15th International LS-DYNA® Users Conference Implicit

An Enhanced Assumed Strain (EAS)

Solid Element for Nonlinear Implicit Analyses
Fredrik Bengzonϯ, Thomas Borrvallϯ and Ushnish Basuϯϯ
ϯ
DYNAmore Nordic AB
ϯϯ
Livermore Software Technology Corporation (LSTC)

Abstract
Historically, the importance of computational efficiency in explicit analysis has driven the element development in LS-DYNA® [1,2]
towards fast and sufficiently accurate formulations. Single point integrated elements with stabilization are well established techniques
in this area. The recent growth of implicit analysis has led to a demand of increased accuracy of the element response, and
consequently more sophisticated formulations have been introduced in recent years. While high order elements provide a better
response, low order elements remain popular due to their simplicity and robustness. An area that has not yet been exploited in
LS-DYNA is the family of enhanced assumed strain (EAS) elements, the reason being the computational cost associated with this
approach. Solid element 18 is a linear Wilson element based on this technology, but is only available for linear implicit analysis. The
goal with this paper is to generalize this to fully nonlinear implicit analysis, and provide information on its merits and drawbacks.

Introduction
Isoparametric Strain Elements
The design of a finite element is essentially the design of its deformation gradient 𝑭𝑭, the mother of all strains.
The deformation gradient is in a continuum mechanical context defined as

𝜕𝜕𝒙𝒙
𝑭𝑭 =
𝜕𝜕𝑿𝑿

where 𝒙𝒙 is the current spatial location for a given reference material coordinate 𝑿𝑿, and is a measure of the
stretch and rotation of an infinitesimal line element in the body. Based on 𝑭𝑭, examples of different strain
measures are

𝜺𝜺 = 𝑙𝑙𝑙𝑙𝑭𝑭𝑇𝑇 𝑭𝑭 (natural strain)

1
𝑬𝑬 = (𝑭𝑭𝑇𝑇 𝑭𝑭 − 𝑰𝑰) (Green strain)
2
𝑫𝑫 = 𝑠𝑠𝑠𝑠𝑠𝑠𝑭𝑭̇𝑭𝑭−1 (rate of deformation)
1
𝒆𝒆 = (𝑰𝑰 − 𝑭𝑭−𝑇𝑇 𝑭𝑭−1 ) (Almansi strain).
2

June 10-12, 2018 1

15th International LS-DYNA® Users Conference Implicit

For a given strain measure we can imagine that all possible strains generated by arbitrary deformations
constitutes a strain space ℰ, so for instance 𝑬𝑬 ∈ ℰ if we happen to work with the Green strain measure. Now,
when a continuum body is discretized into finite elements, each element is in the simplest setting represented by
𝑚𝑚 coordinate points 𝒙𝒙𝑖𝑖 and associated isoparametric shape functions 𝑁𝑁𝑖𝑖 , 𝑖𝑖 = 1, … , 𝑚𝑚. The spatial location of a
material point in the element is written as

𝒙𝒙 = 𝒙𝒙𝑖𝑖 𝑁𝑁𝑖𝑖

and a direct application results in

𝜕𝜕𝑁𝑁𝑖𝑖
𝑭𝑭 = 𝒙𝒙𝑖𝑖 .
𝜕𝜕𝑿𝑿

The consequence of the discretization is that the numerical strains using this latter 𝑭𝑭 are restricted to a finite
dimensional subspace ℰℎ of the space of continuum strains ℰ, and the quality of the element comes down to the
richness of this subspace ℰℎ ⊂ ℰ. To this end, for a given level of discretization ℎ, the subspace is to most
extent defined by the choice of shape functions 𝑁𝑁𝑖𝑖 . It is for instance known that high order polynomials tend to
improve the situation as they appropriately enlarge the space of strains.

Considering low order hexahedral elements, for which the shape functions are the standard tri-linear
isoparametric shape functions, the methodology presented above inevitably leads to locking phenomena, i.e., the
element becomes too stiff. In short, this is explained by the element’s inability to represent a physical
deformation mode without exhibiting spurious strain. Typically, a pure bending deformation is accompanied
with either excessive volumetric or shear strain that absorbs energy and stiffens the response. To use a phrasing
that reconnects to the discussion above, the strain space ℰℎ is too small.

Assumed Strain Elements

A way to deal with locking is to augment the deformation gradient based on reasonable assumptions, leading to
an assumed strain element with deformation gradient 𝑭𝑭𝑎𝑎𝑎𝑎 . There are restrictions associated with such a
modification, a crucial one is to maintain frame invariance by making sure that the assumed velocity gradient
𝑳𝑳𝑎𝑎𝑎𝑎 = 𝑭𝑭̇𝑎𝑎𝑎𝑎 𝑭𝑭−1
𝑎𝑎𝑎𝑎 is skew-symmetric whenever elements undergo rigid body motion. A not so severe restriction is
that the element satisfies various patch tests commonly found in the literature. A first example of such an
approach is to assume that the deformation gradient is constant in an element and given as either its value in the
� . This will on the one hand alleviate locking but on the other
center, 𝑭𝑭𝑎𝑎𝑎𝑎 = 𝑭𝑭0 , or its integrated mean, 𝑭𝑭𝑎𝑎𝑎𝑎 = 𝑭𝑭
render an incomplete element in the sense that non-rigid deformation modes won’t generate strains, i.e., the
symmetric part of 𝑳𝑳𝑎𝑎𝑎𝑎 is zero. Such elements require artificial (hourglass) stabilization and are available as the
type 1 element in LS-DYNA. A complete element is resulted from assuming that only the volumetric part of the
deformation gradient is constant, and the deviatoric part unaffected. This is the so called selective reduced
integration element available as type 2 in LS-DYNA. We either have 𝑭𝑭𝑎𝑎𝑎𝑎 = (𝐽𝐽0 /𝐽𝐽)1/3 𝑭𝑭 or 𝑭𝑭𝑎𝑎𝑎𝑎 = (𝐽𝐽/̅ 𝐽𝐽)1/3 𝑭𝑭
depending on the choice of averaging, where 𝐽𝐽 = 𝑑𝑑𝑑𝑑𝑑𝑑𝑭𝑭 is the jacobian. The choice of 𝐽𝐽 ̅ is preferred over 𝐽𝐽0 as
this element satisfies the patch test, the other does not. This element alleviates volumetric locking but still
exhibits stiff behavior in shear, in particular for elements with poor aspect ratio. A heuristic approach to deal
with the latter is presented in [3], and the corresponding element is available as types -1 and -2 in LS-DYNA.
Those elements do not satisfy the patch test.

June 10-12, 2018 2

15th International LS-DYNA® Users Conference Implicit

Enhanced Assumed Strain Elements

Pioneered by Juan Simo and co-workers in the late 80’s and early 90’s, the concept of enhanced assumed strain
elements emerged in the finite element community [4]. For these elements the deformation gradient is modified
by adding an enhanced contribution 𝑭𝑭𝑒𝑒

𝑭𝑭𝑒𝑒𝑒𝑒𝑒𝑒 = 𝑭𝑭 + 𝑭𝑭𝑒𝑒 .

The enhanced contribution 𝑭𝑭𝑒𝑒 is in turn a function of an independent field 𝜶𝜶 that can be seen as representing
deformation modes in which the element needs to be relaxed, i.e., this is a way to enlarge the strain space. The
presence of 𝜶𝜶 implies the existence of a work conjugate field 𝒇𝒇𝛼𝛼 , and appropriate application of the virtual work
principle results in that an additional set of equations, 𝒇𝒇𝛼𝛼 = 𝟎𝟎, needs to be solved. Fortunately, 𝜶𝜶 is assumed
discontinuous over element boundaries, which means that this compatibility equation can be solved locally for
each element and the size of the global system of equations is thus unaffected. Nevertheless, solving many local
sets of nonlinear equations adds a significant overall cost to evaluating the nodal forces, and is the main reason
why it has not been given much attention in an explicit finite element context. For implicit time integration the
story may be told differently, the price of spending more time in element routines may give payback in the
currency of better results and larger time steps. The purpose of this paper is to investigate this potential.

Theory
General
A framework for the design of enhanced assumed strain elements sufficient for satisfying frame invariance and
patch tests was introduced by Simo and Rafai [4]. Within this framework, Simo et.al. [5] developed some
specific elements with interesting properties, and this is an attempt to summarize the equations behind one of
these. For this we treat a single element occupying a region 𝛺𝛺 in space, and emphasize that the notation used
from hereon is independent of the notation used in previous sections.

Let 𝒙𝒙 denote the coordinate vector for the nodes in the current configuration, 𝒙𝒙𝑛𝑛 the coordinate vector in the
previous (converged) configuration and 𝒙𝒙0 the coordinate vector in the reference configuration. We then let 𝒖𝒖 =
𝒙𝒙 − 𝒙𝒙𝑛𝑛 and 𝒗𝒗 = 𝒖𝒖/Δ𝑡𝑡 be the displacement and velocity vectors, respectively, where Δ𝑡𝑡 is the time step. All
vectors mentioned so far are of dimension 24, 8 nodes times 3 degrees of freedom per node. The element is
characterized by a vector 𝜶𝜶 of dimension 12 representing the enhanced deformation modes, and we likewise
use 𝜶𝜶𝑛𝑛 to denote its value in the previous configuration, and use Δ𝜶𝜶 = 𝜶𝜶 − 𝜶𝜶𝑛𝑛 for its increment. There are 4
enhanced modes, each represented by a triplet 𝜶𝜶𝑖𝑖 , 𝑖𝑖 = 1, … ,4.

The enhanced deformation gradient is denoted 𝑭𝑭 = 𝑭𝑭(𝒙𝒙, 𝜶𝜶), and the constitutive law takes a strain measure as
input to compute or increment the resulting Cauchy stress 𝝈𝝈. Through a work principle, where we use 𝛿𝛿1 to
denote the first variation, the identification of 𝒇𝒇𝑥𝑥 and 𝒇𝒇𝛼𝛼 can be made by the relation

𝛿𝛿1 𝒙𝒙𝑇𝑇 𝒇𝒇𝑥𝑥 + 𝛿𝛿1 𝜶𝜶𝑇𝑇 𝒇𝒇𝛼𝛼 = � (𝛿𝛿1 𝑭𝑭𝑭𝑭−1 ): 𝝈𝝈𝑑𝑑Ω.

Ω

June 10-12, 2018 3

15th International LS-DYNA® Users Conference Implicit

The stiffness matrix terms can likewise be identified through the second variation 𝛿𝛿2 of this work principle

𝜕𝜕𝒇𝒇𝑥𝑥 𝜕𝜕𝒇𝒇𝑥𝑥 𝜕𝜕𝒇𝒇𝛼𝛼 𝜕𝜕𝒇𝒇𝛼𝛼

𝛿𝛿1 𝒙𝒙𝑇𝑇 𝛿𝛿2 𝒙𝒙 + 𝛿𝛿1 𝒙𝒙𝑇𝑇 𝛿𝛿2 𝜶𝜶 + 𝛿𝛿1 𝜶𝜶𝑇𝑇 𝛿𝛿2 𝒙𝒙 + 𝛿𝛿1 𝜶𝜶𝑇𝑇 𝛿𝛿 𝜶𝜶 = ⋯ =
𝜕𝜕𝒙𝒙 𝜕𝜕𝜶𝜶 𝜕𝜕𝒙𝒙 𝜕𝜕𝜶𝜶 2
= � �𝛿𝛿2 𝛿𝛿1 𝑭𝑭𝑭𝑭−1 + (𝛿𝛿2 𝑭𝑭𝑭𝑭−1 )𝑇𝑇 (𝛿𝛿1 𝑭𝑭𝑭𝑭−1 )�: 𝝈𝝈𝑑𝑑Ω + � (𝛿𝛿1 𝑭𝑭𝑭𝑭−1 ): 𝑪𝑪: (𝛿𝛿2 𝑭𝑭𝑭𝑭−1 )𝑑𝑑Ω.
Ω Ω

Worth noticing here is that for this relation to hold, 𝑪𝑪 is the constitutive modulus relating the Truesdell rate of
Cauchy stress to the rate-of-deformation. Following Simo et.al. [5], all integrals above are numerically
integrated using a 9 point integration scheme.

The first of the force vectors, 𝒇𝒇𝑥𝑥 , should balance the other forces in the problem, while the second, 𝒇𝒇𝛼𝛼 , should
solve the compatibility equation 𝒇𝒇𝛼𝛼 = 𝟎𝟎 for the current value of 𝜶𝜶. For this, we use a Newton method and the
𝜕𝜕𝒇𝒇
corresponding Hessian 𝛼𝛼 . The elimination of 𝜶𝜶 on the element level requires that the global tangent stiffness
𝜕𝜕𝜶𝜶
matrix is condensated accordingly, we have

𝑑𝑑𝒇𝒇𝑥𝑥 𝜕𝜕𝒇𝒇𝑥𝑥 𝜕𝜕𝒇𝒇𝑥𝑥 𝜕𝜕𝜶𝜶

𝑲𝑲𝑥𝑥 = = +
𝑑𝑑𝒙𝒙 𝜕𝜕𝒙𝒙 𝜕𝜕𝜶𝜶 𝜕𝜕𝒙𝒙
𝜕𝜕𝜶𝜶
where 𝜕𝜕𝒙𝒙 is obtained through the variation of the compatibility equation

𝑑𝑑𝒇𝒇𝛼𝛼 𝜕𝜕𝒇𝒇𝛼𝛼 𝜕𝜕𝒇𝒇𝛼𝛼 𝜕𝜕𝜶𝜶 𝜕𝜕𝜶𝜶 𝜕𝜕𝒇𝒇𝛼𝛼 −1 𝜕𝜕𝒇𝒇𝛼𝛼

= + = 𝟎𝟎 ⇒ = −� �
𝑑𝑑𝒙𝒙 𝜕𝜕𝒙𝒙 𝜕𝜕𝜶𝜶 𝜕𝜕𝒙𝒙 𝜕𝜕𝒙𝒙 𝜕𝜕𝜶𝜶 𝜕𝜕𝒙𝒙

so

𝜕𝜕𝒇𝒇𝑥𝑥 𝜕𝜕𝒇𝒇𝑥𝑥 𝜕𝜕𝒇𝒇𝛼𝛼 −1 𝜕𝜕𝒇𝒇𝛼𝛼

𝑲𝑲𝑥𝑥 = − � � .
𝜕𝜕𝒙𝒙 𝜕𝜕𝜶𝜶 𝜕𝜕𝜶𝜶 𝜕𝜕𝒙𝒙

Specifics
The assumed deformation gradient can be expressed as

𝜕𝜕𝑁𝑁𝑖𝑖 −1 �𝑗𝑗
𝐽𝐽0 𝜕𝜕𝑁𝑁 �4
𝐽𝐽0 𝜕𝜕𝑁𝑁 𝜕𝜕𝑁𝑁𝑖𝑖 −1
𝑭𝑭 = �𝒙𝒙𝑖𝑖 𝑱𝑱𝜉𝜉 � + �𝜶𝜶𝑗𝑗 � � 𝑱𝑱−1
0 � + ��� � 𝑱𝑱−1
0 𝜶𝜶4 � 𝒙𝒙𝑖𝑖 � 𝑱𝑱 � ,
𝜕𝜕𝝃𝝃 𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺
𝐽𝐽𝜉𝜉 𝜕𝜕𝝃𝝃 𝐽𝐽𝜉𝜉 𝜕𝜕𝝃𝝃 𝜕𝜕𝝃𝝃 0 0
𝑊𝑊𝑊𝑊𝑊𝑊𝑊𝑊𝑊𝑊𝑊𝑊 𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉

where the first term is the standard Galerkin term and the remaining are the enhanced contributions. We here
use 𝒙𝒙𝑖𝑖 to denote the coordinate vector of node 𝑖𝑖, 𝑖𝑖 = 1, … ,8, and 𝜶𝜶𝑗𝑗 to denote the 𝑗𝑗th enhanced variable, 𝑗𝑗 =
1 … ,4. All these vectors are column vectors of dimension 3. The shape functions associated with the enhanced
contributions are the so called Wilson incompatible shape functions

1 2
�𝑗𝑗 =
𝑁𝑁 �𝜉𝜉 − 1�,
2 𝑗𝑗

June 10-12, 2018 4

15th International LS-DYNA® Users Conference Implicit

for 𝑗𝑗 = 1,2,3, to relax bending deformations, and a function intended to alleviate volumetric locking

�4 = 𝜉𝜉1 𝜉𝜉2 𝜉𝜉3 .

𝑁𝑁

We also use 𝑱𝑱𝜉𝜉 /𝐽𝐽𝜉𝜉 and 𝑱𝑱0 /𝐽𝐽0 to denote the jacobian matrix/determinant between the iso-parametric and
reference domains, and subscript 0 means that a quantity is evaluated at the center point of the domain. Noting
𝜕𝜕(°)
that is a row vector of dimension 3, all operations in the expression for 𝑭𝑭 can be viewed as standard matrix
𝜕𝜕𝝃𝝃
notation. The variation operator is denoted 𝛿𝛿1 , and this applied on 𝑭𝑭 becomes

𝜕𝜕𝑁𝑁𝑖𝑖 −1 𝐽𝐽0 𝜕𝜕𝑁𝑁 �4 𝜕𝜕𝑁𝑁𝑖𝑖 −1 �𝑗𝑗

𝐽𝐽0 𝜕𝜕𝑁𝑁
𝛿𝛿1 𝑭𝑭 = 𝛿𝛿1 𝒙𝒙𝑖𝑖 � 𝑱𝑱𝜉𝜉 + �� � 𝑱𝑱−1 𝜶𝜶 � � 𝑱𝑱 � + 𝛿𝛿 𝜶𝜶 �� � 𝑱𝑱−1 �
𝜕𝜕𝝃𝝃 𝐽𝐽𝜉𝜉 𝜕𝜕𝝃𝝃 0 4 𝜕𝜕𝝃𝝃 0 0 1 𝑗𝑗
𝐽𝐽𝜉𝜉 𝜕𝜕𝝃𝝃 0
�4
𝐽𝐽0 𝜕𝜕𝑁𝑁 𝜕𝜕𝑁𝑁𝑖𝑖 −1
+ �� � 𝑱𝑱−1
0 𝛿𝛿1 𝜶𝜶4 � �𝒙𝒙𝑖𝑖 � 𝑱𝑱 �,
𝐽𝐽𝜉𝜉 𝜕𝜕𝝃𝝃 𝜕𝜕𝝃𝝃 0 0

where sum over 𝑖𝑖 is from 1 to 8, while sum over 𝑗𝑗 is from 1 to 3. Because of the product term involving 𝒙𝒙 and
𝜶𝜶, the second variation is non-vanishing, denoting this 𝛿𝛿2 we have

�4
𝐽𝐽0 𝜕𝜕𝑁𝑁 𝜕𝜕𝑁𝑁𝑖𝑖 −1 �4
𝐽𝐽0 𝜕𝜕𝑁𝑁 𝜕𝜕𝑁𝑁𝑖𝑖 −1
𝛿𝛿2 𝛿𝛿1 𝑭𝑭 = ��� � 𝑱𝑱−1
0 𝛿𝛿1 𝜶𝜶4 � 𝛿𝛿2 𝒙𝒙𝑖𝑖 � 𝑱𝑱0 + �� � 𝑱𝑱−1
0 𝛿𝛿2 𝜶𝜶4 � 𝛿𝛿1 𝒙𝒙𝑖𝑖 � 𝑱𝑱 �.
𝐽𝐽𝜉𝜉 𝜕𝜕𝝃𝝃 𝜕𝜕𝝃𝝃 0 𝐽𝐽𝜉𝜉 𝜕𝜕𝝃𝝃 𝜕𝜕𝝃𝝃 0 0

These variations can be inserted into the expressions for nodal forces and stiffnesses in previous sections and
completes the basic theory of the element.

Modifications
Two modifications to the equations given above were done.

In order to alleviate volumetric locking for distorted meshes in the incompressible limit, the following
modification of the Galerkin term of the deformation gradient is used,

𝜕𝜕𝑁𝑁𝑖𝑖 −1 𝐽𝐽0 𝜕𝜕𝓗𝓗 −1

𝑭𝑭𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺𝐺 = 𝒙𝒙𝑖𝑖 � 𝑱𝑱0 + 𝒙𝒙𝑖𝑖 𝜸𝜸𝑖𝑖 � � 𝑱𝑱 ,
𝜕𝜕𝝃𝝃 0 𝐽𝐽𝜉𝜉 𝜕𝜕𝝃𝝃 0

where 𝜸𝜸𝑖𝑖 are the familiar 𝛾𝛾-vectors (row vector of size 4 for each 𝑖𝑖) and 𝓗𝓗 = (𝜉𝜉2 𝜉𝜉3 𝜉𝜉1 𝜉𝜉3 𝜉𝜉1 𝜉𝜉2 𝜉𝜉1 𝜉𝜉2 𝜉𝜉3 )𝑇𝑇
the associated hourglass functions commonly used in the representation of low order hexahedral elements. The
justification behind this modification comes from studying the incompressibility constraint 𝑰𝑰: 𝛿𝛿1 𝑭𝑭 = 0 to see
that nonzero deformations are trivially allowed with this modification, a discussion can be found in [5].

To suppress known numerical instabilities for highly compressive deformations, the theory above is applied to
the incremental deformation gradient ∆𝑭𝑭. To be specific, the total deformation gradient 𝑭𝑭 is given by

𝑭𝑭 = ∆𝑭𝑭𝑭𝑭𝑛𝑛

where 𝑭𝑭𝑛𝑛 is the deformation gradient in the previous time step, treated as an internal history variable, and the
enhanced assumed strain theory is applied to ∆𝑭𝑭 using the last known configuration as the reference
configuration.

June 10-12, 2018 5

15th International LS-DYNA® Users Conference Implicit

Examples
To validate the element we present and discuss some examples.

Clamped Case B

𝑧𝑧 Case A

𝑥𝑥 𝑦𝑦

Figure 1 Distorted beam with load cases

Distorted Beam
The beam shown in Figure 1 is equipped with an elastic material, 𝐸𝐸 = 1500 and 𝜐𝜐 = 0.25, and is an example to
test the element sensitivity to a distorted mesh. The beam, of dimension 2 × 2 × 10, is clamped at one end, and
on the other a bending torque of 40 (case A) and a transverse load of 6 (case B) is applied. The analytical end
tip displacement for case A is 1 and for case B 1.026 and Table 1 summarizes the numerical results.

Table 1 Z-displacement for distorted beam

Element Type Case A Case B

1 – Belytchko-Bindeman HG 0.780 0.829
1 – Cosserat HG 0.976 1.002
2 – S/R integration 0.782 0.869
18 – EAS 0.911 0.934
24 – Quadratic 0.989 1.008

Not surprisingly the high order element provides the most accurate response, followed by the Cosserat element.
The Cosserat element is a low order element with a sophisticated hourglass treatment accounting for the mesh
distortion in the reference configuration, which may explain the good result. Among the other elements the EAS
element presented here is most accurate, and the results compare well with those in [6]. In general low order
elements seem sensitive to the shape of the elements, and even though the Cosserat element performs quite well
it is still an hourglass element and may not exhibit correct stress response for larger deformations.

Near-Incompressible Block
The block shown in Figure 2 is also equipped with an elastic material, but nearly incompressible with 𝜐𝜐 =
0.4999, and 𝐸𝐸 = 210000. This is an example to test how stiff the element as well as its performance under
constrained conditions. The block has a height of 50 and length and width 100, but only one quarter is meshed
due to symmetry. The base of the block is clamped and a uniform distributed pressure of 250, acting on an area
of 20 × 20, is applied in the center of the top. Two meshes are tested, both with 5 × 5 × 5 elements, but one
structured (case A) and one slightly distorted (case B). The vertical displacement of the top center node is
summarized in Table 2 along with results from [6]. In comparison, the EAS element performs well, and has a
reasonable resistance against mesh distortion.
June 10-12, 2018 6
15th International LS-DYNA® Users Conference Implicit

Figure 2 Near-incompressible block subject to a pressure load

Table 2 Z-displacement for near-incompressible block

Element Type Case A Case B
1 – Belytschko-Bindeman HG -1.905e-2 -1.914e-2
2 – S/R integration -1.966e-2 -1.972e-2
18 – EAS -1.892e-2 -1.834e-2
Reference [6] -1.892e-2 -1.840e-2

A shortcoming of EAS methods is that they may exhibit instabilities in a state of finite deformation [7], often
under non-linear or severely constrained conditions. Figure 3 depicts one such situation and shows a cube
equipped with an elasto-plastic material deforming under compression. After a while hour-glassing develops in
the compressed region and convergence breaks down.

Figure 3 Hourglass instability in the EAS element

June 10-12, 2018 7
15th International LS-DYNA® Users Conference Implicit

Clamped

𝑀𝑀

Figure 4 Straight cantilever beam subjected to moment

Nonlinear Cantilever Beam

A straight cantilever beam with dimensions 𝐻𝐻 × 𝐵𝐵 × 𝐿𝐿 = 0.1 × 1 × 10 is discretized into 25 elements along its
𝐸𝐸𝐸𝐸𝐻𝐻 3 𝜋𝜋
length, see Figure 4. It is clamped at one end, and a moment 𝑀𝑀 = 6𝐿𝐿 is applied at the other that analytically
will transform the beam into a perfect cylinder. The material is hyperelastic with 𝐸𝐸 = 12000000 and zero
Poisson ratio.

Type 2

Type 18
and
Type 1
Cosserat

Type -2

Type 1 Belytschko-Bindeman
Figure 5 Final configuration of cantilever beam
Figure 5 shows the deformed configurations for some element types. Noticable is that the fully integrated
element type 2 locks in shear for this element geometry and load case, something that is alleviated by using type
-2. The latter is still not as good as the others, presumably because only one element through the thickness is
used and the constant pressure profile prohibits a proper bending response. The EAS element is together with
the Cosserat formulation giving the best result, albeit not perfect.

June 10-12, 2018 8

15th International LS-DYNA® Users Conference Implicit

Summary
A nonlinear extension of the linear type 18 element in LS-DYNA has been implemented and presented. It is
based on an enhanced assumed strain theory which is renowned for accurate response but suffering from
computationally intense calculations. The main intention is to offer this element as part of the nonlinear implicit
solver, for which the computational expense in element routines is not as devastating as in explicit analysis. At
the time of writing, the element routines are not optimized for speed and the overhead compared to other fully
integrated elements is in the order of 10 times, but there is good hope for improvements on that note. The
examples presented are fairly simple and serves the purpose of comparing the element to other available
formulations, we are yet to exercise it in more complex situations but hope this will be done in the near future.

References

1. LS-DYNA Keyword Manual, Volumes I-II, Livermore Software Technology Corporation.

2. T. Erhart, Review of solid element formulations in LS-DYNA, LS-DYNA Forum, October 12-13, 2011, Filderstadt, Germany.
3. T. Borrvall, A heuristic attempt to reduce transverse shear locking in fully integrated hexahedra with poor aspect ratio, Proc. 7th
European LS-DYNA Conference, Salzburg, 2009.
4. J.C. Simo and M.S. Rafai, A class of mixed assumed strain methods and the method of incompatible modes, Int. J. Numer.
Methods Eng. 29 (1990) pp. 1595-1638.
5. J.C. Simo, F. Armero and R.L. Taylor, Improved versions of assumed enhanced strain tri-linear elements for 3D finite
deformation problems, Comput. Methods in Appl. Mech. Eng. 110 (1993) pp. 359-386.
6. P.M.A. Areias, J.M.A. César de Sá, C.A. Conceição António and A.A. Fernandez, Analysis of 3D problems using a new
enhanced strain hexahedral element, Int. J. Numer. Methods Eng. 58 (2003) pp. 1637-1682.
7. P. Wriggers, Nonlinear Finite Element Methods, Springer Verlag (2008).

June 10-12, 2018 9