1

CELL MODEL PREDICTIONS OF DUCTILE

CRACK INITIATION IN DAMAGED PIPELINES
Abstract This studyexplores the capabilities of acomputational cell frameworkinto a3-Dsetting
to model ductile fracture behavior in tensile specimens and in damaged pipelines. The cell method-
ology provides a convenient approach for ductile crack extension suitable for large scale numerical
analyses which includes a damage criterion and a microstructural length scale over which damage
occurs. Laboratory testing of a high strength structural steel provides the experimental stress-
straindatafor roundbar andcircumferentially notchedtensile specimens to calibrate the cell model
parameters for the material. The present workapplies the cell methodology usingtwo damage crite-
rion to describe ductile fracture in tensile specimens: (1) the Gurson-Tvergaard (GT) constitutive
model for the softening of material and (2) the the stress-modified, critical strain (SMCS) criterion
for void coalescence. These damage criteria are then applied to predict ductile cracking for a pipe
specimen tested under cycling bend loading. While the methodology still appears to have limited
applicability to predict ductile cracking behavior inpipe specimens, the cell model predictions of the
ductile response for the tensile specimens showexcelent agreemeent with experimental measure-
ments.
Claudio Ruggieri
1
, Fernando F. Santos
1
, Mitsuru Ohata
2
and Masao Toyoda
2
1. INTRODUCTION
Large scale straining and plastic deformation effects on fracture behavior of structural
steels play a key role in failure assessments of critical structures, including marine and nu-
clear facilities, oil and gas pipelines in onshore and offshore systems. Very large localized
plastic deformation which arises fromextreme or accidental loads such as, for example, col-
lisions of adjacent risers in deep water floating production systems (FPS) or earthquake-in-
duced loading on onshore and offshore pipelines, most often causes severe material damage
withsignificant reductioninductility and fracture toughness. Crack-like defects that devel-
op in such structures appear after initiation of ductile cracks in highly strained (damaged)
regions subjected to tension stresses. During service or further overload conditions, ductile
crack extension of these defects can lead to catastrophic (brittle) structural failures. Conse-
quently, advanced methodologies for ductile fracture assessments must include the effects
of large scale deformation and plastic straining on material failure. As a step in this direc-
tion, recent research efforts employ small scale tensile specimens to support development
of procedures to assess ductile crack initiation in structural components subjected to large
plastic deformation [1,2].
This study explores the capabilities of a computational cell framework into a 3-Dsetting
to model ductile fracture behavior in tensile specimens and in damaged pipelines. The cell
methodology provides a convenient approach for ductile crack extension suitable for large
1
Department of Naval Architecture and Ocean Engineering,
University of So Paulo, So Paulo, Brazil E-mail: [email protected]
2
Department of Manufacturing Science, Faculty of Engineering,
Osaka University, Suita, Osaka, Japan.
2
scale numerical analyses which includes a damage criterion and a microstructural length
scale over which damage occurs. Laboratory testing of a high strength structural steel pro-
vide the experimental stress-straindatafor roundbar andcircumferentially notchedtensile
specimens to calibrate the cell model parameters for the material. The present work applies
the cell methodology using two damage criterion to describe ductile fracture in tensile speci-
mens: (1) the Gurson-Tvergaard [3-4] constitutive model for the softening of material and
(2) the the stress-modified, critical strain (SMCS) criterion for void coalescence [2,9,10].
These damage criteria are thenappliedto predict ductile cracking for apipe specimentested
under cycling bend loading. While the methodology still appears to have limited applicabili-
ty to predict ductile cracking behavior for the pipe specimen, the cell model predictions of
the ductile response for the tensile specimens showexcellent agreement with experimental
measurements.
2. MICROMECHANICS CRITERION FOR DUCTILE CRACK GROWTH
Ductile fracture inmetals is amultistepmode of material failure incorporating the combina-
tion of microvoid nucleation, growth and coalescence at the microscale level (see, e.g., the
review of Garrinson and Moody [6]). The complex interplay of the key micromechanisms
leading to ductile failure underlies the need of additional consideration to describe the mi-
cromechanics of ductile tearing. Two widely used micromechanics approaches to describe
ductile fracture followed by material damage are: (1) the dilatant plasticity model of Gurson
[3], with modifications by Tvergaard [4], hereafter denoted GT, which predicts the softening
of material ina continuumcontext due to the idealized growthof a spherical voidor aperiod-
ic array of voids, and (2) the stress-modified, critical strain (SMCS) criterion for void coales-
cence of Hancock and co-workers [9,10].
Recent analytical efforts building uponthe GTmodel prompted the development of a nu-
merical approach using computational cells; a term recently coined by Xia and Shih (X&S)
[5]. X&S advocate a computational model for ductile growth which defines a single layer of
void-containing, cubical cells having linear dimension D along the crack plane on which
Mode I growth evolves. The cells have initial (smeared) void volume fraction denoted by f
0
.
The layer thickness (D) introduces a strong length-scale over which damage occurs; else-
where, the background material obeys the flow theory of plasticity without damage by void
growth. Material outside the computational cells, the background material, follows a con-
ventional J
2
flow theory of plasticity and remains undamaged by void growth in the cells.
As a further simplification, the void nucleates from an inclusion of relative size f
0
immedi-
ately uponloading. Progressive voidgrowthandsubsequent macroscopic material softening
in each cell are described with the Gurson-Tvergaard (GT) constitutive model for dilatant
plasticity [3,4] given by
g(
e
,
m
, , f) =

2
+2q
1
f cosh
3q
2

m
2

1 +q
3
f
2

=0 (1)
where
e
denotes the effective Mises (macroscopic) stress,
m
is the mean (macroscopic)
stress, is the current flow stress of the cell matrix material and f defines the current void
fraction. Under multiaxial stress states,
e
=(3S
ij
S
ij
2)
12
where S
ij
denotes the dev-
iatoric components of Cauchy stress. Factors q
1
, q
2
and q
3
introduced by Tvergaardimprove
the model predictions for periodic arrays of cylindrical and spherical voids; here we use
q
1
=1.25, q
2
=1.0 and q
3
=q
2
1
.
Further improvement in the cell methodology introduced by Ruggieri and Dodds (R&D)
[8] enables the model create new traction free surfaces to represent physical crack exten-
sion. When f in the cell incident on the current crack tip reaches a critical value, f
E
, the com-
putational procedures remove the cell thereby advancing the crack tip in discrete incre-
ments of the cell size (Tvergaard[4] refers to this process as the element extinctionor vanish
technique). R&D[8] implements a cell extinction process using a linear-traction separation
model with f
E
typically assigned a value of 0.15~0.20. The final stage of void linkup with
3
the macroscopic crack front then occurs by reducing the remaining stresses to zero in a pre-
scribed linear manner.
For the SMCS criterion, element extinction takes place when the equivalent plastic
strain,
p
, reaches a critical value given by,

p
c
=exp

(2)
where and are material dependent constants. Hancock and Cowling [9] and Mackenzie,
et al. [10] proposed the above form of the SMCS model based on experimental studies of
notched tensile specimens for structural steels. Panontin and Shepard [2] describe a com-
plete study of the calibration process to estimate and from notched-tensile data for an
A516 pressure vessel steel and an HY 80 steel.
3. EXPERIMENTAL MEASUREMENTS
Yasuda et al. [1] recently reported on a series of tension tests conducted on a C-Mn alloy
structural steel. Acylindrical (smooth round bar) andnotched tensile specimens withdiffer-
ent notchradius were extractedfroma 13 mmplate (rolling direction) to measure the stress-
strain response of the tested structural steel. The diameter of the central section for the
round bar tensile specimen is 6 mm with a gage length L=20 mm. The notched specimens
have a circumferential notch with a 1 mm and 2 mm notch root radius (R) and gage length
L=20 mm; the diameter of the central section for this specimen (deepest point of the notch
root) is also 6 mm. These notched tensile specimens are denoted R1 and R2 specimens. Fig-
ure 1 presents the geometry and dimensions of the tensile specimens used in the experi-
ments. The material has 384 MPa yield stress at room temperature (20 C) and relatively
high hardening properties (
u

ys
1.6). Figure 5 shows the measured true stress-logarith-
mic strain curve for the round bar and the notched tensile specimens. The solid symbols
mark the instability point (ductile failure) for each specimen.
To investigate ductile cracking behavior indamagedpipelines, full scale cyclic bendtests
were performed on a 165 mmO.Dtubular specimen with 11 mmwall thickness. The materi-
al is a pipeline steel with very similar mechanical characteristics to the structural steel
employed inthe tensiontests previously described. The stress-strainresponse for this mate-
rial follows relatively closely the tensile response shown inFig. 5 for the round bar specimen
with 315 MPa yield stress at room temperature (20 C) and similar hardening properties
(
u

ys
1.4). The tubular specimen was initially subjected to an indentation by 3-point
bend loading until the measured back surface strain reached 1%. This was followed by a
compressive axial loading with an applied axial displacement of 30 mm to generate large
localized buckling in the dented region. Next, the axial loading was reversed to a traction
loading applied until a visible ductile crack could be observed in the pipe surface. Fig. 2
shows an schematic illustration of the pipe specimen utilized in the cyclic bend tests. Fig.
7(a) presents the measured load vs. axial displacement for the pipe specimen; the solid sym-
bol marks the onset of ductile crack initiation observed in the pipe surface. Yasuda et al. [1]
provides additional details of the experimental program and materials employed.
4. FINITE ELEMENT MODELS
Figures 3(a) shows the finite element model constructed for 3-Danalyses of the notch round
tensile specimens (R2 notched specimen) utilized to measure the stress-strain response for
the structural steel employed inthis study. The geometry and size of the models matchthose
for specimens tested in the experiments previously described (see Fig. 1). To decrease execu-
tion time, the 3-Dfinite element model does not include modeling of the specimen rigid grip
shown in Fig. 1; numerical analyses indicate a negligible effect of this region on the tensile
response for the specimens. The rigid grip conditions are modelled by imposing X-direction
4
Figure 1 Geometry of tensile specimen employed in the experiments:
(a) round bar specimen; (b) circumferentially notched specimen.
(b) (a)
19 19
4
19
4
20 66
6
Unit: mm
19 19
4
19
4
20
6
Unit: mm
10
2
R1 , R2
andY-directionconstraints onthe external surface of the outermost layer (Z=L2). Symme-
try of the geometry and loading conditions enables analyses using only one-quarter, 3-D
model of the specimens. Appropriate constraints are imposed on the symmetry planes. Dis-
placement controlled loading of the model as indicated in Fig. 3(a) permits continuation of
the analyses once the load decreases during necking.
The 3-D model for the circumferentially notched specimen with R=2 mm has 24470,
8-node elements arranged into 28 variable thickness layers over the half-length (L2), as
illustrated in Fig. 3(a). The first 20 layers contain 9480 computational cells (each one mod-
elled by a cube of dimension D2D2D2) in a similar arrangement for the round bar
tensile specimento accommodate the notchroot geometry (see Fig. 1). Here, anapproximate
square of 474 computational cells for the first layer (Z=0) enables modeling of the planar
ductile crack extension well at the longitudinal center plane (deepest point of the notch re-
gion). Very similar mesh details are employed for the 3-D models of the circumferentially
notched specimen with R=1 mm and for the round bar specimen.
Figure 3(b) shows the finite element model constructed for the pipe specimen. The quar-
ter-symmetric model has 17402 8-nodes, 3-D elements with appropriate constraints im-
posed on the symmetry planes. To enable crack growth in the damaged surface of the pipe,
an approximate rectangle of 620 computational cells with dimension D2D2D2 are
arranged in the first layer (Z=0) as indicated in Fig. 3(c). Displacement controlled loading
of the models permits continuation of the analyses once the load decreases during crack
growth.
The analyses utilize a piecewise-linear approximation of the measured true stress-loga-
rithmic straincurve for the roundbar tensile specimenshowninFig. 5, but withthe uniaxial
true stress corrected for necking effects on plastic flow using Davidenkovs equation (see
additional details in Yasuda et al.[1]). Other mechanical properties needed for the analyses
5
Figure 2 Schematic illustration of the pipe specimen and the cyclic bend test.
800
Unit: mm
165
R80
80
Indentation Load
Compression
(Prebending)
Traction
include E=206 GPa and =0.3. The matrix material of the computational cell elements and
the void-free background material are assigned these properties. The three-dimensional
computations for the crack growth analyses reported here are generated using the research
code WARP3D [7].
5. NUMERICAL RESULTS
5.1. Calibration of the Cell Parameters and the SMCS Criterion
The cell size D and initial porosity f
0
define the key parameters coupling the physical and
computational models for ductile tearing. To apply the cell model to aspecific material, these
two parameters must be calibrated to bring the numerical predictions into agreement with
experimental measurements. Here, we choose the cell parameters D and f
0
which provide
the best fit to the experimentally measured stress-strain data for the material. The present
work adopts D2=100 m as the cell size parameter employed in the numerical analyses.
Such value provides a sufficiently refined mesh while, at the same time, avoiding potential
numerical difficulties with too low f
0
-values (the numerical stress-strain response of the
tensile specimen scales almost proportionally with D for fixed f
0
, i.e., a thicker layer re-
quires more total work to reach critical conditions).
With parameter D determined, the initial volume fraction, f
0
, remains as the only pa-
rameter unspecified. The calibration process focuses on determining a suitable value for the
initial volume fraction, f
0
, that produces the best fit to the measured tensile data for the ma-
terial. Figures 4 shows the measured and predicted true stress-logarithmic straincurves for
the R2 tensile specimen with defined by ln(A
0
A
i
) and for f
0
=0.005, 0.0075 and 0.01.
Inthis plot, the predictedtensile responses agree well withthe measured values up to strain
6

Figure 3 Finite element 3-D models for the tested specimens employed in the
analyses: (a) R2 specimen; (b) pipe specimen; (c) cells for growth analysis
in the pipe specimen.
L2
D
0
2
27022 nodes
24470 elements
Cells for Growth Analysis
Z =0
20370 nodes
17402 elements
Cells for Growth Analysis
Z =0
X
Y
Z
(a)
(b)
(c)
7
levels of 40 %for all f
0
-values but we note that the numerical predictions provide slightly
higher stress values for a fixed strain level. Suchbehavior is most likely associated withdis-
crepancies between the (numerical) elastic-plastic hardening model and the actual harden-
ing behavior for the material at higher strains. Most importantly, however, the prediction
of instability by ductile failure (point marked ) in the specimen also varies widely with
initial volume fractions; this point follows the onset of rapid crack growth with a marked
decrease of the load capacity for these specimens. Here, the predicted instability point
agrees well with the experimental measure for f
0
=0.0025. Figure 5(a) provides the mea-
suredandpredictedtrue stress-logarithmic straincurves for all the testedtensile specimens
For the calibrated f
0
-values, the predicted tensile response agrees well with the measured
values. The present analyses then provides the following calibrated values: f
0
=0.005
(round bar specimen) and f
0
=0.0025 (R2 and R1 specimens). While the calibrated f
0
-value
for the round bar specimen differs fromthe calibrated values for the notched specimens, we
consider f
0
=0.0025 as the material specific parameter for these analyses.
Figure 4 Comparison of measured and predicted true stress - logarithmic strain
curve for the R2 notched specimen.
Ductile Failure
(Instability)
f
0
=0.0025
f
0
=0.0075
Experimental Data
(MPa)
1000
1500
0
500
=ln(A
0
A
i
) (%)
0 60 80 100 40

f
0
=0.005
R2 Specimen
20
To provide a simpler extension of the SMCS criterion applicable in 3-D low constraint
configurations, the crackgrowthanalyses for the tensile specimens are nowemployed to cal-
ibrate parameters and . The procedure defines the critical combination of equivalent
plastic strain (
p
c
) and stress triaxiality (h=
m

e
) for the damaged material ahead of the
(internal) macroscopic crack at the instability point of the tensile specimens (solid symbols
inthe plot of Fig. 5). Operationally,
p
c
and hare takenat the cell withcurrent porosity f=0.1
which corresponds to a position between the cell currently undergoing extinction and the
peak stress location. At this position, the internal crack has extended 0.8~1.4 mm (8~14
cells in the X-direction and Y-direction) before final failure of the tensile specimens so that
stresses are decreasing rapidly and the void fraction is increasing sharply. Consequently,
the use of slightly different f values, other than 0.1, to define the crack-tip location for plot-
ting purposes does not appreciably alter the critical combination of
p
c
and h. Fig. 6 presents
the variation of stress triaxiality with equivalent plastic strain for the tested material using
8
Figure 5 Experimentally measured true stress - logarithmic strain curve for
the round bar and circumferentially notched tensile specimens.
Round Bar Specimen
(MPa)
1000
1500
0
500
=ln(A
0
A
i
) (%)
0 50 100 150
R2 Specimen
CMn Structural steel ( 20
o
C)
R1 Specimen
200
f
0
=0.005
f
0
=0.0025
f
0
=0.0025
the crack growth analyses for the tensile specimens. The calibrated parameters for the
SMCS criterion are =3.11 and =0.75.
5.2. Application to Ductile Crack Initiation of Damaged Pipeline
The exploratory application of the micromechanics methodology for ductile crack growth
compares predictions of ductile crackingbehavior andfailure loadfor the pipe specimensub-
jected to the cyclic bending test previously described. As already noted, the material is a
pipeline steel with very similar mechanical characteristics to the structural steel employed
in the tension tests.
Predictions of ductile cracking for the pipe specimen using the computational cell meth-
odology incorporating the GT model require specification of the cell parameters f
0
and D.
Because the cell response is governed primarily by the flowandtearing properties of the ma-
terial, using very similar steel grades withvirtually identical flowproperties might be satis-
factory for the calibration process. Here, we adopt the cell parameters previously calibrated
for the C-Mn structural steel ( f
0
=0.0025 and D=200 m) as the material-specific parame-
ters to predict ductile failure for the pipe specimens. However, the numerical analyses ex-
hibited significant problems related to the behavior of the GTmodel and, consequently, with
ductile crackgrowthinthe denting andbuckling regionof the specimens. Further investiga-
tion revealed very strong effects of the lowtriaxiality regions that develop during the entire
loading history of the tested pipe as the source of such problems. Development of the GT
constitutive model focused on describing void growth under conditions of high stress tri-
axiality, conditions typically found along a growing crack front remote from traction free
surfaces. The yield surface, g, depends exponentially on
m
, while in pure shear it has a
linear dependence on f . For the pipe specimens, very lowlevels of stress triaxiality develop
well at the regionwhere ductile cracking takes place (following the denting andbucking pro-
9
Figure 6 Calibrated variation of stress triaxiality with equivalent plastic
strain using the growth analyses for the tensile specimens.
Round Bar Specimen
h=
m

e
2.0
3.0
0
1.0

p
0 0.5 1.0 1.5
R2 Specimen
CMn Structural steel ( 20
o
C)
R1 Specimen
2.0
2.5
1.5
0.5
e
p
= 3.11 exp

0.75

m

cess). Consequently, void growth in the cells is suppressed during almost the entire loading
history of the specimen even though plastic strains display large values (
p
>50
0
).
The alternative procedure to predict ductile cracking in the pipe specimen utilizes the
SMCScriterionwithparameters and calibratedusing the growthanalyses for the tensile
specimens. Fig. 7(a) shows the load-load point displacement response for the pipe specimen.
Fig 7(b) presents the deformed shape of the pipe after indentation followed by a 30 mmaxial
compressive displacement. In contrast to the numerical analysis using the void growth cri-
terion(GTmodel), use of the SMCScriteriondoes predict ductile crackinitiation andductile
crack growth in the highly damaged area well inside the denting and buckling zone which
is in agreement with experimental observations (see Yasuda et al. [1]). However, the predic-
tion of pipe failure (point marked ) defined as the onset of crack growth still displays large
differences in comparison with the measured value (solid symbol).
6. CONCLUDING REMARKS
This study describes a 3-D computational framework to model ductile failure behavior in
cylindrical tensile specimens suitable to very large-scale computations. Our exploratory 3-D
analyses of round bar and circumferentially notched tensile specimens demonstrate the ca-
pability of the cell model to predict the tensile response and the ductile behavior including
the evolution of void growth. In particular, numerical results for the tensile specimen ana-
lyzed in this work predicts the instability point (ductile failure) in very good agreement with
experimental measurements. However, the methodology still appears to have limited appli-
cability to predict ductile crackingbehavior for pipe specimens whichdevelopvery lowlevels
of stress triaxiality in the damaged region. Clearly, additional experimental and numerical
studies are needed to further validate applications of the methodology based upon the GT
and the SMCS damage criteria. Nevertheless, the present work, when taken together with
previous analyses provide an additional body of results against which the robustness of the
10
Figure 7 Load-load point response for the testedpipe. (b) Deformed shape for
the tested pipe at the end of axial compressive loading.
P
2000
1000
1000
2000
0
40 30 20 10 0
(kN)
(mm)
Onset of Crack Growth

Numerical Prediction
(a)
(b)

11
cell model approach can be weighed. Ongoing work is currently underway to address these
issues for ductile cracking analyses of damaged pipelines.
Acknowledgements
This investigation is partially supported by the State of So Paulo Research Foundation
(FAPESP) through Grant 01/06919-9 and through a graduate scholarship (00/09819-2) pro-
vided for the first author (FFS).
REFERENCES
[1] Yasuda, O., Hirono, M., Ohata, M. and Toyoda, M., Ductile Crack Initiation Behavior of Pre-Strained
Steel, International Institute of Welding, IIW Doc. X-1461, 2000.
[2] Panontin, T. L. and Sheppard, S. D., The Relationship Between Constraint and Ductile Fracture Initia-
tion as Defined by Micromechanical Analyses in Fracture Mechanics: 26th Volume, ASTM STP 1256 (W.
G. Reuter, J. H. Underwood and J. C. Newman, Jr., Eds.), pp. 54-85, 1995.
[3] Gurson, A. L., ContinuumTheory of Ductile Rupture by Void Nucleation and Growth: Part I - Yield Crite-
ria and Flow Rules for Porous Ductile Media, Journal of Engineering Materials and Technology, Vol. 99,
pp. 2-15.
[4] Tvergaard, V., Material Failure by Void Growth to Coalescence, Advances in Applied Mechanics, Vol. 27,
pp. 83-151, 1990.
[5] Xia, L. and Shih, C. F., Ductile Crack Growth - I. A Numerical Study Using Computational Cells with
Microstructurally-Based Length Scales, Journal of the Mechanics and Physics of Solids, Vol. 43, pp.
233-259, 1995.
[6] Garrison, W. M, Jr. and Moody, N. R., Ductile Fracture, Journal of the Physics and Chemistry of Solids,
Vol. 48, pp. 1035-1074, 1987.
[7] Koppenhoefer, K., Gullerud, A., Ruggieri, C., Dodds, R. and Healy, B., WARP3D: Dynamic Nonlinear
Analysis of Solids Using a Preconditioned Conjugate Gradient Software Architecture, Structural Re-
search Series (SRS) 596, UILU-ENG-94-2017, University of Illinois at Urbana-Champaign, 1994.
[8] Ruggieri, C. and Dodds, R. H., Numerical Modeling of Ductile Crack Growth in3-DUsing Computational
Cell Elements, International Journal of Fracture, Vol. 82, pp. 67-95, 1996.
[9] Hancock, J., and Cowling, M., Role of Sate of Stress in Crack-Tip Failure Processes, Metal Science, Aug.-
Sept., pp. 292304,1980.
[10] Mackenzie, A., Hancock, J., and Brown, D., On the Influence of State of Stress on Ductile Fracture Initia-
tion in High Strength Steels, Engineering Fracture Mechanics, Vol. 9, pp. 167188, 1977.