Survey of Four Damage Models For Concrete

SANDIA REPORT
SAND2009-5544
Unlimited Release
Printed August 2009
Survey of Four Damage Models for
Concrete
Rebecca M. Brannon and Seubpong Leelavanichkul
Prepared by
Sandia National Laboratories
Albuquerque, New Mexico 87185 and Livermore, California 94550
Sandia is a multiprogram laboratory operated by Sandia Corporation,
a Lockheed Martin Company, for the United States Department of Energys
National Nuclear Security Administration under Contract DE-AC04-94-AL85000.
Approved for public release; further dissemination unlimited.
Issued by Sandia National Laboratories, operated for the United States Department of Energy
by Sandia Corporation.
NOTICE: This report was prepared as an account of work sponsored by an agency of the United
States Government. Neither the United States Government, nor any agency thereof, nor any
of their employees, nor any of their contractors, subcontractors, or their employees, make any
warranty, express or implied, or assume any legal liability or responsibility for the accuracy,
completeness, or usefulness of any information, apparatus, product, or process disclosed, or rep-
resent that its use would not infringe privately owned rights. Reference herein to any specic
commercial product, process, or service by trade name, trademark, manufacturer, or otherwise,
does not necessarily constitute or imply its endorsement, recommendation, or favoring by the
United States Government, any agency thereof, or any of their contractors or subcontractors.
The views and opinions expressed herein do not necessarily state or reect those of the United
States Government, any agency thereof, or any of their contractors.
Printed in the United States of America. This report has been reproduced directly from the best
available copy.
Available to DOE and DOE contractors from
U.S. Department of Energy
Ofce of Scientic and Technical Information
P.O. Box 62
Oak Ridge, TN 37831
Telephone: (865) 576-8401
Facsimile: (865) 576-5728
E-Mail: [email protected]
Online ordering: http://www.osti.gov/bridge
Available to the public from
U.S. Department of Commerce
National Technical Information Service
5285 Port Royal Rd
Springeld, VA 22161
Telephone: (800) 553-6847
Facsimile: (703) 605-6900
E-Mail: [email protected]
Online ordering: http://www.ntis.gov/help/ordermethods.asp?loc=7-4-0#online
D
E
P
A
R
T
M
ENTOF
E
N
E
R
G
Y
U
N
I
T
E
D
S
T
A
TES OF
A
M
E
R
I
C
A
2
SAND2009-5544
Unlimited Release
Printed August 2009
Survey of Four Damage Models for Concrete
Rebecca M. Brannon and Seubpong Leelavanichkul
Department of Mechanical Engineering
University of Utah
50 S. Campus Dr.
Salt Lake City, UT 84109
[email protected]
[email protected]
Sandia Contract No. 903761
Abstract
Four conventional damage plasticity models for concrete, the Karagozian and Case model (K&C),
the Riedel-Hiermaier-Thoma model (RHT), the Brannon-Fossum model (BF1), and the Contin-
uous Surface Cap Model (CSCM) are compared. The K&C and RHT models have been used in
commercial nite element programs many years, whereas the BF1 and CSCM models are relatively
new. All four models are essentially isotropic plasticity models for which plasticity is regarded
as any form of inelasticity. All of the models support nonlinear elasticity, but with different for-
mulations. All four models employ three shear strength surfaces. The yield surface bounds an
evolving set of elastically obtainable stress states. The limit surface bounds stress states that
can be reached by any means (elastic or plastic). To model softening, it is recognized that some
stress states might be reached once, but, because of irreversible damage, might not be achievable
again. In other words, softening is the process of collapse of the limit surface, ultimately down
to a nal residual surface for fully failed material. The four models being compared differ in
their softening evolution equations, as well as in their equations used to degrade the elastic stiff-
ness. For all four models, the strength surfaces are cast in stress space. For all four models, it is
recognized that scale effects are important for softening, but the models differ signicantly in their
approaches. The K&C documentation, for example, mentions that a particular material parameter
3
affecting the damage evolution rate must be set by the user according to the mesh size to preserve
energy to failure. Similarly, the BF1 model presumes that all material parameters are set to values
appropriate to the scale of the element, and automated assignment of scale-appropriate values is
available only through an enhanced implementation of BF1 (called BFS) that regards scale effects
to be coupled to statistical variability of material properties. The RHT model appears to similarly
support optional uncertainty and automated settings for scale-dependent material parameters. The
K&C, RHT, and CSCM models support rate dependence by allowing the strength to be a function
of strain rate, whereas the BF1 model uses Duvaut-Lion viscoplasticity theory to give a smoother
prediction of transient effects. During softening, all four models require a certain amount of strain
to develop before allowing signicant damage accumulation. For the K&C, RHT, and CSCM
models, the strain-to-failure is tied to fracture energy release, whereas a similar effect is achieved
indirectly in the BF1 model by a time-based criterion that is tied to crack propagation speed.
4
Contents
Nomenclature 11
1 Introduction 13
General formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Shear strength surface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Octahedral prole . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2 The K&C Concrete Model 19
Strength surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Rate and scale dependence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
Damage accumulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Plastic update . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
Shear and bulk moduli . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3 The RHT Concrete Model 27
Plastic update . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4 The BF1 GeoMaterial Model 35
Rate dependence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
Plastic update . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
Evolution equation for pore collapse . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
Evolution equation for backstress . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
5 LS-DYNA Concrete Model 159 (CSCM) 47
Plastic update . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
6 Model comparisons 53
Comparison of theory and implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
5
Numerical comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
Meridional prole . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
Single element test: isotropic compression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
Single element test: uniaxial strain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
Projectile penetration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
Verication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
K&C: drop weight tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
K&C: simulations of penetration and perforation of high performance concrete . . . 69
K&C: explosive wall breaching . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
K&C: vehicle-barrier crash test simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
RHT: simulation of penetration of high performance concrete . . . . . . . . . . . . . . . . . 70
RHT: concrete subjected to projectile and fragment impacts . . . . . . . . . . . . . . . . . . 70
RHT: jumbo jet impacting on thick concrete walls . . . . . . . . . . . . . . . . . . . . . . . . . . 70
7 Conclusions 73
References 74
6
List of Figures
1.1 Left: yield surface without cap. Right: yield surface with cap. . . . . . . . . . . . . . . . . . 16
1.2 Deviatoric section: (left) Willam-Warnke, (center) Mohr-Coulomb, and (right)
Gudehus. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.1 K&C meridian proles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.2 Failure surfaces in 3D stress space. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.3 Experimental data for DIF according to CEB-FIB design model code [26]. . . . . . . 22
2.4 Example of damage function (). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.1 Failure surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.2 An elliptical cap function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.3 RHT octahedral prole and surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.4 Left: Experimental data for DIF according to CEB-FIB design model code [26].
Right: DIF used in AUTODYN: RHT concrete model. . . . . . . . . . . . . . . . . . . . . . . 31
3.5 Bi-linear uniaxial stress-crack opening relationship . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.1 Parameters for residual surface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.2 Illustrations of the generalized Duvaut-Lions rate sensitivity and the scale factor
employed in the BF1 model [14] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
4.3 Standard Weibull distribution plots showing increased Brazilian strength T
br
with
decreased sample size. Here, P
s
is the complementary cumulative probability,
which may be interpreted as the probability that the sample is safe from failure.
The slope of the tted line is the Weibull modulus, which quanties variability in
strength. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.4 BF1 damage function for FSPEED values in the range from 5 to 30. The higher
FSPEED values correspond to the steeper slope. . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.5 Hydrostatic pressure vs. volumetric strain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4.6 Qualitative sketch of shear stress vs. shear strain . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
6.1 Comparison of meridian proles to the K&Cs prole. (The CSCM and BF1 curves
coincide.) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
6.2 p equation of state employed by the RHT model. . . . . . . . . . . . . . . . . . . . . . . . 57
6.3 Equation of state: BF1 and CSCM crush curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
6.4 Comparison of hydrostatic pressure versus volumetric strain . . . . . . . . . . . . . . . . . . 59
6.5 Comparison of stress difference versus pressure under uniaxial strain. (The BF1
and CSCM models results are different not because of model differences but be-
cause the BF1 model was driven with logarithmic strain, whereas the CSCM model
was run using engineering strain.) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
6.6 RHT Model: Contour plots of damage: side, front, and back view of the target (top
to bottom). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
7
6.7 CSCM Model: Contour plots of damage: side, front, and back view of the target
(top to bottom). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
6.8 FEA vs. analytical/numerical elastic wave velocity example . . . . . . . . . . . . . . . . . . 67
8
List of Tables
6.1 Parameters for meridional curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
6.2 P EOS input parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
6.3 Uniaxial strain loading for single element tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
6.4 Material properties for concrete . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
6.5 Input parameters for the steel projectile . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
6.6 Input parameters for the RHT concrete target . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
6.7 Input parameters for the CSCM concrete target . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
6.8 Residual velocity comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
9
10
Nomenclature
f
t
tensile strength
f
c
compressive strength
T
trial elastic stress
S stress deviator
1
,
2
,
3
principal stresses
I
1
, J
2
, J
3
invariants
A
axial stress
L
lateral stress
f yield function
F
f
strength surface
(, I
1
) lode angle function
ratio of the tensile to compressive meridian radii at a given pressure
Y
m
,Y
r
,Y
y
limit, residual, and yield surface
a
k
parameters dening strength surfaces
Effective plastic strain parameter
i j
,
i j
strain tensor and strain increments
, effective strain rate and effective strain rate increments
v
,
p
v
volumetric strain and plastic volumetric strain
,
vp
,
d
stress tensor (general, viscoplastic, damage)
r
f
rate enhancement factor
DIF dynamic increasing factor
(K&C) damage
potential function
scalar factor governing the magnitude of the plastic ow
C stiffness tensor
11
G shear modulus
K, K
U
, K
L
bulk modulus, unloading bulk modulus, loading bulk modulus
scale factor for bulk modulus used in K&C model
Poissons ratio
, rate effects exponents used in RHT model
p
spall
concrete spall strength used in RHT model surface
F
c
elliptical cap function
X cap location
cap initiation
BQ brittle to ductile transition factor
D
1
, D
2
(RHT) damage parameters
w
u
element characteristic length (crack width)
l characteristic crack length
G
F
Fracture energy
b
k
, g
k
(BF1) nonlinear bulk and shear moduli parameters
, shifted stress tensor and backstress
h
k
,C
H
hardening modulus
H
k
hardening tensor
N (BF1) isotropic hardening shift factor
G() decay function
characteristic time
L
,
H
Low and high rate stresses
viscoplastic interpolation
N
H
(CSCM) hardening initiation
b
brittle damage threshold
s
ductile damage threshold
r
ob
, r
od
, r
o
initial damage threshold (brittle, ductile, general)
(CSCM) rate effect uidity parameter
s
,
s
.
s
,
s
DIF parameters
12
Chapter 1
Introduction
A persistent challenge in simulating damage of concrete structures is the development of efcient
and accurate constitutive models. The desired models need to produce a smooth transition from a
linear or nonlinear elastic range to a non-linear hardening regime and ultimately a weak post-peak
state. This report compares four concrete material models: (1) the Karagozian and Case model
(K&C), (2) the Riedel-Hiermaier-Thoma model (RHT), (3) the Brannon-Fossum model (BF1),
and (4) the Continuous Surface Cap Model (CSCM). To facilitate discussion of these models, a
common terminology will be adopted for concepts common to all four models. The performance
of each model is assessed by making comparison between some simulations.
Unconned concrete tensile strength ( f
t
) can be as much as 92% lower than the compressive
strength ( f
c
) [11]. The ultimate strength of concrete depends on the pressure and shear stresses. At
low pressure, the inelastic behavior of concrete material is not related to the motion of dislocations
as for metallic materials. In uniaxial loading, deformation is approximately linear in the elastic
regime. As the deformation increases, the cracks increase in size and number, and then eventually
propagate through the material to culminate in ultimate failure.
In extension, the active crack planes are orthogonal to the load direction. In compression, they
are parallel to the load direction (i.e., misaligned cracks will kink in this direction). In either case,
crack planes tend to form orthogonal to the direction of the least compressive (or most tensile)
principal stress. A peak stress is reached at a point where microcracking has caused sufcient
degradation of stiffness such that the material would become unstable if loaded in stress control. If
hydrostatic pressure is present, a fully damaged material in compression retains a residual strength
such as that of a granular medium.
A distinctive behavior of concrete and other quasibrittle materials is the phenomenon of dilata-
tion (i.e., volume increase) under inelastic compressive loading. Although compressive stresses
initially induce a volume reduction, continued compression induces material damage in the form
of shear cracking. Subsequent dilatation is typically attributed to geometrically necessary intro-
duction of void space associated with crack kinking. Standard concrete exhibits volume expansion
under compressive loading at low conning pressure, but does not dilatate at high conning pres-
sure greater than 100MPa [11].
For triaxial tests conducted under sufciently high conning pressure, crack growth tends to
be negligible in comparison to porosity changes. For purely hydrostatic loading, a porous equation
of state is usually employed to model three different phases: elastic deformation, compaction, and
13
solidication. During the compaction phase, pores in the material collapse. In the nal solidica-
tion phase, the material is approximately homogeneous (because pore space has been fully crushed
out), and the volumetric response is once again elastic.
All four models in this report fall loosely in the category of generalized isotropic plasticity the-
ory. Specically, they all presume existence of an elastic domain. The boundary of this domain in
stress space is called the yield surface even though mechanisms of inelasticity are not necessar-
ily associated with dislocations. Because all four models are isotropic, the yield surface in stress
space has a certain degree of symmetry about the hydrostat. The radial distance from the hydrostat
is a measure of equivalent shear stress. The detail of the surfaces used in these models is discussed
in upcoming chapters.
General formulation
The implementation of the concrete models under investigation can be broken into (1) elastic and
plastic updates, (2) strength surface formulations, (3) rate and scale effects, and (4) damage ac-
cumulation. The models differ in their approaches to these areas. This report does not cover the
details of elastic and plastic updates used in each model because all of them follow the typical
elasticity and plasticity theories. All of the models under investigation currently presume that the
concrete is initially isotropic. The BF1 and CSCM models support developed anisotropy in the
rudimentary form of kinematic hardening. Neither the K&C nor RHT models support intrinsic
(i.e. pre-existing) elastic anisotropy, but the BF1 model includes support for joints in the concrete
and the CSCM model includes support for rebar. All models support nonlinear elasticity, imple-
mented in incremental form such that stress increments are linear and isotropic in strain increments
with the tangent bulk and shear moduli varying with deformation or stress.
Depending on the type of loads, the concrete will eventually yield or fail. The yield threshold
is dened by the yield surface that is described in the following chapter. During compaction, the
material is tentatively presumed to be elastic thus giving a trial elastic stress
T
. If
T
is found
to lie outside the yield surface, the tentative assumption is rejected, and the loading increment is
re-evaluated using plastic update equations. When damage occurs and begins to accumulate, the
strength of the concrete is reduced by appropriately collapsing the strength surface in stress space.
High loading rates are well known to lead to an apparent increase in strength. In the K&C
and RHT models, this behavior is accommodated by expanding the yield surface so that higher
stress levels are required to reach it. The BF1 and CSCM models account for rate dependence
through a viscoplastic approach that better matches stress transients prior to reaching the steady
state strength.
14
Shear strength surface
To include the effects of material strength and resistance to shear distortion, one can work with
the stress deviator S, which is dened as the difference between the total stress, , and a uniform
hydrostatic pressure, p,
S = pI or S =
1
3
tr()I. (1.1)
where the hydrostatic pressure (or mean stress) can be represented by one third of the rst invariant,
I
1
, of the total stress.
p =
1
3
(
1
+
2
+
3
) =
1
3
tr =
I
1
3
.
The second and the third invariants are given by
J
2
=
1
2
tr(S
2
), and J
3
=
1
3
tr(S
3
). (1.2)
All four concrete models investigated in this report rely heavily on axisymmetric compressive
stress data. The mechanics invariants for axisymmetric loading having an axial stress
A
and two
equal lateral stresses
L
are
I
1
=
A
+2
L
, J
2
=
1
3
(
A
L
)
2
, J
3
=
2
27
(
A
L
)
3
. (1.3)
For elastic distortion, after loading and unloading, all the distortion energy is recovered and the
material returns to its initial conguration. However, when the distortion is large enough that the
material reaches its elastic limit, only elastic distortion energy is recovered. The material suffers
permanent plastic strain and can therefore no longer return to its initial conguration. Hence, a
yield function is used to describe the material elastic limit and the subsequent transition to plastic
ow.
The yield criterion for all four concrete models in this report can be written in the form
J
2
= F
f
(I
1
, , ), (1.4)
where

J
2
, I
1
, and are stress invariants, and stands for one or more internal variables. The
stress invariant I
1
is proportional to pressure (specically, I
1
= 3p) and also proportional to the
axial coordinate of the stress state along the hydrostat in principal stress space. The stress invariant
15
Figure 1.1. Left: yield surface without cap. Right: yield surface
with cap.
J
2
is proportional to equivalent shear stress and also proportional to the radial distance of the
stress state from the hydrostat in stress space. The Lode angle stress invariant serves as an
alternative to the third invariant, J
3
, and it quanties the angular coordinate of the stress state in
principal stress space.
The above yield criterion corresponds to the following yield function
f = J
2
[F
f
(I
1
, , )]
2
= 0, (1.5)
Equation (1.4) and (1.5) are the basic forms employed by the K&C and RHT models. The
BF1 and CSCM models include extra terms that account for backstress, but can also be reduced
to the same expression as the other two models. The full expressions of the yield function used
in the BF1 and CSCM models are given later in Chapters 4 and 5, respectively. In addition to
the yield surface, all four concrete models use two additional surfaces to describe the peak stress
limit of the material. The limit surface bounds the set of stress states that are achievable at
least once. After a stress state at the limit surface has been reached, irreversible damage occurs
in the material causing the boundary of achievable stress states to shrink until ultimately reaching
a residual surface corresponding to a fully damaged state. These three strength surfaces are
sketched in Fig. 1.1. The limit and residual surfaces are stationary, while the current yield surface
evolves in response to evolution of internal variables that directly or indirectly account for porosity
and microcracks. The equations representing these surfaces are given in more detail in Chapters 2,
3, and 4.
Figure 1.1(right) shows the capped yield surface shape that is typical of models that account
for porosity. Not only does the cap introduce an elastic limit in pure hydrostatic compression, it
also allows porosity to affect the shear strength. This approach is adopted in the RHT, BF1, and
16
CSCM models. However, the K&C model allows for a hydrostatic elastic limit only through an
equation of state, which does not include the effect of porosity on shear strength.
Octahedral prole
Cylindrical Lode coordinates (r, , z) represent an alternative invariant triplet that can be obtained
from the conventional invariant triplet (I
1
, J
2
, and J
3
) as follows [12]:
r =
_
2J
2
, sin3 =
J
3
2
_
3
J
2
_
3/2
, z =
I
1
3
. (1.6)
With this denition of the Lode angle, triaxial compression corresponds to a Lode angle of 30
.
The Lode angle dependence in Eqs. (1.4) and (1.5) is accomplished by multiplying the compressive
meridian, F
f
(I
1
, 30
, ) in Fig. 1.1, by a scale function (, I

1
). This Lode angle scale function is
a ratio of the radius of the octahedral cross section at an angle to the radius of the compressive
meridional prole at angle 30
. Various formulations for the scale function have been used in the
literature. Of these, the following are common choices:
1. Willam-Warnke function [38]:
() =
4
_
1
2
_
cos
2
+(12)
2
2(1
2
)cos
+(2 1)
_
4(1
2
)cos
2
+5
2
4
, (1.7)
where
=/6+, (0.5 2),

2. Mohr-Coulomb [33]:
() =
2
3
33
_
1
1+
_
_
_
cos
3
_
1
1+
_
sin
3
_
_
(0.5 2), (1.8)
3. Gudehus [15]:
() =
1
2
_
1+sin3 +
1
(1sin3)
_
(7/9 9/7). (1.9)
Here, is the ratio of the radius, r
t
, at the tensile meridian (where =30
) to the radius, r
c
, of
the compressive meridian. The octahedral cross-sections corresponding to these Lode angle scale
functions are illustrated in Fig. 1.2 for various values of the strength ratio parameter . Note that
the functions are normalized to coincide at the compressive meridian. All four models investigated
in this report support Willam-Warnke Lode angle function. The BF1 model also allows the options
of Mohr-Coulomb and Gudehus.
17
Figure 1.2. Deviatoric section: (left) Willam-Warnke, (center)
Mohr-Coulomb, and (right) Gudehus.
18
Chapter 2
The K&C Concrete Model
Strength surfaces
The K&Cmodel uses stress differences to describe the yield surface (
y
), the limit surface (
m
),
and the residual surface (
r
). In view of Eq.(1.3), where the stress difference is (
A
L
), the
stress difference can be written as
3J
2
, which allows generalization of the theory to general stress
states. During the initial loading or reloading, the stresses are elastic until an initial yield surface
is reached. The initial yield surface hardens to the limit surface or softens to the residual surface,
depending on the nature of loading or on the material state. Three xed surfaces are used in the
K&C model and are dened as
Y
y
=
y
= a
0y
+
p
a
1y
+a
2y
p
(yield surface), (2.1)
Y
m
=
m
= a
0m
+
p
a
1m
+a
2m
p
(limit surface), (2.2)
Y
r
=
r
= a
0f
+
p
a
1f
+a
2f
p
(residual surface), (2.3)
where the a-parameters are user inputs, and a
0f
= 0 for concrete. The three surfaces are dened
by different values of the a-parameters. To facilitate comparing the K&C model to other models,
Eqs. (2.1) to (2.2) can be recast in terms of standard invariants as follows:
J
2
= F(I
1
), (2.4)
where
F(I
1
) =
1
3
_
a
0m
+
I
1
3a
1m
+a
2m
I
1
_
The K&C concrete model uses Willam-Warnkes Lode-angle function () shown in Eq. (1.7)
to describe the octahedral cross section of the surfaces. If tensile data are available instead of
19
Figure 2.1. K&C meridian proles
compressive data, the compressive meridian can be approximated by dividing the tensile meridian
by the relative distance between the compression and tension meridian (p) at each pressure
p. The above equations apply only for compressive pressures. For tensile pressure, these equations
are replaced by
=
3
2
(p+ f
t
) or F(I
1
) =
3
2
_
I
1
3
+ f
t
_
. (2.5)
Equation (2.5) ensures conditions given in [8, 25] are met (i.e., passes through (p, ) =
(f
t
, 0) under triaxial test and (p, ) = (f
t
/3, f
t
) for uniaxial test). Illustrated below is a com-
plete linear piecewise denition of (p) as given in [28] for this model:
(p) =
_
_
1
2
: p 0,
1
2
+
3
2
_
f
t
f
c
_
: p = f
c
/3,
f
c
a
0
+
2 f
c
/3
a
1
+2a
a
f
c
/3
: p = 2 f
c
/3,
0.753 : p = 3 f
c
,
1 : p 8.45 f
c
,
(2.6)
where is a scalar factor multiplying f
c
to denote the location where the failure occurs. The
function given in Eq. (2.6) is linear between the specied points. For example, Kupfer et al.[25]
showed in biaxial compression tests that the failure occurred at (
1
,
2
,
3
) = (0, f
c
, f
c
), with
1.15. Even though the K&C model allows to be pressure-dependent, a slope discontinuity
is present due to the piecewise representation of .
20
Figure 2.2. Failure surfaces in 3D stress space
Rate and scale dependence
The K&C model uses rate effects to handle shear damage accumulation. A strain rate enhance-
ment factor r
f
is used to scale the strength surface when the material is subjected to a high load-
ing rate. This strength enhancement factor is called the dynamic increasing factor (DIF) in the
CEB-FIP model code 90 (Comit e Euro-International du B eton and F ed eration International de la
Pr econtrainte) [35], see Fig. 2.3
When pressure p is returned from the equation of state, an unenhanced pressure p/r
f
and
unenhanced (i.e., quasistatic) shear strength F(I
1
/r
f
) are computed. Multiplying the strain rate
enhancement factor r
f
(or DIF) to F(I
1
/r
f
), a new enhanced limit surface at the current pressure
p is obtained:
e
= r
f
_
p
r
f
_
or F
e
(I
1
) = r
f
F
_
I
1
r
f
_
. (2.7)
To include the strain rate enhancement factor r
f
(or DIF), a modied effective plastic strain is
dened as
= h
_
2
3
p
i j
p
i j
, (2.8)
21
Figure 2.3. Experimental data for DIF according to CEB-FIB
design model code [26].
where
h =
_
_
1
r
f
1+
p
r
f
f
t
b
1
for p 0 (compression),
1
r
f
1+
p
r
f
f
t
b
2
for p < 0 (tension).
(2.9)
Equation (2.9) allows the damage accumulation to be different in tension and compression. The
b
1
and b
2
parameters are tted to experimental data. The input scalar parameter b
2
governs the
softening part of the unconned uniaxial tension stress-strain curve as the stress point moves from
the limit to the residual surface, while b
1
governs the softening for compression.
Damage due to isotropic tensile stressing is handled by adding a volumetric damage accumu-
lation that is computed by incrementing the effective plastic strain parameter by according
to
= b
3
f
d
k
d
(
v
v,yield
), (2.10)
where b
3
is an input parameter,
v
is volumetric strain,
v,yield
is the volumetric strain at yield. The
factor f
d
limits the effect of this change according to proximity of the stress state to the hydrostat.
Specically,
f
d
=
_
1
|
3J
2
/p|
0.1
: 0 |
3J
2
/p| < 0.1,
0 : |
3J
2
/p| 0.1.
(2.11)
Determination of the input parameters b
1
, b
2
, and b
3
is described in [28]. The parameter b
2
is
computed iteratively using the data from the unconned uniaxial tensile test until the area under
22
the stress-strain curve coincides with G
F
/w
u
where G
F
is the fracture energy, and w
u
is crack
front width (which equals the element size). Therefore, different values of b
2
must be used for
different element sizes; otherwise the computed energy release will be incorrect. Whether or not
this adjustment of b
2
is automated is unclear.
Damage accumulation
Once the initial yield surface is reached, the stress state is evolved by interpolating between
Eq. (2.1) and Eq. (2.2) according to
= (
m
y
) +
y
or F(I
1
) = [F
m
(I
1
) F
y
(I
1
)] +F
y
(I
1
), (2.12)
where a user dened damage function indicates the location of the current yield surface relative
to limit surface and is a function of an effective plastic strain parameter,
=
t
0
_
2
3

p
:
p
dt (2.13)
The damage is initially zero at = 0 and increases to unity at a user-specied value
m
marking
the onset of softening. During softening, , which now decreases as increases, is used to inter-
polate the current surface between limit and residual surfaces, Eq. (2.2) and Eq. (2.3), respectively
according to
= (
m
r
) +
r
or F(I
1
) = [F
m
(I
1
) F
r
(I
1
)] +F
r
(I
1
), (2.14)
Typical () used in the K&C model has behavior as illustrated in Fig. 2.4.
The experimental data presented in [8, 25] showed that the principal stress difference should be
approximately f
t
for biaxial compression and triaxial tension tests. To be able to reach this point,
the K&C model initially sets a value of pressure cutoff p
c
to f
t
. This choice is consistent with
a maximum principal stress criterion at tensile pressures. If stresses reach the failure threshold in
the negative pressure range, the parameter is then used to move the pressure cutoff from f
t
to
zero in a smooth fashion. This is done by checking the pressure returned by the equation of state
(EOS), and resetting it to p
c
using the following conditions
p
c
=
_
f
t
if the limit surface has not been reached (hardening),
f
t
if the limit surface has been reached (softening).
23
Figure 2.4. Example of damage function ().
Plastic update
The cut-off pressure is reduced during the process of softening, and can cause a segment of the
meridian in the negative pressure portion to become very steep. To avoid a steep slope in this
region, the limit surface Y(p, ) in Eq. (2.2) is modied according to
Y
1
(p, ) =
_
m
(p)
p
f
p
p
f
p
c
()
m
(p
c
)
_
= 3(p+ f
t
), (2.15)
where
m
= 3(p+ f
t
) is the nominal limit surface in compression,
p
f
= 0 is the intersection of the residual surface with the pressure axis, and
p
c
= f
t
is the intersection of the limit surface with the pressure axis.
Hence, the current modied limit surface during the softening can be written as
Y(p, ) =
_

m
(p) +(1)
f
(p) for p > 0,
3(p+ f
t
) for p 0,
(2.16)
where
f
is the current unmodied failure surface. When the radial rate enhancement is used,
the surface is computed as a function of p/r
f
and then multiplied by r
f
as shown in Eq. (2.7).
At any time step, the shear strength changes with both pressure p and damage . The current
strength Y is initially updated only according to the current pressure. The fully updated surface
is then determined iteratively accounting for the updated damage . Let Y
be the strength corre-

sponding to the updated pressure but before the value of is determined. Then the fully updated
strength Y
n+1
is determined as a result of the change in according to
Y
n+1
Y
=
Y
d =
Y
d
d
d (2.17)
24
Using Eq. (2.8),
Y
n+1
Y
=
Y
d
d
h()
_
2
3
d
p
i j
: d
p
i j
. (2.18)
As is typical in plasticity models, the strain increment is decomposed into elastic and plastic parts
(d = d
e
+d
p
). A conventional regular associated ow rule is adopted, and the stress state is
updated in a conventional manner (see, e.g., [1]).
Shear and bulk moduli
Prior to yielding, Hookes law is used for the elastic stress-strain relationship. The K&C model
supports nonlinear elasticity by permitting the moduli to vary with pressure. The shear modulus is
computed from a user specied Poissons ratio and the bulk modulus. It was commented in [28]
that when the difference between the loading and unload/reload bulk moduli is large, a negative
effective Poissons ratio may occur. Therefore, the bulk modulus is entered as part of the EOS input
set and is scaled within the the K&C model using a factor depending on how far the pressure
is below the virgin curve [28] (loading portion of the users specied pressure vs. volumetric
curve).
=

+(pp
f
)/K
U
, (2.19)
where =
v,min
v
,
v
is volumetric strain, and
K
U
is the unload/reload bulk modulus from the EOS.
The shear modulus is then calculated from the scaled bulk modulus K
as
G =
(1.53)K
1+
, (2.20)
where K
= (K
L
K
U
)e
5.55
+K
U
, and K
L
is the loading modulus.
The constant 5.55 is chosen such that the K
increases half way to unload/reload value when

p dropped 1/8 of the way from the virgin curve to p = p
f
(p
f
= 0 for concrete). According to
Eq. (2.20), a user is required to input only a Poissons ratio for the K&C model to compute the
shear modulus.
25
26
Chapter 3
The RHT Concrete Model
This chapter discusses a material model for concrete using the RHT concrete model that has the
following capability associated with brittle material
Pressure hardening,
Strain hardening,
Strain rate hardening,
Third invariant dependence for compressive and tensile meridians,
Damage effects (strain softening), and
Crack-Softening.
The above terminology was taken directly from ANSYS AUTODYN product features [6]. It is
unclear what is meant by pressure hardening, but we suspect that it merely refers to increased
shear strength with pressure. If so, the term hardening is used here in a nonstandard way.
Strength surfaces
Similar to the K&C model, three strength surfaces are used in the RHT material model (see
Fig. 3.1). The RHT concrete model expresses these surfaces in terms of the compressive meridian
Y
TXC
(p), a rate factor r
f
= r
f
( ) (denoted as F
rate
( ) in [18]), and the ratio of compressive and
tensile radii (). Willam-Warnkes Lode-angle function () is used in this model.
The strength along the compressive meridian is expressed as a triaxial compression normalized
to the unconned compression strength f
c
Y
TXC
=
Y
TXC
f
c
= a
1
_
p
f
p
spall
f
c
r
f
_
a
2
or F(I
1
) =
a
1
3
_
I
1
3
r
f
p
spall
_
a
2
, (3.1)
27
Figure 3.1. Failure surfaces
where
r
f
=
_
_
_
_

0
_
: p > f
c
, with
0
= 3010
6
s
1
,
_

0
_
: p < f
c
, with
0
= 310
6
s
1
,
a
1
= Initial slope of failure surface,
a
2
= Pressure dependence of failure surface,
P
spall
= Spall strength,
p = Pressure,
= Material constant,
= Material constant.
Unlike the K&C model, which apparently does not model porosity effects on strength, the RHT
model provides an option of an elliptical cap function F
c
(p) that closes the yield surface at high
pressure, see Fig. 3.2.
F
c
(p) =
_
_
1 : p ,
_
1
_
p
X
_
: < p < X,
0 : p X.
(3.2)
where is a pressure at which the uniaxial compression path intercepts with the elastic surface, and
X is the pressure where the yield surface intersects with the hydrostat axis. In the RHT model, X =
f
c
/3, which is close to the pore crush pressure. This feature is not available in the K&C material
model. The yield surface in the RHT model is determined through three parameters: the ratio of
28
Figure 3.2. An elliptical cap function
initial shear modulus to the modulus after the elastic limit has been passed, the ratio between the
compressive yield strength and the compressive ultimate strength, and the ratio between the tensile
yield strength and the ultimate tensile strength.
Similar to the K&C model, third invariant dependence corresponding to a noncircular octa-
hedral prole is obtained by using the Willam-Warnke function Eq. (1.7) as a scaling factor; see
Fig.3.3. Unlike the K&C concrete model, where the ratio of a material tensile strength to compres-
sive strength (p) is represented by a piecewise linear function, the RHT concrete model denes
(p) as
(p) =
0
+BQ
p
f
c
. (3.3)
where
0
is the tensile to compression meridian ratio, and BQ is a brittle to ductile transition factor.
By default, the model assigns a value of 0.6805 to
0
and 0.0105 to BQ.
The RHT model implements a strain rate law that uses a dynamic increase factor (DIF) for tension
at varying strain rates. The DIF is represented by a ratio of dynamic and static tensile strength,
29
Figure 3.3. RHT octahedral prole and surfaces
and can be expressed as [35]
DIF =
f
ct
f
cts
=
_
_
_

s
_
1.016
s
for 30 s
1
,
s
_

s
_
1/3
for > 30 s
1
,
(3.4)
s
=
1
10+6 f
c
/ f
co
, with f
co
= 10 MPa,
s
= 10
7.112
s
2.33
,

s
= 310
6
s
1
, (3.5)
where f
ct
is dynamic tensile strength at , and f
cst
is the static tensile strength at a reference rate,

s
. The strain rate can be any value between 10
6
to 160 s
1
. The parameter
s
is adjusted such
that Eq. (3.4) approximates the DIF curve that complies with experimental data given in CEB-FIB
Model Code [35], see Fig. 3.4.
For projectile and fragment impacts, cracking, spalling and scabbing are mainly inuenced by
the tensile strength, fracture energy, and strain rate in tension. Penetration, on the other hand,
is inuenced by the pressure and the strain rate in compression. When the sudden increase in
strength occurs at lower strain rates, Unosson [37] pointed out that a scabbing in the simulation
can be reduced by a using DIF value in tension. Hence, to predict the correct behavior of the
penetration, spalling, and scabbing, DIF data for tension and compression are required.
The RHT model handles the scale effect similar to the K&C model, namely scaling of the
fracture energy. A linear [19] or bilinear [16, 20] softening law, which is based on the crack
opening, can be included in the RHT model post-failure response under tension [26] when the
stress reduces to zero and the real crack is formed. The fracture energy G
F
and tensile strength f
t
are used to compute the crack width, w
u
, as shown in Fig. 3.5. In the AUTODYN implementation
30
Figure 3.4. Left: Experimental data for DIF according to CEB-
FIB design model code [26]. Right: DIF used in AUTODYN:
RHT concrete model.
Figure 3.5. Bi-linear uniaxial stress-crack opening relationship
31
of the RHT model, the maximum cracking strain is related to the maximum crack opening using a
smeared crack approach as
u
=
w
u
l
=
4G
F
f
t
l
, (3.6)
where l is a characteristic length typically set equal to the cube root of element volume. The slopes
in Fig. 3.5 are dened as
k
1
=
f
2
t
G
F
for
1
6
u
, (3.7)
k
2
=
f
2
t
10G
F
for >
1
6
u
, (3.8)
where is the cracking strain and
u
is the ultimate cracking strain. This approach is used when the
erosion option is selected in AUTODYN. The implementation of the bilinear softening law to the
RHT model is presented in [26]. In the current commercial release of AUTODYN, however, only
a linear softening is available for the RHT concrete model. Its linear softening slope is dened as
k =
f
2
t
2G
F
. (3.9)
Damage accumulation
Once material begins to harden or soften, the damage factor D is used to determine the value of
the current strength surface. The damage factor is dened using
D =

f
, (3.10)
where
p
is the accumulated plastic strain, and
f
is the failure strain given by
f
= D
1
_
p
f
p
spall
f
c
_
D
2
, (3.11)
and parameters D
1
and D
2
are user input material constants. Damage causes a reduction in strength,
hence, the strength surface is modied by shifting the surface from an initial surface to a current
damage one. Similar to the K&C model, the current damaged surface during softening is interpo-
lated between the limit and residual surfaces as
Y
= (1D)Y
m
+DY
r
or F(I
1
) = (1D)F(I
1
)
m
+DF
r
(I
1
), (3.12)
32
and the residual surface is dened as
Y
r
= a
1f
_
p
f
c
_
a
2 f
or F
r
(I
1
) =
a
1f
3
_
I
1
3
_
a
2 f
, (3.13)
(3.14)
where
a
1f
= Initial slope of residual surface,
a
2f
= Residual strength exponent, pressure dependence for residual surface.
Equation (3.12) represents the interpolation between the undamaged material (D=0) and dam-
aged material (D = 1) at the limit surface.
Plastic update
Similar to the K&C concrete models, a conventional regular associated ow rule is adopted by the
RHT model. Therefore, the details on plastic update of this model is not covered in this report.
The numerical schemes provided by the RHT models developers can be found in [31].
Similar to the K&C model, the shear and bulk moduli are used and specied through the EOS
provided by the host code(ANSYS AUTODYN). Several options are provided by ANSYS AUTO-
DYN; for example, linear, polynomial, and p EOS. The bulk and shear moduli are controlled
via the EOS similar to the K&C model. However, details on any form of modications through
scaling factors are not provided in the RHT or the ANSYS AUTODYN documentations.
33
34
Chapter 4
The BF1 GeoMaterial Model
The BF1 model is a version of the Sandia GeoModel [12] that has been enhanced to support soft-
ening. The BF1 softening model was originally designed to emulate and, where possible, enhance
the softening approaches used in the Johnson-Holmquist ceramic models, JH1 and JH2 [21, 22].
Like the K&C and RHT models, the softening algorithm is based on strength reduction through
collapse of the limit surface and a phenomenological damage reduction of elastic properties.
Strength surfaces
Like the K&C and RHT models, the BF1 model uses three failure surfaces as shown in Fig. 2.1,
and the corresponding yield criteria are
Y
y
=
_
J
2
=
(F
m
(I
1
) N)F
c
(I
1
, )
(, I
1
)
, (yield surface), (4.1)
Y
m
=
_
J
2
=
F
m
(I
1
)
(, I
1
)
, (limit surface), (4.2)
Y
r
=
_
J
2
=
F
r
(I
1
)
(, I
1
)
, (residual surface), (4.3)
where F(I
1
) is taken to be an afne-exponential spline:
F
m
(I
1
) = a
1
a
3
exp(a
2
I
1
) +a
4
I
1
, (4.4)
F
r
(I
1
) = a
1f
a
3f
exp(a
2f
I
1
), (4.5)
and = S is a shifted stress tensor ( is backstress).
Similar to the RHT model, the BF1 model provides an option for porosity effects, and the cap
function is
F
c
(I
1
, ) =
_
1 : I
1
<,
1
_
I
1
X
_
2
: otherwise,
(4.6)
35
In Eqs. (4.1)(4.3), J
2
is the second invariant of the shifted stress (stress minus backstress) and
N is the maximum allowed translation of backstress. The BF1 limit surface is comparable to
the K&C and RHT models, and the distance between the yield surface and the limit surface is
controlled by the value N. For nonzero N, the motion of the yield surface towards the limit surface
is accomplished by kinematic hardening, which accounts for the Baushinger effect and does not
appear to be supported by the other two models.
Like the RHT model, the BF1 cap function F
c
(I
1
, ) accommodates material weakening caused
by porosity. As in Fig. 3.2, the variable marks the point where F
c
branches (smoothly) from a
constant value of unity at low pressure to begin its descent along an elliptical path to the value zero
at the hydrostatic compression elastic limit where the yield surface crosses the hydrostat at I
1
= X.
Like the K&C and RHT concrete model, the BF1 material model supports the Willam-Warnke
() function for third invariant dependence, but it also provides Mohr-Coulomb and Gudehus
options. The ratio (I
1
) used in this model can be a constant or it can be determined automatically
within the BF1 code as a pressure-dependent function coupled to pressure dependence of the TXC
strength:
(I
1
) =
1
1+
3A(I
1
)
. (4.7)
The ratio (I
1
) is determined automatically based on the slope of the compressive meridian, A(I
1
).
If this meridional slope is zero, then = 1. The meridional slope steepens (with decreasing
pressure) to a maximum allowed value. Thus, the yield surface smoothly varies from a von Mises
character at high pressure to a maximum principal stress at low pressure. When used with the
Willam-Warnke option, this gives the pressure varying octahedral prole similar to Fig. 2.2 and
Fig. 3.3 for the K&C and RHT models, respectively.
Like the K&C and RHT models, the BF1 model allows the limit surface to collapse down to
a residual surface as damage increases. Both the initial limit surface and residual surface are de-
scribed using the form in Eq. (4.4). They merely use different aparameters. The morphing of the
limit surface between them as damage progresses relies on an internal alternative parameterization
of Eq. (4.4). A limit surface of the form given in Eq. (4.4) can be viewed as bounded by the dashed
lines in Fig. 4.1. The user species values for the indicated slopes residual surfaces, from which
the code computes corresponding aparameters.
As damage proceeds, each of the four limit surface parameters is interpolated linearly between
intact and residual values. If for example, the user wishes to emulate a loss of hydrostatic tensile
strength similar to the K&C model, then PEAKI1 for the residual surface is zero. To emulate
damage similar to that of the Johnson-Holmquist damage (JH1 or JH2), the user would give re-
duced residual values for FSLOPE, STREN, and PEAKI1 (and YSLOPE = 0 for both intact and
residual). In the absence of data for failed strength, the BF1 model defaults the residual strength
parameters to that of sand (PEAKI1 = 0, FSLOPE = 0.18, YSLOPE = 0, STREN = intact value).
An upcoming release of BF1 includes numerical efciency enhancements that give 40% speed
up by a return algorithm in a lower-dimensional space for which the result is then projected into
36
Figure 4.1. Parameters for residual surface
six dimensional space, similar to the approach of Bicanic and Pearce [3]. The new version also
includes support for yield or limit surface vertices and new handling for pathological yield surface
contours near the hydrostatic tensile limit.
The plastic ow used by BF1 is rate dependent. Under high strain rates, elastic material response
occurs almost instantaneously, but accumulated damage is retarded by the materials inherent vis-
cosity, which prohibits observable inelasticity to proceed instantaneously. Thus, at high rates,
the material will appear to be more elastic than it would at low rates. Until sufcient time has
elapsed for damage to accumulate, the stress will lie transiently outside the yield surface, or even
outside the limit surface.
Rate dependence
Unlike the K&C and RHT models that rely on DIF data to account for rate effect, the BF1 model
uses a generalized Duvaut-Lions [10] rate-sensitive formulation, which computes two limiting
solution for the updated stress:
1. the low-rate (quasi-static) solution
L
that is found by solving the rate-independent equa-
tions.
37
Figure 4.2. Illustrations of the generalized Duvaut-Lions rate
sensitivity and the scale factor employed in the BF1 model [14]
2. the high-rate solution
H
corresponding to insufcient time for any plastic response to de-
velop so that it is simply the trial elastic stress.
To a good approximation, as illustrated in Fig. 4.2, The Duvaut-Lions rate formulation updates
stress using interpolation between the low-rate quasi-static plasticity solution
L
and the high-rate
purely elastic solution
H
,

L
+(
H
L
), (4.8)
where is a scale factor that varies from 1 at high strain rates to 0 at low strain rates as shown in
Fig. 4.2. To handle transients properly, the implementation in BF1 [12] is actually more sophisti-
cated than Eq. (4.8).
The abscissa in Fig. 4.2 is normalized by the materials characteristic response time . A
time interval t is considered long if t , and there is sufcient time for material to fully
develop plastic response and yield a solution that coincides with the quasi-static solution
L
. In
contrast, the time interval is deemed short if t .
Like the K&C and RHT concrete models, which rely on the empirical DIF data, the BF1
also uses empirical data to provide exibility in matching high strain rate for a wide range of
material types. However, the similarity ends there. The dynamic increasing factor (DIF) that is
used in the former models is a function of the strain rate, making it jump discontinuously if there
is a jump in strain rate (as in the arrival of a shock). In the BF1 model, there is an effective
38
DIF that is a functional of the strain rate that corresponds to stress states that can lie outside
the yield surface but cannot arrive or depart from such transient states instantaneously. In the
BF1 formulation, experimental data for apparent increase in strength is interpreted as the steady
state stress displaced from the yield surface under constant strain rate. The distinction between
views is that the DIF function approach fails to capture the transients prior to reaching steady
state. Moreover, the DIF function approach can cause numerical problems because it is capable
of discontinuities. As detailed in the BF1 users guide, the characteristic time is not a constant but
instead may itself depend on the strain rate and on the position of the stress on the yield surface so
that rate dependence of pore collapse can differ from that of cracking [5].
The BF1 model supports different levels of rate sensitivity depending on the mechanism of
inelasticity. Specically, pore collapse can be more rate sensitive than cracking, as has been ob-
served in laboratory data [13, 5]. Rate dependence for softening is a relatively new addition to
the BF1 model that is not documented in [12]. Softening is viewed as arising from crack growth.
Since cracks tend to grow at a xed speed regardless of the loading rate, BF1 treats softening rate
dependence as a time-based process for which material scale effects enter naturally by recognizing
that the amount of time required for a crack to propagate a xed distance (e.g. the distance to the
next crack to begin crack coalescence at the onset of catastrophic failure) must be itself xed if
crack speed is a constant. BF1 detects the onset of softening by stress reaching the limit surface,
but it delays the subsequent degradation in elastic and strength properties until the required amount
of time has passed. As mentioned, this delay time is viewed as the time needed to propagate a xed
distance and therefore its value is scale dependent.
Specically, when the BF1 model is run with the BFS
1
enhancement, it is proportional to a
characteristic length of the nite element similar to the one used by the K&C and RHT models.
The BF1 model is used as a base damage model that is premised on the assumption that its
material parameters have been assigned values appropriate to the scale of the nite element for a
homogeneously deformed domain. As illustrated in Fig. 4.3, such tests ideally would be conducted
for multiple specimen sizes to directly measure scale effects as well as inherent variability in
measured properties. If

T is the time-to-failure observed for a laboratory sample of volume

V, then
the time-to-failure T assigned to a nite element of volume V is
T =

T
_
V
V
_
1/3
. (4.9)
One appeal of a time-based damage progression model is that it naturally leads to a dependence
of the effective damage energy on loading rate. The amount of strain that can accumulated between
t
grow
= 0 and t
grow
= T is higher at high strain rates, thus leading to higher stresses, higher failure
energy, and therefore an increase in the number of failed elements. This trend is consistent with
fragmentation behavior observed in the laboratory where samples impacted at high rate produce a
larger number of fragments than those impacted at low rates. This feature distinguishes BF1 from
the K&C and RHT models, which apparently use rate insensitive fracture energies.
1
The BFS model [5, 29] is a model for automatically assigning scale appropriate BF1 parameters based on the size
of the nite element relative to the size of the specimen used in laboratory model calibration tests.
39
Figure 4.3. Standard Weibull distribution plots showing in-
creased Brazilian strength T
br
with decreased sample size. Here, P
s
is the complementary cumulative probability, which may be inter-
preted as the probability that the sample is safe from failure. The
slope of the tted line is the Weibull modulus, which quanties
variability in strength.
40
Figure 4.4. BF1 damage function for FSPEED values in the
range from 5 to 30. The higher FSPEED values correspond to the
steeper slope.
Damage accumulation
During calculations, if the stress is at or above the limit stress, a time-of-growth variable is incre-
mented as
t
n+1
grow
=t, (4.10)
where t is the time step. Otherwise,
t
n+1
grow
=t
n
grow
. (4.11)
Given the current value of t
grow
, an adjustable phenomenological damage parameter is evaluated
using a function of the form illustrated in Fig. 4.4.
The smoothness of the transition of damage from 0 to 1 is controlled by a user parameter
FSPEED such that large values of FSPEED would correspond to a nearly step discontinuity from
D = 0 to D = 1. The FSPEED option was added merely as a convenience to allow BF1 to emulate
the JH1 damage model [21] using large FSPEED or JH2 [22] using smaller values of FSPEED to
allow more gradual development of damage.
Whereas many damage models might evolve damage as a function of accumulated plastic
strain, the function in Fig. 4.4 evolves damage as a function of time. The rationale behind this
choice is discussed below.
41
As damage progress from D=0 to D=1, the tangent shear and bulk moduli degrade from their
initial values to residual values that are currently assigned internally in the code. Assuming that
the residual state corresponds to a rubble-like state, the shear modulus is reduced to zero. The bulk
modulus is reduced to a small fraction of its initial value if the pressure is tensile, but it equals its
intact value if pressure is compressive. The reason why the tensile bulk modulus is not allowed to
reach zero is not physical, but instead tied to the method for tracking of volumetric strain to detect
recompression. This method is currently under revision to allow zero bulk modulus in tension.
The stiffness degradation component of the BF1 model is regarded by even its own developers
as nothing more than an ad hoc means of achieving qualitatively correct behavior. A more physi-
cally based stiffness degradation model would allow development of induced anisotropy reecting
the tendency for quasi-brittle materials to develop orthotropically oriented cracks. Future revisions
of BF1 are anticipated to support induced elastic anisotropy either by retrotting its existing ability
to model orthotropic rock joints or by introducing a directional damage theory based on the work
of Dienes [9] and Kachanov [23]. At present, however, the shear modulus G and bulk modulus K
are degraded as follows:
K = K
intact
(1d
K
), (4.12)
G = G
intact
(1d
G
), (4.13)
where
d
K
=
_
0 if p > 0 (compression),
D
if p < 0 (tension),
(4.14)
and
d
G
= D
(dF/dI
1
)
p=p
current
(dF/dI
1
)
p=0
. (4.15)
The expression for d
G
is designed to allow full recovery of the shear stiffness as conning pressure
increases. Recognizing from the work of Dienes [9] and Kachanov [23] that elastic compliance is
related to the cube of crack size and recalling that crack propagation speeds tend to be constant,
the D
used in the above formulas is based on the isotropic part of the anisotropic crack-degraded
stiffness formulas of Dienes, and is expressed as
D
= 1
1
1+
_
1
1D
1
_
3
. (4.16)
Using D
instead of D will cause the bulk modulus to hang on close to its initial value for a
while to reect the fact that small cracks do not signicantly alter stiffness. A signicant and
sudden drop in stiffness can be seen only when cracks become large. As mentioned, the power of
3.0 dependence reects trends predicted in microphysical theories for stiffness of a cracked body.
42
Plastic update
During the initial hardening phase (before onset of damage) the BF1 model updates the material
state using standard techniques of classical plasticity theory. As already mentioned, for example,
the strain rate is decomposed into elastic and plastic parts, with the elastic part determined from
elastic unloading data and the plastic part being a multiple of the ow potential gradient,

p
=
_
_
, (4.17)
where is called the consistency parameter because its value is set to ensure consistency with
the requirement that stress remain on the yield surface during plastic loading. The consistency
parameter can be determined by satisfying
f =
f
: +
f
+
f
: = 0. (4.18)
Unlike the K&C and RHT models, the BF1 allows for kinematic hardening as well as isotropic
hardening. As is typical in conventional plasticity modeling, closure of the governing equations
(i.e. obtaining enough equations to solve, as detailed in [1]) requires specication of evolution
equations for all internal variables. For BF1, there are two internal state variables: for pore
collapse and for kinematic hardening.
The isotropic hardening is governed by the rate variable , where is the pressure at which
void begins to collapse, see Fig. 3.2. The kinematic hardening is governed by . It was shown in
[12] that these rates are proportional to the consistency parameter and can be expressed as
= h
k
and = H
, (4.19)
where h
k
is an isotropic hardening modulus, and H
is a kinematic hardening tensor, each of which

is determined from laboratory data as described below.
Evolution equation for pore collapse
If a material is capable of permanent volume change, then the material likely contains voids. Recall
from Eq. (4.6) that the branch point and the hydrostat intercept X are used to characterize the
effect of void collapse. The hydrostatic intercept X
o
equals the value of I
1
(which is proportional
to pressure p = I
1
/3) at which pores begin to collapse, as indicated by point A in Fig.4.5. There
after, increasing pressure is required to continue pore collapse. If this pressure is provided, the
yield hydrostat, X = 3p, intercept moves outward until all pores have been crushed out, at which
43
Figure 4.5. Hydrostatic pressure vs. volumetric strain
point the load and unload curves in hydrostatic compression become tangent, as indicated by point
B in Fig. 4.5.
If the pressure is released at any point during hydrostatic compression, there will be permanent
residual plastic volume change,
p
v
, as labeled in Fig. 4.5. Thus, since a relationship between X and
p
v
is directly measured in the laboratory, it might seem natural to use the hydrostat X as an internal
variable. However, for numerical reasons, it proves to be more convenient to use as the internal
state variable. Given the measured relationship between X and
p
v
and the relationship between
and X in Eq. (4.6), and
p
v
are implicitly related. Then, by the chain rule
=
d
d
p
v

p
v
=
d
dX
dX
d
p

p
v
. (4.20)
Moreover, since
p
v
= tr
p
, Eq. (4.19) leads to
= h
k
where h
k
= 3
d
dX
dX
d
p
v
I
1
(4.21)
Evolution equation for backstress
When kinematic hardening is enabled, a shifted stress tensor = S is used in the yield func-
tion instead of the actual stress. The backstress (deviatoric tensor) is initially zero and evolves
44
Figure 4.6. Qualitative sketch of shear stress vs. shear strain
proportionally to the deviatoric part of the plastic strain rate according to
= HG
()
p
and
p
= dev(
p
) =
_
_
, (4.22)
where H is a material constant and G
() is a scalar-valued decay function designed to limit the

kinematic hardening such that G
0 as second invariant of approaches a user-specied maxi-

mum allowed value N. This behavior gives results similar to power law hardening as qualitatively
illustrated in Fig. 4.6. The BF1 model uses the following decay function
G
() = 1
tr
2
2N
. (4.23)
Using Eqs. (4.19) and (4.22), the kinematic hardening modulus tensor is
H
= HG
()dev
_
_
. (4.24)
While the K&C model presumes a constant Poissons ratio (from which nonlinear shear modulus
can be computed from nonlinear bulk modulus), the BF1 model computes the shear and bulk
45
modulus using input parameters that are obtained by curve tting hydrostatic and triaxial data to
nonlinear functions. The BF1 model supports linear and nonlinear hypoelasticity. Unlike the K&C
and RHT model, where tabulated data are used, the nonlinear elasticity in BF1 is implemented by
allowing the tangent moduli to vary with the stress according to
K = b
0
+b
1
exp
_
b
2
|I
1
|
_
, (4.25)
G = g
0
_
_
1g
1
exp
_
g
2
J
1/2
2
_
1g
1
_
_
, (4.26)
where b
0
, b
1
, g
0
, g
1
and g
2
are material parameters tted to experimental data. The model is linear
elastic if b
0
and g
0
are specied, and all other elastic parameters are zero (or unspecied). Unlike
the K&C and RHT models, additional terms (with additional parameters) are available to support
elastic-plastic coupling, where plastic hardening changes elastic properties (e.g. pore collapse
induces elastic stiffening).
A disadvantage of Eqs. (4.25) and (4.26) is that they are difcult to parameterize because
neither is integratable to obtain a closed form analytical expression for stress as a function of
elastic strain. Alternatives are therefore under investigation. Other concerns are that Eqs. (4.25)
and (4.26) have not been well validated in tension, and the current (2008) implementation does
not consistently incorporate a z-tensor of the type discussed by Brannon [14] that is required in
incremental plasticity with elastic-plastic coupling.
Like the K&C and RHT models, BF1 uses an isotropic elastic stiffness except that some pre-
existing initial elastic anisotropy is optionally available for joints. However, since damage gener-
ally induces signicant anisotropy in stiffness, all three models currently rely on very ne meshing
so that deformation-induced anisotropy is approximated through explicit (mesh resolved) hetero-
geneity.
46
Chapter 5
LS-DYNA Concrete Model 159 (CSCM)
This chapter presents an additional overview of a new concrete material found in LS-DYNA, Con-
crete Model 159 (referred to as CSCM in this report). It was developed for DYNA3D Analysis
Tools for Roadside Safety Applications II (2007) program by the U.S. Department of Transporta-
tion [11]. Since this model is not a part of this survey projects original Statement of Work (SOW),
only a brief summary of its features will be given here.
Strength surfaces
The CSCM model uses strength surfaces similar to the previous models investigated in this report.
The failure surface is dened by the three invariants together with the cap hardening parameter
similarly to the RHT and BF1 models. The yield function is expressed by
f (I
1
, J
2
, J
3
, ) = J
2
(, I
1
)F
2
f
F
c
(5.1)
where (, I
1
) is the Rubin third-invariant factor. The cap function is the same as in the RHT and
BF1 models (see Chapters 3 and 4). The same afne-exponential spline used by the BF1 model is
used to describe the limit surface,
F
m
(I
1
) = a
1
a
3
exp
a
2
I
1
+a
4
I
1
(5.2)
The initial yield stress is then determined from the limit surface using
F
y
(I
1
) = N
H
(a
1
a
3
exp
a
2
I
1
+a
4
I
1
) (5.3)
where N
H
is a factor ranging between 0.7 < N
H
1, which governs the location of the initial yield
surface. Therefore, the CSCM model is similar to the RHT models use of a multiplier to specify
the separation between the initial yield surface, whereas the BF1 model species this separation
additively.
The CSCM model supports kinematic hardening very similar to the BF1 model. The translation
of the yield surface is done via the back stress . The total stress is updated by summing the initial
47
stress and the backstress. The hardening rule used by this material model is based on stress to
ensure that the shear surface coincides with the limit surface. The rate of kinematic hardening is
controlled by a user input C
H
, and the incremental back stress is expressed as
=C
H
G()( ) t. (5.4)
The quantity G() is used to limit the increment such that the yield surface cannot move beyond
the limit surface as discussed in Chapter 4 for the BF1 model. In fact, the above equation can
be compared with Eq. (4.78) in [12]. No simulations were performed to determine if the CSCM
kinematic hardening model is identical to the BF1 model in all respects.
The strength of the model is increased with increasing strain rate. The CSCM model applies rate
effects to the limit surface, residual surface, and the fracture energy as shown in the previous
sections. A modied Duvaut-Lions formulation is applied to the yield surface such that the high-
rate stress is an interpolation between the quasistatic low-rate stress and the elastic stress:

L
+(
H
L
), where =
t/
1+t/
. (5.5)
A similar equation is given in the BF1 models introduction to rate dependence, but the BF1 doc-
umentation points out that additional terms are needed once the dynamic stress lies outside the
quasistatic yield surface.
As discussed in Chapter 4, the viscoplastic stress is bounded between the current rate-independent
stress and the elastic trial stress at each time step. The high strain rate is handled by modifying
using
=

o

n
(5.6)
Equation (5.6) allows for user input parameters,
o
and n, that can t rate effects data at high and
low strain rates. These parameters are used to represent DIF specications given by CEB [35]
similar to the K&C and RHT models. Input parameter can be determined using the following
relationships as given in CEB-FIP [35],
Tension:
DIF =
_

o
1.016
s
for 30s
1
s

o
1/3
for > 30s
1
,
(5.7)
48
where
s
=
1
10+6 f
c
/ f
c
o
,
log
s
= 7.112
s
2.33,
f
c
o = 10 MPa,

o
= 3010
6
s
1
, and
f
c
= concrete compressive strength.
Compression:
DIF =
_

o
1.026
s
for 30s
1
s

o
1/3
for > 30s
1
,
(5.8)
where
s
=
1
5+9 f
c
/ f
c
o
log
s
= 6.156
s
2,

o
= 3010
6
s
1
, and
f
c
= concrete compressive strength.
Despite having a similar base-line formulation to handle the rate effect as the BF1 model (using
Duvuat-Lions rate sensitivity formula), the CSCM model has a built-in feature that allows a user
to include DIF data for the rate effect. In addition, a user is also given an option to apply the DIF
to the static fracture energy (enabled by the models defaults), which causes the fracture energy to
be strain rate dependent. The CSCM models default DIF is based on the developers experience,
and is different from those given in CEB-FIP model code.
Similar to the K&C and RHT models, the CSCM model handles scale effects by incorporating
an element characteristic length w
u
(cube root of the element volume). The model calculates
the damage parameters as a function of element size. Regardless of element size, the fracture
energy, G
f
, remains constant. The fracture energy is regulated separately between brittle and
ductile softening, and is computed by integrating the stress-displacement curve.
G
f
=
_
r
0b
w
_
1+b
ab
_
log(1+b) for brittle softening
2r
0d
w
_
1+d
cd
_
log(1+d) +2w
_
1+d
c
2
_
0
ye
y
1+ce
y
dy for ductile softening,
(5.9)
where
y = c
_
x
x
o
_
_
f
w
,
x = displacement,
x
o
= displacement at peak strength f
.
49
Using Eqs. (5.10) and (5.9), the softening parameters a and c are computed according to the el-
ement characteristic length w, while b and c remain user input parameters. The value G
f
is ap-
proximated by the model from three fracture energy inputs: (1) from tensile stress, G
f t
, (2) shear
stress, G
f s
, and (3) compressive stress, G
f c
. When rate effects are considered, the fracture energy
is scaled using
G
vp
f
= G
f
_
r
o
r
s
_
n
. (5.10)
The range of 0.5 n 1 is recommended by the developer. When n equals to 1, the G
vp
f
is
proportional to the increase in strength with rate effects.
Damage accumulation
Both strain softening and modulus reduction are accounted for in the damage formulation based
on [34]. The damage stress,
d
, is computed by
d
= (1D)
vp
(5.11)
where D is a damage parameter ranging from 0 to 1, and
vp
is a stress tensor without damage,
which is updated from the viscoplasticity algorithm. This algorithm structure of applying damage
after evaluation of the non-damaged stress update is identical to what was used in the versions
of the BF1 model predating 2008. (Starting in 2008, the damage part of the BF1 algorithm was
integrated within the stress update subcycles.)
The CSCM model handles damage using a strain-based energy approach. When this energy
exceeds a material damage threshold, damage is initiated and accumulated via the parameter D.
The damage threshold is determined using two different formulations for brittle and ductile dam-
age. Unlike the BF1 model, which allows damage in compression, brittle damage accumulates
in the CSCM model only when the pressure is tensile. Its damage threshold,
b
, depends on the
maximum principal strain
b
=
_
E
2
max
. (5.12)
The brittle damage initiates when
b
> r
0b
, where r
0b
is the initial brittle damage threshold. Such
behavior is supported as a special case in the BF1 model.
Ductile damage, on the other hand, accumulates when the pressure is compressive, and the
damage threshold,
d
depends on the total strain components and is expressed as
d
=
_
1
2
: . (5.13)
50
It initiates when the initial ductile damage threshold, r
0d
, is exceeded.
The rate effect is accounted for by shifting the damage threshold using
r
0
=
_
1+
E
r
s
E
_
r
s
, (5.14)
where r
o
is the shifted threshold with viscoplasticity, r
s
is the damage threshold before the appli-
cation of viscoplasticity, and is rate effects. The shifted damage threshold allows the delay of the
damage initiation while the plasticity accumulates. This approach appears to be unrelated to the
damage delay strategies of the other three models.
Damage accumulation during softening is computed as a function of the damage threshold
using
D() =
_
_
_
0.999
b
_
1+b
1+bexp
a(r
0b
)
1
_
for brittle =
b
,
D
max
d
_
1+d
1+d exp
c(r
0d
)
1
_
for ductile =
d
(5.15)
The parameters a, b, c, and d are determined by curve-tting Eq. (5.15) to the softening portion of
a stress-strain plot. The parameter D
max
represent a maximum attainable damage and is dened as
D
max
=
_
_
3J
2
I
1
_
1.5
if

3J
2
I
1
< 1
1 if otherwise.
(5.16)
When considering rate effects, the above equation is scaled using
D
max
= D
max
max
_
1,
_
1+
r
0
r
s
_
1.5
_
(5.17)
In the current release of the CSCM material model, the exponent 1.5 is set internally by the devel-
oper based on examination of single element simulations. The maximum increment of the damage
in a single time step is 0.1, and it is also set internally by the model.
Incidentally, Eq. (5.15) allows exibility in the damage accumulation behavior that is similar
to the time-based approach used in the BF1 model.
Plastic update
Similar to the other concrete models, a conventional regular associated ow rule is adopted by
the CSCM model. Therefore, the details on plastic update of this model are not covered in this
51
report. According to the developers, the model efciency is improved by using subincrementation
instead of iteration to return the stress state to the yield surface. It is invoked only when the current
strain increment exceeds a maximum strain limit specied by a user or the models default. The
numerical schemes provided by the CSCM models developers can be found in [30, 32].
The CSCM model computes the default shear and bulk moduli using
G =
E
2(1+)
and K =
E
3(12)
, (5.18)
where E is the Youngs modulus and is the Possions ratio. The Youngs modulus E is determined
from an equation in CEB-FIP:
E = E
C
_
f
c
10
_
1/3
, (5.19)
where E
C
is the Youngs modulus of a 10 MPa concrete. The Youngs modulus E
C
is not used in
the prepeak hardening part of the simulations.
52
Chapter 6
Model comparisons
This chapter presents a list of the comparison of theory and implementation of each model. In
addition, the four models are also compared numerically for several case studies.
Comparison of theory and implementation
Listed below are the features comparison of each model:
The K&C concrete model relies most heavily on a set of empirically obtained data and
curves. The accuracy of the result depends on a good set of data for yield stress factor and
effective plastic strain , which are obtained via trial and error.
The RHT, BF1, and CSCM models support a cap on the yield surface, while the K&C model
does not appear to have this option (although the K&C model does allow porosity effects in
the compaction EOS).
The BF1 model supports intrinsic anisotropy through an elastic joint model. The CSCM
model supports intrinsic elastic anisotropy associated with rebar. Both BF1 and CSCM
models support deformation induced anisotropy thru kinematic hardening. None of the other
models support anisotropy in any form. All of these models, including the BF1 and CSCM
models, neglect the anisotropy that one might expect from oriented crack growth.
The BF1 model determines the TXE/TXC strength ratio automatically based on the slope
of the compressive meridian. If this slope is zero, = 1. As the slope increases to a maxi-
mum allowed value (corresponding to a principal stress criterion), =1/2. The RHT model
appears to be handling this issue in a similar manner. On the other hand, the K&C model
denes as a piecewise function of pressure, where is always 1/2 in the tensile pressure
range.
All four models offer defaults for unspecied parameters. The K&C model defaults values
to that for 45MPa concrete. The RHT model offers defaults via loading preset parameters
for 35MPa and 140MPa concretes. The BF1 model offers preset parameters for 23MPa con-
crete and conventional strength Portland cement as well as several other materials such as
limestone and ceramic. Moreover, when building a new parameter set from scratch, the BF1
53
model defaults to reasonable guesses. For example, if no ow parameters are given, they
default to associative ow parameters. The CSCM model automatically generates parame-
ters for concretes having strengths that range between 20 and 58 MPa via built in quadratic
tting equations.
Softening is best handled in the BF1 model by using it in combination with statistical het-
erogeneity and scale effects available in the BFS enhancement that can be regarded as an
add-on available only in certain host codes. The K&C, RHT, and CSCM models appear to
use only scale effects, but they could be modied to be run with statistics as well if they use
an input parameter change of variables similar to the BF1 model use of the parameters in
Fig. 4.1 as alternatives to its a-parameters. Parameter re-denition is needed when running
with uncertainty to ensure that admissible realizations of the failure surfaces are generated.
The K&C, RHT, and CSCM models deal mainly with concrete material, and may seem to
require fewer user input parameters than the BF1 model, but this seems to be simply because
parameters for those models default to concrete values. The greater concern is that some
built-in parameters (i.e., not controllable by the user) may limit usefulness of the K&C and
RHT models for other concrete-like materials such as rocks.
Although the BF1 model has many possible user input parameters, it automatically sets
defaults for any unspecied parameters. The BF1 model might be capable of simulating
responses of a wider range of materials. For example, user input parameters can be mod-
ied to model simpler elastic materials, as well as classical plasticity idealizations such as
von Mises, Tresca, and Mohr-Coulomb theories. The ability to reduce to simpler models
has allowed the BF1 models solution algorithm to be rigorously tested against analytical
solutions.
To model rate sensitivity, the K&C, RHT and CSCM models use a strength enhancement
factor (or DIF), which follows the concrete data given in CEB-FIP design code. The K&C
model allows for data of DIF to be entered in a tabulated form. The RHT model, on the other
hand, cannot cover the whole range of strain rate given in CEB-FIP because it represents DIF
vs. log only in a linear form.
Both BF1 and CSCM models handle rate effects using a Duvaut-Lions overstress formulation
that is time based rather than strain rate based, which therefore allows for more accurate
prediction of transients. These models incorporate a characteristic material response time.
The CSCM model includes a means of setting the response time to match CEB-FIP data,
whereas no automated support is currently available in the BF1 model to match specied
data for rate effects in uniaxial strain loading.
Both K&C and RHT models support user-specied EOS. The RHT supports a wide range of
EOS forms that includes linear, polynomial, p, and user tabulated compaction data, etc.
The K&C models allow EOS to be specied only in a tabulated form. On the other hand,
the BF1 and CSCM models provides a built-in nonlinear EOS designed to match data over
pressure ranges ordinarily encountered in low to moderate pressure hydrostatics tests, and
does not allow for any user specied EOS methods.
54
Figure 6.1. Comparison of meridian proles to the K&Cs pro-
le. (The CSCM and BF1 curves coincide.)
The BF1 model includes elastic-plastic and elastic-damage coupling (plasticity or damage
changes the bulk modulus). The K&Cmodel has no coupling of elastic properties to inelastic
loading history. However, it scales the bulk modulus depending on pressure using a scaling
factor. The coupling feature is unclear for the RHT model because this information is lacking
in its user manual. All three models support nonlinear bulk modulus.
The K&C, BF1, and CSCM models provide well-documented user manuals (in English),
which also explain the schemes used in detail. On the other hand, the user manual for the
RHT model is not available with the packaged model.
Numerical comparison
The data used for the RHT, BF1, and CSCM models are tuned to yield results similar to those
presented by Unosson [37] using the K&C model . The material models default values are used
for the parameters when their input information are insufcient. This section refers to the users
inputs using symbols given in AUTODYN and LS-DYNA [6, 27].
Meridional prole
This section presents a numerical comparison of the meridional curves obtained from the four
investigated models. The curves from the RHT, BF1, and CSCM models are constructed based on
the parameters for the 153MPa concrete used for the K&C model in [37]. These parameters are
shown in Table 6.1. The comparisons of the curves from each model are shown in Fig.6.1.
55
Table 6.1. Parameters for meridional curves
Model parameters value
All Compressive strength, f
c
153 MPa
All Tensile strength, f
t
8.2 MPa
K&C limit surface parameter, a
0
50.6 MPa
1
0.465
2
65.7 GPa
1
K&C residual surface parameter, a
0f
0.0 MPa
1f
0.465
2f
65.7 GPa
1
RHT limit surface initial slope, A 2.2
RHT limit surface exponent, N 0.5
RHT residual surface initial slope, B 1.8
RHT residual surface exponent, M 0.55
BF1 limit surface parameter, a
1
722 MPa
2
4.9910
10
3
713 MPa
4
none
BF1 limit surface parameter, FSLOPEI 0.36
BF1 limit surface parameter, STRENI 722 MPa
BF1 limit surface parameter, PEAKI1I 24.6 MPa
BF1 residual surface parameter, FSLOPEF 0.3
BF1 residual surface parameter, STRENF 733 MPa
BF1 residual surface parameter, PEAKI1F 246 Pa
CSCM limit surface parameter, 722 MPa
CSCM limit surface parameter, 4.9910
10
CSCM limit surface parameter, 713 MPa
CSCM limit surface parameter, none
56
Figure 6.2. p equation of state employed by the RHT model.
Table 6.2. P EOS input parameters
Input parameters values
Porous density (kg/m
3
),
p
2572
Porous sound speed (m/s), c
p
3200
Initial compaction pressure (MPa), p
e
200
Solid compaction pressure (MPa), p
s
2000
Compaction exponent, n 2
Bulk modulus (GPa), K 28.43
Single element test: isotropic compression
Single element isotropic compression tests are conducted for each model, and hydrostatic pressure
versus volumetric strain are obtained. The results are compared to the experimental data for a
153MPa concrete provided in [37]. Both K&C and RHT models require the EOSs to be specied.
The K&C model requires a tabular EOS. The RHT model provides options for various functional
forms of the EOS. For porosity effects, our test simulations employ the following p model:
= 1+(
p
1)
_
p
s
p
p
s
p
e
_
n
. (6.1)
The parameters
p
, p
s
and p
e
used here are dened and shown in Fig. 6.2 and Table 6.2.
The BF1 and CSCM models have built-in mechanical EOS and they both employ the following
relationship between porosity and plastic volumetric strain (which is an indirect measure of pore
57
Figure 6.3. Equation of state: BF1 and CSCM crush curve
Table 6.3. Uniaxial strain loading for single element tests
time (s) displacement (m) velocity (m/s)
K&C, BF1, CSCM RHT
0.0 0.0 -0.3
1.0 -0.3 -0.3
1.000001 - 0.3
2.0 0.0 0.3
collapse):
p
3

p
v
= p
3
e
(p
1
+p
2

)

, where

=

X p
0
= 3(pP
E
). (6.2)
The parameters p
1
, p
2
, p
3
and P
E
are dened in Fig. 6.3. The parameters required by the RHT
p, BF1 and CSCM crush curves are approximated by curve tting the EOS data given in [37]
to the K&C model. The comparison of the hydrostatic compression curves from each model is
illustrated in Fig. 6.4
Single element test: uniaxial strain
Single element uniaxial strain tests are conducted using the data shown in Tables 6.1 and 6.4. The
loading data for each model are given in Table 6.3. Figure 6.5 shows a comparison of the stress
difference
3J
2
versus pressure. All of the material models show similar behavior except the K&C
model. In this test, the pressure exceeds the maximum pressure data given in the EOS table used
58
Figure 6.4. Comparison of hydrostatic pressure versus volumet-
ric strain
Table 6.4. Material properties for concrete
Property value
Density (kg/m
3
) 2770
Uniaxial compressive strength (MPa) 153
Modulus of elasticity (GPa) 58
Poissons ratio 0.16
Bulk modulus (GPa) 28.4
Shear modulus (GPa) 25
by the K&C model, and extrapolated data are added to the EOS table for the K&C model to obtain
results beyond 2 GPa.
Projectile penetration
Here we present numerical simulations of the perforation of a high performance concrete with a
75 mm steel projectile using the RHT and CSCM material models. The goal of these simulations
is to investigate the results obtained using these two models and to compare them to the results
from the K&C model given in [37]. ANSYS: AUTODYN is used for the investigation of the RHT
concrete model, while LS-DYNA is used for the CSCM model. The data used in the RHT and
CSCM models are tuned to give similar results to Unossons as discussed in the single elements
sections. These data are shown in Tables 6.5 - 6.7. A simplied Johnson-Cook strength model is
used for the steel projectile to ensure that the pressure in the target concrete does not damage the
projectile.
59
Figure 6.5. Comparison of stress difference versus pressure un-
der uniaxial strain. (The BF1 and CSCM models results are differ-
ent not because of model differences but because the BF1 model
was driven with logarithmic strain, whereas the CSCM model was
run using engineering strain.)
Table 6.5. Input parameters for the steel projectile
Property value
Density (kg/m
3
) 7800
Modulus of elasticity (GPa) 207
Initial velocity (m/s) 617
Shear modulus (GPa) 81.8
Yield stress (MPa), A 1539
Hardening constant (MPa), B 477
Hardening exponent, n 0.18
Strain rate constant, C 0.012
60
Table 6.6. Input parameters for the RHT concrete target
Property value
Density (kg/m
3
), 2770
Shear modulus (GPa), G 25
Compressive strength (MPa) 153
Tensile strength, f
t
/ f
c
0.054
Shear strength, f
s
/ f
c
0.3
Limit surface parameter, A 2.2
Limit surface parameter, N 0.5
Tensile/compressive meridian ratio, Q
o
0.6805
Brittle to ductile transition, BQ 0.0105
G
elastic
/G
elastic - plastic
, PREFACT 2.0
Elastic strength/ f
t
, TENSRAT 0.8
Elastic strength/ f
c
, COMPRAT 0.75
Cap on yield surface enable
Residual surface parameter, B 1.8
Residual surface parameter, M, 0.55
Compressive strain rate exponent, 0.01
Tension strain rate exponent, , 0.013
Maximum fracture strength ratio 110
20
Damage constant, D
1
0.04
Damage constant, D
2
1.0
Minimum strain to failure 0.01
Residual shear modulus ratio, SHRATD 0.13
Principal tensile failure stress (Pa) 8.210
6
Maximum principal stress difference/2 110
20
Fracture energy (J/m
2
) 162
Flow rule radial return
Stochastic failure none
61
Table 6.7. Input parameters for the CSCM concrete target
Property value
Density (kg/m
3
), 2770
Maximum strain increment default
Rate effect disable
Elements erosion criterion 1.5
Modulus recovery in compression, enable
Cap retraction none
Preexisting damage none
Shear modulus (GPa), G 25
Bulk modulus (GPa), K 28.43
TXC surface parameter (Pa), 4.1910
8
TXC surface parameter (1/Pa), 110
9
TXC surface parameter (Pa), 3.8910
8
TXC surface parameter, 0.0433
TOR surface parameter,
1
0.761
TOR surface parameter (1/Pa),
1
0
TOR surface parameter,
1
0
TOR surface parameter (1/Pa),
1
210
5
TXE surface parameter,
2
0.68
TXE surface parameter (1/Pa),
2
0
TXE surface parameter,
2
0
TXE surface parameter (1/Pa),
2
2.2910
5
Hardening initiation N
H
and rate C
H
0.0 and 0.0
Cap aspect ratio R and initial location X
o
(Pa) 5.0, 610
8
Maximum plastic volume compaction, W 0.7434
Linear shape parameter (1/Pa), D
1
610
10
Quadratic shape parameter (1/Pa
2
), D
2
0
Ductile and brittle softening parameter, B and D 100 and 0.1
Fracture energy in uniaxial compression (N/m), G
f c
16200
Fracture energy in uniaxial tension (N/m), G
f t
162
Fracture energy in pure shear (N/m), G
f s
162
Shear-to-compression transition parameter, pwrc 5.0
Shear-to-tension transition parameter, pwrt 1.0
Modify moderate pressure softening parameter, pmod 1.0
Compressive rate effect and exponent,
0c
and N
0c
1.8310
4
and 0.504
Tensile rate effect and exponent,
0t
and N
0t
1.7610
5
and 0.560
Maximum overstress allowed in compression (Pa), overc 1.0510
8
Maximum overstress allowed in tension (Pa), overt 7.7610
6
Ratio of uidity parameters, srate 1.0
Fracture energy rate effect parameter 1.0
62
The contour plots of damage prediction by the RHT and CSCM models are illustrated in
Figs. 6.6 and 6.7. Despite claims that inclusion of scale effects eliminates mesh dependence,
it is clear from the pattern of radial cracking and from the shape of the perforation that the mesh
texture strongly inuences how the damage is distributed. As explained in [5], deterministic dam-
age models of the type surveyed in this report are incapable of predicting realistic radial cracking
even if they include scale effects for damage. This fact is why the BF1 model does not include
scale effects. Instead, the BFS extension of the BF1 model incorporates scale effects as a natu-
ral consequence of statistical heterogeneity of the material strength, thus leading to a statistically
mesh independent prediction for the number and orientation of radial cracks. Specically, the BFS
extension generates random realizations of the BF1 parameters based on scaling the size of the
nite element relative to the specimen size used in the BF1 model parameterization.
As mentioned earlier, our perforation case study aims to reproduce the analysis in [37], which
used the K&C model in an older version of LS-DYNA. Similar number of nodes and elements are
used in our simulations. However, in the current version of LS-DYNA, this problem fails to run
to completion even though it uses identical parameters, geometry, and boundary conditions. When
tting the data of DIF given in [37] to both models, perforations in the concrete targets are not
predicted. The penetrations in both RHT and CSCM simulations stop at approximately 2/3 of the
targets thickness.
When the DIF values are close to unity, the simulations using the RHT and CSCM models
predict perforation. The size of the impact crater predicted by the RHT model is much smaller than
that in [37] for the K&C model. The CSCM, on the other hand, shows the crater of a comparable
size. The RHT, CSCM and the previously published K&C simulation [37] exhibit double cracks in
the coordinate direction, and single cracks at 45
. Such spurious behavior is typical of deterministic

smeared damage models. As the mesh is rened, it is expected that the number of radial cracks will
increase, thus indicating non-predictiveness of smeared damage models even when they include
scale effects for fracture energy. The severe mesh dependence of smeared damage models is also
noted in a systematic study by Timmel, et al. [36].
As shown in Table 6.8, none of the smeared damage models are able to predict the residual
velocity of 291 m/s that was observed in the experiment.
Table 6.8. Residual velocity comparison
Material models impact velocity (m/s) residual velocity (m/s)
Experiment [37] 617 291
K&C [37] 617 320
RHT 617 264
CSCM 617 327
63
Figure 6.6. RHT Model: Contour plots of damage: side, front,
and back view of the target (top to bottom).
64
Figure 6.7. CSCM Model: Contour plots of damage: side, front,
and back view of the target (top to bottom).
65
Verication
Verication is evidence that the equations are solved correctly. We could nd no evidence of sys-
tematic verication protocols for either the K&C, RHT, or CSCM models. Some potential for
verication issues have been identied, but not aggressively investigated by these developers. For
example, existence of a vertex in the failure surface of the K&C models was noted [28], but dis-
missed with the comment that this does not violate any fundamental theoretical requirement. In
fact, due to the use of a Prandtl-Reuss ow rule as implemented with the radial return algorithm, it
creates no numerical difculties either. Very little is said in the K&C, RHT, and CSCM publica-
tions about mesh dependence other than an unsubstantiated claim that inclusion of a length scale
in fracture energy eliminates mesh dependency.
In contrast to the K&C, RHT, and CSCM models, the BF1 model uses the following verication
practices:
Demonstration of mesh dependency that led to an enhanced form of the BF1 model (called
BFS) that accounts for scale dependent statistical uncertainty of strength, thus mitigating
(not eliminating) mesh dependency while also reproducing the massive uncertainty in exper-
imental data for concrete.
Identication of three possible types of vertices in the failure surface and implementation of
vertex handling that has been veried against analytical solution.
Run time monitoring of solution quality with remote debugging. Without access to the K&C,
RHT, and CSCM source codes, it is uncertain how these models deal with inadmissible
results such as an erroneous prediction of a negative consistency parameter. If the BF1
model encounters such a problem, it terminates the calculation by issuing a debugging le
that can be sent to the model developer for a bug x followed by a rapid re-release of a
corrected model (often within 48 hours).
Automated scripts that allow assembly of version-controlled releases of the model in forms:
an F77 version, an F90 version, and two stand-alone model driver versions. Automation of
the release ensures identical source code on multiple platforms.
Driver regression and benchmark testing involving approximately 30 test problems using
three material data sets (thus totaling 90 tests) that must either pass before a new version
is released or, if they do not pass, a known bug comment must be included in the release
documentation.
Genuine verication testing against analytical solutions. Unlike the K&C and RHT models,
BF1 is capable of reducing to simpler idealized models for which agreement with analytical
solutions has been conrmed.
Pseudo verication testing for expected trends in solution as parameters are changed.
Robustness verication achieved through benchmarks that target common problem areas
such as loading past a vertex, loading that is highly tangential to the failure surface, running
66
Figure 6.8. FEA vs. analytical/numerical elastic wave velocity
example
with an inordinately large time step to ensure proper subcycling, dramatic transients in the
loading direction, massive strains, etc.
Rate of convergence testing with and without subcycling demonstrating rst order conver-
gence. Note: second-order convergence has not been attempted since it would require not
only strain rate , but also strain acceleration .
Limited eld-scale verication testing to reveal issues not attributable to the constitutive
model. In situations where analysts have suspected a verication or validation problem with
the BF1 model, the BF1 developers have designed verication tests that ultimately showed
the bug to reside in the host code, not the constitutive model. The good agreement of the
BF1 model for the elastic problem in Fig. 6.8 for example, was not achieved until articial
viscosity was repaired in the host codes that were attempting the inelastic validation of that
problem. The BF1s remote debugging capabilities have also led to discovery of errors in
host code advection of tensor elds as well as more routine problems such as user input
mistakes (most commonly confusion between uniaxial effective strength and effective shear
strength).
Extensive documentation. The BF1 model has two components: the base model for the
nonsoftening case and damage theory for softening. A detailed users guide for the nonsoft-
ening component that documents the theory for relatively new users is available [12]. It also
67
outlines the numerical solver; summarizes verication and validation tests, provides step-
by-step parameterization instructions, summarizes input parameter symbols and keywords
(with cross-referencing to relevant theory equations); provides input sets for several mate-
rials (including high- and low-strength concrete), and shows how to set input for reducing
the model to simpler idealizations (Hookes law, von Mises plasticity, Mohr-Coulomb, etc)
for verication testing. The more recent BFS changes to the BF1 model have not yet been
documented.
Run-time input sanity checking with automated defaults. The BF1 model has 40 possible
inputs keywords, but the number of inputs that are actually required scales appropriately
with the complexity of the problem. If, for example, the user wishes to run linear elasticity,
then only two parameters (bulk and shear modulus) are needed. All user input is checked
at run time for physical or mathematical admissibility (e.g. bulk modulus must be positive).
Reasonable defaults are set for unspecied parameters (e.g. associativity is imposed unless
the user explicitly requests non-associativity).
Portability. The BF1 model conforms to the Model Interface Guidelines [4], which has
allowed it to be implemented in ve different host codes, all using identical source code.
An installation of the model at LLNL took only one day of the BF1s developers time.
Moreover, this broad use of the model has exercised it on numerous computing platforms
using multiple compilers, thereby correcting minor numerical issues such as convergence
criteria, along the way.
The above list of software SQA practices is perhaps what distinguishes the BF1 model from
other models more than any other metric. Rising up to these SQA standards might lead to apparent
delays in the implementation of new model features but it is believed that they ultimately reduce
the total time required to achieve trustworthy, physically meaningful, simulations.
It should be noted that similar to the BF1 model, the CSCM model provides extensive docu-
mentation that gives an overview of the theory, as well as examples of how to used the model in
LS-DYNA. The K&C model has documentation that provide such information, but the detailed
documentation is available only for the Release II of the model. The K&C Concrete Release III
is the latest release, which has a user guide that focuses on helping users getting started. No
theoretical details are given with this current release.
Validation
As parts of commercial packages, the K&C and RHT models have been used and validated by
many authors. This chapter presents and summarizes the ndings from some of these works.
68
K&C: drop weight tests
In drop weight tests performed at FOI (FOA), the K&C models ability to predict the material and
structural response was evaluated [17]. In the tests, four reinforced concrete beams were subjected
to impact loading using a drop weight. Hansson et al. [17], showed that the simulation was able to
replicate the cracks in mode I, similar to the test results, but also predicted a mode II crack from
the center point of impact that was not observed in the experiment.
K&C: simulations of penetration and perforation of high performance con-
crete
Simulations of penetration and perforation of three types of high performance concrete targets were
performed to assess the K&C models ability to predict depth of penetration or residual velocity. A
6.3kg armor piercing steel projectile with an ogive nose radius of 127mm, length of 225mm, and
diameter of 75mmwas red at the targets at zero impact angle and 620m/s velocity. The targets had
a diameter of 1400mm. The targets lengths of 400mm and 800mm were used for the perforation
and penetration tests, respectively. Unosson [37] showed that the model yielded good agreement
with test data for perforation, but unsatisfactory results for penetration. Different element size was
not able to be used in one material denition, and the damage curve was tted only to one specic
element size. As a result, an incorrect fracture energy release was computed, resulting in excessive
amounts of shear damage in the element transition zone. When DIF (see Section 2) was used in
the model, its effect lasted only to half the targets depth, beyond which changes in DIF had no
effect.
K&C: explosive wall breaching
Double-reinforced concrete slabs having dimensions of 2.42.40.2m were subjected to the ex-
plosive loading environments produced by contact detonation charges. Akers et al. [2], showed
in their work that the K&C model predicted a hole in the slabs close to the experimental value.
However, the damage in the slabs was found to be excessive. No detailed explanations regarding
the incorrect damage prediction were given in [2]. However, the tendency of this class of model to
over-predict damage is well known and believed by some researchers [5] to be attributable to de-
terminism (i.e. lack of statistical variability) that prevents localized failure in the form of discrete
crack networks. Incorporating both scale effects statistical variability in properties dramatically
reduces the phenomenon of excessive damage.
K&C: vehicle-barrier crash test simulations
Numerical simulations for triaxial loading were conducted on plain concrete and compared to the
experimental data. The calibrated K&C parameters were used to simulate a vehicle impact/crash
69
into a concrete barrier. Under triaxial tests, the parameters were tuned until the model showed
correct linear and nonlinear response of the concrete in comparison to the experimental data [39].
Based on these well tuned parameters, a vehicle-barrier crash test was simulated. Yonten et al.
[39] presented the comparison of the simulations using various concrete models available within
the LS-DYNA program; each displayed different prediction from one another. No experimental
data were used in assessing the vehicle-barrier crash test simulation results.
RHT: simulation of penetration of high performance concrete
Simulations of a steel projectile penetrating a concrete target were presented and compared to
experimental results in [18]. Hansson et al. [18] showed the inuence of impact geometry on
the calculated penetration path. For instance, a very small yaw angle had a large effect on the
penetration path and exit velocity of the projectile in the simulation. The input data were tuned
to match the penetration experiments but there was no mention of whether or not the alteration
of parameters corrupted agreement with the original data on which the original parameters were
based. It was also shown that the failure surfaces are dependent on pressure, strain rate, and triaxial
stress state, and hence could handle most cases that are of importance for projectile penetration
into brittle material like concrete. In addition, Hansson et al. [18] pointed out the difculty of
assessing the parameters that describe damage evolution, residual strength, and the tensile behavior
and that newtesting methods to determine damage development in concrete and strength of damage
or partly damaged concrete still need to be developed. Some progress on measurement of damaged
strength has been made by Chen et al. [7]. The BF1 model acknowledges this lack of data by
defaulting an unspecied residual yield surface to that of sand.
RHT: concrete subjected to projectile and fragment impacts
Lepp anen [26] showed a simulation using the RHT model where a single fragment was shot against
a concrete wall with impact velocities of 1024 m/s, 1163 m/s and 1238 m/s. At 1024 and 1163
m/s impact velocity, the calculation did not show the spalling evident in the experimental data, but
scabbing behavior was correctly described. On the other hand, at 1238 m/s impact velocity, both
spalling and scabbing behavior were correctly predicted.
RHT: jumbo jet impacting on thick concrete walls
The RHT model was used by Katayama et al. [24] to simulate impact of a Boeing 747 passenger
jet into a concrete wall. This three dimensional simulation was assessed for its ability to predict
observed perforation, non-perforation, and spalling of the concrete. The model was able to predict
the qualitative response of the concrete wall, but quantitative accuracy was not discussed in detail.
These are a few examples of existing simulations performed using the K&C and RHT concrete
models. Both models appear to be very promising despite some weaknesses and existing aws.
70
With parameter tuning, they seem to be able to capture the qualitative response of the concrete
under impact but fail to accurately capture the damage in detail. Both models rely heavily on the
data extracted from correct/meaningful experiments, which are not often available.
According to works cited here, the K&C model appears to work well in predicting perforation
but not penetration. The excessive damage from penetration is largely caused by the element size
dependency according to [37]. However, several improvements have been made since the model
(LS-DYNA material 72) was evaluated, and a scaling factor that adjusts the fracture energy release
depending on the element size has been included in the newer version (material 72R3). The K&C
model includes tables and functions that are sophisticated enough to allow very accurate tting
to standard materials characterization experiments. However, predictiveness in general loading is
still lacking.
For the RHT model, many benchmarks were performed to provide information for the models
basic parameters. However, these data are still not sufcient for the model to accurately capture
the response at various strain rates. Similar to the K&C model, penetration is inuenced heavily
by the empirical data. While none of the works using the K&C model mentioned above discussed
the effect of spalling and scabbing, simulations performed using the RHT model showed that these
effects can be predicted to some degree.
71
72
Chapter 7
Conclusions
Four conventional damage plasticity models for concrete, K&C, RHT, BF1, and CSCM are com-
pared. All four models use very similar elastic and inelastic computational procedures. All of the
models support nonlinear elasticity, but with different formulations. All four models employ three
strength surfaces: (1) yield surface, (2) limit surface, and (3) residual surface. Differences are seen
in mathematical expressions dening the proles of these strength surfaces.
The four models being compared differ in their softening evolution equations, as well as in
their equations used to degrade the elastic stiffness. For all four models, scale effects are modeled
differently. The K&C model requires that a particular material parameter affecting the damage
evolution rate must be set by the user according to the mesh size to preserve energy to failure.
Similarly, the BF1 model presumes that all material parameters are set to values appropriate to
the scale of the element, and automated assignment of scale-appropriate values is available only
through an enhanced implementation of BF1 (called BFS) that regards scale effects to be coupled to
statistical variability of material properties. The RHT model appears to similarly support optional
uncertainty and automated settings for scale dependent material parameters.
K&C and RHT support rate dependence by allowing the strength to be a function of strain rate,
whereas the BF1 and CSCM models use Duvaut-Lion viscoplasticity theory to give a smoother
prediction of transient effects. During softening, all four models require a certain amount of strain
to develop before allowing signicant damage accumulation. For K&C, RHT, and CSCM, the
strain-to-failure is tied to fracture energy release, whereas a similar effect is achieved indirectly in
the BF1 model by a time-based criterion that is tied to crack propagation speed.
All models investigated in this report rely heavily on well tuned parameters to reproduce ob-
served material response. The BFS enhancement to the BF1 model automates the assignment of
parameters based on experimentally justied scaling effects. There appears to be no barrier to
adopting the best features of any of these models into one another. Therefore, any of these models
could be used as a starting point to rectify deciencies of smeared damage models and to then
further enhance the physics of damage.
73
74
References
[1] Shock Wave Science and Technology Reference Library, chapter Elements of Phenomenolog-
ical Plasticity: Geometrical Insight, Computational Algorithms, and Topics in Shock Physics.
Springer Berlin, 2007.
[2] S. Akers, R. Weed, D. Rickman, and Danielson K. Numerical simulations of explosive wall
breaching. In Proceedings of the Users Group Conference, 2005.
[3] N. Bicanic and C. J. Pearce. Computational aspects of a softening plasticity model for plain
concrete. Mechanics of Cohesive-Frictional Materials, 1:7594, 1998.
[4] R. M. Brannon and Wong M. K. MIG version 0.0 Model Interface Guidelines: Rules to Ac-
celerate Installation of Numerical Models Into Any Compliant Parent Code. Sandia National
Laboratories Report SAND96-2000, Albuquerque, 1996.
[5] R. M. Brannon and J. M. Wells. Validating theories for brittle damage. Metallurgical and
Materials Transactions A, 2007.
[6] Century Dynamics Ltd., Horsham. AUTODYN: Theory Manual, 4.3 edition, 2000.
[7] W. Chen and H. Luo. Dynamic compressive responses of intact and damaged ceramics from
a single split hopkinson pressure bar experiment. Experimental Mechanics, 44(3):295299,
2004.
[8] W. F. Chen. Plasticity in Reinforced Concrete. McGraw Hill, New York, 1982.
[9] J. K. Dienes and Q. H. Zuo. Impact initiation of explosives and propellants via statistical
crack mechanics. Journal of Mechanics and Physics of Solids, 54(10):22352240, 2006.
[10] G. Duvaut and Lions J. L. Les inequations en mecanique et en physique. Springer-Verlag,
Dunod, Paris, 1976. English transl.
[11] Federal Highway Administration, USA. User Manual for LS-DYNA Concrete Material, May
2007.
[12] A. F. Fossumand R. M. Brannon. The SANDIA GEOMODEL Theory and Users Guide. San-
dia National Laboratories Report SAND2004-3226, Albuquerque, unlimited release edition,
August 2004.
[13] A. F. Fossum and R. M. Brannon. Unied compaction/dilation, strain-rate sensitive, con-
stitutive model for rock mechanics structural analysis applications. ARMA/NARMS, pages
04546, 2004.
75
[14] A. F. Fossum and Brannon R. M. On a viscoplastic model for rocks with mechanism-
dependent characteristic times. Acta Geotechnica, 1(2):89106, September 2006.
[15] G. Gudehus. Elastoplastische stoffgleichungen fr trockenen sand. Ingenieur Arch, 42:151
169, 1973.
[16] K. Gylltoft. Fracture mechanics models for fatigue in concrete structures. PhD thesis,
Lule a University of Technology, 1983.
[17] H. Hansson and P. Skoglund. Structural Protection for Stationary/Mobile Tactical Behaviour
2001. Technical Report. FOI Swedish Defence Research Agency: Weapon and Protection,
Swedish Defence HQ, 2001.
[18] H. Hansson and P. Skoglund. Simulation of Concrete penetration in 2D and 3D with the RHT
Material Model. Technical Report. FOI Swedish Defence Research Agency: Weapon and
Protection, Swedish Defence HQ, 2002.
[19] A. Hillerborg. Analysis of fracture by means of the ctitious crack model, particularly for
bre reinforced concrete. The Int. J. Cement Composites, 2(4):177184, 1980.
[20] A. Hillerborg. The theoretical basis of a method to determine the fracture energy g
f
of
concrete. Rilem Technical Commttees, 18(106):291296, 1985.
[21] G. R. Johnson and T. J. Holmquist. An improved computational constitutive model for brittle
materials. In High-pressure science and technology, Myrtle Beach, SC, 1993.
[22] G. R. Johnson and T. J. Holmquist. Response of boron carbide subjected to large strains, high
strain rates, and high pressures. J. Applied Phys., 85:80608073, 1999.
[23] M. Kachanov. Anisotropic materials with interacting arbitrarily oriented cracks. In Proceed-
ing of the 14th U.S. Army Symposium on Solid Mechanics, Myrtle Beach, SC, 1996.
[24] M. Katayama, M. Itoh, and R. Rainsberger. Numerical simulation of jumbo jet impacting
on thick concrete walls - effects of reinforcement and wall thickness. In The Second Asian
Conference of High Pressure Research, Nara, Japan, 2004.
[25] H. Kupfer, H. Hilsdorf, and H. Rush. Behavior of concrete under biaxial stresses. ACI J.,
66:656666, 1969.
[26] J. Lepp anen. Concrete subjected to projectile and fragment impacts: Modelling of crack
softening and strain rate dependency in tension. Int. J. Impact Engng, 19(9-10):847873,
1997.
[27] Livermore Software Technology Corporation, USA. LS-DYNA: Keyword Users Manual,
May 2007.
[28] L. J. Malvar, J. E. Crawford, J. W. Wesevich, and D. Simons. A plasticity concrete material
model for dyna 3d. Int. J. Impact Engng, 19(9-10):847873, 1997.
76
[29] H. W. Meyers, Jr., and R. M. Brannon. A Model for Statistical Variation of Fracture Proper-
ties in a Continuum Mechanics Code. Army Research Laboratory, U.S.A., 2007.
[30] Y. D. Murray and B. A. Lewis. Numerical simulation of damage in concrete. Technical
report, The Defense Nuclear Agency by APTEK, 1995.
[31] W. Riedel. Beton unter dynamischen lasten, Meso- und makromechanische modelle und ihre
parameter. PhD thesis, EMI-Bericht 6/00, 2000.
[32] L. E. Schwer and Y. D. Murray. A three invariant smooth cap model with mixed hardening.
International Journal for Numerical and Analytical Methods in Geomechanics, 18:657688,
1994.
[33] P. E. Senseny, A. F. Fossum, and T. W. Pfeie. Non-associative constitutive laws for low
porosity rocks. Int J Num Anal Meth Geomech, 7:101115, 1983.
[34] J. C. Simo and J. W. Ju. Strain and stress based continuum damage model. Int J of Solids and
Structures, 23(7), 1987.
[35] Thomas Telford. CEB-FIB Model Code 1990. Lausanne, Switzerland, design code edition,
1993.
[36] M. Timmel, S. Kolling, P. Osterrieder, and P. A. Du Bois. A nite element model for impact
simulation with laminated glass. International Journal of Impact Engineering, 34:1465
1478, 2007.
[37] M. Unosson. Numerical simulation of penetration and perforation of high performance con-
crete with 75mm steel projectile. User Report. Defence Research Establishment: Weapons
and Protection Division, Tumba, Sweden, 2000.
[38] K. J. Willam and E. P. Warnke. Constitutive model for the triaxial behavior of concrete. In
ISMES Seminar on Concrete Structures Subjected to Triaxial Stresses, Bergamo, Italy, 1975.
[39] K. Yonten, M. Manzari, A. Eskandarian, and D. Marzougui. An evaluation of constitutive
models of concrete in ls-dyna nite element code. In 15th ASCE Engineering Mechanics
Conference, New York, NY, 2002.
77
DISTRIBUTION:
Mechanical Engineering Department
University of Utah
50 S. Central Campus Drive, 2134 MEB
Salt Lake City, UT 84112
Professor Rebecca M. Brannon, [email protected]
Seubpong Leelavanichkul, [email protected]
Lawrence Livermore National Laboratory
P.O. Box 808, M/S L-206
Livermore, CA 94551
Tarabay Antoun, Ph.D., [email protected]
Lawrence Livermore National Laboratory
P.O. Box 808, L-231
Livermore, CA 94551-9900
Joseph P. Morris, Ph.D., [email protected]
Los Alamos National Laboratory
W-13, MS A142
Los Alamos, NM 87545
Matt Lewis, Ph.D., [email protected]
Center for Advanced Metallic and Ceramic Systems
Latrobe 122
Johns Hopkins University
3400 North Charles St.
Baltimore, MD 21218
Professor K. T. Ramesh, [email protected]
Northwestern University
2145 Sheridan Road, CEE Rm A135
Evanston, Illinois 60208-3109
Professor Zdenek P. Bazant, [email protected]
78
University of Colorado
Department of Civil, Environmental, and Architectural Engineering
1111 Engineering Dr.
428 UCB, ECOT 441
Boulder, CO 80309-0428
Professor Richard Regueiro, [email protected]
Sandia Internal:
Sandia National Laboratories
P. O. Box 5800
Albuquerque, NM 87185
Computational Shock & Multiphysics
MS 0378
R. M. Summers, [email protected]
O. E. Strack, [email protected]
Computational Solid Mechanics & Structural Dynamics
MS 0380
J. Jung, [email protected]
M. W. Heinstein, [email protected]
B. W. Spencer, [email protected]
Geomechanics
MS 0751
S. J. Bauer, [email protected]
T. Dewers, [email protected]
S. R. Sobolik, [email protected]
Solid Mechanics
MS 0372
J. V. Cox, [email protected]
H. E. Fang, [email protected]
K. W. Gwinn, [email protected]
K. E. Metzinger, [email protected]
J. T. Ostien, [email protected]
W. M. Scherzinger, [email protected]
Strategic Initiatives
MS 0372
J. Guadalupe Arguello, [email protected]
J. E. Bean, [email protected]
J. E. Bishop, [email protected]
79
A. S. Gullerud, [email protected]
J. F. Holland, [email protected]
J. S. Rath, [email protected]
J. M. Redmond, [email protected]
C. M. Stone, [email protected]
Mechanical Environments
MS 0847
S. W. Attaway, [email protected]
Mechanics of Materials
MS 9042
M. L. Chiesa, [email protected]
National Security Applications
MS 0751
L. S. Costin, [email protected]
Penetration Systems
MS 1160
D. A. Dederman, [email protected]
V. K. Luk, [email protected]
Performance Assessment & Decision Analysis
MS 1395
M. Y. Lee, [email protected]
Structural Integrity & Licensing
MS 0744
M. F. Hessheimer, [email protected]
Surety & GTS
MS 9042
E. P. Chen, [email protected]
Technical Library, 9536 (electronic copy)
MS 0899
80
v1.31

Survey of Four Damage Models For Concrete

Uploaded by

Copyright:

Available Formats

Survey of Four Damage Models For Concrete

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Survey of Four Damage Models For Concrete

Uploaded by

Copyright:

Available Formats

SANDIA REPORT

, ) in Fig. 1.1, by a scale function (, I

=/6+, (0.5 2),

be the strength corre-

increases half way to unload/reload value when

is a kinematic hardening tensor, each of which

() is a scalar-valued decay function designed to limit the

0 as second invariant of approaches a user-specied maxi-

. Such spurious behavior is typical of deterministic

You might also like