You are on page 1of 26

Engineering Fracture Mechanics 73 (2006) 1653–1678

www.elsevier.com/locate/engfracmech

Evaluation of six fracture models in high velocity perforation


X. Teng *, T. Wierzbicki
Impact and Crashworthiness Lab, Room 5-218, Massachusetts Institute of Technology, 77 Massachusetts Avenue,
Cambridge, MA 02139, USA

Received 31 August 2005; received in revised form 4 January 2006; accepted 17 January 2006
Available online 28 February 2006

Abstract

A systematic evaluation of six ductile fracture models is performed to identify the most suitable fracture criterion for
high velocity perforation problems. Included in the paper are the Wilkins, the Johnson–Cook, the maximum shear stress,
the modified Cockcroft–Latham, the constant fracture strain, and the Bao–Wierzbicki fracture models. These six fracture
models are implemented into ABAQUS/Explicit by means of a user material subroutine (VUMAT), and applied to model
the failure processes of a steel and an aluminum target plate impacted by a projectile. The numerical simulations are exam-
ined by comparing with the experimental results published in the open literature. The Wilkins fracture model predicts
spallation of the impacted zone of the target plate beneath the projectile. This unrealistic result is due to its power law
form of the pressure term. The maximum shear stress fracture criterion fails to capture the shear plugging failure pattern
in a wide range of the impact velocity. Material fracture properties cannot be fully characterized with the constant fracture
strain and the modified Cockcroft–Latham fracture models. Various tensile tests on round bars do not give a consistent
critical damage. The calculated residual velocities of the projectile are sensitive to the magnitude of the fracture parame-
ters. The Johnson–Cook and the Bao–Wierzbicki fracture models formulated in the space of the stress triaxiality and the
equivalent plastic strain to fracture are capable of predicting the realistic fracture patterns and at the same time the correct
residual velocities. Finally, the limitations of the Johnson–Cook fracture model are discussed.
 2006 Elsevier Ltd. All rights reserved.

Keywords: Ductile fracture criteria; High velocity perforation; Element erosion

1. Introduction

Recently, Wierzbicki et al. [1] evaluated seven ductile fracture models from the aspect of calibration.
Included in the analysis were maximum shear stress, constant fracture strain, Johnson–Cook [2], Wilkins
et al. [3], Cockcroft–Latham [4], CrachFEM, and Xue–Wierzbicki fracture models. As a continuation of
the project, the present paper investigates the applicability of these fracture models to high velocity perfora-
tion problems.

*
Corresponding author. Tel.: +1 617 253 6055; fax: +1 617 253 8125.
E-mail address: xteng@mit.edu (X. Teng).

0013-7944/$ - see front matter  2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.engfracmech.2006.01.009
1654 X. Teng, T. Wierzbicki / Engineering Fracture Mechanics 73 (2006) 1653–1678

Nomenclature

A, B, n, C, q material constants in the Johnson–Cook constitutive model


cv specific heat
D accumulated damage
D1, D2, D3 three material constants in the Johnson–Cook fracture model
Dcr critical damage value
d diameter of the projectile
d0 initial diameter of the round bars
df diameter of the round bars at fracture
E Young’s modulus
h thickness of the target plate
J3 third invariant of the stress deviator
l diameter of the target plate
M0 mass of the projectile
p pressure, p = rh
pcr cut-off value for the tensile stress
R notch radius
Rcr critical volume in the Wilkins model
s1, s2, s3 three deviatoric principal stresses
T0 room temperature
Tm melting temperature
DT temperature rise
V0 initial impact velocity
Vr residual velocity
w1 the first term related to pressure in the Wilkins model
w2 the second term related to ratio of deviatoric principal stresses in the Wilkins model
hÆi Macauley bracket
a, c, b material constants in the Wilkins model
v ratio of plastic work converted to thermal energy
m Poisson’s ratio
q mass density
rij stress components
r1 maximum principal stress
rh hydrostatic stress
r
 equivalent stress
smax maximum shear stress
scr critical shear stress
epl effective plastic strain
ef effective plastic strain to fracture
e_ 0 reference strain rate
e_ pl current strain rate
deijpl increment of plastic strain components
n deviatoric state variable

Although many ductile fracture models were proposed in the literature long time ago, only few are cur-
rently available in material libraries of leading commercial finite element codes. For example, the constant
fracture strain and the Johnson–Cook fracture criterion have been incorporated into ABAQUS/Explicit
and LS-DYNA. The latest version of PAM-CRASH and LS-DYNA introduced the Wilkins fracture model.
Some of commercial finite element codes also provide interfaces that allow users to implement various
X. Teng, T. Wierzbicki / Engineering Fracture Mechanics 73 (2006) 1653–1678 1655

advanced material constitutive models and/or fracture models. The first objective of this research is to imple-
ment the fracture models above into ABAQUS/Explicit through a user material subroutine (VUMAT).
The second objective of the present paper is to identify the most suitable ductile fracture models for high
velocity impact situations. This task is fulfilled through a finite element analysis of the failure process of a tar-
get plate under projectile impact using various fracture models. One will be able to judge the correctness of the
numerical prediction by comparing with experimental results published in the open literature. Børvik et al.
[5–7] studied experimentally and numerically ballistic behaviors of Weldox steel plates of various thickness
impacted by a hard projectile. Børvik et al. [8,9] also investigated in a systematic way strength and fracture
properties of Weldox 460 E steel by performing tensile tests on round bars under different strain rates and
elevated temperature. The complete experimental data provide a solid platform to assess various fracture
models qualitatively and at the same time quantitatively.
The other material considered is 2024-T351 aluminum alloy, which is less ductile than Weldox 460 E steel.
Bao and Wierzbicki [10] published a comprehensive study on fracture properties of 2024-T351 aluminum alloy
by performing a number of tensile, compressive, and shearing tests and corresponding numerical simulations.
Based on these experimental and numerical investigations, Wierzbicki et al. [1] calibrated material constants
for the seven fracture models mentioned above. These fracture parameters will be directly used to predict fail-
ure patterns of a 2024-T351 aluminum alloy plate impacted by a hard, flat-nosed projectile.
Due to insufficient experimental data, The Xue–Wierzbicki and the CrachFEM fracture models will not be
evaluated here. For each fracture model considered, material parameters are first calibrated from the round
bar tensile tests. Then these fracture criteria are used to simulate perforation processes of the target plates
in a wide range of initial velocities. Finally, this paper concludes that the Bao–Wierzbicki and the John-
son–Cook fracture models formulated in the space of the stress triaxiality and the effective fracture strain,
are the most capable of predicting the fracture response of high velocity perforation problems.

2. Problem formulation

As a benchmark problem, chosen here is an extensive experimental and numerical study on ballistic behav-
iors of Weldox 460 E steel plates performed by Børvik et al. [5–7]. Perforation capability of three types of pro-
jectiles of flat, round, and conical noses were compared. The thickness of the target plates ranged from 6 mm to
30 mm and the impact velocity varied from 70 m/s to 400 m/s. For the purpose of the present study, only one
type of impact tests is considered in this paper: a circular target plate of thickness h = 12 mm impacted by a flat-
nosed projectile moving at various velocities from 184.3 m/s to 399.5 m/s. Two types of target materials are
introduced. One is Weldox 460 E steel which represents a relatively ductile material and the other is 2024-
T351 aluminum alloy, which is a less ductile material. Based on a number of component tests under large plastic
deformation, high strain rates, and elevated temperature, Børvik et al. [8,9] calibrated the Johnson–Cook plas-
ticity and fracture models for Weldox 460 E steel. The plasticity parameters taken from these papers are directly
used in the present investigation, see Table 1. The notations of the variables in Table 1 are defined in Section 3.
The fracture parameters will be calibrated for each of the six fracture models using the tensile test data pre-
sented in Ref. [8]. The plasticity data for 2024-T351 aluminum alloy are taken from a previous publication
of the present authors [11] (see Table 2) while the constants of the six fracture models are taken from Ref. [1].
The configuration of the target–projectile system is illustrated in Fig. 1. A two-dimensional finite element
model was built using axisymmetric elements with reduced integration (CAX4R), see Fig. 2. The integral
viscoelastic approach, which is the default option in ABAQUS/Explicit, was used to control hourglass

Table 1
Material constants for Weldox 460 E steel
E (GPa) m q (kg/m3) e_ 0 ðs1 Þ C cv (J/kg K)
200 0.33 7850 5.00 · 104 0.0123 452
Tm (K) T0 (K) q A (MPa) B (MPa) n
1800 293 0.94 490 383 0.45
1656 X. Teng, T. Wierzbicki / Engineering Fracture Mechanics 73 (2006) 1653–1678

Table 2
Material constants for 2024-T351 aluminum alloy
E (GPa) m q (kg/m3) e_ 0 ðs1 Þ C cv (J/kg K)
74.66 0.3 2700 3.33 · 104 0.0083 875
Tm (K) T0 (K) q A (MPa) B (MPa) n
775 293 1.0 352 440 0.42

Fig. 1. Geometry of a circular plate impacted by a flat-nosed cylindrical, hard projectile.

Fig. 2. Finite element model of the target plate impacted by the flat-nosed projectile.

deformation. In all the calculations, the total ‘‘artificial’’ strain energy associated with hourglass control is less
than 5% of the total plastic dissipation. The ‘‘hard’’ kinematic and frictionless contact condition was pre-
scribed between the projectile and the impacted zone of the target plate. The Johnson–Cook thermo-plasticity
model described in Section 3, was used to characterize the material properties of both Weldox steel and 2024-
T351 aluminum alloy. The projectile is assumed to be undeformable and is simply modeled as a rigid surface,
since it is much harder than the target.
Because elements undergo thermal softening in high velocity impact and element removal is used to
describe crack formation and propagation, finite element solutions would be sensitive to mesh size. The pres-
ent authors [12] investigated this issue and found that compared to ‘‘local’’ response such as stresses and
strains at a crack tip, ‘‘global’’ variables such as energy absorption and residual velocities are less dependent
on the element size. A similar conclusion was also drawn by Børvik et al. [6] for high velocity perforation
problems. In the present finite element model, 60 elements were defined through the thickness in the impacted
zone and thus the minimum element size is 0.2 mm · 0.2 mm. Such an element size is able to give a correct
prediction of the residual velocity in a wide range of the initial impact velocity [12].
In high velocity impact, a target plate undergoes complex loading. A shock stress wave generated immedi-
ately upon impact propagates through the target thickness. Compression dominates over the impacted
zone beneath a projectile. As the impacted zone moves downward, shear stresses develop through the target
X. Teng, T. Wierzbicki / Engineering Fracture Mechanics 73 (2006) 1653–1678 1657

thickness. Eventually, compression and shear will give way to bending and subsequently axial stretching. A
target plate may be either sheared off or torn apart. Hence, high velocity perforation is a very demanding test
to examine the applicability of a given fracture model.

3. Implementation into ABAQUS/Explicit

ABAQUS/Explicit provides two options for introducing a ductile fracture model into a computational
scheme. One option is the Johnson–Cook fracture model. The second option allows one to specify the fracture
strain as an arbitrary function of the stress triaxiality. While the above choices provide a fair amount of flex-
ibility in predicting fracture, many ductile fracture criteria to be evaluated in this paper do not belong to these
two classes. Therefore, for the purpose of the present comparison, these fracture models have to be imple-
mented into ABAQUS/Explicit by means of a vectorized user material subroutine (VUMAT). Since the intro-
duction of a fracture model interferes with the calculation of stresses and strains, a material constitutive model
also has to be implemented at the same time with a fracture model in the subroutine.
A user material subroutine basically calculates new stress components and updates state variables such as
the effective plastic strain and the damage indicator for each element at each time step. The present formula-
tion assumes that there is an additive relationship between elastic and plastic strain rates. The increment of the
effective plastic strain is determined using an elastic predictor, and radial return-mapping algorithm.
The target plate during the perforation process is subjected to large plastic deformation, high strain rates
and elevated temperature. Johnson and Cook [13] developed a constitutive model to describe material prop-
erties under dynamic loading. The von Mises yield surface with the associated flow rule is used in this material
model. Its isotropic hardening law includes effects of strain rates and temperature rise, defined by
h i  
e_ pl

T  T0
q 
n
 ¼ A þ Bepl 1 þ C ln
r 1 ; ð1Þ
e_ 0 Tm  T0
where r is the von Mises stress; epl is the effective plastic strain; A, B, n, C, and q are five material constants,
which need to be calibrated from experiments; e_ pl and e_ 0 are the current and reference strain rates; Tm and T0
are the melting and room temperature, respectively. This model accounts for isotropic strain hardening, strain
rate hardening, and temperature softening in the uncoupled form.
Due to high strain rates, heat generated by plastic work in a localized region does not have sufficient time
to escape to the surrounding material. The impact process considered in the present paper is assumed to be
adiabatic. The temperature increase DT owing to plastic work is written as
Z
v ij ij
DT ¼ r depl ; i; j ¼ 1; 2; 3; ð2Þ
qcv
where q is the density; cv is the specific heat; v = 0.9 is the ratio of plastic energy converted to heat; rij and deijpl
are the stress and incremental plastic strain components.
All the fracture models are assumed to be uncoupled with the material constitutive model in that the
condition for fracture is checked at each step outside the loop of stress and strain calculation. When an
accumulated damage indicator reaches a critical value, an element suddenly fails and completely loses its
load-carrying capability. Failed elements are removed to illustrate the formation and growth of cracks.
To verify the newly implemented material model, comparison with the existing Johnson–Cook’s model in
ABAQUS/Explicit was made on a single element subjected to uniaxial tension, compression, and finite shear,
as well as the necking response of a smooth round bar under dynamic tension. The newly implemented John-
son–Cook’s material model gives almost identical results to the one already existing in the ABAQUS/Explicit
material library.

4. Calibration of Weldox 460 E steel

Børvik et al. [8,9] performed a series of tensile tests on smooth and notched round bars to determine
strength and fracture properties of Weldox 460 E steel. These tests provided the global load–displacement
curves and the instants at the point of fracture represented by critical diameter reduction. However, fracture
1658 X. Teng, T. Wierzbicki / Engineering Fracture Mechanics 73 (2006) 1653–1678

Fig. 3. Geometry of the smooth and notched round bars (unit: mm).

is controlled by local stress and strain states as well as their histories. Hence, the tests alone are not sufficient to
identify material parameters of various fracture models. To this end, detailed numerical simulations of the ten-
sile tests on one smooth and three notched round specimens were conducted using ABAQUS/Standard.
Fig. 3 shows the geometry of the smooth and notched round bars designed by Børvik et al. [8]. The notches
of various radii introduce different levels of hydrostatic tensile stresses in the necking zone and lead to distinct
effective plastic strains to fracture. Four two-dimensional axisymmetric finite element models were built with
the minimum element size 0.2 mm · 0.2 mm in the gauge section, see Fig. 4. This element size is the same as

Fig. 4. Finite element models of the smooth and notched round bars under quasi-static tension.
X. Teng, T. Wierzbicki / Engineering Fracture Mechanics 73 (2006) 1653–1678 1659

Fig. 5. Tensile loads vs. normalized diameter reduction of the gauge section. The test points were taken from Børvik et al. [8].

that in the impacted zone of the target plate where cracks are expected to be generated. One end of the finite
element models was fixed and the other end was prescribed with a tensile displacement. Plots of tensile load vs.
diameter reduction of the gauge section are given in Fig. 5 showing very good agreement with experimental
results. It indicates that the plasticity parameters were correctly determined for the present model.
From the numerical simulations, one is able to track the development of stresses and strains of critical sites
until the instant at the point of fracture. In this paper, the fracture instant is determined by the experimentally
measured maximum (fracture) diameter reduction of the specimens. This approximately corresponds to the
formation of macroscopic cracks. These numerical results will serve as the input data for the calibration of
each fracture model.
Before calibration, the onset site of fracture need to be identified from the specimens. For a smooth round
bar under tension, a crack always initiates at the center of the necking zone, which experiences the largest plas-
tic deformation and the highest stress triaxiality at the same time. The situation becomes complex for notched
round bars. There are two competing sites of fracture initiation, i.e. the root of the notch and the center of the
gauge section, depending on the radius of notches, the microstructure of materials, the quality of surface, etc.
In the present paper, the fracture parameters for each model will be determined by checking in which of the
two possible sites the fracture criteria are met first.
Table 3 gathers the experimentally measured and numerically calculated local fracture quantities, in which
ef;N is the fracture strain of the center of the necking zone from the numerical simulation; ef;T is calculated from
the area reduction in terms of the diameter of the fracture cross-section df and the initial diameter of the gauge
section of the round bar d0
df
ef;T ¼ 2 ln ; ð3Þ
d0
r is the stress triaxiality of the gauge section defined by the ratio of the hydrostatic stress rh to the equiv-
rh =
alent stress r
; the subscript ‘‘av’’ indicates the average value of the stress triaxiality during a loading process:

Table 3
Fracture strains and stress triaxialities of the center of the necking zone for the four tensile tests
ef,T ef,N ðrrh ÞB ðrrh Þav
R=1 1.64 1.72 0.333 0.678
R = 4 mm 1.13 1.18 0.652 0.998
R = 2 mm 1.01 0.98 0.893 1.200
R = 0.8 mm 0.86 0.66 1.389 1.485
1660 X. Teng, T. Wierzbicki / Engineering Fracture Mechanics 73 (2006) 1653–1678

r  Z ef;N
h 1 rh
¼ depl ; ð4Þ
r
 av ef;N 0 r

ðrh =
rÞB is the calculated stress triaxiality of the center of the necking zone from the initial notch radius R using
Bridgman’s formula [14]
r   
h 1 d0
¼ þ ln 1 þ . ð5Þ
 B 3
r 4R

One can see large differences in the fracture strain and the stress triaxiality between the values determined from
the tests and numerically calculated at the point of crack formation.

5. Wilkins fracture model

Wilkins et al. [3] proposed a cumulative ductile fracture criterion as an integral function of the effective
plastic strain epl weighted by two terms w1 and w2 related, respectively, to the pressure and the ratio of devi-
atoric principal stresses,
Z epl
D¼ w1 w2 depl in Rcr ; ð6Þ
0

where D is the damage indicator. In the present application, the magnitude of the critical volume Rcr is taken
equal to the size of an element. The term w1 represents the effect of hydrostatic pressure p = rh on damage
accumulation, defined by
 c
1
w1 ¼ ; ð7Þ
1 þ ap
where a and c are two material constants. Note, that a has the dimension of Pa1. In contrast to the Wilkins
et al.’s formulation, the effect of pressure on material ductility is usually described in the literature through
the stress triaxiality, e.g. Johnson and Cook [2], Bao and Wierzbicki [10]. The second weighting term is
defined by
b
w2 ¼ ð2  AÞ ; ð8Þ
where b is a material constant and A is the ratio of deviatoric principal stresses s
 
s2 s2
A ¼ max ; ; s3 6 s2 6 s1 . ð9Þ
s3 s1
It can be shown that A = 1 at the symmetry axis of a round bar, and A = 0 for a round bar under torsion or a
plane-strain specimen. As Wierzbicki and Xue [15] pointed out, the term w2 is related to the normalized third
invariant of the deviatoric stress tensor.
Fracture occurs when the damage D reaches a critical value Dcr. This critical value is considered to be mate-
rial characteristics, independent of loading conditions, geometry, or size of a specimen. An element is also con-
sidered to fail if these two conditions are satisfied: p < pcr and epl > 0:0 where pcr is the critical hydrostatic
tensile stress, given by
1
pcr ¼  . ð10Þ
a
Mathematically, the first term w1 becomes a complex number if the hydrostatic tensile stress is higher than this
absolute value of pcr. Physically, the critical value may correspond to the occurrence of the spalling fracture
mode, see Fig. 6. Spallation is an outstanding failure mode in impact and explosive loading situations.
Under a restricted loading path of an axisymmetric specimen (A = 1) subjected to a constant hydrostatic
pressure, Eq. (6) could be integrated to give
ef ¼ Dcr ð1 þ apÞc ¼ Dcr ð1  arh Þc . ð11Þ
X. Teng, T. Wierzbicki / Engineering Fracture Mechanics 73 (2006) 1653–1678 1661

Fig. 6. Simplified Wilkins fracture model: the fracture strain as a function of pressure. Here the pressure is a constant during the loading
process and A = 1.

Plots of the fracture strain vs. the hydrostatic stress are shown in Fig. 6. The Wilkins fracture model indeed
represents well the fundamental characteristics of material ductility that the fracture strain decreases with the
increasing hydrostatic tensile stress. However, the asymptotic behavior for large hydrostatic tensile stresses is
not reasonably formulated, which will be demonstrated below.

5.1. Calibration

In the Wilkins fracture model, four material constants need to be identified: a, c, Dcr, and b. The first three
constants can be determined from the round bar tensile tests. If fracture initiates at the center of the necking
zone (the symmetry axis) where A = 1, the second term w2 does not contribute to damage accumulation. The
value of b can be simply obtained from either torsion or shear tests. But none of such tests were performed for
Weldox 460 E steel. Here, the value of b is calibrated from a perforation test using an inverse method.
Fig. 7 shows the relation between the pressure and the effective plastic strain of the point at the center of the
necking zone. These curves are input to solve Eq. (6) with w2 = 1. There are four tests available to determine

Fig. 7. Pressure vs. effective plastic strain up to fracture for the points at the center of the necking zone.
1662 X. Teng, T. Wierzbicki / Engineering Fracture Mechanics 73 (2006) 1653–1678

Fig. 8. Accumulative damage as a function of effective plastic strain for the points at the center of the gauge section and the notch root of
the round bar specimens.

three material constant Dcr, a, and c. A short MATLAB optimization code was programmed, which gives a set
of the fracture constants: Dcr = 5.16, a = 0.373 · 109 Pa1, and c = 3.99.
The value of b was determined from the perforation test of V0 = 200.4 m/s in combination with numerical
simulations. By forcing the numerically predicted residual velocity of the projectile to be as close as possible to
the experimentally measured one (Vr = 44.9 m/s), it turns out that b = 1.0 gives a satisfactory prediction of the
residual velocity (Vr = 44.5 m/s). However, the present inverse method is not general and the same parameter
b may lead to large errors in calculating residual velocities at different impact velocities. An additional shear
test for Weldox 460 steel is being planned as a joint project between Dr. Børvik and the present investigating
team.
In the above identification of the fracture parameters, it is assumed that the onset site of fracture is located
at the center of the necking zone. Another possible site is the notch root because of stress concentration. The
calibrated constants should be, therefore, examined by the calculation of damage accumulation at these notch
roots. Fig. 8 shows the evolution of the damage both at the notch root and at the center of the necking zone in
the four cases. It appears that in terms of the Wilkins fracture model the maximum damage always takes place
at the center of the necking zone, though the second term participates in damage accumulation at the notch
roots.

5.2. Application

Fig. 9 shows the numerically predicted failure patterns of the target plate in the cases of V0 = 200.4 m/s and
303.5 m/s using the Wilkins fracture model. Both cases exhibit spalling in the impacted zone where a number
of elements fail. Immediately upon impact, a shock wave of high magnitude is generated at the proximal sur-
face beneath the projectile. The shock wave is reflected as a tensile stress wave from the free distal surface.
Voids nucleate, grow, and coalesce in this wave propagation phase. Cracks may be generated inside the
impacted zone depending on spalling strength of materials and impact velocities. However, this is not the case
for the present problem. The spallation predicted by the Wilkins fracture model is not observed in the
retrieved plugs [6,7]. As shown in Fig. 10, the ejected plugs did not break into several pieces in a wide range
of impact velocities.
The unrealistic prediction of the fracture pattern is due to the formulation of the ‘‘pressure’’ term w1. Fol-
lowing Eq. (11), the accumulated damage has a vertical asymptote when pressure tends to the critical value pcr.
By contrast, the asymptotic behaviors of the Johnson–Cook and the Bao–Wierzbicki fracture models give a
more physically correct response.
X. Teng, T. Wierzbicki / Engineering Fracture Mechanics 73 (2006) 1653–1678 1663

Fig. 9. Predicted failure patterns of the target plate using the Wilkins fracture model. Here the failed elements are assumed to be still
capable of transmitting pressure but not tensile or shear stresses.

Fig. 10. Retrieved plugs and projectiles from a series of perforation tests on the target plates of various thickness [7].

6. Johnson–Cook fracture model

Johnson and Cook [2] proposed that material ductility can be expressed by an exponential function of the
stress triaxiality
 r 
h
ef ¼ D1 þ D2 exp D3 ; ð12Þ
r

where D1, D2, and D3 are three material constants. The expression of Eq. (12) was first suggested by Hancock
and Mackenzie [16]. This formulation that the fracture strain depends on the stress triaxiality can be traced
back to McClintock’s [17] and Rice and Tracey’s [18] pioneer theoretical studies on the enlargement of voids.
In particular, in the Rice and Tracey’s theoretical derivation, D3 =  1.5, and D2 is left free for calibration.
1664 X. Teng, T. Wierzbicki / Engineering Fracture Mechanics 73 (2006) 1653–1678

Another limiting case of Eq. (12) is the constant fracture strain for which D2 = 0. This case will be studied in a
separate section.
In a general case, the stress triaxiality is not a constant but varies during a loading process. Johnson and
Cook [2] assumed that damage accumulates in a linear way, i.e.
Z epl
1
D¼ depl . ð13Þ
0 
e f

An element fails when D reaches unity, i.e. Dcr = 1.0.


The constant D1 is an asymptotic value of the fracture strain as the stress triaxiality goes to infinity. This
value would be important if the JC fracture criterion is used to predict spalling. Recently, the JC fracture
model was further extended by Bao and Wierzbicki [10] as well as Wierzbicki and Xue [15]. The BW fracture
locus is similar to the JC fracture model for high stress triaxialities but has a different form than the JC model
in the range of low and negative stress triaxialities, see Fig. 26.
Due to the simplicity of formulation, the ease of calibration, and the wide availability of material constants
for many metals, the JC fracture model has found numerous applications in the literature. For example,
Børvik et al. [6] used the JC fracture criterion to simulate the present perforation problem. In contrast to
the approach taken by Børvik et al. [6], the present paper adopts the uncoupled approach, i.e. damage accu-
mulation is not taken into account in the calculation of stresses and strains.

6.1. Calibration

The JC fracture model can be calibrated from the tensile tests on the smooth and notched round bars in
combination with either theoretical analysis or finite element simulations. Here, the latter method is used.
The numerical solutions provide accurate information on the time history of stress and strain components
at the critical point. The fracture strains at the center of the necking zone for each test are given in Table 3.
In the present paper, a new optimization procedure is proposed to directly identify the three constants of
the JC fracture model. This procedure does not require to define the average magnitude of the stress triaxiality.
It is based on an error minimization technique. The nonlinear least-squares method incorporated in the MAT-
LAB Optimization Toolbox was used to solve D1, D2, and D3. The objective function is defined by
X
4
min fi2 ðxÞ; ð14Þ
x
i¼1

where x = [D1, D2, D3] is a vector; the subscripts i = 1, 2, 3, 4 correspond to the smooth bar and the notched
bars with R = 4 mm, 2 mm, and 0.8 mm, respectively; and fi is
Z ef;i
1
fi ðxÞ ¼ 1:0  depl;i ; ð15Þ
0 D1 þ D 2 expðD ri Þ
3 rh;i =

where the function of fi is numerically integrated using the trapezoidal method. A short MATLAB code was
programmed and gives the optimized set of the fracture constants: D1 = 0.06, D2 = 4.08, and D3 =  1.27.
This procedure is principally the same as the one proposed earlier by Bao et al. [19]. Comparison of fracture
loci calibrated from various approaches will be presented in another publication.
A plot of Eq. (12) with the above calibrated values of D1, D2, and D3 is shown in Fig. 11 as a smooth con-
tinuous curve. On the same graph, presented are the histories of the stress triaxialities of all the four specimens.
Note, that tensile tests on smooth and notched round bars cover only the range of stress triaxialities higher
than 1/3. In application, the fracture locus was extrapolated to the range of low and/or even negative stress
triaxialities following the exponential function. This extrapolation remains to be validated since neither com-
pression nor shear dominated tests on Weldox 460 E steel are currently available. Also, note that the cali-
brated JC fracture locus has a rather steep gradient so that the fracture strains at the zero and negative
stress triaxialities are quite large. This prevents from the development of the artificial erosion of the elements
in the impacted zone.
X. Teng, T. Wierzbicki / Engineering Fracture Mechanics 73 (2006) 1653–1678 1665

Fig. 11. Newly calibrated Johnson–Cook fracture locus for Weldox 460 E steel as well as the evolution of the effective plastic strain and
the stress triaxiality for the four tensile tests.

The development of the above JC fracture locus is based on the assumption that fracture would initiate at
the center of the necking zone. This point has the largest effective plastic strain and the highest stress triaxiality
at the same time. It is true for the smooth round bar and the notched round bar of R = 4 mm. For the notched
round bar of R = 2 mm, the effective plastic strain of the center of the necking zone is a little bit lower than
that of the notch root while the stress triaxiality of the former is three times higher than that of the latter.
Hence, for the case of R = 2 mm fracture would start at the center of the necking zone rather than at the notch
root. The problem becomes complicated for the case of R = 0.8 mm. A crack may start either at the center or
at the notch root, or cracks occur simultaneously at both sites. The time histories of the effective plastic strain
and the stress triaxiality of both sites were plotted in Fig. 11. It appears that the notch root has a higher plastic
strain but a lower stress triaxiality than the center of the gauge section at the point of fracture. The trajectory
of the center first reaches the Johnson–Cook fracture locus. Therefore, a crack would be formed first at the
center of the gauge section in the case with R = 0.8 mm in terms of the Johnson–Cook fracture model. Note,
that the same conclusion was obtained using the Wilkins fracture model.

6.2. Application

The JC fracture model predicts two clean cuts through the target thickness and the whole impacted zone is
ejected as a plug, see Fig. 12. As mentioned above, the large fracture strain under high compression prevents
the artificial element erosion in the impacted zone. The newly calibrated fracture locus gives ef ¼ 6:29 under
uniaxial compression ðrh = r ¼ 1=3Þ. The sharp increase of the fracture strain with the pressure effectively acts

Fig. 12. Numerically predicted failure pattern of the target plate at V0 = 285.4 m/s using the JC fracture model.
1666 X. Teng, T. Wierzbicki / Engineering Fracture Mechanics 73 (2006) 1653–1678

Fig. 13. Initial impact velocity vs. residual velocity predicted by the JC fracture model.

as the cut-off value for the negative stress triaxiality. The concept of the cut-off value was first proposed and
elaborated by Wierzbicki and Bao [20,21], see also Section 10. The captured failure patterns are consistent
with experimental observations. In general, the numerically calculated residual velocities are close to the
experimentally measured ones, see Fig. 13. However, the difference in the residual velocity increases as the ini-
tial impact velocity approaches the ballistic limit. The maximum error reaches 35% at V0 = 200.4 m/s.

7. Maximum shear stress fracture criterion

Following Tresca’s idea on yielding under multi-axial loading, the maximum shear stress fracture criterion
assumes that a crack initiates and grows in a body when the maximum shear stress of a material point reaches
a critical value. The maximum shear stress fracture criterion is commonly used, especially for brittle materials
that fail with little plastic deformation. In high velocity perforation, shear stresses are mostly responsible for
the formation and growth of through-thickness cracks. Therefore, it would be of interest to investigate the
applicability of the maximum shear stress fracture criterion to impact problems, which to the best of the
authors’ knowledge has not been reported in the literature.
Crack formation for a round bar under tension is mainly controlled by hydrostatic tensile stresses, although
shear stresses assist in crack propagation and take over in the formation of the cup-and-cone fracture surface.
At the symmetry axis of the specimens where fracture usually initiates, shear stresses are zero. It is not, there-
fore, correct to identify the critical value of the maximum shear stress from round bar tensile tests. No results
were found in the literature on shear or torsional fracture tests of Weldox 460 E steel. A perforation test is
used instead to determine the critical value of the shear parameter scr by forcing the numerically predicted
residual velocity to be as close as possible to the experimentally measured one.
The residual velocity is a decreasing function of the critical maximum shear stress. However, an interesting
discontinuity of the above relationship has been detected in the present research around scr = 525 MPa.
Fig. 14 shows an expanded scale plot of the numerically predicted residual velocity vs. the critical value of
the maximum shear stress for the case with V0 = 244.2 m/s. As the critical value of the maximum shear stress
slightly increases from 525.2775 MPa to 525.28 MPa, the predicted residual velocity suddenly jumps from
Vr = 174.3 m/s to 0.0. The numerically predicted residual velocity is always larger than the measured value
of Vr = 139.8 m/s no matter how much the fracture parameter is adjusted. The minimum relative error is
25% and there is no way to reduce it.
The extreme sensitivity of the numerically predicted residual velocity to the critical value of the maximum
shear stress can be clearly explained by tracking down the time history of the maximum shear stress of an ele-
ment at a potential crack path, see Fig. 15. In high velocity impact, elements undergo strain softening due to
X. Teng, T. Wierzbicki / Engineering Fracture Mechanics 73 (2006) 1653–1678 1667

Fig. 14. Numerically predicted residual velocity as a function of the critical value of the maximum shear stress for the case with
V0 = 244.2 m/s.

Fig. 15. The evolution of the maximum shear stress as a function of the effective plastic strain for a typical element at V0 = 244.2 m/s.

thermal effects. Instead of monotonically increasing with the effective plastic strain, shear stresses decrease
after reaching a maximum point. There exists a peak value of the maximum shear stress for an element.
For a simple shear case, the peak value can be theoretically determined from Eqs. (1) and (2), see, e.g.
Teng and Wierzbicki [11], and Bai and Dodd [22]. For a general three-dimensional problem, a peak value
indeed exists and can be obtained from numerical simulations. The peak value of the maximum shear stress
often serves as an indication of the formation of adiabatic shear bands [22]. For example, Mason and Wors-
wick [23] used the maximum shear stress fracture criterion to predict adiabatic shear banding and following
fracture for punch problems, in which the maximum shear stress was automatically determined by a finite ele-
ment code.
If a critical value of the maximum shear stress is higher than the peak value in the calculation (solid line in
Fig. 15), an element will never fail, though it continues to deform plastically. Fig. 16 shows the predicted fail-
ure pattern of the plate using the maximum shear stress fracture criterion. The remainder of the ligament
undergoes very large plastic distortion but the target does not completely fracture. Usually, cracks would
be formed after adiabatic shear bands have been fully developed in high velocity impact. If the critical value
of scr is lower than the peak value (dash line in Fig. 15), the target would prematurely fail.
1668 X. Teng, T. Wierzbicki / Engineering Fracture Mechanics 73 (2006) 1653–1678

Fig. 16. Fracture pattern of the target plate for the case with V0 = 244.2 m/s. The critical value of the maximum shear stress implemented
in the calculation is 525.3 MPa. The upper part of the ligament has failed.

It can be concluded that the maximum shear stress is not a sufficient fracture criterion for high velocity
impact problems. It may also be true that stress-based fracture criteria would be unable to correctly simulate
ductile failure processes under dynamic loading.

8. Modified Cockcroft–Latham fracture criterion

Cockcroft and Latham [4] postulated that ductile fracture is controlled by an integral of the maximum prin-
cipal stress r1 with respect to the effective plastic strain,
Z ef
Dcr ¼ hr1 i depl ; ð16Þ
0

where hr1i is the Macauley bracket, which renders the value of r1 if r1 > 0, and zero if r1 6 0. Note, this dam-
age indicator has the physical dimension of energy density. Oh et al. [24] modified this fracture criterion by
normalized the maximum principal stress by the equivalent stress
Z ef
hr1 i
Dcr ¼ depl . ð17Þ
0 r

This modified CL fracture model will be used in the present paper. The original CL fracture criterion and its
various modifications have found a wide application in metal bulk forming processes but not much in impact
engineering.
Under the plane-stress condition, Wierzbicki and Werner [25] successfully transformed Eq. (17) to the space
of the equivalent fracture strain and the stress triaxiality:
Dcr
ef ¼ . ð18Þ
1 þ 3rh =
r
As shown in Fig. 17, the set of upsetting tests corresponding to the range of the negative stress triaxiality down
to 1/3 can be nicely approximated by the modified CL fracture criterion (the solid curve). It indicates that
compressive tests on short cylinders of various ratios of height to diameter will give a consistent critical dam-
age for this fracture model. However, the critical values calibrated from tensile tests on smooth and notched
round bar specimens scatter in a wide range [26].
The existence of two distinct branches of a fracture envelope is supported by the observation that there are
two basic failure mechanisms: void growth and ‘‘shear decohesion’’. Specimens usually fail by void growth in
the range of high stress triaxialities and by shear fracture in the negative range. One single critical damage
would be impossible to represent fracture properties of a material under different loading conditions. The
modified fracture model may work well in the negative range of the stress triaxiality or in a narrow positive
range while in impact problems, the stress triaxiality constantly changes from large negative to large positive
values.
X. Teng, T. Wierzbicki / Engineering Fracture Mechanics 73 (2006) 1653–1678 1669

Fig. 17. Fracture points calibrated from a series of upsetting tests on short cylinders and tensile tests on round bars for 2024-T351
aluminum alloy. These data were taken from Ref. [10].

8.1. Calibration

Recognizing the above fundamental difficulty, it is still interesting to see what will happen if the modified
CL fracture criterion is calibrated from the tensile tests. Fig. 18 shows plots of the normalized maximum prin-
cipal stress vs. the effective plastic strain up to fracture. According to Eq. (17), the areas below each curve rep-
resent the critical value of the criterion, which are listed in Table 4. Similar to the case with 2024-T351
aluminum alloy, the tensile tests on Weldox 460 E steel do not give a unique value of the critical damage

Fig. 18. Normalized maximum principal stresses vs. effective plastic strains at the center of the necking zone for the smooth and notched
round bars under tension.

Table 4
The critical damage of the modified CL fracture model for Weldox 460 E steel
Smooth R = 4 mm R = 2 mm R = 0.8 mm
Section center Notch root Section center Notch root Section center Notch root
2.32 1.96 1.11 1.83 1.15 1.42 1.56
1670 X. Teng, T. Wierzbicki / Engineering Fracture Mechanics 73 (2006) 1653–1678

Fig. 19. Residual velocity vs. initial impact velocity based on the modified CL fracture model.

for the modified CL fracture model. There is a large difference in the critical damage parameter depending on
the value of the notch radius. It can be concluded that only one material constant is not sufficient to charac-
terize fracture properties of a material in the range of high stress triaxialities.
Table 4 also includes the maximum damage at the notch root of the specimens. It appears that for the case
with R = 0.8 mm, the maximum damage at the notch root is a little bit larger than at the center of the necking
zone, i.e. fracture would initiate at the notch root in terms of the modified CL fracture model. This conclusion
is contradict to those obtained from the Wilkins and the JC fracture criteria.

8.2. Application

A question arises what is the sensitivity of the solution on the assumed damage parameter. The numerical
simulation was performed taking two extreme values of the damage parameter. It is found that the modified
CL fracture model is able to yield satisfactory fracture patterns of the target plate. Since we implicitly assume
that damage would not accumulate if r1 6 0, the artificial erosion of the elements in the impacted zone, where
compression is dominant and the maximum principal stress may be negative, does not take place. This
assumption is equivalent to the cut-off value for the negative stress triaxiality proposed by Wierzbicki and
Bao [20,21]. Two cracks are generated through the target thickness for each case with the initial impact velo-
city higher than the ballistic limit. The fracture pattern is similar to the one shown in Fig. 12.
It can be observed from Fig. 19 that Dcr = 1.42 gives the closest agreement with the experimental data,
while the residual velocities, especially the ballistic limit, were underestimated in the case of Dcr = 2.32. The
difference in the predicted residual velocity is shaded in Fig. 19. The large difference occurs near the ballistic
limit and the difference gradually disappears as the impact velocity increases. It can be concluded the CL frac-
ture criterion provides a correct qualitative solution (correct failure mode) because the effect of the cut-off
value is accounted for automatically. At the same time quantitative results may have large errors considering
the non-uniqueness in the determination of the fracture parameter.

9. Constant fracture strain

Due to simplicity, a constant effective plastic strain to fracture is widely used to simulate crack formation
and growth. This fracture criterion is indeed able to fulfill the task in cases where the stress triaxiality varies
spatially and temporally in a narrow range. It becomes unsuitable for complex loading cases such as high
velocity perforation.
The critical value of this fracture model can be calibrated from the round bar tensile tests. However, the
fracture strain obtained is different depending on which test is taken for calibration. This same type of
X. Teng, T. Wierzbicki / Engineering Fracture Mechanics 73 (2006) 1653–1678 1671

Fig. 20. Predicted failure process of the target plate using the constant fracture strain ef ¼ 0:66 without the cut-off value at V0 = 285.4 m/s.

difficulty as encountered in the CL fracture criterion is magnified by the non-existence of the cut-off value
property for the negative stress triaxiality. The fracture strain determined from the smooth bar is almost three
times higher than its counterpart from the notched bar with the smallest notch radius, see Table 3. This section
will investigate the sensitivity of numerically predicted residual velocities and fracture patterns to the magni-
tude of the fracture strain. Calculations were run for two limiting cases: ef ¼ 1:72 and 0.66.
Fig. 20 shows the predicted failure process of the target plate at V0 = 285.4 m/s using the constant fracture
strain ef ¼ 0:66. It appears that a number of elements around the sharp corners of the projectile are sequen-
tially eroded. This fracture pattern is not realistic. The artificial erosion of the elements in the impacted zone

Fig. 21. Residual velocity vs. initial impact velocity based on the constant fracture strain.
1672 X. Teng, T. Wierzbicki / Engineering Fracture Mechanics 73 (2006) 1653–1678

does not occur in the case with ef ¼ 1:72. The fracture model with such a critical value predicts a reasonable
fracture pattern, similar to the one by the JC fracture criterion shown in Fig. 12. However, the residual veloc-
ities of the projectile are overestimated, see Fig. 21. By artificially raising the magnitude of the fracture strain,
one would possibly predict realistic fracture patterns and correct residual velocities at the same time. But this
fracture criterion would lack generality. For instance, the tensile fracture of the smooth and notched round
bars would not be correctly simulated if the fracture strain is higher than 1.72.
One of the main reasons that the simple constant critical strain fracture criterion is inadequate in high
velocity impact situations is that no distinction is made between material ductility in tension and compression.
This deficiency has been corrected by Bao and Wierzbicki [10] as explained in the next section.

10. Bao–Wierzbicki fracture criterion

Wierzbicki and Bao [20] performed a careful revisit of classical Bridgman’s experiments [14] on the depen-
dence of material ductility on hydrostatic pressure. They found that a ductile material would never fail if the
stress triaxiality is less than 1/3. A similar conclusion also follows an analysis of upsetting tests [21]. This
feature makes the Bao–Wierzbicki fracture locus distinct from other models. They also found that the fracture
locus would exhibit three branches in the whole range of the stress triaxiality as a result of the competition of
two failure mechanisms: void growth and ‘‘shear decohesion’’, see Fig. 26 for an example.
Since neither shear- nor compression-dominated tests were performed, the full Bao–Wierzbicki fracture
locus covering the whole range of the stress triaxiality cannot be constructed for Weldox 460 E steel at the
moment. Here, a simplified Bao–Wierzbicki fracture locus is proposed that the cut-off value for the negative
stress triaxiality is combined with a constant equivalent strain to fracture in the range of rh = r > 1=3, see
Fig. 22. In contrast to the conventional constant fracture strain criterion in the preceding section, this fracture
model always leads to a correct qualitative description of the perforation process. The predicted failure pattern
is almost identical to the one based on the JC fracture criterion, see Fig. 12. There are still differences in the
calculated residual velocities. It can be seen from Fig. 23 that the results corresponding to the smooth bar frac-
ture strain of ef ¼ 1:64 give very good agreement with the test results.

11. Application to 2024-T351 aluminum alloy plate

Bao and Wierzbicki [10] performed a series of tensile, compressive, and shearing tests on 2024-T351 alumi-
num alloy. From these tests and corresponding numerical simulations, Wierzbicki et al. [1] identified the frac-
ture constants of the six fracture models considered in the preceding sections. Interested readers can be

Fig. 22. Simplified Bao–Wierzbicki fracture locus with the cut-off value for the negative stress triaxiality.
X. Teng, T. Wierzbicki / Engineering Fracture Mechanics 73 (2006) 1653–1678 1673

Fig. 23. Residual velocity vs. initial impact velocity based on the simplified Bao–Wierzbicki fracture locus.

referred to that paper for the detailed calibration procedures for each fracture model. Since no perforation
tests on 2024-T351 aluminum alloy were reported in the literature, this section focuses attention only on
the prediction of the fracture pattern but not on the residual velocity. Using the notations of the present paper,
the magnitude of the fracture constants of the above six models are given in Table 5.
Similar to the case with Weldox 460 E steel, the Wilkins fracture model also predicts spallation in the
impacted zone for 2024-T351 aluminum alloy at V0 = 245 m/s, see Fig. 24. The spalling failure mode was
not observed in a perforation test at such a level of impact velocity. Fig. 25 shows the fracture pattern of a
2024-T351 aluminum plate impacted by a flat-nosed steel projectile at V0 = 245 m/s reported by Mescall [27].
It should be noted that the parameters for the JC fracture model were taken from Ref. [1] rather than the
original paper by Johnson and Holmquist [28]. In this paper the fracture envelope was determined from only
one shear test with some additional assumptions. The present authors do not think these assumptions are

Table 5
Fracture parameters of the six fracture models for 2024-T351 aluminum alloy
Original Wilkins model and Enhanced Wilkins Kamoulakos model (EWK) Dcr = 0.93, a = 1.2 · 109 Pa1, c = 2.15, and b = 2.18
Johnson–Cook model D1 = 0.0, D2 = 1.0, and D3 = 1.94
Maximum shear stress scr = 380 MPa
Modified Cockcroft–Latham model Dcr = 0.058–0.485
Empirical Bao–Wierzbicki model See Fig. 26
Constant fracture strain ef ¼ 0:21

Fig. 24. Predicted failure pattern of the 2024-T351 aluminum alloy plate at V0 = 245 m/s using the Wilkins fracture model. The failed
elements are assumed to be capable of transmitting pressure but neither shear nor tension.
1674 X. Teng, T. Wierzbicki / Engineering Fracture Mechanics 73 (2006) 1653–1678

Fig. 25. Failure pattern of a 2024-T351 aluminum alloy plate impacted by a steel projectile at V0 = 245 m/s [27].

justified. Therefore the present simulation was performed based on our own tests. Comparison of the three
fracture loci discussed above is given in Fig. 26. The difference between the newly calibrated JC and the
BW fracture loci is shown by the shaded area in Fig. 26. Both fracture loci give almost identical result in terms
of the fracture pattern, see Fig. 27.
The application of the maximum shear stress predicts the formation of a cylindrical plug. However, the
plug is always attached by a thin ligament to the target, see Fig. 28. Therefore, the maximum shear stress frac-
ture criterion is also unable to predict correctly the failure pattern of the aluminum plate.

Fig. 26. Three fracture loci for 2024-T351 aluminum alloy.

Fig. 27. Predicted failure pattern of the 2024-T351 aluminum alloy plate at V0 = 245 m/s using the JC fracture model calibrated by
Wierzbicki et al. [1]. The empirical three-branch Bao–Wierzbicki fracture locus gives an almost identical fracture pattern.
X. Teng, T. Wierzbicki / Engineering Fracture Mechanics 73 (2006) 1653–1678 1675

Fig. 28. Predicted failure pattern of the 2024-T351 aluminum alloy plate at V0 = 245 m/s using the maximum shear stress fracture model
with scr = 380 MPa.

A constant fracture strain ef ¼ 0:21 for 2024-T351 aluminum alloy was determined in Ref. [1] from a tensile
test on a transverse plane-strain specimen, where a slant fracture surface was generated due to shear stresses.
Applying this fracture criterion to the present perforation problem, one can see that the failure pattern pre-
dicted is completely wrong not only a large volume of materials (artificial eroded elements) disappear but
the pattern of cracks does not correspond to reality, see Fig. 29. This simulation of the aluminum alloy plate
is actually much worse than the case of Weldox steel.
Finally, no surprise is found in the case of the modified CL fracture model. Qualitatively, the numerical
prediction of the perforation process is correct but the residual velocity depends on the magnitude of the frac-
ture parameter. Bao and Wierzbicki [26] found that the critical values calibrated from different types of tests
scatter in a wide range. The tensile test on a smooth round bar gives the critical damage 0.485, approximately
10 times higher than the data obtained from an upsetting test on a short cylinder.

Fig. 29. Predicted failure pattern of the aluminum plate using the constant fracture strain ef ¼ 0:21 w=o the cut-off value at V0 = 245 m/s.
The blank area represents the artificially eroded elements.
1676 X. Teng, T. Wierzbicki / Engineering Fracture Mechanics 73 (2006) 1653–1678

12. Discussion and conclusions

In terms of predicting failure patterns of high velocity perforation problems, it is the asymptotic behaviors of
fracture models that are important. If the asymptotic value of a fracture locus is established incorrectly in the
range of high stress triaxialities, the unrealistic spallation of a target may occur. Similarly, the asymptotic value
of a fracture locus in the range of negative stress triaxialities should either have a cut-off value or a sufficiently
large magnitude of ductility. Otherwise, the artificial element erosion would occur in the impacted zone of a
target beneath a projectile. In the intermediate range of the stress triaxiality, a fracture locus must be deter-
mined from several tests rather than a single test in order to get a good prediction of residual velocities.
The Wilkins fracture model predicts spallation of the impacted zone for both Weldox 460 E steel and 2024-
T351 aluminum alloy. Such a failure mode was not observed in the tests. This unrealistic prediction is due to
the power law form of the pressure term with a vanishing ductility at a specified pressure. The formulation of
the Wilkins fracture model need to be further examined experimentally and theoretically.
The JC fracture model is able to predict the realistic fracture patterns and at the same time the correct resid-
ual velocities for the present perforation problem. It should be pointed out that the newly developed calibra-
tion procedure leads to the very steep fracture locus, i.e. high fracture strains in the negative range of the stress
triaxiality. This prevents the artificial erosion of the elements in the impacted zone. By contrast, the JC frac-
ture criterion calibrated in a conventional way, such as the one proposed by Børvik et al. [8], may lose its
applicability in a loading case dominated by compression, e.g. the Taylor test with fracture [29].
At the same time, some issues related to the JC fracture model remain to be clarified. For example, is one
single monotonically decreasing fracture locus really able to represent ductile fracture properties in the whole
range of the stress triaxiality? McClintock [30] compared equivalent plastic strains to fracture between in ten-
sion and in torsion for eight metals. Six cases have a greater fracture strain in tension than in torsion. The
fracture loci of these materials would have three distinct branches in the space of the effective fracture strain
and the stress triaxiality. As shown in Fig. 26, there exists large difference between the empirical Bao–Wierzb-
icki fracture locus and the JC fracture locus for 2024-T351 aluminum alloy. This JC fracture criterion would
not be able to correctly predict crack formation in the case dominated by shear/compression.
The maximum shear stress fracture criterion is not able to give realistic shear plugging fracture patterns in a
wide range of impact velocities. Because of thermal softening, stresses do not monotonically increase with
strains under dynamic conditions. A specific critical value for the maximum shear stress would predict either
premature failure or no complete failure of the target plate. Stress-based ductile fracture criteria do not seem
promising for high velocity impact problems.
The modified Cockcroft–Latham fracture model uses only one constant to characterize fracture properties
of a material. However, it is difficult to decide what type of test should be used in calibration; upsetting, shear,
or tension, etc. Due to this ambiguity, errors in predicting residual velocities are large for both types of high
and low ductility materials. Therefore, this fracture model, as applied to high velocity impacts, lacks generality
in the aspect of either calibration or application.
The constant fracture strain predicts the artificial erosion of the elements in the impacted zone at a high
impact velocity. The simplistic version of the Bao–Wierzbicki fracture model, which introduces the cut-off
value for the negative stress triaxiality, can remedy this deficiency. However, the magnitude of the fracture
strain has a strong influence on the calculation of the residual velocity. Similar to the modified Cockcroft–
Latham fracture model, both the conventional constant critical strain and the simplified Bao–Wierzbicki frac-
ture models fails to give satisfactory results for a wide range of problems.
The Bao–Wierzbicki fracture model of three branches has correct asymptotic behaviors for both large neg-
ative and positive stress triaxialities. It also correctly represents the material ductility in the entire range of the
stress triaxiality. This fracture locus has been applied to many impact problems: rigid mass-to-beam/plate
impact, the Taylor tests with fracture, ductile crack formation within adiabatic shear bands, etc. A number
of fracture patterns encountered in impact engineering have been successfully recreated. These results were
presented in the PhD thesis of one of the present authors [31]. The downside of this fracture model is that
requires several different tests for calibration.
At the very end, it has been noted that all the fracture criteria formulated in the space of the equivalent
fracture strain and the stress triaxiality are unable to distinguish between different material ductility in axisym-
X. Teng, T. Wierzbicki / Engineering Fracture Mechanics 73 (2006) 1653–1678 1677

Fig. 30. Time history of the third invariant at the shear crack path at V0 = 303.5 m/s for Weldox 460 E steel.

metric and transverse plane-strain conditions. Wierzbicki and Xue [15] attributed this disparity to the effect of
the deviatoric state variable n, defined by
27 J 3
n¼ ; ð19Þ
3
2 r
where J3 = s1s2s3 is the third invariant of the stress deviators. The third invariant is lacking among all the frac-
ture models except the Wilkins fracture criterion. The research is in-progress to clarify this issue.
As the impact velocity approaches the ballistic limit of the target plate, the difference in the residual velocity
between the numerical prediction and the experimental data increases. The error reaches as large as 35% at
V0 = 200.4 m/s. Note, that the constants of the present JC fracture model are calibrated from the symmetry
axis of the round bars, where n = ±1, while the stress state at the shear crack path is much close to the plane-
strain condition (n = 0), see Fig. 30. It is not clear whether the prediction of the residual velocity can be
improved by incorporating the third invariant into the JC fracture model. This will be addressed in a separate
publication.

References

[1] Wierzbicki T, Bao Y, Lee YW, Bai Y. Calibration and evaluation of seven fracture models. Int J Mech Sci 2005;47(4–5):719–43.
[2] Johnson GR, Cook WH. Fracture characteristics of three metals subjected to various strains, strain rates, temperatures and pressures.
Engng Fract Mech 1985;21(1):31–48.
[3] Wilkins ML, Streit RD, Reaugh JE. Cumulative-strain-damage model of ductile fracture: simulation and prediction of engineering
fracture tests. Tech. rep. UCRL-53058, Lawrence Livermore Laboratory, University of California, Livermore, CA 94550, 1983.
[4] Cockcroft MG, Latham DJ. Ductility and the workability of metals. J Inst Metals 1968;96:33–9.
[5] Børvik T, Langseth M, Hopperstad OS, Malo KA. Perforation of 12 mm thick steel plates by 20 mm diameter projectiles with flat,
hemispherical and conical noses. Part I: Experimental study. Int J Impact Engng 2002;27(1):19–35.
[6] Børvik T, Hopperstad OS, Berstad T, Langseth M. Perforation of 12 mm thick steel plates by 20 mm diameter projectiles with flat,
hemispherical and conical noses. Part II: Numerical simulations. Int J Impact Engng 2002;27(1):37–64.
[7] Børvik T, Hopperstad OS, Langseth M, Malo KA. Effects of target thickness in blunt projectile penetration of Weldox 460 E steel
plates. Int J Impact Engng 2003;28(4):413–64.
[8] Børvik T, Hopperstad OS, Berstad T, Langseth ML. A computational model of viscoplasticity and ductile damage for impact and
penetration. Euro J Mech: A/Solids 2001;20:685–712.
[9] Børvik T, Hopperstada O, Deya S, Pizzinatoc E, Langsetha M, Albertinic C. Strength and ductility of Weldox 460 E steel at high
strain rates, elevated temperatures and various stress triaxialities. Engng Fract Mech 2005;72:1071–87.
[10] Bao Y, Wierzbicki T. On fracture locus in the equivalent strain and stress triaxiality space. Int J Mech Sci 2004;46(1):81–98.
[11] Teng X, Wierzbicki T. Effects of fracture criteria on high velocity perforation of thin beams. Int J Comput Meth 2004;1(1):171–200.
[12] Teng X, Wierzbicki T. Numerical study on crack propagation in high velocity perforation. Comput Struct 2005;83:989–1004.
[13] Johnson GR, Cook WH. A constitutive model and data for metals subjected to large strains, high strain rates and high temperatures.
In: Proceedings of the seventh international symposium on ballistics. Hague, Netherlands, 1983. p. 541–47.
1678 X. Teng, T. Wierzbicki / Engineering Fracture Mechanics 73 (2006) 1653–1678

[14] Bridgman PW. Studies in large plastic flow and fracture. Cambridge: Harvard University Press; 1964.
[15] Wierzbicki T, Xue L. On the effect of the third invariant of the stress deviator on ductile fracture. Tech. rep. no. 138, Impact and
Crashworthiness Lab, MIT, Cambridge, MA, 2005.
[16] Hancock JW, Mackenzie AC. On the mechanisms of ductile failure in high strength steels subjected to multi-axial stress-states.
J Mech Phys Solids 1976;24:147–69.
[17] McClintock FA. A criterion for ductile fracture by the growth of holes. J Appl Mech 1968;35:363–71.
[18] Rice JR, Tracey DM. On the ductile enlargement of voids in triaxial stress fields. J Mech Phys Solids 1969;17:201–17.
[19] Bao Y, Bai Y, Wierzbicki T. Calibration of A710 steel for fracture. Tech. rep. no. 135, Impact and Crashworthiness Lab,
Massachusetts Institute of Technology, Cambridge, MA, 2005.
[20] Wierzbicki T, Bao Y. Bridgman revisited: On the history effects on ductile fracture. J Mech Phys Solids, submitted for publication.
[21] Bao Y, Wierzbicki T. On the cut-off value of negative triaxiality for fracture. Engng Fract Mech 2005;72(7):1049–69.
[22] Bai Y, Dodd B. Adiabatic shear localization: occurrence, theories and applications. New York: Pergamon Press; 1992.
[23] Mason C, Worswick MI. Adiabatic shear in annealed and shock-hardened iron and in quenched and tempered 4340 steel. Int J Fract
2001;111:29–51.
[24] Oh S, Chen CC, Kobayashi S. Ductile failure in axisymmetric extrusion and drawing. Part 2: Workability in extrusion and drawing.
J Engng Indust 1979;101:36–44.
[25] Wierzbicki T, Werner H. Cockcroft and latham revisited. Tech. rep. 16, Impact and Crashworthiness Lab, Massachusetts Institute of
Technology, Cambridge, MA, 1998.
[26] Bao Y, Wierzbicki T. A comparative study on various ductile crack formation criteria. J Engng Mater Technol 2004;126(3):314–24.
[27] Mescall JF. Computer simulation of penetration. In: Sih GC, editor. Recent advances in engineering science. Proceedings of the 14th
annual meeting of the society of engineering science. Pennsylvania: Lehigh University Publication; 1977. p. 81–93.
[28] Johnson GR, Holmquist TJ. Test data and computational strength and fracture model constants for 23 materials subjected to large
strain, high-strain rates, and high temperatures. Tech. rep. LA-11463-MS, Los Alamos National Laboratory, 1989.
[29] Teng X, Wierzbicki T, Hiermaier S, Rohr I. Numerical prediction of the Taylor test with fracture. Int J Solids Struct 2005;42(9/10):
2929–48.
[30] McClintock FA. Plasticity aspects of fracture. In: Liebowitz H, editor. Fracture: an advanced treatise, vol. III. New York: Academic
Press; 1971.
[31] Teng X. High velocity impact fracture. PhD thesis, Massachusetts Institute of Technology, Cambridge, MA, December 2004.

You might also like