d_5*R_p0.2
0 0
0 1 2 3 4 5 0 1 2 3 4 5
VL [mm] δ5 [mm]

Fig. 3.7: Relation between the J –integral Fig. 3.8: Relation between the J-integral and
and elongation VL CTOD

Plotting J vs. ∆a, Fig. 3.9, yields resistance curves, JR, which are supposed to characterise
the material’s resistance to crack extension.27 Since the F(VL) curves of the two tests differ, so
the JR curves do, which is inconsistent with the CTOD R-curves, Fig. 3.2, because of the
reasons discussed above. There are also some oscillations beyond ∆a > 60 mm which should
be regarded as deficient.
Finally, the CTOD R-curves, Fig. 3.2, appear as reliable and significant experimental
reference data for identifying the cohesive parameters via numerical simulations.

26
The coefficient of proportionality appeared to be 1, correspondent to plane stress.
27
JR curves appeared to be geometry dependent, however, a problem generating a countless
number of publications.

report_Brocks-Rabbolini, wbrocks, 24 June 2015, - 27 -


1200
M(T)150 #1.1.3

J [N/mm]
1000 M(T)150 #2.1.8

800

600

400

200

0
Fig. 3.9: JR-curves from F(VL, ∆a), eqs. 0 20 40 60 80
(2.6) and (3.1) ∆a [mm]

3.2. FE model

The FE model is basically the same as in exercise #2, see Figs. 2.5 and 2.6. Instead of
fixing the vertical displacements in the ligament, b ≤ [B ≤ H, [C = 0, cohesive elements28
are applied in the ligament between b ≤ [B ≤ ab*:ž to allow for crack extension.

constraint equations cohesive parameters


&‚) &Ÿ)
ZC = −ZC delta_N = δc,

&‚) &Ÿ)
ZB = ZB
traction_N = σc/2

Fig. 3.10: Cohesive elements at symmetry line: Constraint equations and input variables
&‚) &Ÿ)
Symmetry29 is realised by applying linear constraint equations, ZC = −ZC , Fig. 3.1,
&Ÿ) &‚) &Ÿ)
which guarantee a symmetric opening of the cohesive element, š = ZC − ZC = 2ZC ,
30
implying delta_N = δc. These constraint equations induce extra nodal forces , however,
which reduce the stresses in the respective element to one half, so that the cohesive strength
traction_N = σc/2.

28
SCHEIDER, I.: The cohesive model – Foundation and implementation, Release 2.0.3, Report,
Helmholtz Centre Geesthacht, 2011.
29
BROCKS, W.; ARAFAH, D.; MADIA, M.: Exploiting symmetries of FE models and application to
cohesive elements, Report Milano / Kiel, November 2013.
http://www.tf.uni-kiel.de/matwis/instmat/departments/brocks/brocks_homepage_en.html
30
“Linear constraint equations introduce constraint forces at all degrees of freedom appearing in
the equations. These forces are considered external, but they are not included in reaction force
output. Therefore, the totals provided at the end of the reaction force output tables may reflect
an incomplete measure of global equilibrium.” - ABAQUS Analysis User's Manual (6.11): 33.2.1
Linear constraint equations.

report_Brocks-Rabbolini, wbrocks, 24 June 2015, - 28 -


The variables of user elements cannot be displayed by ABAQUS/Viewer. They are hence
transferred to the adjacent continuum elements as user output variables by providing an
element mapping between cohesive and continuum elements which is given in the .cohes file,
*ELEMENT MAP
cont_elem_1, coh_elem_1
cont_elem_2, coh_elem_2

cont_elem_n, coh_elem_n
The cohesive law of SCHEIDER26 is applied

( ) ( )
δ δ 2
2 δn1 − δn1 for δ n ≤ δ1


σ n (δ n ) = σ c ⋅  1 for δ1 < δ n ≤ δ 2 , (3.5)

( ) ( )
3 2
δ −δ δ n −δ 2
1 + 2 δnc −δ 22 − 3 δ c −δ 2 for δ 2 ≤ δ n ≤ δ c

with the cohesive energy


δc
1  2 δ1 δ 2 
Γ c = ∫ σ n (δ n ) dδ n = σ cδ c 1- + . (3.6)
0
2  3 δc δc 
For plane stress structures the thickness change on the ligament must be taken into
account31, which is calculated based on the in-plane strain components assuming isochoric
deformation,
= &1 + EE ) = &1 − BB − CC ). (3.7)
The keyword
*THICKNESS DEPENDENCE
assigns the plane stress cohesive elements to be thickness dependent. Since the thickness is
determined from the adjacent continuum elements, the keyword *ELEMENT MAP must be
used together with *THICKNESS DEPENDENCE.
A user defined output variable (UVARM9) in the UEL subroutine indicates completely
failed cohesive elements by taking an integer value of -1. The number of failed elements can
be counted, and crack extension, ∆a, is calculated by multiplying it with the element length of
0.125mm.
A far-field J-integral is calculated as described above in Part #2 for the contour shown in
Fig. 3.8.

31
SCHEIDER, I.; BROCKS, W.: Cohesive elements for thin-walled structures. Comp. Mat. Science
37 (2006), 101-109.

report_Brocks-Rabbolini, wbrocks, 24 June 2015, - 29 -


Fig. 3.11: Contour (domain) definition
for J-integral
integral calculation

3.3. Simulation Results - Comparison with Test Data.


Data

The first set of cohesive parameters suggested above is taken from SCHEIDER et al.
[2006]32. Different from the referenced publication, they did not yield a good approximation
of the experimental data,, however, see green curves in Figs. 3.12 and 3.13.
3.13 Remember that
the F(VL) curve of specimen #1.1.3 is the proper reference.

200 5

δ5 [mm]
F [kN]

4
150

100
2
M(T)150 #1.1.3
M(T)150 #1.1.3
M(T)150 #2.1.8
M(T)150 #2.1.8 FE (par set 1)
50 FE (par set 1) 1 FE (par set 2)
FE (par set 2) FE (par set 3)
FE (par set 3) FE (par set 4)
FE (par set 4) 0
0 0 20 40 60 80
0 1 2 3 4 5
VL [mm] ∆a [mm]

point displacement Fig. 3.13: CTOD δ5 vs. crack extension


Fig. 3.12: Force vs. load-point
parameter set δ1 / δc δ1 / δc σc [MPa] δc [mm] Γc [kJ/m2]

#1 0.05 0.5 560 0.024 10.0

#2 0.05 0.5 560 0.0365 14.8

#3 0.05 0.5 560 0,043 17.5

#4 0.05 0.5 560 0.049 20.0

Table 3.1:: Cohesive parameters applied in the simulations

32
SCHEIDER, I.; SCHÖDEL, M.; BROCKS, W.; SCHÖNFELD, W.: Crack propagation analyses with
CTOA and cohesive model: Comparison and experimental validation.
validation Engng. Fract. Mech.
Mech 73
(2006), 252–263.

report_Brocks-Rabbolini, wbrocks,
wbrocks 24 June 2015, - 30 -
The reason for the deviation between test data and the simulation is the substantial
overestimation of crack extension, Fig. 3.14. Further simulations were hence run with
increased values of the cohesive energy, Γc, as listed in Table 3.1, below. Increasing Γc
implies an increase of the critical separation, δc, according to eq. (3.6). Parameter set #4 failed
to converge at VL = 3.4 mm (∆a = 33.9 mm).
80 5
M(T)150 #1.1.3
∆a [mm]

δ5 [mm]
M(T)150 #1.1.3
M(T)150 #2.1.8
M(T)150 #2.1.8
FE (par set 1)
4 FE (par set 1)
60 FE (par set 2)
FE (par set 2)
FE (par set 3) FE (par set 3)
FE (par set 4) FE (par set 4)
3
40

20
1

0
0
0 1 2 3 4 5
0 1 2 3 4 5
VL [mm] VL [mm]

Fig. 3.14: Crack extension vs. load-point Fig. 3.15: CTOD vs. load-point displace-
displacement ment
The increase of “ductility” in the TSL delays crack extension, Fig. 3.14, and makes the
simulations approach the test curves, F(VL) and δ5(∆a). The dependence of the local CTOD
on the global deformation is much less affected by the cohesive parameters, Fig. 3.15.
Increasing Γc does also affect maximum force and the initiation values of J and CTOD, see
Table 3.2. More or less, the initiation values of the simulations lay within the scatter of the
measured values, and the overestimations of maximum force is less than 3%. Considering the
unavoidable errors in measuring initiation values, the simulations yield quite satisfactory
results, after all.
test #1.1.3 test #2.1.8 simul #1 simul #2 simul #3 simul #4
Fmax [kN] 172 172 169 174 175 177
¡šj | .B** DmmF 0.16 0.06 0.09 0.13 0.15 0.16
¡r| .B** DkJ⁄m F
C 32.3 20.4 21.2 27.8 29.5 39.2
Table 3.2: Effect of cohesive energy on Fmax and initiation values33 of J and CTOD
Fig. 3.16 shows the relation between the cohesive parameters δc and Γc and the “initiation”
values33 of the CTOD and J resistance curves.

33
“Engineering” definition of initiation at ∆a = 0.1 mm obtained from interpolation of data.
report_Brocks-Rabbolini, wbrocks, 24 June 2015, - 31 -
4
J_i / Γ_c
3,5 δ_i / δ_c

2,5

1,5

Fig. 3.16: Ratio of cohesive parameters δc 0,5


and Γc and the respective
0
initiation values33 of CTOD and J 1 2 3 4

The plots of equivalent plastic strain, εp, at ∆a = 40 mm and 80 mm in Fig. 3.17 indicate
that no contour can be found for calculating J which bypasses the plastic zone.
zone

Fig. 3.17: Plastic zones at ∆a = 40 mm and 80 mm


Due to the dissipation of external work by plastic deformation, thethe “far field” J-integral
values calculated for the contour shown in Fig. 3.11, have always to be less or equal the
limiting values of the (global) energy release rate calculated from the F(VL,∆a) curves, which
is confirmed by Fig. 3.18 for all parameter sets.
The evaluation of J as energy release rate from the F(VL,∆a)) curves requires the
calculation of _% = _% − _%3 according to eq. (3.3) and hence the elastic compliance CL. As
was found in Fig. 2.7,, already, the elastic compliance of the FE model is slightly lower,
lower i.e. its
stiffness is higher, than the value resulting from TADA’s equation. However, the compliance is
increased due to the cohesive elements, d% ‡ˆ = 0.98d%Š+‹+ , see Fig. 2.7(b).
2.7(b) Nevertheless,
applying TADA’s equation would yield artificial negative plastic deformations in the elastic
range. Hence, the compliance for calculating _%3 has been modified,
¡§ ©ª «

d% &b⁄H ) = ¦§ ®¯°¯ &&gy⁄6) ± d%Š+‹+ &b⁄H )
¬ →
. (3.8)
y 6

report_Brocks-Rabbolini, wbrocks,
wbrocks 24 June 2015, - 32 -
1000 1000
M(T)150 #1.1.3 M(T)150 #1.1.3
J [N/mm]

J [N/mm]
M(T)150 #2.1.8 M(T)150 #2.1.8
800 J_FE (contour) 800 J_FE (contour)
J_FE (en.rel.rate) J_FE (en.rel.rate)
par set #1 par set #2
600 600

400 400

200 200

0 0
0 1 2 3 4 5 0 1 2 3 4 5
VL [mm] VL [mm]

(a) (b)
1000 1000
M(T)150 #1.1.3 M(T)150 #1.1.3
J [N/mm]

J [N/mm]

M(T)150 #2.1.8 M(T)150 #2.1.8


800 J_FE (contour) 800 J_FE (contour)
J_FE (en.rel.rate) J_FE (en.rel.rate)
par set #3 par set #4
600 600

400 400

200 200

0 0
0 1 2 3 4 5 0 1 2 3 4 5
VL [mm] VL [mm]

(c) (d)
Fig. 3.18: J calculated as contour integral (Fig. 3.11) and energy release rate, cohesive
parameter sets #1 (a) to #4 (d)
The linear relationship between J and δ5 according to eq. (3.4) is confirmed by the
simulations independent of the cohesive parameters, Fig. 3.18.
Fig. 3.19 finally shows the simulated JR curves. The simulation applying parameter set #4
coincides with the experimental curve of specimen #1.1.3 up to a crack extension of more
than 30 mm.

report_Brocks-Rabbolini, wbrocks, 24 June 2015, - 33 -


1000
J [N/mm] 1000

J [N/mm]
800 800

600 600

400 FE (par set 3) 400


M(T)150 #1.1.3
M(T)150 #2.1.8 M(T)150 #2.1.8
FE (par set 1) FE (par set 1)
200 FE (par set 2) 200 FE (par set 2)
FE (par set 3) FE (par set 3)
FE (par set 4) FE (par set 4)
0 0
0 1 2 3 4 5 0 20 40 60
δ5 [mm] ∆a [mm]

Fig. 3.18: J-integral vs. CTOD Fig. 3.19: JR curves

Epilogue and Acknowledgement

The report presents the results of an exercise for the students of a PhD course on
Computational Fracture Mechanics held at the Dipartimento di Meccanica of the Politecnico
di Milano in March and April 2015. Beyond the direct results of the task which have been
used as benchmark for the evaluation of the students’ reports, additional information and
results are included which have not been part of the definition of the project but may serve as
background information.
The report can hopefully also give guidance to engineers and scientists working in the field
of numerical simulations of fracture problems how to achieve reliable results. Unfortunately,
even current contributions to international scientific journals contain erroneous conceptions of
the elastic-plastic J-integral and its path dependence.
The report also intends to show that not only numerical simulations have to be examined
critically but also experimental results have to be checked with respect to consistency. The
advantage of a simulation is, at least, that it yields reproducible results.
The authors thank MANFRED SCHÖDEL for making the test data available.

report_Brocks-Rabbolini, wbrocks, 24 June 2015, - 34 -

View publication stats

S-ar putea să vă placă și