On the Importance

Wooseok Ji
of Work-Conjugacy
Research Fellow

Anthony M. Waas1
and Objective Stress
Felix Pawlowski Collegiate Professor
e-mail: [email protected]
Rates in Finite Deformation
Department of Aerospace Engineering,
Composite Structures Laboratory,
Incremental Finite Element
University of Michigan,
Ann Arbor, MI 48109 Analysis
This paper is concerned with two issues that arise in the finite element analysis of 3D solids.
Zdenek P. Bazant The first issue examines the objectivity of various stress rates that are adopted in incremen-
Walter Murphy Professor, tal analysis of solids. In doing so, it is revealed that large errors are incurred by an
Department of Civil Engineering, improper choice of stress rate. An example problem is presented to show the implications of
Northwestern University, the choice of stress rate. The second issue addresses the need to maintain work-conjugacy
Evanston, IL 60208 in formulating and solving bifurcation buckling problems of 3D elastic solids. Four popular
commercial codes are used to obtain buckling loads of an axially compressed thick sand-
wich panel, and it is shown that large errors in buckling load predictions are incurred as a
result of violating the requirement of work-conjugacy. Remedies to fix the errors in the
numerical solution strategy are given. [DOI: 10.1115/1.4007828]

1 Introduction the tenets pointed out in Bazant [1] and later exemplified in
Bazant and Cedolin [2] with respect to the proper use of objective
The finite element method is a key computational tool in solid
stress rates and with respect to computing buckling loads of sol-
mechanics, and it has now become the mainstay of problems
ids, leading to large discrepancies in computed outcomes. Simple
involving any of the broad phenomena of material deformation—
remedies to fix these deficiencies are suggested. The implications
elasticity, plasticity, and damage. However, its utility for prob-
of the reported findings are significant for structural analysts,
lems involving finite deformation of highly anisotropic materials,
since the errors pointed out are quite large for a range of problems
such as laminated composites, sandwich composites, and materi-
that are of practical significance in structural design that use built-
als with distributed and aligned damage (sets of parallel micro-
up structures or modern composite materials that exhibit strong
cracks), requires careful consideration in the definition and utility
anisotropy.
of various stresses and stress rates and associated work-conjugate
This paper is organized as follows: We first present various
strains and strain rates. In earlier work by Bazant [1], the require-
stress rates that are adopted in formulating incremental equations
ments on the proper work definition that are necessary for the
for analyzing nonlinear mechanics problems using the finite ele-
equivalence between different mathematical formulations for the
ment method. This is followed by studying an example problem
incremental elastic stability of a 3D solid were presented. Among
of shearing a block using four popular commercial codes. The
the various mathematical formulations available to address the in-
results are compared against each other and an available analytical
finitesimal elastic stability of a 3D solid, Bazant showed that,
solution. Next, we consider the bifurcation buckling problems of a
when a certain finite strain measure is selected to describe the
homogenized orthotropic strip and a thick orthotropic sandwich
incremental deformation, the corresponding conjugate incremen-
panel, analytically and within the context of using the finite ele-
tal stress and the constitutive model associated with those choices
ment method to compute buckling loads. The results predicted by
of stress and strain must be used in order to properly define the in-
the commercial codes are compared against each other and the an-
ternal work due to transitions in the equilibrium state. He also
alytical solutions. The paper concludes with suggestions to fix the
derived a unified general formulation in terms of a parameter m,
errors that are present in the commercial codes for the two types
with different values of m leading to different mathematical for-
of problems discussed.
mulations. His unified formulation provides consistency and cor-
rectness for obtaining the work-conjugate relations between the
finite strain and the incremental stress and the corresponding con-
stitutive model. 2 Basic Finite Element Formulation
Modern commercial codes have formulated their finite element Using Objective Stress Rates
equations for certain classes of problems using stress rates. As is The equilibrium equation of a deformable solid can be written
well known, the rate form of the resulting equations needs to be in a rate form using the nominal stress in the initial, reference state
objective with respect to coordinate transformations. In this paper, [3].
it is shown that many popular commercial codes violate many of
$X  N_ þ b_ ¼ 0 (1)
1
Corresponding author.
Manuscript received April 1, 2012; final manuscript received July 22, 2012;
where $X  N_ ¼ divergence of N_ with respect to the reference
accepted manuscript posted October 10, 2012; published online May 23, 2013. coordinate system, N_ ¼ rate of nominal stress change after an in-
Editor: Yonggang Huang. finitesimal time, and b_ ¼ body force rate per unit volume of the

Journal of Applied Mechanics Copyright V

C 2013 by ASME JULY 2013, Vol. 80 / 041024-1

Downloaded From: http://appliedmechanics.asmedigitalcollection.asme.org/ on 07/24/2013 Terms of Use: http://asme.org/terms

 initial configuration. The corresponding traction boundary condi- objective rates of Kirchhoff stress. For example, m ¼ 2 gives the
tion is objective Truesdell rate of Kirchhoff stress and m ¼ 0 corre-
r
sponds to the Jaumann rate. s_ and s ðTKÞ are connected through the
t_ ¼ n  N_ (2) definition of the objective Truesdell rate of Kirchhoff stress as

where t_ ¼ nominal traction rate prescribed on a surface in the ini- r ðTKÞ

s ¼ s_  L  s  s  LT (10)
tial state and n ¼ unit normal of the surface in the reference con-
figuration. The principle of virtual work with virtual velocity r
fields dv is employed to formulate the finite element equations The corresponding tangential moduli of s ðTKÞ can be derived
corresponding to Eqs. (1) and (2). Integrating the scalar product from the constitutive model associated with the second Piola–
between Eq. (1) and the virtual velocity fields over the volume in Kirchhoff stress, S, and Green’s Lagrangian strain, E ¼ 12ðC  IÞ,
the reference configuration yields where C ¼ FT F is the right Cauchy deformation tensor and I is
the identity tensor of rank 2. When U is the strain energy per unit
ð
  initial volume of a solid, S can be obtained from
$X  N_ þ b_  dv dV B ¼ 0 (3)
VB @U
S¼ (11)
where B represents quantities associated with the reference, base @E
state. From the chain rule,
or in the form of a stress-strain relation,
ð  
  @dv _
$X  N_  dv  N_ : þ b  dv dV B ¼ 0 (4) S¼C:E (12)
VB @X
where C is the fourth order material elasticity tensor paired with
Utilizing the Gauss theorem with the first term, the Green’s Lagrangian strain defined as
ð ð  
@dv _ @2U
n  N_  dv dSB þ N_ : þ b  dv dV B ¼ 0 (5) CIJKL ¼ (13)
SB VB @X @EIJ @EKL

From Eq. (2), the final form of the virtual work equation is The rate form of Eq. (12) is
obtained as
ð   ð ð S_ ¼ C : E_ (14)
_N : @dv dV B ¼ _t  dv dSB þ b_  dv dV B ¼ 0 (6)
VB @X SB VB The Kirchhoff stress can be written in terms of the second Piola–
Kirchhoff stress as
Using the Kirchhoff stress, s ¼ F  N, where F is the deforma-
tion gradient, the rate of nominal stress change is expressed as s ¼ F  S  FT (15)

N_ ¼ F1  s_ þ F_ 1  s and the rate of change of the Kirchhoff stress is

(7)
¼ F1  s_  F1  L  s
s_ ¼ F_  S  FT þ F  S  F_ T þ F  S_  FT
  (16)
where L ¼ F_  F1 ¼ velocity gradient. Now, the internal virtual ¼ L  s þ s  LT þ F  C : E_  FT
_ is expressed in terms of the rate of Kirchhoff stress,
work rate, dU,
_ the velocity gradient, L, and the virtual velocity gradient, dL,
s, From the definition of the Truesdell rate in Eq. (10), Eq. (16) can
such that be rearranged and rewritten in indicial notation as
ð
@dv B rðTKÞ
dU_ ¼ N_ : dV sij ¼ FiI FjJ FkK FlL CIJKL Dkl (17)
B @X
ðV
 1  @dv B
¼ F  ðs_  L  sÞ : dV or
V @X
ð B   (8)
r ðTKÞ
@dv 1 s ¼ CðTKÞ : D (18)
¼ Tr  F  ðs_  L  sÞ dV B
B @X
ðV
  ðTKÞ
¼ s_  s  LT : dLdV B where Cijkl ¼ FiI FjJ FkK FlL CIJKL is the spatial (or Eulerian) elas-
VB ticity tensor. Now, Eq. (9) is expressed as

where Tr denotes the trace of a tensor. It is noted here that the r ðmÞ 1
symmetric property of the Kirchhoff stress is used in deriving s ¼ CðTKÞ : D þ ð2  mÞðD  s þ s  DÞ (19)
2
_ in Eq. (8)
Eq. (8). The rate of change of the Kirchhoff stress, s,
can be expressed with various types of objective stress rates,
which can be written in a unified form as or, utilizing dij (identity tensor of rank 2), it can be written in indi-
cial notation as
r ðmÞ r 1

s ¼ s ðTKÞ þ ð2  mÞðD  s þ s  DÞ (9) rðmÞ ðTKÞ 1 
2 sij ¼ Cijkl þ ð2  mÞ sik djl þ sjl dik Dkl
2


r ðTKÞ 1  
where s ðTKÞ is the objective Truesdell rate of Kirchhoff
 stress and ¼ Cijkl þ ð2  mÞ sik djl þ sjk dil þ sil djk þ sjl dik Dkl
D is the rate of deformation defined as D ¼ 12 L þ LT . The pa- 4
rameter “m” in Eq. (9) corresponds to different, specific types of (20)

041024-2 / Vol. 80, JULY 2013 Transactions of the ASME

Downloaded From: http://appliedmechanics.asmedigitalcollection.asme.org/ on 07/24/2013 Terms of Use: http://asme.org/terms

 with addition of the minor symmetry property of the fourth order internal work needs to be second order accurate to describe the
tensor. Hence, the unified objective stress rate is written in the equilibrium state of a solid after infinitesimal increments, as
form of elasticity as shown in Ref. [2] (see Chapter 11.3). In this case, higher order
terms in Eq. (18) can be neglected and the tangential moduli are
r ðmÞ simplified to the infinitesimal form of
s ¼ CðmÞ : D (21)
ð2Þ
where Cijkl ’ diI djJ dkK dlL CIJKL ¼ Cijkl (28)

ðmÞ ðTKÞ 1  
Cijkl ¼ Cijkl þ ð2  mÞ sik djl þ sjk dil þ sil djk þ sjl dik (22) 2.2 Formulation B: Truesdell Rate of Cauchy Stress. Next,
4 the Truesdell rate of Cauchy stress is utilized to express the princi-
ple of virtual work in the current deformed configuration. With
From Eqs. (9), (10), and Eq. (21), Eq. (8) is rewritten as
m ¼ 2, Eq. (25) is written as
ð

1 ð h ð ð
i
r ðmÞ
dU_ ¼ s  ð2  mÞðD  s þ s  DÞ þ L  s : dLdV B r ð2Þ 
B 2 r : dD þ r : LT  dL dV ¼ t_0  dv dS þ b_0  dv dV
ðV h
r ðmÞ  i V S V
¼ s : dD  ð2  mÞs : ðD  dDÞ þ s : LT  dL dV B (29)
VB
(23) The Truesdell rate of Cauchy stress is defined as
Thus, the principle of virtual work in the reference configuration r ð2Þ r
is written as r ¼ r ðTCÞ ¼ r_  L  r  r  LT þ TrðLÞr (30)
ð h
r ðmÞ  i
s : dD  ð2  mÞs : ðD  dDÞ þ s : LT  dL dV B and it can be written in terms of the Truesdell rate of Kirchhoff
VB stress as
ð ð
¼ t_  dv dSB þ b_  dv dV B ¼ 0 (24) r ðTCÞ r
SB VB r ¼ J 1 s ðTCÞ (31)
or, in the current (deformed) configuration, it reads r
ð h Thus, the spatial elasticity tensor for r ðTCÞ is obtained as
r ðmÞ  i
r : dD  ð2  mÞr : ðD  dDÞ þ r : LT  dL dV CðTCÞ ¼ J 1 CðTKÞ . However, when we impose the same assump-
V tion on the need for second order accurate internal work due to
ð ð (25)
¼ t_0  dvdS þ b_0  dv dV ¼ 0 stress increments, the Truesdell rate of Cauchy stress can be
S V directly derived from the simplification of the Truesdell rate of
r ðmÞ
Kirchhoff stress, such that
1 r ðmÞ
where r ¼J s and r is the Cauchy stress that is related to
the Kirchhoff stress through r ðTKÞ _  L  ðJrÞ  ðJrÞ  LT
s ¼ ðJrÞ
s ¼ Jr (26) ’ r_ þ TrðLÞr  L  r  r  LT (32)
B
where J ¼ dV=dV is the determinant of the deformation gradient. r ðTCÞ
¼r
We now conclusively provide the basis of the rate-based finite
element formulation. This formulation consists of Eq. (21) for stress
increments, Eq. (22) for the corresponding constitutive model, and    the Jacobian is approximated as J ’ 1 þ uk;k and
where
Eqs. (24) or (25) for the equilibrium statement in the initial or _ ij ’ r_ ij þ vk;k rij . Higher order terms associated
J rij ¼ J r_ ij þ Jr
current deformed configuration, respectively. The following with uk;k are neglected in deriving Eq. (32), since they are not im-
sections (Secs. 2.1–2.6) discuss the different formulations that portant in defining incremental internal work to second order ac-
employ different stress rates depending on the parameter “m”. It is r
curacy. Therefore, the elasticity tensor for r ðTCÞ is obtained as
noted here that, when stresses are being updated in finite element
analysis, it is crucial to maintain a consistent formulation between CðTCÞ ¼ C, where the internal work is preserved to second order
Eq. (21) and Eq. (22) if the principle of virtual work is correctly accuracy.
implemented using any choice of stress rate. The choice of different
stress rate types is solely dependent on the efficiency of solving a 2.3 Formulation C: Jaumann Rate of Kirchhoff
particular type of problem. Depending on the value of “m”, the Stress. The Jaumann rate is a popular choice in modern computa-
resulting mathematical expressions may be different, but they are tional continuum mechanics, because it is relatively easy to imple-
mutually equivalent to each other, providing the same end result. ment into a finite element framework (for example, the spin tensor
W is readily available as a by-product of D) and leads to symmet-
ric tangential moduli. The Jaumann rate of Kirchhoff stress is
2.1 Formulation A: Truesdell Rate of Kirchhoff defined as
Stress. The formulation based on the objective Truesdell rate of
Kirchhoff stress is obtained when m ¼ 2. From Eq. (9), r ð0Þ r
r ð2Þ r s ¼ s ðJKÞ ¼ s_ þ s  W  W  s (33)
s ¼ s ðTKÞ . The principle of virtual work in the initial configu-
ration, Eq. (24), is written as
ð ð ð Substitution of m ¼ 0 into Eq. (24) yields the principle of virtual
r ðTKÞ   work based on the Jaumann rate of Kirchhoff stress,
s : dD þ s : LT  dL dV B ¼ t_  dv dSB þ b_  dv dV B
ð h
VB SB VB
r ð0Þ  i
(27) s : dD  2s : ðD  dDÞ þ s : LT  dL dV B
VB
ð ð
and the corresponding spatial elasticity tensor is ¼ t_  dvdSB þ b_  dv dV B (34)
ð2Þ ðTKÞ
Cijkl ¼ Cijkl ¼ FiI FjJ FkK FlL CIJKL . In many practical cases, the SB VB

Journal of Applied Mechanics JULY 2013, Vol. 80 / 041024-3

Downloaded From: http://appliedmechanics.asmedigitalcollection.asme.org/ on 07/24/2013 Terms of Use: http://asme.org/terms

 and the corresponding tangential moduli are Bazant [1] showed that the Jaumann rate of Cauchy stress is not
energetically conjugate to any strain measure. Ji et al. [4] demon-
ð0Þ ðJKÞ ðTKÞ 1  strated that a significant error is induced in computing buckling
Cijkl ¼ Cijkl ¼ Cijkl þ sik djl þ sjk dil þ sil djk þ sjl dik (35)
2 loads if a constant material tensor is used with the Jaumann rate of
Cauchy stress. The deficiency of the Jaumann rate of Cauchy
When we define the second order accurate internal work, only the stress can be corrected by redefining the elasticity tensor as
first order terms in the constitutive model in Eq. (35) are required
ðJCÞ ðBCÞ
[2]. Since the Jacobian can be approximated as J ¼ 1 þ uk;k , Cijkl ¼ Cijkl  rij dkl (45)
Eq. (35) is now reduced to
1  as shown in page 727 of Ref. [2], where it is given as Eq. 11.4.6.
ð0Þ
Cijkl ¼ Cijkl þ rik djl þ rjk dil þ ril djk þ rjl dik (36) Equation (45) converts Eqs. (44) to (37). The use of this transfor-
2 mation in an incremental analysis to extract buckling loads, for
2.4 Formulation D: Biezeno–Hencky Rate of Cauchy example, is described in Ref. [5].
Stress. The principle of virtual work in terms of the objective
Biezeno–Hencky rate of Cauchy stress can be obtained from Eq. 2.6 Summary of Formulations. The principle of virtual
(25) with m ¼ 0, work expressed using various types of stress rates are summarized
ð h here with the corresponding elasticity tensors in the infinitesimal
r ð0Þ  i form based on defining internal work to second order accuracy. It
r : dD  2r : ðD  dDÞ þ r : LT  dL dV
V should be noted that the different formulations described in this
ð ð
presentation result in the same internal virtual work rate defined
¼ t_0  dv dS þ b_0  dv dV (37) in Eq. (8). The end result of a boundary value problem solution
S V
will be the same, regardless of the type of objective stress rate
The objective Biezeno–Hencky rate of Cauchy stress is defined as used in carrying out the solution.
• Formulation A: Truesdell rate of Kirchhoff stress
r ð0Þ r
r ¼ r ðBCÞ ¼ r_ þ r  W  W  r þ TrðLÞr (38) ð ð ð
r ðTKÞ  
and it is related to the Jaumann rate of Kirchhoff stress through s : dD þ s : LT  dL dV B ¼ t_  dvdSB þ b_  dvdV B
VB SB VB
r ðBCÞ r (46a)
r ¼ J 1 s ðJKÞ (39)
ðTKÞ
C ¼C (46b)
Thus, the material tensor for the Biezeno–Hencky stress rate of
Cauchy stress is expressed as • Formulation B: Truesdell rate of Cauchy stress
ð h ð ð
1  r ðTCÞ  i
ðBCÞ
Cijkl ¼ J 1 Cijkl þ
ðTKÞ
rik djl þ rjk dil þ ril djk þ rjl dik (40) r : dD þ r : LT  dL dV ¼ t_0  dvdS þ b_0  dvdV
2 V S V
(47a)
Again, when we define the internal work of stress increments to
preserve second order accuracy, the Jaumann rate of Kirchhoff CðTCÞ ¼ C (47b)
stress is simplified to
• Formulation C: Jaumann rate of Kirchhoff stress
r ðJKÞ
_ þ ðJrÞ  W  W  ðJrÞ
¼ ðJrÞ ð h
s r ðJKÞ  i
’ r_ þ TrðLÞr þ r  W  W  r (41) s : dD  2s : ðD  dDÞ þ s : LT  dL dV B
VB
r ðBCÞ
ð ð
¼r ¼ t_  dv dSB þ b_  dv dV B (48a)
S V
and thus the material tensor for the Biezeno–Hencky stress rate of ðJKÞ 1 
Cijkl ¼ Cijkl þ rik djl þ rjk dil þ ril djk þ rjl dik (48b)
Cauchy stress is expressed as 2

ðBCÞ 1  • Formulation D: Biezeno–Hencky rate of Cauchy stress

Cijkl ¼ Cijkl þ rik djl þ rjk dil þ ril djk þ rjl dik (42)
2 ð h
r ðBCÞ  i
r : dD  2r : ðD  dDÞ þ r : LT  dL dV
2.5 Formulation E: Jaumann Rate of Cauchy Stress. The V
Jaumann rate of Cauchy stress can be regarded as a special case of ð ð
the Biezeno–Hencky formulation where the term associated with ¼ t_0  dv dS þ b_0  dv dV (49a)
S V
the volumetric deformation, TrðLÞr, in Eq. (38) is neglected. The
Jaumann rate of Cauchy stress is defined as ðBCÞ 1 
Cijkl ¼ Cijkl þ rik djl þ rjk dil þ ril djk þ rjl dik (49b)
2
r ðJCÞ
r ¼ r_ þ r  W  W  r (43) • Formulation E: Jaumann rate of Cauchy stress
Following steps similar to those adopted so far, the principle of ð h i
r ðJCÞ  
virtual work in ther current deformed configuration can be r : dD  2r : ðD  dDÞ þ r : LT  dL þ TrðLÞr : dD dV
expressed in terms of r ðJCÞ and r, and it reads V
ð ð
ð h i ¼ t_0  dvdS þ b_0  dvdV (50a)
r ðJCÞ   S V
r : dD  2r : ðD  dDÞ þ r : LT  dL þ TrðLÞr : dD dV
V
ð ð ðJCÞ 1 
Cijkl ¼ Cijkl  rij dkl þ rik djl þ rjk dil þ ril djk þ rjl dik
¼ t_0  dv dS þ b_0  dv dV (44) 2
S V (50b)

041024-4 / Vol. 80, JULY 2013 Transactions of the ASME

Downloaded From: http://appliedmechanics.asmedigitalcollection.asme.org/ on 07/24/2013 Terms of Use: http://asme.org/terms

 Fig. 1 Configuration of the simple shear test problem

3 Finite Element Analysis Demonstrating Incorrect

Formulation
First, we will demonstrate the effect of using an incorrect for-
mulation, resulting in incorrect tangential moduli (Eq. (22)). As
discussed in Sec. 2, when an objective stress rate is used with the
associated constitutive material model, the internal virtual work is
correctly defined. Here, the formulation using the Jaumann rate of
Cauchy stress with constant moduli is employed to demonstrate
errors caused by the incorrect internal work definition. The Fig. 2 Numerical and theoretical solutions of the simple shear
problem from different types of objectivity stress rates with
Jaumann stress rate is chosen because many commercial finite constant tangential moduli
element analysis (FEA) software packages utilize the Jaumann
rate in their FE formulations. The formulation using the Jaumann
stress rate of the Cauchy stress with constant moduli is denoted as formulation, which is based on the Jaumann rate of the Cauchy
“Formulation F”, and the resulting solutions using Formulation F stress, corresponds to Formulation F, where the material elasticity
are compared against an analytical solution and the correct FEA tensor is assumed to be constant (which is incorrect). Here, formu-
solution. The commercial code ABAQUS is employed here to lation F is implemented in ABAQUS through a user subroutine. In
demonstrate errors caused by the incorrect definition of the inter- the analytical solution based on the Jaumann rate, constant tan-
nal virtual work. ABAQUS uses the Jaumann rate and defines the gential moduli have been used. Furthermore, there is no “correct”
exact consistent Jacobian as theoretical solution that can employ the Jaumann rate, because it
is not energetically work-conjugate to any strain measure, as
~ ¼ 1 @DðJrÞ pointed out in Ref. [2], except in certain classes of problems
C (51) where incompressibility can be used as an additional assumption.
J @De
On the other hand, the Truesdell rate with the correct tangential
where De ¼ logarithmic strain increment. Formulation F is imple- moduli (which are constant) does indeed provide the expected
mented through a user subroutine to show that it corresponds to shear response.
the incorrect internal virtual work definition as used in the built-in When more than one element is used for the simple shear prob-
ABAQUS procedure. lem, proper traction boundary conditions must be prescribed on
The simple shear test proposed by Dienes [6] is used as the the lateral, vertical edges to maintain the deformation gradient for
boundary value problem. It is known that the Jaumann rate of the simple shear over the entire body [9]. Alternatively, with peri-
Cauchy stress exhibits shear stress oscillations [6,7]. Thus, this odic boundary conditions applied on the lateral surfaces (surface
example will be used to identify the type of objective stress rate OB and AC in Fig. 1), the representative shear stress can be
used in ABAQUS and also to demonstrate the effect of using the obtained from the element at the center, regardless of the number
incorrect tangential moduli in the formulation. The incorrect tan- of elements used. Figure 3 shows the shear response of a rectangu-
gential moduli also lead to errors in updating the stress in an incre- lar strip when the horizontal length of the undeformed block is 10
mental analysis. The initially square plate, OACB, is subjected to times longer than the undeformed vertical length. Clearly, for the
simple shear, as illustrated in Fig. 1. Initially, only one plane multiple element cases, all the erroneous formulations do provide
strain element is used to avoid any interaction between elements. the same consistent, nonphysical result as in the one element case,
The bottom horizontal surface is restricted from moving in the x establishing that the error incurred is not dependent on mesh
and y directions, and the top surface, CB, is constrained from effects or size effects. Although results from ABAQUS are used
moving in the y direction. The top surface CB is subjected to a here to discuss the effect of the boundary conditions on the edges,
fixed displacement in the x direction. other FE codes show the same behavior, but those results are
Figure 2 shows the shear response predicted using various com- omitted for the sake of brevity.
mercial FE codes as well as the correct theoretical solution, which
uses the Truesdell rate. Analytical solutions using Green–Naghdi
and Jaumann stress rates implemented with constant moduli are 4 Bifurcation Buckling Computations of Thick
also plotted in Fig. 2. The theoretical solutions for the simple Orthotropic Structures—Comparisons Between
shear problem are available in Ref. [8]. Nonlinear static solvers Popular Commercial Codes
are used from each FE program with the effect of the large strain
deformation accounted for. It appears that NASTRAN is based on 4.1 Buckling Load Prediction for an Orthotropic Strip. Ji
the Truesdell stress rate, and its output is in agreement with the et al. [4] demonstrated errors caused by the improper and incorrect
correct analytical solution. The ANSYS code is based on the formulations in a homogeneous, 2D orthotropic strip-buckling
Green–Naghdi stress rate formulation, as shown in Fig. 2. LS- analysis. They showed that errors in predicting buckling load are
Dyna produces an oscillatory shear response, implying that its as large as 100% when constant tangential moduli are used with
nonlinear static solver is based on the Jaumann stress rate formu- the Jaumann rate of Cauchy stress. The effect of the incorrect for-
lation. Furthermore, it is confirmed that the ABAQUS built-in mulation in using the constant moduli together with the Jaumann

Journal of Applied Mechanics JULY 2013, Vol. 80 / 041024-5

Downloaded From: http://appliedmechanics.asmedigitalcollection.asme.org/ on 07/24/2013 Terms of Use: http://asme.org/terms

 Table 1 Material properties of a orthotropic strip for Case 1

Property Value

Exx ðGPaÞ 2000  Gxy

Eyy ðGPaÞ 2  Gxy
Gxy ðGPaÞ 7.17
xy 0.29
h (m) 1
L/h 19

Table 2 Material properties of a orthotropic strip for Case 2

Property Value

Exx ðGPaÞ 1  1000  Eyy

Eyy ðGPaÞ 2  Gxy
Gxy ðGPaÞ 7.17
xy 0.29
h (m) 1
L/h 3

Fig. 3 Shear stress obtained from a long strip FE model using

periodic boundary conditions

rate of the Cauchy stress has been ignored in the FEA community.
The stress rate-based FEA has been extensively developed and
utilized for characterizing the mechanical behavior of
elastoplastic-type materials, whose elastic deformation part is usu-
ally assumed to be small compared to the plastic deformation.
Consequently, since the problem then becomes dominated by the
assumption of incompressibility invoked in standard plasticity for-
mulations, the error caused by using the incorrect stress rate
becomes masked. However, when an orthotropic elastic material
is considered in a buckling problem, the incorrect formulation
does lead to a significant overprediction in computing a buckling
load [4]. Ji et al. showed that, when the FE formulation for a buck-
ling problem fails to account for the correct pairing relations
between stress, strain, and tangential moduli, errors in predicting
the buckling load could be as large as 100%. The missing volu-
metric term pointed out in Ref. [4] results in a significant error
because the orthotropic material exhibits directional-dependent
deformation. They found that the popular commercial package,
ABAQUS, overpredicts the buckling loads. Here, other popular
commercial codes are also examined for the same buckling prob-
lem that was considered in Ref. [4]. Buckling loads computed Fig. 5 Buckling load predictions from various commercial FEA
from ANSYS, NASTRAN, and LS-Dyna are compared with those programs as a function of the aspect ratio
from the correct work-conjugate FE formulation [4] as well as an-
alytical solutions based on elasticity theory [10].
The elastic and homogeneous orthotropic strip, as shown in Fig. summarized in Table 1 and Table 2, respectively. In both cases, the
4, is uniformly compressed from one end, while the other end is out-of-plane properties are Ezz ¼ Eyy , Gxz ¼ Gxy , Gyz ¼ 5:96 GPa,
restrained from axial deformation. The right edge remains straight xz ¼ 0:0159, and yz ¼ 0:49:
but is free to move in the y-direction. Buckling behavior of the Figure 5 shows the buckling load predictions from various FEA
orthotropic strip is examined for two cases: first when the aspect ra- programs, as the aspect ratio of the strip is varied. The orthotropic
tio between the length and height of the strip varies and second strip is assumed to be deformed in a plane strain setting, and linear
when the ratio between the longitudinal and transverse direction quadrilateral plane strain elements are used for the buckling FE
stiffness changes. Material properties for the two cases are analyses. FEM A and FEM B are the finite element programs that
are based on the study by Ji et al. [4]. FEM A denotes the finite
element method based on the m ¼ 2 formulation and FEM B is
based on the m ¼ 0 formulation, but used with the constant tan-
gential moduli incorrectly. In FEM B, the constant tangential
moduli (that should be paired with the m ¼ 2 formulation) are
intentionally used to demonstrate the effect of the improper use of
the nonwork conjugate pairs. Various commercial FEA packages
are also used to compute the buckling loads of the orthotropic
strip, whose results are shown in Fig. 5. *BUCKLE and CPE4 ele-
ments are used in the ABAQUS 6.11 version. In other codes, for
the buckling analysis, ANTYPE 1 of ANSYS 13.0 is employed
Fig. 4 Orthotropic strip under uniform axial compression with static analysis to obtain the prebuckling stress state of the

041024-6 / Vol. 80, JULY 2013 Transactions of the ASME

Downloaded From: http://appliedmechanics.asmedigitalcollection.asme.org/ on 07/24/2013 Terms of Use: http://asme.org/terms

 Fig. 6 Errors in predicting buckling loads shown in Fig. 5 Fig. 8 Errors in predicting buckling loads for a thick ortho-
tropic strip shown in Fig. 7

Fig. 9 Configuration of an axially compressed sandwich panel

analytical solutions and other FEA programs. In the FEM B

formulation with m ¼ 0 and constant tangential moduli, an extra
volumetric term enters in the total integration for the internal vir-
tual work [4,10]. When the strip has strong orthotropic properties,
the effects from this nonvanishing term become significant, as
Fig. 7 Buckling load predictions from various commercial FEA shown in Fig. 8, since the volumetric deformation is directionally
programs as a function of the stiffness ratio between the longi- dependent.
tudinal and transverse direction

4.2 Buckling Load Prediction for a Sandwich Structure

strip. SOL 600 SEBUCKLE of NASTRAN 2011 with the Marc With Orthotropic Constituents. A sandwich beam or panel is a
solver is utilized to account for the orthotropic material properties typical orthotropic structure that is widely used in practical engi-
in a plane strain deformation. LS-Dyna also provides bifurcation neering applications. Since finite element analysis is now com-
buckling analysis through *CONTROL_IMPLICIT_BUCKLE. monly used to evaluate the mechanical performance of sandwich
Analytical buckling loads are also obtained from the theoretical structures, commercial codes that are used to perform structural
elasticity-based formulation shown in Ref. [10]. analysis containing orthotropic sandwich structures need careful
As shown in Fig. 5, both ABAQUS and ANSYS overpredict examination. Figure 9 illustrates the loading and boundary config-
the buckling loads, but produce the same identical results as those uration of the sandwich panel to be studied here. These are similar
from FEM B. Note that ABAQUS is based on the Jaumann rate of to the orthotropic strip studied before. Throughout the presenta-
the Cauchy stress with constant moduli, as we identified in Sec. 3. tion, the superscript “f ” represents quantities associated with the
Figure 6 shows the errors in predicting the buckling loads of face sheets and “c” with the core. The face sheets and core are
Fig. 5. The relative errors are calculated with respect to the solu- both orthotropic. The top and bottom face sheets are assumed to
tions from elasticity theory. The errors increase as the strip have the same geometric and material properties. The sandwich
becomes thicker, i.e., as the significance of the orthotropic proper- panel is subjected to a plane strain deformation in the xy plane.
ties become more important. The FE model is created using quadrilateral plane strain elements.
Figure 7 shows the buckling load predictions from various FEA The results from an exact 2D elasticity solution [10,11], which
programs when the material properties vary. Figure 8 shows the serves as a benchmark, are used to evaluate the results obtained
relative errors in the computed buckling loads with respect to the from the FE commercial codes.
analytical solutions. Again, the buckling loads predicted from Figure 10 compares buckling loads computed as a function of
FEM B, ABAQUS, and ANSYS are higher than those from the the stiffness ratio (between the longitudinal and transverse

Journal of Applied Mechanics JULY 2013, Vol. 80 / 041024-7

Downloaded From: http://appliedmechanics.asmedigitalcollection.asme.org/ on 07/24/2013 Terms of Use: http://asme.org/terms

 Fig. 11 Comparison of the buckling load of a sandwich panel
Fig. 10 Comparison of the buckling load of a sandwich panel
as a function of the panel aspect ratio
as a function of the stiffness ratio of the core between the trans-
verse and longitudinal direction
Table 4 Material properties of the face sheets and the core for
the sandwich panel in Fig. 11
Table 3 Material properties of the face sheets and the core in
Fig. 10 Property Value
Property Value Efxx ðGPaÞ 107
Efxx ðGPaÞ 107 Efyy ðGPaÞ 15
Efyy ðGPaÞ 15 Face sheet Gfxy ðGPaÞ 4.3
f
Face sheet Gfxy ðGPaÞ 4.3 xy 0.3
f
xy 0.3 hf (mm) 0.1
f
h (mm) 0.1 Ecxx ðGPaÞ 480  Gcxy
Ecyy ðGPaÞ 4  Gcxy Ecyy ðGPaÞ 4  Gcxy

Core Gcxy ðGPaÞ Gfxy =200 Core Gcxy ðGPaÞ Gfxy =200
c
c
xy 0.25 xy 0.25
c
h ðmmÞ 0.8 hc ðmmÞ 0.8

problem, becomes significant when the core deformation is not

direction) of the core. The material properties used for the results negligible. Ji and Waas [10,11] have shown that, when the core
in Fig. 10 are listed in Table 3. The orthotropic material properties has almost the same order of stiffness as the face sheets, predicted
of the face sheets are obtained from Ref. [12]. The aspect ratio of buckling loads from improper formulations result in large errors.
the sandwich structure, L=h, is set to 3 for this example. Buckling One cannot simply neglect the axial load-carrying capacity of the
loads in Fig. 10 are normalized by PB defined as core. Furthermore, when the constituents of a sandwich structure
are orthotropic, the incorrect formulation is again seen to greatly
PE overpredict buckling loads, even for sandwich panels with rela-
PB ¼ (52)
1 þ PE =GA tively soft cores, as shown in the examples here.

where PE ¼ 4p2 EI=L2 ¼ Euler’s buckling load, EI ¼ equivalent

bending stiffness of a sandwich panel, and GA ¼ equivalent shear 5 Conclusions
resistance of a sandwich panel [13]. In Fig. 10, the buckling loads We have shown that neglecting to account for proper stress
from FEM B, ABAQUS, and ANSYS are seen to be higher than rates and work-conjugacy can lead to significant errors in finite
the results from FEM A, NASTRAN, and LS-Dyna as the core element-based analysis when used for problems that require an
stiffness ratio increases. The latter results agree with the exact 2D incremental approach to obtaining a solution. By way of two
elasticity benchmark. It appears that, with increasing orthotropy examples, we have demonstrated that nonphysical solutions and
of the core, the errors from the improper formulation (FEM B) incorrect critical loads are predicted by popular commercial codes
become significant. that are now routinely adopted in structural analysis. We have
Figure 11 shows the buckling load comparison as a function of shown that these codes can easily be fixed by abandoning the Jau-
the aspect ratio (panel length to height ratio). Table 4 lists the ma- mann rate of Cauchy stress and instead using the Truesdell rate
terial properties used in Fig. 11. FEM B, ABAQUS, and ANSYS with the correct, constant tangential moduli. Various types of
again overpredict the buckling loads as the sandwich structure stress rates and the corresponding internal work expressions have
becomes thicker. As discussed earlier, the effect of the nonvanish- also been presented, and it is noted that most stress rates are
ing term, resulting from the incorrect formulation of the buckling paired with tangential moduli that are dependent on the current

041024-8 / Vol. 80, JULY 2013 Transactions of the ASME

Downloaded From: http://appliedmechanics.asmedigitalcollection.asme.org/ on 07/24/2013 Terms of Use: http://asme.org/terms

 stress. The internal work will be defined correctly only if proper References
tangential moduli are used when a specific type of stress rate is [1] Bazant, Z. P., 1971, “Correlation Study of Formulations of Incremental Defor-
chosen. The choice among the different types of stress rates could mation and Stability of Continuous Bodies,” ASME J. Appl. Mech., 38(4), pp.
be arbitrary, but the internal work should be defined properly, 919–928.
[2] Bazant, Z. P., and Cedolin, L., 1991, Stability of Structures: Elastic, Inelastic,
regardless of the choice. It is shown that buckling loads of thick Fracture and Damage Theories, Oxford University Press, New York.
orthotropic structures can be significantly overpredicted if the [3] Hill, R., 1959, “Some Basic Principles in Mechanics of Solids Without Natural
proper pairing for the internal work definition is not used. Various Time,” J. Mech. Phys. Solids, 7(3), pp. 209–225.
commercial FEA programs are examined here, and ABAQUS and [4] Ji, W., Waas, A. M., and Bazant, Z. P., 2010, “Errors Caused by Non-Work-
Conjugate Stress and Strain Measures and Necessary Corrections in Finite Ele-
ANSYS are identified as codes that adopt erroneous formulations ment Programs,” ASME J. Appl. Mech., 77(4), p. 044504.
based on the Jaumann rate of the Cauchy stress with constant tan- [5] Bazant, Z. P., Gattu, M., and Vorel, J., 2012, “Work Conjugacy Error in Com-
gential moduli. As shown here, this causes significant errors in the mercial Finite Element Codes: Its Magnitude and How to Compensate for It,”
class of examples studied here. On the other hand, NASTRAN Proc. R. Soc. A, 468(2146), pp. 3047–3058.
[6] Dienes, J. K., 1979, “On the Analysis of Rotation and Stress Rate in Deforming
with the Marc solver and LS-DYNA have internal formulations Bodies,” Acta Mech., 32(4), pp. 217–232.
that correctly predict the buckling loads of thick orthotropic struc- [7] Lin, R., 2002, “Numerical Study of Consistency of Rate Constitutive Equations
tures; however, LS-DYNA predictions are inaccurate for the nonlin- With Elasticity at Finite Deformation,” Int. J. Numer. Methods Eng., 55, pp.
ear incremental analysis related to the simple shear problem. 1053–1077.
[8] Johnson, G. C., and Bammann, D. J., 1984, “A Discussion of Stress Rates in Fi-
NASTRAN, which is widely adopted for analyzing thin-wall struc- nite Deformation Problems,” Int. J. Solids Struct., 20(8), pp. 725–737.
tures, for which the significance of the errors that have been pointed [9] Rivlin, R. S., 1948, “Large Elastic Deformations of Isotropic Materials. IV. Fur-
out are not important, is seen to implement the correct FE formula- ther Developments of the General Theory,” Philos. Trans. R. Soc. London, Ser.
tion for both classes of problems that have been examined in this pa- A, 241, pp. 379–397.
[10] Ji, W., and Waas, A. M., 2009, “2D Elastic Analysis of the Sandwich Panel
per. We emphasize again that the switch to a fully work-conjugate Buckling Problem: Benchmark Solutions and Accurate Finite Element For-
formulation would be easy and this is highly recommended [4]. mulations,” Z. Angew. Math. Phys., 61(5), pp. 897–917.
[11] Ji, W., and Waas, A. M., 2012, “Accurate Buckling Load Calculations of a
Thick Orthotropic Sandwich Panel,” Compos. Sci. Technol., 72(10), pp.
1134–1139.
Acknowledgment [12] Fagerberg, L., 2004, “Wrinkling and Compression Failure Transition in Sand-
wich Panels,” J. Sandwich Struct. Mater., 6(2), pp. 129–144.
We are grateful to Professor R. McMeeking, University of Cali- [13] Bazant, Z. P., and Beghini, A., 2006, “Stability and Finite Strain of Homoge-
fornia, Santa Barbara, for thoughtful comments on an earlier ver- nized Structures Soft in Shear: Sandwich or Fiber Composites and Layered
sion of the paper that led to improvement of our presentation. Bodies,” Int. J. Solids Struct., 43(6), pp. 1571–1593.

Journal of Applied Mechanics JULY 2013, Vol. 80 / 041024-9

Downloaded From: http://appliedmechanics.asmedigitalcollection.asme.org/ on 07/24/2013 Terms of Use: http://asme.org/terms