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

net/publication/327595782

An efficient and robust VUMAT implementation of elastoplastic constitutive

laws in Abaqus/Explicit finite element code

Article  in  Mechanics and Industry · January 2018

DOI: 10.1051/meca/2018021

CITATIONS READS

5 1,204

2 authors, including:

Olivier Pantalé
Ecole Nationale d'Ingénieurs de Tarbes
47 PUBLICATIONS   560 CITATIONS   

SEE PROFILE

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

Habilitation a diriger des recherches View project

Surfaces Properties View project

All content following this page was uploaded by Olivier Pantalé on 04 December 2018.

The user has requested enhancement of the downloaded file.

 An efficient and robust VUMAT implementation of
elastoplastic constitutive laws in Abaqus/Explicit finite
element code
Lu Ming, Olivier Pantalé

To cite this version:

Lu Ming, Olivier Pantalé. An efficient and robust VUMAT implementation of elastoplastic constitutive
laws in Abaqus/Explicit finite element code. Mechanics

Industry, EDP Sciences, 2018, 19 (3), pp.308. <10.1051/meca/2018021>. <hal-01905414>

HAL Id: hal-01905414

https://hal.archives-ouvertes.fr/hal-01905414
Submitted on 25 Oct 2018

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents
entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non,
lished or not. The documents may come from émanant des établissements d’enseignement et de
teaching and research institutions in France or recherche français ou étrangers, des laboratoires
abroad, or from public or private research centers. publics ou privés.
 Open Archive Toulouse Archive Ouverte

OATAO is an open access repository that collects the work of Toulouse

researchers and makes it freely available over the web where possible

This is an author’s version published in: http://oatao.univ-toulouse.fr/20809

Official URL: http://doi.org/10.1051/meca/2018021

To cite this version:

Ming, Lu and Pantalé, Olivier An efficient and robust VUMAT

implementation of elastoplastic constitutive laws in Abaqus/Explicit
finite element code. (2018) Mechanics & Industry, 19 (3). 308. ISSN
2257-7777

Any correspondence concerning this service should be sent

to the repository administrator: [email protected]
 An efficient and robust VUMAT implementation of elastoplastic
constitutive laws in Abaqus/Explicit finite element code

Lu Ming, Olivier Pantalé∗

Laboratoire Génie de Production, Université de Toulouse, INP-ENIT, Toulouse, France

Abstract
This paper describes the development of an efficient and robust numerical algorithm for the im-
plementation of elastoplastic constitutive laws in the commercial nonlinear finite element soft-
ware Abaqus/Explicit through a VUMAT FORTRAN subroutine. In the present paper, while
the Abaqus/Explicit uses an explicit time integration scheme, the implicit radial return mapping
algorithm is used to compute the plastic strain, the plastic strain rate and the temperature at the
end of each increment instead of the widely used forward Euler approach. This more complex
process allows us to obtain more precise results with only a slight increase of the total computa-
tional time. Corrector term of the radial return scheme is obtained through the implementation
of a safe and robust Newton-Raphson algorithm able to converge even when the piecewise de-
fined hardening curve is not derivable everywhere. The complete method of how to implement a
user-defined elastoplastic material model using the radial return mapping integration scheme is
presented in details with the application to the widely used Johnson-Cook constitutive law. Five
benchmark tests including one element tests, necking of a circular bar and 2D and 3D Taylor im-
pact tests show the efficiency and robustness of the proposed algorithm and confirm the improved
efficiency in terms of precision, stability and solution CPU time. Finally, three alternative con-
stitutive laws (the TANH, modified TANH and Bäker laws) are presented, implemented through
our VUMAT routine and tested.
Keywords: Radial return algorithms, Elastoplasticity, Finite element, Strain-dependent,
Thermomechanical

1 Introduction
Over the last past years, numerous studies have been concentrated on dynamic mechanical
properties of materials. Constitutive laws, the relation between internal forces and deformations,
have also received much attention. Under large deformations and high deformation rates such as
forming, machining processes and impact analyses, elastoplastic constitutive laws play a signifi-
cant role in predicting the mechanical behavior of materials.
Although many kinds of constitutive laws have been implemented into the commercial non-
linear finite element software Abaqus [1], in some conditions they cannot fit well some complex

∗ Correspondingauthor
Email address: ❖❧✐✈✐❡%✳'❛♥*❛❧❡❅❡♥✐*✳❢% (Olivier Pantalé)
URL: ❤**♣✿✴✴✇✇✇✳❡♥✐*✳❢% (Olivier Pantalé)
behaviors of materials such as the deformation process involving large strain, high strain rates,
temperature softening, etc... Therefore, Abaqus provides the ability for users to implement only
the constitutive flow, or the complete constitutive behavior law by themselves via FORTRAN
user subroutines: UHARD or VUHARD for the hardening flow law and UMAT or VUMAT for
the complete behavior law. While the user hardening implementation is quite simple and easy,
the user material implementation needs more expertise. The major difference from VUMAT to
UMAT is that VUMAT subroutines do not need the evaluation of the so-called consistent Jaco-
bian matrix, but only the evaluation of the stress tensor σ1 at the end of the increment. The usual
approach used to write a VUMAT subroutine, because the time increment is very small due to the
explicit time integration scheme, is to use a straightforward evaluation of the equivalent plastic
strain increment ∆ε p without requiring any local iterative algorithm [2, 3].
In this paper a different approach has been chosen and consist in evaluating, in the VUMAT
subroutine, the plastic strain increment ∆ε p using the implicit radial return algorithm presented
in section 2. In order to illustrate the proposed approach and to be able to validate it, the Johnson-
Cook constitutive flow law [4], presented in section 3, has been selected to be implemented into
the Abaqus/Explicit code through both a VUHARD and a VUMAT subroutines. The complete
implementation of the Johnson-Cook elastoplastic constitutive law through a VUMAT subrou-
tine is presented in details in section 4. Numerical efficiency and precision of the proposed
implementation are finally investigated, in section 5, through some dynamic benchmarks tests
and have shown the efficiency and robustness of the proposed integration scheme. Section 5 also
present some alternative constitutive laws results for the Taylor impact test.

2 Time integration algorithm of J 2 plasticity

Time integration algorithms based on finite difference methods are widely used in finite el-
ement solutions of mechanics. In these algorithms, time is discretized on a finite grid, and the
distance between consecutive points on the grid is defined as the time step ∆t. With the positions
and some of the time derivatives at time t, the integration scheme calculates the same quantities
at a later time t +∆t. Through the iteration of the procedure, the time evolution of those quantities
can be traced.

2.1 J2 plasticity model

The elastic stress-strain equation is usually written in the following incremental form:
∇
σ= H : De (1)
∇
where σ is an objective stress rate in order to take into account the objectivity in large defor-
mations, : is the double dot product, H is the linear isotropic fourth order elastic tensor given
by:
1
H = K 1 ⊗ 1 − 2G(I − 1 ⊗ 1) (2)
3
with 1 the second order identity tensor, I the fourth order identity tensor, ⊗ the Dyadic product,
G the shear modulus and K the Bulk modulus linked to the Young’s modulus E and Poisson’s
ratio ν by the following:
E E
G= and K = (3)
2(1 + ν) 3(1 − 2ν)
2.1 J2 plasticity model
e
and D is the so-called elastic part of the rate of deformation tensor D generally assumed [5] for
hypo-elastic material to conform to the following additive decomposition equation:
D = De + D p (4)
p
where D is the plastic part of the rate of deformation tensor D. According to the literature, J2
plasticity, also called von Mises yield criterion, was proposed by Huber in 1904 and von Mises
in 1913 [6]. It is widely applied in mechanical engineering and metal forming. It assumes the
existence of a scalar yield function f given by:
q
f = σ − σy (ε p , ε p , T ) = 0 (5)
where σ is the von Mises equivalent stress, or effective stress, defined from the deviatoric stress
s = σ − 13 tr [σ]1 as: r
3
σ= s:s (6)
2
and σy is the so-called current yield stress of the material, depending in our kind of applications
q
on the the equivalent plastic strain ε p , the equivalent plastic strain rate ε p and the temperature T
given by:
ˆ t q ˆ tr
2 p
p
ε = p
ε dt = D : D p dt (7)
0 0 3
η
σ : Dp
q
T= (8)
ρC p
where η is the Taylor-Quinney [7] coefficient defining the amount of plastic work converted into
heat energy, C p is the specific heat coefficient and ρ is the density of the material. The so-called
yield function f defines the yield surface (when f = 0), the elastic domain (when f < 0) and the
no-access domain (when f > 0). Assuming an associative plastic flow rule, the plastic strain rate
D p can be expressed with the following equation:
D p = γn (9)
where γ is a scalar representing the flow intensity and n is a second order tensor (the unit normal
to the flow stress determined exclusively in terms of the trial elastic stress [8]) representing the
flow direction given by:
n= √
s (10)
s:s
From equation (7) one can easily obtain:
r r ˆ t r
q
p 2 p 2 2
ε = γ and ε = γdt = Γ (11)
3 3 0 3
The plasticity model described here above has to be integrated in time with respect to an in-
cremental objective algorithm. Different expressions of the time derivative of the Cauchy stress
∇
tensor σ used in equation (1) can be used, leading to some noticeable differences in the final
results. Abaqus/Explicit for example in case of solid elements uses a Jaumann stress rate for all
Built-In constitutive models and a Green-Naghdi stress rate in case of VUMAT user subroutines
[1]. This can lead to some difficulties when one wants to compare VUMAT results with Built-In
models as we will see further.
2.2 Radial return mapping algorithm

2.2 Radial return mapping algorithm

A widely used method for the time integration of J2 plasticity with isotropic hardening is
the return mapping algorithm introduced by Wilkins [9] and Maenchen and Sack [10]. This
algorithm is now extensively used and detailed in numerous books such as [6, 11] or papers like
Simo et al. [8] or Ponthot [12]. We assume that, thanks to the Finite Element formulation, the
strain increment between time t0 (the last equilibrated configuration) and t1 (the new equilibrated
configuration with t1 = t0 + ∆t) is ∆E and the deviatoric stress tensor at time t0 is s0 . The trial
deviatoric part of the stress tensor str and the final hydrostatic pressure p1 are calculated from
the strain increment ∆E assuming that the whole step is fully elastic in a first time, so that:

str = s0 + 2G dev ∆E
h i
(12)

and: h i
p1 = p0 + K tr ∆E (13)
Conforming to equation (5), at time t1 , the yield function f is therefore:
q
f = σtr − σy0 (ε0p , ε0p , T 0 ) (14)

with: r
3
σtr = str : str (15)
2
and σy0 the yield stress of the last equilibrated configuration at the beginning of the current incre-
ment (t = t0 ). In order to know if the predicted trial stress str is admissible or not, a test has to
be performed on the sign of the yield function f leading to the two options here after:

• If f ≤ 0, the whole step is fully elastic, which means that the predicted stress is admissible
and we conclude that s1 = str .

• If f > 0, the trial stress is not admissible and a plastic correction has to be performed in
order to compute the final value of s1 .

The plastic correction is computed enforcing the (discrete) generalized consistency parameter
f (Γ) = 0 at the end of the current increment (so that time t = t1 ):
r
3 q
f (Γ) = s1 : s1 − σy1 (ε1p , ε1p , T1 ) = 0 (16)
2
with: 




s1 = str −q2GΓn
ε p = ε0p + 23 Γ



 q1



p 1
q
2 (17)
ε = 3Γ





 1 ∆t
 q
 T1 = T0 + η y y
 2
3 (σ0 + σ1 )Γ


2ρC p

Combination of equations (16) and (17) leads to the following final form of the consistency
parameter f (Γ): √
f (Γ) = σtr − 6GΓ − σy1 (Γ) = 0 (18)
2.3 Root-finding of the non-linear equation

where the only unknown value is the scalar Γ, already defined in equation (11). If one assume
here a linear hardening of the material during the time-step, as usually done thanks to the explicit
integration scheme used in Abaqus/Explicit, the following analytical solution to equation (16)
can be written, as proposed for example by Gao et al. [2] or the Abaqus manual [3]:
r y
3 σtr − σ0
Γ= (19)
2 3G + h
with h = dσy0 /dε0p , the slope of the hardening law at the current point, assumed to be constant
during the time increment ∆t. This approach is usually adopted in VUMAT implementations
because of its simplicity as this later doesn’t require any iteration to solve the proposed problem.
Unfortunately, as it will be presented further, this lead to many instabilities because of the ap-
proximation proposed by this approach when the non-linear terms of the constitutive flow law
becomes important. In our approach, we have therefore chosen to solve equation (18) using an
approach similar to the one presented in Zaera et al. [13], the so called root finding methods used
to solve equation (11) will be presented in section 2.3.
Once the correct final value of the plastic corrector Γ has been obtained, the final values of the
deviatoric part of the stress tensor s1 and the updated internal variables are computed thanks to
equations (17). At the end of the proposed algorithm, the final stress at the end of the increment
is computed using the following equation:

σ1 = s1 + p1 1 (20)

2.3 Root-finding of the non-linear equation

Several root-finding methods can be used to find the numerical solution of equation (18). In
our kind of applications, after a correct selection of the bounds of the searching interval for the
solution, we know that the requested root is bracketed in a given interval. Once we know that the
proposed interval contains a root, several classical procedures are available to refine it. Among
the proposed methods, some of them are known to have fast convergence rate, precision and/or
robustness [14, 15]. Unfortunately, the methods that are guaranteed to converge are known to
be the slower ones, while those that exhibit a fast convergence rate can also dash rapidly to
infinity without warning if measures are not taken to avoid such behavior. Among the large
list of available methods (Chord, Bisection, Regula-Falsi, Ridders’, Newton-Raphson), only two
of them have finally been selected for their efficiency and robustness and are presented here
after: the bisection method, a slow and sure method and the Newton-Raphson method a fast but
sometimes failing method.

2.3.1 The bisection method

The first root-finding method presented in this paper is the so-called bisection method. This
method is one that cannot fail, but with a slow rate of convergence. It’s an iterative root-finding
method where the predicted interval of the root is successively halved until it becomes sufficiently
small, which is also called interval halving method. This process is repeated until the interval is
small enough, which satisfies:
|xi+1 − xi | ≤ εbis (21)
where εbis is the error tolerance of the bisection method. Assuming the root to be found is ini-
tially bracketed in an interval [x0 , x1 ] satisfying f (x0 ) f (x1 ) < 0, the method converges linearly,
which is comparatively slow, but the main advantage of this method is that it only requires the
2.3 Root-finding of the non-linear equation

evaluation of the function f (x) itself and not its derivative f ′ (x) as in other methods such as the
Newton-Raphson. Therefore, this method can become competitive when the computation of the
derivative of the function f (x) becomes long to compute; i.e. when the expression of f (x) is
complex.

2.3.2 The safe version of Newton-Raphson method

The Newton-Raphson method [14, 15] is the best known method of finding roots because of
its simplicity and efficiency (very high rate of convergence). The only apparent drawback of this
method is that it requires the evaluation of the derivative f ′ (x) of the function f (x), so it’s only
usable in problems where f ′ (x) can be readily computed, or numerically evaluated as we will see
further. From Taylor series expansion of f (x) about x, we obtain the following expression:
f (xi )
xi+1 = xi − (22)
f ′ (xi )
The Newton-Raphson is an iterative process starting with a first guess x0 for a root of the function
f (x) and the process is repeated until the convergence criterion given by:

|xi+1 − xi | ≤ εNR (23)

where εNR is the error tolerance of the Newton-Raphson method, is reached.

In some situations, the Newton-Raphson method appears to have poor global convergence,
because the tangent line is not always an acceptable approximation of the function. But more
important, when f (x) is a piecewise function defined by two different expressions on the left and
right side of a given point x s , f (x) will not be differentiable at x s , leading to numerical difficulties
and divergence of the Newton-Raphson algorithm when the seeking solution xi crosses the point
x s . As a result, the so-called safe version of Newton-Raphson, described in Figure 1, resulting
from a combination of the Newton-Raphson and the bisection methods is proposed here after.
If there is a root in the interval [x0 , x1 ] satisfying f (x0 ) f (x1 ) < 0, the safe version of Newton-
Raphson method regards the midpoint of [x0 , x1 ] as the first guess of the root and the Newton-
Raphson iteration starts. After each iteration, the interval is updated by replacing one of the two
boundaries with the new solution. If the iteration is out of the interval, it is disregarded and
replaced with a bisection step to re-position the initial guess. The Newton-Raphson algorithm
requires to calculate the derivative f ′ (Γ) of the yield function f (Γ) with respect to the Γ parameter
when solving equation (18), so that:
√ dσy (Γ)
f ′ (Γ) = − 6G − (24)
dΓ
with the following definition for the derivative of the yield function:
q
dσy (Γ) ∂σy dε p ∂σy d ε p ∂σy dT
= + +
dΓ ∂ε p dΓ dΓ ∂T dΓ
q
∂ εp
r  
2  ∂σy 1 ∂σy ησy ∂σy 
= + q + (25)
3  ∂ε p ∆t
 
ρC p ∂T 
∂ εp
A common method to calculate the derivatives of the yield function σy with respect to ε p ,
q
ε p and T is to use an analytical method, which is to calculate the analytical expression for every
2.3 Root-finding of the non-linear equation

get f (x), f ′ (x), x0 , x1

1
x = 2
(x0 + x1 )

(
x1 = x if f (x). f (x0 ) < 0
x0 = x if f (x). f (x0 ) > 0
′
∆x = − f (x)/ f (x)
x ← x + ∆x

Yes
x ∈ [x0 , x1 ]

No
1
∆x = (x1 − x0 )
2
x = x0 + ∆x

No
|∆x| < εNR

Yes

Return x

Figure 1: Flowchart of the Safe-NR algorithm

partial derivative based on the hardening flow law of the material. However, for most yield
functions, it may be difficult to obtain derivatives using the analytical method. Therefore, in
such cases, we propose here after a numerical solution as an alternative. This numerical solution
is obtained by adding a small increment to the equivalent plastic strain, plastic strain rate and
y y y
temperature respectively in order to calculate the three partial derivatives ∂σ
∂ε p
, ∂σqp and ∂σ
∂T in
∂ε
equation (25):
q q
∂σy σy (ε p + ∆ε p , ε p , T ) − σy (ε p , ε p , T )
= (26)
∂ε p ∆ε p
q q q
∂σy σy (ε p , ε p +∆ ε p , T ) − σy (ε p , ε p , T )
q = q (27)
∂ εp ∆ εp
q q
∂σy σy (ε p , ε p , T + ∆T ) − σy (ε p , ε p , T )
= (28)
∂T ∆T
q
Of course, accurate results depends on a correct choice for the three increments ∆ε p , ∆ ε p
and ∆T . In all following numerical tests, the three increments have been arbitrary fixed to the
same value ∆x = 10−6 .
3 Hardening flow law

3.1 The Johnson-Cook hardening flow law

In this paper, the Johnson-Cook hardening flow law [4, 16] has been selected to be imple-
mented in Abaqus/Explicit through VUMAT and VUHARD subroutines. Although the proposed
approach can be applied to implement other elastoplastic constitutive laws, the Johnson-Cook
law is the best choice for validating the proposed approach, because it has already been imple-
mented in Abaqus/Explicit code, which means the benchmark tests can be conducted between
the Abaqus Built-In constitutive law and our FORTRAN implementations.
The Johnson-Cook hardening flow law is probably the most widely used flow law for the
simulation of high strain rate deformation processes taking into account plastic strain, plastic
strain rate and temperature effects. Since a lot of efforts have been made in the past to identify
the constitutive flow law parameters for many materials, it is implemented in numerous Finite El-
q
ement codes such as Abaqus [1]. The general formulation σy (ε p , ε p , T ) is given by the following
equation:
  q  " !m #
   n
 ε p  T − T0
y p 
σ = A + Bε 1 + C ln  q  1 −
 
 
 
 (29)
ε0 Tm − T0
q
where ε0 is the reference strain rate, T 0 and T m are the reference temperature and the melting
temperature of the material respectively and A, B, C, n and m are the five constitutive flow law
parameters. The multiplicative formulation of this flow law allows for the following form:
 n
q
σy = A + Bε p σyq(ε p )σyT (T ) (30)
ε
q
where, according to [4, 13, 17], the dependence on the plastic strain rate σyq(ε p ) is only taken into
q q ε
account if ε p ≥ε0 :  q qp ! q q
 y p ε p
σ ( ε ) = 1 + C ln if ε ≥ε0




 ε q q

 q ε0
q q (31)
 σyq(ε p ) = 1 if ε p <ε0



ε

and the dependence on temperature σyT (T ) is defined so that, if T < T 0 there is no temperature
dependence of the yield stress and if T ≥ T m the material is assumed to behave like liquid:
 y  m



 σT (T ) = 1 − TTm−T
−T
0
0
if T 0 ≤ T ≤ T m
 y

 σT (T ) = 1 if T < T 0 (32)
 σy (T ) = 0


if T ≥ T m
T

Those two conditions defined by equations (31) and (32) lead to some discontinuities in the
q q
hardening relation σy (ε p , ε p , T ), its derivative h(ε p , ε p , T ) and the yield function itself resulting
in numerical difficulties in the iterative solving procedure. The yield function is therefore not
q
differentiable at ε0 and T 0 . Nevertheless, the robust Newton-Raphson procedure proposed in
section 2.3.2 has been found sufficiently robust to overcome those problems. Analytical forms
3.2 VUHARD implementation in Abaqus/Explicit
q
for the derivatives of the Johnson-Cook flow law σy with respect to ε p , ε p and T are given by the
three following equations:
∂σy pn−1 y p
q
= nBε σ ( ε )σyT (T ) (33)
∂ε p
q
ε
 y q q
pn
  y p
∂σ C



 ∂ε pq = q A + Bε σ T (T ) if ε ≥ε0
εp

 y
q q (34)
 ∂σqp = 0 if ε p <ε0



∂ε
  n
−m A+Bε p m
y
q 
y p
 ∂σ ) TTm−T

= σ ( ε 0
if T ∈ [T 0 , T m ]


∂T T −T 0 ε
q −T 0 (35)
 ∂σy = 0



∂T if T < [T 0 , T m ]
n−1
The last problem usually encountered is that, because of the term nBε p in equation (33), h
tends to infinity when the first plastic increment occurs (i.e. when ε p = 0). Therefore, we will
have to apply a special treatment in the hardening parameter computation on the first plastic step.
This will be presented further in section 4.3.

3.2 VUHARD implementation in Abaqus/Explicit

The VUHARD subroutine is a straightforward approach to implement a new constitutive flow
law in Abaqus/Explicit by just implementing a FORTRAN subroutine to compute the yield stress
q q
of the material σy (ε p , ε p , T ) and its derivatives with respect to ε p , ε p and T . The main part of
the Built-In constitutive law is used for time integration of the stress, for a given time increment,
and the provided user subroutine is used to compute the hardening flow law. Very few details are
given about this implementation in the Abaqus documentation, but some useful informations are
available in Jansen van Rensburg et al. [18].
The numerical implementation of the VUHARD subroutine for the Johnson-Cook flow has
q
been done through a FORTRAN program defining the hardening flow law σy (ε p , ε p , T ) according
q
to equation (29) and the three analytical derivatives of σy with respect to ε p , ε p and T defined
by equations (33-35). This part of the code is exactly the same as the one that will be used in the
VUMAT subroutine presented in the next section of this paper.

4 VUMAT implementation in Abaqus/Explicit

All details concerning the implementation of the radial return mapping algorithm using a
Fortran VUMAT subroutine for the Johnson-Cook flow law are detailed in this section. The
detailed flowchart algorithm is illustrated in Figure 2. The first block of the proposed algorithm
“Start of VUMAT” is used to get the material properties defined as user material constants.
Abaqus/Explicit provides the user subroutine VUMAT the quantities defined here after:
• The strain increment ∆E for the current time step,
• The stress tensor σ0 and the temperature T 0 at the beginning of the current increment,
• The time increment ∆t corresponding to the current time step,
• A table of solution dependent state variables (SDVs) used to store important data such as
q
ε p , ε p , Γ and transfer them from one increment to the other.
 Start of VUMAT

Abaqus package subroutine

Yes σ1 = σ0 +h i
t = 0
2G dev [∆E] + K tr ∆E 1

No

p0 = 1
3 tr [σ0 ]; s0
= σ0 − p0 1
h i
End of VUMAT
p1 = p0 + K tr ∆E
str = s0 +q2G dev ∆E
h i

2 str : str
3
σtr =

No
σtr > σy0

Yes
σ
Γ = Γ0 ; Γmin = 0; Γmax = √ tr
6G

q
ε1p = ε0p + 2
Γ
q q 3
p 1 2
ε1 = ∆t 3 Γ
T 1 = T 0 + √6ρCη
(σy0 + σy1 )Γ s1 = str
p
q σy1 = σy0
σy1 = σy1 (ε1p , ε1p , T 1 ) Γ = 0
√
f (Γ) = σtr − 6GΓ − σy1 (Γ)
′ √ dσy1 (Γ)
f (Γ) = − 6G − dΓ
′
∆Γ = − f (Γ)/ f (Γ)

(
Γmax = Γ if f (Γ) < 0
Γmin = Γ if f (Γ) > 0
Γ ← Γ + ∆Γ

Yes
Γ ∈ [Γmin , Γmax ]

No

∆Γ = 12 (Γmax − Γmin )
Γ = Γmin + ∆Γ

No
|∆Γ| < εNR

Yes

n = ksσtrk
s1 = str − 2GΓn

σ1 = s1 + p1 1
1
e1 = e0 + 2ρ (σ0 + σ1 ) : ∆E
ω1 = ω0p + √16ρ (σy0 + σy1 )Γ
p

End of VUMAT

Figure 2: Flowchart of VUMAT implementation

4.1 Abaqus package subroutine

The VUMAT subroutine must compute and return the value of the stress σ1 and the SDVs
variables at the end of the increment for each integration point. The internal and dissipated
energies have also to be evaluated in order to compute the temperatures in the model.

4.1 Abaqus package subroutine

The package part in the Abaqus software is a mandatory step used to compute the starting
values (compute the value of the time increment ∆t for example by taking into account the ma-
terial behaviour) for a reference time t = 0. For this we need to compute the elastic stress due to
a virtual strain increment ∆E provided by Abaqus using the following expression:
h i h i
σ1 = σ0 + 2G dev ∆E + K tr ∆E 1 (36)

The computed stress σ1 will be discarded after the end of the package step.

4.2 Elastic predictor

The aim of this subsection is to calculate the deviatoric part of the predicted elastic stress str
and the von Mises equivalent stress σtr and compare it with the yield stress at the beginning of
the increment σy0 in order to test if the current step is fully elastic or partly plastic. Therefore, we
need firstly to decompose the initial stress σ0 into its hydrostatic pressure (p0 ) and its deviatoric
(s0 ) parts:
p0 = 13 tr [σ0 ]
(
(37)
s0 = σ0 − p0 1
Then, we compute the new pressure (p1 ) and the deviatoric part of the trial stress tensor (str )
using:  h i
 p1 = p0 + K tr ∆Eh i


(38)
 str = s0 + 2G dev ∆E


q
and compare the yield stress at the beginning of the increment σy0 (ε0p , ε0p , T 0 ) to the von Mises
equivalent trial stress σtr (see equation (15) for this later). Depending on this test, we will then
correct or not the final stress.

• If σtr ≤ σy0 , the plastic corrector is zero, so the plastic correction steps can be skipped. We
assume that the predicted stress is the final one s1 = str , the plastic corrector Γ = 0, the
yield stress remains unchanged σy1 = σy0 and we can go directly to the final computations
in section 4.4.

• If σtr > σy0 , the step is at least partly plastic and the plastic corrector described in section
4.3 has to be computed in order to draw back the predicted stress onto the yield surface of
the material at the end of the current increment.

4.3 Plastic corrector

The safe version of the Newton-Raphson algorithm is used to recover the stress in accordance
with the elastoplastic constitutive behavior law. This Newton-Raphson method is used to com-
pute the Γ parameter defining the correction due to the increase of the strain. In order to enhance
the computations efficiency, we initialize the value of the Γ parameter to its value at the end of
the previous increment (Γ = Γ0 ). If the current increment is the first plastic increment (i.e. if
4.4 Final computations

ε0p = 0), as we cannot compute the value of h(Γ) when the plastic strain is zero, we initialize Γ
to an initial value different of zero (Γ = 10−8 ). The interval for the bisection part is initialized
h √ i q
so that Γ ∈ 0, σtr / 6G . The predicted equivalent plastic strain ε1p , plastic strain rate ε1p and
temperature T 1 at the end of the increment are computed thanks to the system of equations (17).
The yield stress σy1 , the yield function f (Γ) and its derivative f ′ (Γ) are then computed. Then, we
test the convergence of the Newton-Raphson algorithm by computing the increment ∆Γ of the Γ
parameter:
f (Γ)
∆Γ = − ′ (39)
f (Γ)
and comparing it to the Newton-Raphson precision εNR defined by the user:
• If ∆Γ > εNR we need to iterate to compute the correction of the Γ value using the following
relation:
Γ ← Γ + ∆Γ (40)
′
and re-evaluate the values of f (Γ) and f (Γ).
• If ∆Γ ≤ εNR we have obtained the final value of the Γ parameter.
The final deviatoric part of the stress tensor s1 is computed from the predicted value str and the
correction term Γ using:
s1 = str − 2GΓn (41)
with n, within the framework of the radial return algorithm, defined by:

n= p
str (42)
str : str
4.4 Final computations
The main work of the final computations is to update the state variables (SDVs) and the en-
ergies. The final stress tensor at the end of the increment σ1 is computed from the hydrostatic
pressure p1 and the deviatoric part of the stress tensor s1 thanks to equation (20). The equivalent
q
plastic strain ε1p , equivalent plastic strain rate ε1p and final temperature T 1 at the end of the incre-
ment are stored in their respective SDVs variables for subsequent re-use in the next increment.
We also have to compute the new specific internal energy e1 from:
1
e1 = e0 + (σ0 + σ1 ) : ∆E (43)
2ρ
and the dissipated inelastic energy ω1p from:
1
ω1p = ω0p + √ (σy0 + σy1 )Γ (44)
6ρ
At this point, the VUMAT subroutine comes to an end. At this point of the flowchart, the
final temperature is not computed, since the software uses a subsequent thermal step to evaluate
the temperature raise due to the dissipated inelastic energy and the conduction part. The internal
energy is modified during this thermal step and this seems to be taken into account during this
extra step (out of the VUMAT subroutine) since the tests on 1 element with imposed displacement
of all nodes gives the exact same results as the Build-in routine.
 Table 1: Material properties of the 42CrMo4 steel
E ν A B C
(Gpa) (MPa) (MPa)
206.9 0.29 806 614 0.0089
q
n m ε0 T0 Tm
(s−1 ) (°C) (°C)
0.168 1.1 1 20 1540
ρ λ Cp η
(kg/m3 ) (W/m°C) (J/Kg°C)
7830 34.0 460 0.9

5 Validation of the proposed implementations

In this section, the performance of the Johnson-Cook law programmed in the VUMAT sub-
routine is compared with the performance of the Abaqus/Explicit Built-In Johnson-Cook law and
of the VUHARD Johnson-Cook implementation, in order to validate the proposed implementa-
tion approach. The benchmark tests consist of two different one element tests (tensile test and
shear test), the necking of a circular bar and the well known Taylor impact test. The same mate-
rial, a 42CrMo4 steel, has been selected for all those tests, and material properties are reported in
Table 1. The proposed benchmarks are using the “Dynamic Temperature-displacement, Explicit”
procedure of Abaqus explicit FEM code. The inelastic heat fraction parameters has been set to its
default value of η = 0.9. This thermomechanical coupling option allows heat to be generated by
plastic dissipation or viscoelastic dissipation. No other thermal boundary conditions are applied
i.e. a globally adiabatic situation.
All benchmarks tests have been solved using Abaqus/Explicit v.6.14 on a Dell Precision
T7500 computer running Ubuntu 16.04 64bits with 12Gb of Ram and two 4 core E5620 Intel
Xeon Processors. All computations have been done using the double precision option of Abaqus,
with one CPU and the VUMAT and VUHARD subroutines have been compiled using the Intel
Fortran 64 v.14 compiler. The following models have been tested for each of the four presented
benchmarks:

• A-N-R model: VUMAT model with Newton-Raphson procedure and an analytical com-
putation of the derivatives using equations (33-35).

• N-N-R model: VUMAT model with Newton-Raphson procedure and a numerical compu-
tation of the derivatives using equations (26-28).

• Direct model: VUMAT model with a direct evaluation of Γ using equation (19) as pre-
sented in other papers [2].

• VUHARD model: Only the Johnson-Cook constitutive flow law defined by equation (29)
and its three analytical derivatives (33-35) have been implemented using a FORTRAN
subroutine.

• Built-In model: native implementation of the Built-In Johnson-Cook constitutive law, in

order to compare the results with a reference solution.
5.1 One element benchmark tests

Figure 3: Numerical models for the one element tensile and shear tests

Tests have also been done using the Bisection method, but we have made the choice to omit them
in this section to reduce the number of results because those results are very closed to the A-N-R
model but computing time is 10% to 30% longer than the A-N-R and the N-N-R models. At this
point, it has to be noted that, Abaqus/Explicit doesn’t use the same objective stress rate when
using the Built-In implementation and the VUMAT subroutine. As reported in documentation
[1] and confirmed by our own tests based on a one element hyperelastic shear test, a Jaumann
rate is used for both the Built-In formulation and the VUHARD subroutine while a Green-Naghdi
rate is used for the VUMAT subroutine. Therefore, a straightforward comparison of Built-In and
VUMAT results is not possible when large rotations occur.

5.1 One element benchmark tests

One element test, also called single element test, is a very simple and practical method to
investigate the accuracy and sensitivity of the behavior of an element to the external loading.
In this subsection, the deformation of a 4-node bilinear displacement and temperature, reduced
integration with hourglass control element CPE4RT will be simulated using the proposed VU-
MAT subroutine and the Abaqus Built-In model. As only one under-integrated element is used,
we only have one integration point located at the center of the element. In this kind of test, all
nodes of the element are constrained with a prescribed displacement. As the geometry change is
exactly the same in each of the two following benchmarks, it will be easy to compare the results
in terms of plastic deformations, stresses and temperatures. The original size of the element is
10 mm × 10 mm.

5.1.1 One element tensile test

In this first benchmark, the two left nodes of the element are encastred and a prescribed
horizontal displacement d = 10 mm is applied on the two right nodes of the same element as
illustrated in Figure 3. As we are using an explicit integration scheme, the total simulation
time is set to t = 0.01 s. A perfect match between all five results has been found for the one
element tensile test, with a final value of the plastic strain ε p = 0.457 and a final temperature
T = 164.09°C. Differences in computational time are not appreciable in this test since the final
result is obtained in less than 1 s.

5.1.2 One element shear test

This second benchmark is similar to the previous one, but now, the two bottom nodes of the
element are encastred and a prescribed horizontal displacement d = 10 mm is applied on the two
top nodes of the same element as illustrated in Figure 3. Again, a perfect match between all five
results has been found, with a final value of the plastic strain ε p = 0.572 and a final temperature
T = 192.22°C.
5.2 Necking of a circular bar

L/3

6.413 6.350

L=26.67
Figure 4: Numerical model for the necking of a circular bar

5.2 Necking of a circular bar

The necking of a circular bar test is useful to evaluate the performance of VUMAT subroutine
for materials in presence of plasticity and large deformation [12, 8]. Because of the symmetric
structure, an axisymmetric quarter model of the specimen is established. Dimensions of the
specimen are reported in Figure 4. The loading is realized through an imposed total displacement
of 7 mm along the #» z axis on the left side of the specimen while the radial displacement of the
same edge is supposed to remain zero. On the opposite side the axial displacement is restrained
while the radial displacement is free. The mesh consists of 400 CAX4RT elements with a refined
zone of 200 elements on the right side on 1/3 of the total height. Again, the total simulation time
is set to t = 0.01 s because of the explicit integration scheme adopted.
Figure 5 shows the equivalent plastic strain contourplot of the deformed bar for two models:
the Built-In model (left side) and the N-N-R model (right side). The maximum equivalent plastic
strain ε p is located into the center of the bar (so we have chosen to record the time-history
evolutions for the red element in Figure 4) and all the models give quite the same values as
reported in Table 2 for ε p , σ and T . Figure 6 shows the evolution of the von Mises stress σ with
the displacement applied at the end of the specimen for the different models. As reported in this
figure, the Built-In model, the VUHARD model and the two versions of the Newton-Raphson
model give almost the same results. A slight difference can be seen in the Direct model, where
there is a little over-estimation of the von Mises stress in the middle of the simulation.
Table 2 reports some results at the end of the computation concerning the total number of
increments and the total computing time for all models. As reported in this table, the results of
the two versions of the Newton-Raphson model are identical, this tends to prove that there is no
difference in using numerical or analytical derivatives of the flow law. The T/I column show
the ratio of the total computing time over the total number of increments, i.e. the computing
time/increment. As reported in Table 2, the native procedure corresponds to the fastest algorithm
in terms of computing time. This is easy to explain because this internal procedure is natively
optimized within the Abaqus code. The simple fact of transferring all the data to a VUMAT or
VUHARD subroutine increases the computing time independently of the optimization of the im-
plemented stress evaluation algorithm. This translates in particular into the fact that the increase
in CPU time due to the use of the VUHARD subroutine is about 18%, while the increase for the
different versions of the VUMAT routines is about 12 to 15% in terms of T/I ratios.
It is noticeable in this case that the total number of increments needed to perform the whole
simulation is lower for the Newton-Raphson models than for the three other models. This differ-
ence is illustrated in Figure 7 where the evolution of the time increment ∆t during the computa-
5.2 Necking of a circular bar

PEEQ SDV1
(Avg: 75%) (Avg: 75%)
+2.157e+00 +2.179e+00
+1.980e+00 +2.000e+00
+1.804e+00 +1.822e+00
+1.627e+00 +1.643e+00
+1.450e+00 +1.465e+00
+1.274e+00 +1.286e+00
+1.097e+00 +1.108e+00
+9.205e-01 +9.294e-01
+7.439e-01 +7.509e-01
+5.672e-01 +5.725e-01
+3.906e-01 +3.940e-01
+2.140e-01 +2.155e-01
+3.735e-02 +3.708e-02

Built-in N-N-R model

Johnson-Cook Johnson-Cook
in Abaqus/Explicit Subroutine

Figure 5: Equivalent plastic strain ε p contourplot for the necking of a circular bar
5.3 Taylor impact test

Figure 6: Von Mises stress σ vs. displacement for the necking of a circular bar

tion is reported. The smoother variation of the time increment, and greatest values, are obtained
with both versions of Newton-Raphson, leading to the minimal number of increments to com-
plete the simulation. Using the Built-In model and the VUHARD model, a reduction of the stable
time increment from ∆t = 6.5 10−8 s to ∆t = 5.2 10−8 s is noticed after a displacement of 2.52 mm,
while using the Direct method, a large reduction of the stable increment to ∆t = 3.4 10−8 s with
some residual oscillations is encountered after a displacement 2.03 mm. The time increment us-
ing both versions of the Newton-Raphson model is constant up-to a displacement of 3.36 mm and
reduces smoothly after this point because of the striction of the specimen. During the last third
of the simulation time increments of the N-R models are almost equal to the one of the Built-In
model and the VUHARD model.
From those results we can conclude that the integration of the constitutive equation has an
influence on the stable time increment ∆t. We can also note, from the results reported in Table
2 that, using the VUMAT Newton-Raphson doesn’t increase a lot the total computational time
(around 10% more time in this case) as the total number of increments has been reduced by a
factor of 3.8%. We can also note that the VUMAT subroutine is faster than the VUHARD sub-
routine. Of course, computational cost of the VUMAT model can be reduced by optimizing the
FORTRAN routine (removing the numerous number of tests inside the subroutine, optimizing
the computations and the mathematical expressions, ...). It has also to be noted that the requested
precision for the Newton-Raphson subroutine εNR = 10−8 has also an influence on the computa-
tional time, reducing it reduces also the computational time.

5.3 Taylor impact test

5.3.1 Axisymmetric Taylor impact test
Finally, the performance of the proposed VUMAT subroutine is validated under high de-
formation rate with the simulation of the Taylor impact test [19]. In the Taylor impact test, a
5.3 Taylor impact test

Figure 7: Time increment ∆t vs. displacement for the necking of a circular bar

Table 2: Comparison of results for the necking of a circular bar benchmark

T/I σ T
Model Incr. Time εp
(µs) (MPa) (°C)
A-N-R 191 655 1m 31s 469 2.158 1048.15 583.64
N-N-R 191 653 1m 32s 480 2.158 1048.15 583.64
Direct 267 964 2m 08s 478 2.161 1052.60 588.18
VUHARD 200 182 1m 39s 494 2.157 1048.14 583.60
Built-In 199 402 1m 23s 416 2.137 1051.28 579.79
5.3 Taylor impact test

3.2

32.4
Figure 8: Model of the Taylor impact test

cylindrical specimen is launched to impact a rigid target with a prescribed initial velocity. The
numerical model, reported in Figure 8 is established as axisymmetric. The height is 32.4 mm
and the radius is 3.2 mm. The axial displacement is restrained on the right side of the specimen
while the radial displacement is free (to figure a perfect contact without friction of the projectile
onto the target). A predefined velocity of Vc = 287 m/s is imposed on the specimen. The mesh
consists of 250 CAX4RT elements (5 × 50 elements). The total simulation time for the Taylor
impact test is t = 8.0 10−5 s.
Figure 9 shows the equivalent plastic strain contourplot of the deformed rod for two models:
the Built-In model (left side) and the N-N-R model (right side). The distributions of the equiva-
lent plastic strain are almost the same for both models. The maximum equivalent plastic strain ε p
is located into the center element of the model (the red element in Figure 8) and the models give
quite the same value as reported in Table 3 for ε p , T and the final dimensions of the specimen
L f (final length) and D f (final diameter of the impacting face). Figure 10 shows the evolution of
the equivalent plastic strain ε p with time for the different models for the element at the center of
the impacting face (the red element in Figure 8). As reported in this figure and according to the
results presented in Table 3, all models give almost the same results.
It can be noticed from this later that the Direct model (i.e. the classic way to program a
VUMAT subroutine assuming that the explicit time integration scheme allows to use a straight-
forward evaluation of the plastic strain) doesn’t provide results in accordance with the other
models concerning the internal fields and the final geometry of the specimen. This tends to prove
that if one wants to perform an inverse characterization of material constitutive parameters for
dynamic applications based on a Taylor impact test and a FORTRAN VUMAT subroutine for
implementing an exotic constitutive law, this can be source of errors in the identification of the
constitutive equation parameters.

5.3.2 3D Taylor impact test

In order to have relatively longer computational times while remaining within the framework
of the numerical simulation of the Taylor impact test, the choice was made for a 3D modeling.
Therefore, a 3D quarter model of the Taylor cylindrical specimen, with the same dimensions as
the Taylor 2D model, as been set-up as reported in Figure 11. The new mesh consists of 37 422
C3D8RT elements which is a quite large simulation benchmark.
The results concerning the CPU times (increments, time and T/I ratio), the final values of de-
formations and temperatures as well as the geometrical results are shown in Table 4. We can note
5.3 Taylor impact test

PEEQ SDV1
(Avg: 75%) (Avg: 75%)
+1.797e+00 +1.810e+00
+1.647e+00 +1.659e+00
+1.498e+00 +1.508e+00
+1.348e+00 +1.357e+00
+1.198e+00 +1.206e+00
+1.048e+00 +1.056e+00
+8.985e-01 +9.049e-01
+7.488e-01 +7.541e-01
+5.990e-01 +6.032e-01
+4.493e-01 +4.524e-01
+2.995e-01 +3.016e-01
+1.498e-01 +1.508e-01
+0.000e+00 +0.000e+00

Built-in N-N-R model

Johnson-Cook Johnson-Cook
in Abaqus/Explicit Subroutine

Figure 9: Equivalent plastic strain ε p contourplot for the Taylor impact test

Table 3: Comparison of results for the Taylor impact test

T Lf Df
Model Incr. εp
(°C) (mm) (mm)
A-N-R 3 938 1.810 561.09 26.56 11.10
N-N-R 3 939 1.810 561.09 26.56 11.10
Direct 3 967 1.802 570.44 26.62 11.04
VUHARD 3 849 1.808 561.55 25.55 11.11
Built-In 3 834 1.797 560.17 26.57 11.12
5.3 Taylor impact test

Figure 10: Equivalent plastic strain ε p vs. time for the Taylor impact test

32.4

3.2
x y

Figure 11: Model of the 3D Taylor impact test

Table 4: Comparison of results for the 3D Taylor impact test

Time T/I T Lf Df
Model Incr. εp
(ms) (°C) (mm) (mm)
A-N-R 20 702 23m 59s 69.5 8.99 1507 26.42 12.07
N-N-R 20 834 24m 38s 70.9 9.04 1512 26.42 12.06
Direct 19 872 23m 23s 70.6 8.54 1500 26.47 11.95
VUHARD 22 498 24m 46s 66.1 7.53 1434 26.40 11.73
Built-In 22 671 22m 44s 60.2 7.67 1451 26.41 11.80
5.4 VUMAT implementation of the alternative constitutive laws

a very good correlation of the results concerning geometric dimensions, whereas the results in
terms of deformation and temperature (very localized results on a very constrained element) dif-
fer due to a better taking into account of the constitutive law by the VUMAT subroutine (this has
been confirmed by refining the mesh). Similar to the results established in section 5.2 concern-
ing the necking of a circular bar, the T/I ratio is again increased by about 15% for the different
versions of the VUMAT subroutines, whereas the overall increase in CPU time is only about 5%
due to the reduction of the total number of increments required for simulation by about 10%.
Concerning the VUHARD approach, the increase in the T/I ratio is around 10% with approxi-
mately the same number of required increments as the Built-In method, leading to a total CPU
increase of about 10% also. This proves again the efficiency of the proposed VUMAT algorithm.

5.4 VUMAT implementation of the alternative constitutive laws

In this section, the Johnson-Cook constitutive law and three other constitutive laws are imple-
mented in the N-N-R VUMAT and the Direct VUMAT subroutines to simulate Taylor compres-
sion test, in order to validate the application of the proposed algorithm to alternative elastoplastic
constitutive
 laws which follow J2 plasticity and isotropic hardening and have the general form
q 
y p p
of σ ε , ε , T . The main advantage of using our N-N-R approach is that it does not require
to compute explicitly the derivatives of the hardening law, they are numerically computed using
equations (26-28). The other constitutive laws are the TANH constitutive law, modified TANH
constitutive law and Bäker constitutive law.

• Calamaz et al. [20] proposed the so called TANH constitutive law by adding a term mod-
eling the strain softening to the Johnson-Cook constitutive law, given by:
1
σy = σ JC [D + (1 − D) tanh( )] (45)
ε p + ε0
with: !q !
pε p T − T0
D=1− p tanh (46)
1 + pε T rec − T 0
in which σ JC represents the original Johnson-Cook constitutive law. The constant ε0 can
modulate the strain corresponding to the peak stress, p and q are the additional constitutive
law parameters, and T rec is the onset temperature for the strain softening phenomenon.

• Hor et al. [21] proposed later a constitutive law by modifying the TANH constitutive law,
given by:
q
σy = σε p (ε p , T )σT (T )σ qp (ε p , T ) (47)
ε
where:
pn
 p
 
1


 σ ε p (ε , T ) = A + Bε [D + (1 − D) tanh( ε p +ε 0
)]

  T −T
 m2
 σT (T ) = 1 − m1 T −T

 0

 q m 0
qp ! (48)
 p
 ε
 σεqp (ε , T ) = 1 + C(T ) ln q



ε0

with:  
pε p T − T0
! C1 exp C2 TTm
D=1−( ) tanh and C(T ) = (49)
1 + pε p T rec − T 0 T
T m
5.5 Discussion on the benchmarks

and A, B, C1 , C2 , m1 , m2 , n and p are the constitutive law parameters. As with the TANH
model, the constant ε0 can modulate the strain corresponding to the peak stress and T rec is
the onset temperature for the strain softening phenomenon.

• The last constitutive law we have implemented using the VUMAT subroutine is the one
proposed by Bäker [22, 21] and given by:
  q 
 p 
1 + C ln  ε  f (T )
 pn0 f (T )
 
y
σ = Aε   q  (50)
ε0

with: " !m #
T
f (T ) = exp − (51)
Tm
q
where A, n0 and m are the temperature-dependent material parameters, and C and ε0 are
constants.

The 2D model used for the simulation is the same as the one introduced in Section 5.3.1. The
material used in this case is the 42CrMo4 steel with ferrito-perlitic (referred to as 42CrMo4-FP).
The parameters of the four constitutive laws for 42CrMo4-FP steel were proposed by Hor et al.
[21]. The predefined velocity is set to Vc = 200m/s. The equivalent plastic strain contour-plots
of the deformed rod for the four models are shown in Figure 12.
The plastic deformation is concentrated at the bottom, and maximum equivalent plastic strain
is located in the center element of the specimen. More details concerning the total number of
increments, total computing time, maximum equivalent plastic strain ε p , final length L f , final
radius R f of the bottom and maximum temperature T are reported in Table 5. The results of the
Johnson-Cook and TANH models are almost identical. Compared with the previous two models,
the modified TANH model achieves larger equivalent plastic strain at the bottom, and corre-
spondingly it also achieves larger final radius and higher temperature at the bottom. However, as
we can see, this model has less plastic deformation in the other part. No matter in terms of the
maximum equivalent plastic strain and temperature or in terms of the geometric responses, the
Bäker model has the least plastic deformation. Concerning the Direct approach, it can be seen
that this one usually overestimates the plastic deformation and the temperature for the tree first
models. In terms of global results, the three first models give almost the same results, while the
Bäker model gives quite different results with regards to the previous ones.

5.5 Discussion on the benchmarks

In all the proposed benchmark tests, it is noteworthy that the results of the VUMAT subrou-
tines and the Abaqus Built-In model are very closed but some slight differences can be pointed.
Those differences can mainly be explained by the following remarks:

• The Johnson-Cook constitutive law implemented through the VUMAT subroutine doesn’t
have exactly the same expression as the Abaqus/Explicit Built-In model because the de-
pendence on the deformation rate has been modified in our implementation as discussed
before in section 3. This can be seen in the Bar Necking benchmark where the VUHARD
and the Built-In models doesn’t provide exactly the same results.
5.5 Discussion on the benchmarks

SDV1 SDV1 SDV1 SDV1

(Avg: 75%) (Avg: 75%) (Avg: 75%) (Avg: 75%)
+1.242e+00 +1.267e+00 +1.742e+00 +6.657e-01
+1.138e+00 +1.161e+00 +1.597e+00 +6.102e-01
+1.035e+00 +1.056e+00 +1.452e+00 +5.548e-01
+9.315e-01 +9.500e-01 +1.306e+00 +4.993e-01
+8.280e-01 +8.444e-01 +1.161e+00 +4.438e-01
+7.245e-01 +7.389e-01 +1.016e+00 +3.883e-01
+8.709e-01 +3.329e-01
+6.210e-01 +6.333e-01 +2.774e-01
+5.175e-01 +5.278e-01 +7.258e-01
+5.806e-01 +2.219e-01
+4.140e-01 +4.222e-01 +1.664e-01
+3.105e-01 +3.167e-01 +4.355e-01
+2.903e-01 +1.110e-01
+2.070e-01 +2.111e-01 +1.452e-01 +5.548e-02
+1.035e-01 +1.056e-01 +0.000e+00 +0.000e+00
+0.000e+00 +0.000e+00

(a) Johnson-Cook (b) TANH (c) Mod. TANH (d) Bäker

Figure 12: Equivalent plastic strain contour-plot of the Taylor specimen for the four constitutive models using the N-N-R
VUMAT

Table 5: Results for the Taylor compression test

Model Incr. εp T Lf Df
(°C) (mm) (mm)
JC N-N-R 2 355 1.24 307.7 28.26 9.74
JC Direct 2 346 1.23 316.5 28.38 9.63
TANH N-N-R 2 370 1.26 308.1 28.27 9.78
TANH Direct 2 427 1.26 316.4 28.36 9.68
Mod. TANH N-N-R 3 060 1.74 459.5 29.24 10.33
Mod. TANH Direct 3 528 1.76 471.1 29.33 10.25
Bäker N-N-R 1 376 0.67 281.7 29.77 8.32
Bäker Direct 1 925 0.941 263.8 28.23 9.09
 • The Built-In model and the VUHARD model are integrated through an explicit central-
difference time integration rule, while the radial return method, which belongs to an im-
plicit integration algorithm, is employed in the VUMAT subroutines.
• As the root of the function is not an exact solution, but an approximation, the choice of the
precision tolerance has an effect on the final results.

• The Jaumann stress rate is used for the Built-In model and the VUHARD subroutine while
the VUMAT routine employs the Green-Naghdi stress rate. This can cause differences in
the results in large deformations models when finite rotation of a material point is accom-
panied by finite shear [1].

6 Conclusions

In this paper, an approach has been proposed for the implementation of elastoplastic con-
stitutive laws in Abaqus/Explicit through VUMAT subroutine. The proposed implementation
is robust and efficient, and can be applied to simulate the behavior of various materials defined
by any elastoplastic constitutive model following J2 plasticity and isotropic hardening. In the
proposed algorithm, an implicit time integration scheme based on the radial return method is
employed, which is unconditionally stable and can ensure that the von Mises yield criterion is
always satisfied during computation. Validated through numerical benchmarks, a robust and
efficient root-finding method, the so-called safe version of Newton-Raphson method, has been
implemented to compute the plastic corrector term.
This algorithm can now be re-used by only changing the expression of the flow law in order
to implement alternative constitutive law, such as the revised form of the Johnson-Cook law
proposed by Rule et al. [16], or any other one of the same family. Applications of this kind of
developments mainly concerns inverse identification of constitutive law parameters based on a
dynamic impact tests, or programming of alternative constitutive laws in Abaqus/Explicit. The
proposed implementation has been found more precise with regards to other kind of VUMAT
or VUHARD routines based on a direct evaluation of the plastic strain rate in order to provide
accurate results for the identification procedure.

Acknowledgements

This work is partly supported by the scholarship from China Scholarship Council (CSC)
under Grant CSC N°201406290010.

References
[1] Abaqus v. 6.14 documentation, Dassault Systemes Simulia Corporation.
[2] C. Gao, Fe realization of a thermo-visco-plastic constitutive model using vumat in abaqus/explicit program, in:
Computational Mechanics, Springer, 2007, pp. 301–301.
[3] Abaqus v. 6.14 user subroutines reference guide, Dassault Systemes Simulia Corporation.
[4] G. R. Johnson, W. H. Cook, A constitutive model and data for metals subjected to large strains, high strain rates
and high temperatures, in: Proceedings of the 7th International Symposium on Ballistics, Vol. 21, The Hague, The
Netherlands, 1983, pp. 541–547.
[5] S. Nemat-Nasser, On finite deformation elasto-plasticity, International Journal of Solids and Structures 18 (10)
(1982) 857–872.
[6] M.-H. Yu, Generalized plasticity, Springer Science & Business Media, 2006.
 [7] G. I. Taylor, H. Quinney, The latent energy remaining in a metal after cold working, Proceedings of the Royal
Society of London. Series A, Containing Papers of a Mathematical and Physical Character 143 (849) (1934) 307–
326.
[8] J. C. Simo, T. J. R. Hughes, Computational inelasticity, Springer, 1998.
[9] M. L. Wilkins, Calculation of elastic-plastic flow, Tech. rep., Alder, B. (Ed.), Methods in Computational Physics,
Vol. 3. Academic Press, pp. 211-263 (1963).
[10] G. Maenchen, S. Sacks, The tensor code, in methods of computational physics, 3, 181-210. b. alder, s. fernback
and m. rotenberg editors (1964).
[11] F. Dunne, N. Petrinic, Introduction to computational plasticity, Oxford University Press New York, 2005.
[12] J.-P. Ponthot, Unified stress update algorithms for the numerical simulation of large deformation elasto-plastic and
elasto-viscoplastic processes, International Journal of Plasticity 18 (1) (2002) 91–126.
[13] R. Zaera, J. Fernández-Sáez, An implicit consistent algorithm for the integration of thermoviscoplastic constitutive
equations in adiabatic conditions and finite deformations, International journal of solids and structures 43 (6) (2006)
1594–1612.
[14] F. S. Acton, Numerical methods that work, Maa, 1990.
[15] W. H. Press, Numerical recipes 3rd edition: The art of scientific computing, Cambridge university press, 2007.
[16] W. K. Rule, S. Jones, A revised form for the johnson–cook strength model, International Journal of Impact Engi-
neering 21 (8) (1998) 609–624.
[17] L. Schwer, Optional strain-rate forms for the johnson cook constitutive model and the role of the parameter epsilon
0, in: Proceedings of the 6th European LS-DYNA Users Conference, Anwenderforum, Frankenthal, Germany,
2007.
[18] G. Jansen van Rensburg, S. Kok, Tutorial on state variable based plasticity: an abaqus uhard subroutine.
[19] G. I. Taylor, The testing of materials at high rates of loading, Journal of the institution of civil engineers 26 (8)
(1946) 486–519.
[20] M. Calamaz, D. Coupard, F. Girot, Numerical simulation of titanium alloy dry machining with a strain softening
constitutive law, Machining Science and Technology 14 (2) (2010) 244–257.
[21] A. Hor, F. Morel, J.-L. Lebrun, G. Germain, Modelling, identification and application of phenomenological consti-
tutive laws over a large strain rate and temperature range, Mechanics of Materials 64 (2013) 91–110.
[22] M. Bäker, Finite element simulation of high-speed cutting forces, Journal of Materials Processing Technology
176 (1) (2006) 117–126.

View publication stats