You are on page 1of 22

DISCUSSION ON THE APPLICATION OF ADVANCED FINITE ELEMENT

TECHNOLOGY TO DAM ENGINEERING

Mohammadreza Mostafa, PhD, PE, PMP 1


Vik Iso-Ahola, PE, PMP 2

ABSTRACT

The Finite Element (FE) analysis method has been widely used over the past few decades
to analyze and design a variety of structures including dams and appurtenant structures.
The use of increasingly robust and sophisticated finite element software combined with
high-speed computers have enabled engineers to model and analyze very complicated
structures, including most types of nonlinearities. Within available commercial FE
modeling software, there are multiple element types to choose with varying capabilities,
where each have been developed for solving specific problems.

Depending on the element type and the associated kinematic descriptions used in its
development, FE analysis results could vary notably. An engineer must understand the
behavior of each element type in order to apply the correct judgment in selecting the Finite
Element type and results that most closely represent the natural response of the analyzed
structure. As an example, “solid elements” have been vastly used to model three
dimensional structures such as concrete dams. Solid elements are susceptible to various
types of deficiencies, including “locking phenomena”, which may produce erroneous
outputs. There exist several methods to circumvent the potential deficiencies, which have
been implemented in commercial software as “options”. Each method has limitations that
should be considered and understood prior to implementation in the analysis of a structure.
Many engineers choose different options without fully considering and understanding the
purpose of each and the associated limitations.

This paper presents an overview of some deficiencies identified in the most common
elements that are used in commercial software, along with the theories of the methods that
are used to circumvent them. The goal of the paper is to provide a better understanding of
available options that exist in common FE programs. The paper concludes with an example
FE model for a hydraulic structure, demonstrating how different element types impact
results.

INTRODUCTION

The FE method has become a robust tool to analyze variety of structures including dams
and appurtenances over the past few decades. High speed computers along with
sophisticated FE software programs have reduced the analysis time for complicated
problems, from months and weeks to a few hours. The FE software programs have become
very user-friendly and do not require much knowledge of FE methods for performing
analyses. FE software employ many default settings (assumptions) and perform various

1
Stantec, 3301 C Street, Suite 1900, Sacramento, CA 95816, mohammadreza.mostafa@stantec.com.
2
Stantec, 2121 N. California Blvd., Suite 600, Walnut Creek, CA 94596, vik.iso-ahola@stantec.com.

Copyright © 2018 U.S. Society on Dams. All Rights Reserved. 1


hidden solution techniques to make the problem run and converge to a solution more easily.
For example, some software programs have “weak springs” to stabilize the models that
have some geometric instability. As another example, there is a common option in most
software programs as a default setting, or “program controlled”, which leaves it to the
program to decide on the procedure to use to solve a specific problem. Reliance upon
default settings could lead to applying procedures that are not appropriate for the problem
of interest, often without the engineer’s knowledge.

In addition, commercial FE programs consist of variety of elements, solution algorithms,


and constitutive relations for different materials, which are suitable for solving specific
problems, yet are not necessarily restricted, nor warn the user on the appropriate application
to certain problems.

This paper presents the inherent deficiencies of most common elements used in commercial
software for modeling structures, as well as the solutions provided to circumvent these
deficiencies. It should be noted that the topic of this paper has been investigated by many
researchers over the past five decades, and there are many papers, books and technical
reports on this subject matter. Accordingly, it is not possible to review or summarize all
the research and the corresponding mathematical formulations; the reader is encouraged to
further investigate through review of the listed references as well as general literature on
the subject. Also, it should be noted that the examples in this paper are limited to linear
elastic analyses, as well as some basic principles that are commonly missed by some users.

Element Technology

Solid Elements: Solid elements are the elements most commonly used to model structures
with two or three dimensional geometries. However, when applied to modeling of
structures, the well-known displacement-based formulation (linear displacement
interpolation) used in conventional solid elements is not rich enough to capture the
kinematics associated with the expected physical behavior of the structure. As a result,
solid elements are susceptible to several kinds of “locking phenomena” [1].

The locking problem first arose for four node quadrilateral and other lower-order elements,
which was circumvented with the development of higher-order elements, such as the 2D
eight node quadrilateral elements. Although these higher-order elements provided a good
solution to locking phenomena when initially developed, due to lack of fast computers and
lengthy analysis time associated with running models with these higher-order elements,
researchers started using techniques to prevent locking of lower-order elements from the
1970’s and onwards. However, even today, while there are some solutions or “tricks” for
the locking problem, its origins are not fully understood. The effort of alleviating the
locking problem was also extended to 3D solid elements with 8 nodes.

Locking can occur for a number of reasons and, for some element types, can depend on the
geometry of the element. In general, locking happens when an element cannot interpolate
a field property (e.g. bending or membrane behavior) correctly with the element’s shape
functions [1] (bi-linear and tri-linear for 2D and 3D elements respectively). Another source

Copyright © 2018 U.S. Society on Dams. All Rights Reserved. 2


for the locking problem could be the “mapping” of the geometry from Cartesian to Natural
coordinates during the development of the finite element formulation. The following
provides an overview of several locking problems of the 3D solid element:

• Membrane locking: Triggered when the element is subjected to in-plane


longitudinal or transverse shear loads, where low order shape functions are
incapable of capturing the correct in-plane bending response. This behavior is
exacerbated by element in-plane distortion as well as by high aspect ratios see
(Figure 1).

Figure 1. Tensile force (F) applied to patch of distorted mesh.

• Transverse shear locking: Occurs when the element thickness tends to zero, or
equivalently, the element has a high aspect ratio for length (L) versus thickness (t).
This is a very common deficiency when modeling thin structures (walls, slabs, thin
arches) subject to out-of-plane loads. In this case the structure becomes “stiff” and
the FE analysis does not properly simulate the structure’s physical response.

Figure 2. Moment (M) applied to thin solid element.

• Poisson’s thickness locking: Occurs when the across-the-thickness displacement is


assumed to vary linearly, hence the thickness strain becomes constant. However,
due to Poisson’s ratio effect, that strain couples with in-plane strains that vary
linearly across the thickness. This can occur when modeling thick shell-like
structures using elements with tri-linear shape functions.

• Curvature or trapezoidal locking: Occurs when the element edges in the thickness
are not perpendicular to the element mid-plane. This happens when modeling
curved structures (thin arch or multiple arch dams) with thickness-tapered solid
elements.

Figure 3. Out-of-plane force (F) applied to curved structure.

• Volumetric locking: Occurs when dealing with incompressible material with


Poisson’s ratio tending to 0.5. Modeling rubber materials and steel plasticity are

Copyright © 2018 U.S. Society on Dams. All Rights Reserved. 3


examples of cases where the elements suffer from volumetric locking. In these
cases, the structure becomes too stiff and the analysis converges to wrong results.
This also occurs when modeling solid elements to represent impounded reservoirs.

Models that have complicated geometry may experience several of the aforementioned
locking issues, which can result in the divergence of the solution from the real response
of the structure.

Shell Elements: Shell elements suffer from similar locking problems to solid elements
except for trapezoidal locking. Shell elements are used for modeling thin structures such
as radial gates, powerhouse walls and slabs, etc. Shell elements may be classified into
three broad categories – (i) classical shell elements, (ii) degenerate shell elements and (iii)
solid-shell elements. Classical shell elements are based on the conventional theories of
plates and shells, and have nodes with rotational Degrees of Freedom (DOFs) [2-6].
Degenerated shell elements start from the continuum theory but impose appropriate
constraints to express the kinematics in terms of the shell mid-surface state [7-13]. These
elements also have nodes with rotational DOFs. Several shell formulations have a zero
thickness stress (plane stress assumption), while for others the thickness stretch is taken
into account (more suitable for thick shell structures with aspect ratio between 0.1 and 0.3).

Solid-shell (continuum shell) elements resemble solid elements in that their nodes, which
are located on the top and bottom surfaces, possess only translational DOF [e.g. 14-16].
Special measures must be taken to overcome the aforementioned locking problems, which
result from the high ratio of thickness vs. the other two spatial dimensions. Nevertheless,
solid-shell elements are widely used and include the following attractive features:

• Modeling simplifications that result from avoiding 3D rotational DOFs, especially


when dealing with nonlinearities.

• Realistic representation of 3D boundary conditions without need for additional


kinematic assumptions. For example, distinctions can be made as to whether
supports are applied to the top or bottom surface.

• Modeling shell-like structures with bending behavior with only one layer of
elements across the thickness.

Solid-shell (continuum shell) elements have become popular and have been developed in
commercial FE software such as ANSYS [17] and ABAQUS [18]. As an example,
SOLSH190 element of ANSYS is a solid-shell element that possesses the continuum solid
element topology and features eight-node connectivity with three DOFs at each node:
translations in the nodal x, y, and z directions. It is noted that the solid-shell elements are
different than solid elements in that the element top and bottom faces and, hence, the
element normal, stacking direction, and thickness direction should be defined prior to the
analysis. All other features of the solid and solid-shell elements are identical. A good
example of solid-shell application is the modeling of multiple arch dam with only one

Copyright © 2018 U.S. Society on Dams. All Rights Reserved. 4


element across the thickness of the dam to capture the bending response of the structure.
This will reduce the number of DOF’s, and consequently the analysis time.

Remedies to Element Formulation Deficiencies: For a large number of structures that are
analyzed (e.g. dams), there is no analytical solution to compare against FE model results,
which does not allow for verification of results. Accordingly, the alternative is to improve
the model performance; one method is to use elements with mid-side nodes (using 8-node
quadrilaterals for 2D or 20-node solid elements for 3D models). The advantage of higher-
order elements is that the stress fields are continuous between the adjacent elements, but
they are computationally more expensive than low-order elements. In order to make the
best use of low-order elements, researchers have suggested some remedies that improve
the element performance by circumventing locking problems, as described below:

(The reader should note that the indented paragraphs below provide a summary of the
formulations for the methodologies used in improving element performance against
aforementioned locking problems. Examples of the benefits of each method are provided
after the indented section)

Reduced Integration (RI) and Selective Reduced Integration (SRI) methods: The
RI and SRI methods could be applied when dealing with incompressible materials
such as rubber or steel plasticity and fluid elements. The two mentioned methods
attack the volumetric locking when the material’s Poisson’s ratio approaches 0.5.
The constitutive relation for isotropic materials is as follows:
𝜎𝜎𝑖𝑖𝑖𝑖 = 𝜆𝜆𝑢𝑢𝑘𝑘,𝑘𝑘 𝛿𝛿𝑖𝑖𝑖𝑖 + 2𝜇𝜇𝑢𝑢𝑖𝑖,𝑗𝑗
Where 𝜆𝜆 and 𝜇𝜇 are the Lamme parameters for isotropic materials as:
𝐸𝐸𝐸𝐸 𝐸𝐸
𝜆𝜆 = (1+𝜐𝜐)(1−2𝜐𝜐) , 𝜇𝜇 = 2(1+𝜐𝜐)
1 𝑖𝑖 = 𝑗𝑗
𝛿𝛿𝑖𝑖𝑖𝑖 = �
0 𝑜𝑜𝑜𝑜ℎ𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒
3

𝑢𝑢𝑘𝑘,𝑘𝑘 = � 𝜀𝜀𝑘𝑘𝑘𝑘
𝑘𝑘=1

The 𝜆𝜆 term becomes unbounded if 𝜐𝜐=0.5. The RI or Uniform Reduced Integration


is applied on both terms of the constitutive equation, while SRI is applied only on
the 𝜆𝜆 term. The advantage of uniform reduced integration is the economy of the
formulation, i.e. fewer number of Gauss points. Its disadvantage is the rank
deficiency of the stiffness matrix, resulting in hourglass modes. Hourglass modes
are nonphysical modes of deformation that occur in one-point quadrature elements
and produce no stress. Solid and shell elements have 12 and 5 hourglass modes,
respectively.

The easiest, yet not the most efficient way to alleviate the hourglass mode is mesh
refinement. Hourglassing can usually be controlled by applying internal forces to
resist hourglass modes via one of several hourglass control algorithms [19]. There
are many other methods in the literature to alleviate the hourglass defect, many of
which have been implemented in commercial software such as LS-Dyna [20].

Copyright © 2018 U.S. Society on Dams. All Rights Reserved. 5


The Selective Reduced Integration method retains the full rank of the element
stiffness. This method should be used if steel plasticity, metal forming or rubber-
like materials are under investigation.

Reduced and Selective Reduced integration formulations also circumvent the shear
locking problems of FE formulations. As examples, shear locking of plate/shell
elements and 8-node solid elements have been addressed using SRI and RI methods
in references [21] and [22] respectively.

Assumed Natural Strain (ANS) Method. Assumed strain finite elements were first
used by MacNeal [23] in 1978. Several variations of this idea evolved over the
following two decades, including an important ANS variant that was introduced by
Park and Stanley [9] for doubly curved thin shells. The ANS method circumvents
the trapezoidal and curvature locking issues of 3D solid elements. Considering 8-
node 3D solid elements, the Cartesian and Natural strains in the Voigt notation [24]
are as follows:

𝑒𝑒 𝑐𝑐 = �𝑒𝑒𝑥𝑥𝑥𝑥 𝑒𝑒𝑦𝑦𝑦𝑦 𝛾𝛾𝑥𝑥𝑥𝑥 𝑒𝑒𝑧𝑧𝑧𝑧 𝛾𝛾𝑦𝑦𝑦𝑦 𝛾𝛾𝑥𝑥𝑥𝑥 �𝑇𝑇


𝑒𝑒 𝑁𝑁 = �𝑒𝑒𝜀𝜀𝜀𝜀 𝑒𝑒𝜂𝜂𝜂𝜂 𝛾𝛾𝜀𝜀𝜀𝜀 𝑒𝑒𝜁𝜁𝜁𝜁 𝛾𝛾𝜂𝜂𝜂𝜂 𝛾𝛾𝜀𝜀𝜀𝜀 �𝑇𝑇

The assumed strain fields are defined at element level and in the natural coordinate
system. A natural coordinate system is a local coordinate system that allows the
specification of a point within the element by a set of dimensionless numbers whose
magnitude is always smaller than unity. This coordinate system is effective in
developing the element in finite element formulation. This system is defined in a
way such that the magnitude at nodal points will have unity or zero. It also
facilitates the integration to calculate element stiffness. The Cartesian strains are
transformed to natural strains through a transformations matrix, which has been
provided in reference [15].

For 3D solid elements (displacement elements) with tri-linear shape functions, the
shear locking impacts the natural shear strains (𝛾𝛾𝜀𝜀𝜀𝜀 , 𝛾𝛾𝜀𝜀𝜁𝜁 )and the trapezoidal
locking impact the (𝑒𝑒𝜁𝜁𝜁𝜁 ) factor from the strain tensor. In order to avoid trapezoidal
locking, 𝑒𝑒𝜁𝜁𝜁𝜁 is evaluated at four collocation points (K=A,B,C,D shown in Figure
4), and then the evaluated strains at the collocation points are interpolated by using
bilinear shape functions within the element, as follows:
𝐷𝐷
𝐴𝐴𝐴𝐴𝐴𝐴 𝐴𝐴 𝐵𝐵 𝐶𝐶 𝐷𝐷 𝐾𝐾
𝑒𝑒𝜁𝜁𝜁𝜁 = 𝑁𝑁𝐴𝐴 𝑒𝑒𝜁𝜁𝜁𝜁 + 𝑁𝑁𝐵𝐵 𝑒𝑒𝜁𝜁𝜁𝜁 + 𝑁𝑁𝐶𝐶 𝑒𝑒𝜁𝜁𝜁𝜁 + 𝑁𝑁𝐷𝐷 𝑒𝑒𝜁𝜁𝜁𝜁 = � 𝑁𝑁𝐾𝐾 𝑒𝑒𝜁𝜁𝜁𝜁
𝐾𝐾=𝐴𝐴
1
𝑁𝑁𝐾𝐾 = (1 + 𝜉𝜉𝐾𝐾 𝜉𝜉)(1 + 𝜂𝜂𝐾𝐾 𝜂𝜂)
4
𝜉𝜉𝐾𝐾 , 𝜂𝜂𝐾𝐾 and 𝜁𝜁𝐾𝐾 are the natural coordinates corresponding to each of the considered
collocation points.

Copyright © 2018 U.S. Society on Dams. All Rights Reserved. 6


Figure 4. Collocation points for ANS approach.

To alleviate transverse shear locking, the natural shear strains (𝛾𝛾𝜂𝜂𝜂𝜂 ;𝛾𝛾𝜉𝜉𝜉𝜉 ) are
evaluated at four collocation points and interpolated with bilinear shape functions
as follows:
𝐻𝐻
𝐴𝐴𝐴𝐴𝐴𝐴 𝐸𝐸 𝐹𝐹 𝐺𝐺 𝐻𝐻 𝐾𝐾
𝛾𝛾𝜂𝜂𝜂𝜂 = 𝑁𝑁𝐸𝐸 𝛾𝛾𝜂𝜂𝜂𝜂 + 𝑁𝑁𝐹𝐹 𝛾𝛾𝜂𝜂𝜂𝜂 + 𝑁𝑁𝐺𝐺 𝛾𝛾𝜂𝜂𝜂𝜂 + 𝑁𝑁𝐻𝐻 𝛾𝛾𝜂𝜂𝜂𝜂 = � 𝑁𝑁𝐾𝐾 𝛾𝛾𝜂𝜂𝜂𝜂
𝐾𝐾=𝐸𝐸
1
𝑁𝑁𝐾𝐾 = (1 + 𝜉𝜉𝑘𝑘 𝜉𝜉)(1 + 𝜁𝜁𝑘𝑘 𝜁𝜁)
4
𝑀𝑀
𝐴𝐴𝐴𝐴𝐴𝐴 𝐽𝐽 𝐾𝐾 𝐿𝐿 𝑀𝑀 𝐾𝐾
𝛾𝛾𝜉𝜉𝜉𝜉 = 𝑁𝑁𝐽𝐽 𝛾𝛾𝜉𝜉𝜉𝜉 + 𝑁𝑁𝐾𝐾 𝛾𝛾𝜉𝜉𝜉𝜉 + 𝑁𝑁𝐿𝐿 𝛾𝛾𝜉𝜉𝜉𝜉 + 𝑁𝑁𝑀𝑀 𝛾𝛾𝜉𝜉𝜉𝜉 = � 𝑁𝑁𝐾𝐾 𝛾𝛾𝜉𝜉𝜉𝜉
𝐾𝐾=𝐽𝐽
1
𝑁𝑁𝐾𝐾 = (1 + 𝜂𝜂𝑘𝑘 𝜂𝜂)(1 + 𝜁𝜁𝑘𝑘 𝜁𝜁)
4

For 𝛾𝛾𝜂𝜂𝜂𝜂 , the collocation points are E, F, G, and H, and for 𝛾𝛾𝜉𝜉𝜉𝜉 , the collocation
points are J, K, L, and M, as shown in Figure 4. The ensuing solid FE developed
by using the above ANS method, behaves notably better than the displacement-
based solid element when it is subject to shear and trapezoidal locking issues (i.e.
when modeling thin or curved structures under bending forces).

Enhanced Assumed Strain (EAS) Method or Method of Incompatible Modes: This


method, which was first introduced by Wilson [25, 26] and continued by Simo-Refi
for linear problems [27] and Simo-Armero for geometrically nonlinear problems
[28], is based on the three-field method of Hu-Washizu [29]. This method alleviates
all the defects that are associated with the low-order displacement solid or shell
elements. The EAS method can be considered an improved version of the
incompatible modes method, where the enhanced strain approximations derived
from incompatible mode fields automatically satisfy the Taylor, et. al. patch test
condition [30]. Depending on how the enhanced strains are defined, the in-plane or
out-of-plane responses of the element could be improved. As an example in the
following solid formulation, the EAS approach has been incorporated to attack the
membrane, Poisson’s thickness, and volumetric locking problems of a solid
element formulation:

Considering the 8-node solid element shown in Figure 4, one extra


enhancing DOF has been added to the thickness strain (𝑒𝑒𝜁𝜁𝜁𝜁 ) to alleviate the
Poisson’s thickness locking, and five extra DOFs have been added to in-

Copyright © 2018 U.S. Society on Dams. All Rights Reserved. 7


plane strains to circumvent the membrane locking issue. The total Cartesian
strain vector is obtained by:

(𝑒𝑒) (𝑒𝑒)
𝑒𝑒 𝑐𝑐 = [𝐵𝐵 𝑐𝑐 𝑐𝑐
𝐵𝐵𝑒𝑒𝑒𝑒ℎ ] �𝑈𝑈(𝑒𝑒) � = 𝑩𝑩(𝑒𝑒) �𝑈𝑈(𝑒𝑒) �
𝛼𝛼 𝛼𝛼
𝑐𝑐 𝑛𝑛
𝐵𝐵𝑒𝑒𝑒𝑒ℎ = 𝑇𝑇 0 𝐵𝐵𝑒𝑒𝑒𝑒ℎ

𝜉𝜉 0 0 0 𝜉𝜉𝜉𝜉 0
⎡ ⎤
⎢ 0 𝜂𝜂 0 0 −𝜉𝜉𝜉𝜉 0 ⎥
𝑛𝑛
𝐵𝐵𝑒𝑒𝑒𝑒ℎ = ⎢0 0 𝜉𝜉 𝜂𝜂 𝜉𝜉 2 − 𝜂𝜂2 0⎥
⎢0 0 0 0 0 𝜁𝜁 ⎥
⎢0 0 0 0 0 0⎥
⎣0 0 0 0 0 0⎦

Where 𝛼𝛼 (𝑒𝑒) is the vector containing the six extra DOFs, 𝑇𝑇 0 is the
transformation matrix between the Cartesian and Natural coordinates
evaluated at ξ=η=ζ=0, and 𝐵𝐵 𝑐𝑐 is the compatible strain–displacement
operator derived from the tri-linear shape functions. The first three rows
𝑛𝑛
of 𝐵𝐵𝑒𝑒𝑒𝑒ℎ , which correspond to the in-plane response, are picked from Simo’s
EAS quadrilateral element in reference [27]. Following Simo’s work, the
𝑛𝑛
formulation is stable owing to the fact that the columns of 𝐵𝐵𝑒𝑒𝑒𝑒ℎ are linearly
independent, and it passes the constant strain patch test because the
following relation is satisfied:

𝑛𝑛
� 𝐵𝐵𝑒𝑒𝑒𝑒ℎ 𝑑𝑑𝑑𝑑 = 0

The rows and columns of the element local stiffness matrix that correspond
to extra DOFs (𝛼𝛼 (𝑒𝑒) ) will be removed from by doing static condensation in
the global stiffness matrix.

The solid element that was developed using the kinematic enforcements described
above (ANS and EAS) yield acceptable results as compared to the solid element
formulation which is derived from tri-linear shape functions. It is also noted that
the described solid element also falls into the context of solid-shell finite element
since it is capable of capturing the bending response of very thin structures. Figure
5 summarizes the development of the solid element with ANS and EAS approaches.

Copyright © 2018 U.S. Society on Dams. All Rights Reserved. 8


Figure 5. Summary of element development using ANS and EAS methods [16].

There have been many solid element formulations with different EAS parameters
in the literature. The above solid was an example of a solid formulation for which
the EAS approach has been used to circumvent the membrane, Poisson’s thickness,
and volumetric locking problems. There are several elements in commercial
software for which the EAS method or “incompatible displacement” formulation
has been applied. Examples within ABAQUS include elements CPS4I with 4 extra
DOFs; CPE4I with 5 extra DOFs; and C3D8I with 13 extra DOFs. These elements
are computationally more expensive than regular displacement elements due to the
existence of extra DOFs; however, the additional DOFs do not substantially
increase the size of the global stiffness matrix since they can be eliminated at the
element level.

Hybrid element formulation: In general, there are two different multi-field


principles in the FE terminology summarized below:

a) Mixed principle: In solid mechanics, the mixed formulations are based on the
variational principles in which the combination of any of the three internal
fields (stress, displacement and strain) is applied for the element formulation.
For example, the combination of displacement and strain as master fields yields
the Hellinger-Reissner formulation [31], and combination of displacement,
strain and stress as the master fields develops the Huwashizu (three-field)

Copyright © 2018 U.S. Society on Dams. All Rights Reserved. 9


formulation [29], which is the basis for the EAS method that was described in
the previous section.

b) Hybrid principle: In the Hybrid principle the master fields are of different
dimensionality. For example, the combination of displacement (as an internal)
and pressure (as an external) field could define a hybrid formulation. Hybrid
formulation is mostly used to circumvent the volumetric locking problem of
incompressible materials in solid mechanics. Most of solid materials have
incompressible behavior when they undergo large deformations such as rubber-
like material, metal forming, etc. The first hybrid stress element in the context
of FE was developed by Pian in 1964 [32], which was formulated for plane
stress finite elements with stress assumptions, but with displacement DOFs.
Later on, many solid elements were developed for which the hybrid stress
formulation was used to alleviate the volumetric locking problem [e.g. 33]. The
volumetric locking is alleviated by treating the pressure stress as an
independently interpolated master field, coupled to the displacement solution
through a constitutive relationship and a compatibility condition. The
independent interpolation of pressure stress and its treatment as a master field
is the basis of “hybrid” elements suitable for modeling of incompressible
materials. It is noted that the “hybrid formulation” is different from “mixed
formulation” in that one of the master fields is an external field, such as
boundary traction or pressure stress. See reference [34] for Hybrid variational
principles and formulation. Hybrid stress elements have been implemented in
commercial software such as ABAQUS and ANSYS. The main application of
these formulations is to alleviate the behavior of the structures with
incompressible materials.

Numerical Examples. This section presents numerical examples to show how modeled
structural behavior changes using solid elements with different formulations (ANS, EAS,
Hybrid, etc.), when they are subject to different locking problems.

1. Hemispherical shell with 18ı hole [35]. In this example, the solid element described in the
previous sections (summarized in Figure 5) is used to model a thin curved structure. The
arch structure shown in Figure 6 is subject to shear locking, Poisson’s thickness and
curvature locking problems. The radius of the curved shell is R=10.0 mm and the thickness
is t=0.04 mm. The model has a hole making an angle θ=18° with the Z-axis as shown in
the figure. Because of symmetry, only a quadrant of the hemisphere is modeled. The
equator represents a free edge, and the XZ and YZ planes are the symmetry planes. The
structure is subject to two pairs of diametrically opposite loads along the X and Y axes,
respectively, of magnitude F=1.0 N. The material behavior is modeled as isotropic linear
elastic with E= 6.825x107 N/mm2 and υ=0.3. The displacements of node A are normalized
to the reference value 𝑢𝑢𝑥𝑥𝐴𝐴 =.0940mm [35] and presented in Figure 7 along with the results
for solid element formulation with tri-linear shape functions without any additional
kinematic enforcement (such as ANS or EAS approaches). The comparison of convergence
rates between the two formulations show how the EAS and ANS approach alleviate the
response of the solid element when modeling thin-curved structures.

Copyright © 2018 U.S. Society on Dams. All Rights Reserved. 10


Figure 6. Hemisphere with 18° hole.

Hemisphere with Hole Convergence rate


1.2
Normalized displacement of point A

1
0.8
0.6
0.4
0.2
0
0 4 8 12 16 20 24 28 32
Number of element (NxN)

ANS-EAS formulation Tri-linear formulation

Figure 7. Hemisphere with 18° hole- Convergence rate study.

2. Multiple arch dam analysis: In this example, a multiple arch dam is analyzed for
hydrostatic loading using 3D solid elements with different element formulations, and the
results are compared to those obtained by using solid elements with fine mesh (surface area
of element is 2’x 2’ with incompatible displacement formulation). The multiple arch dam
shown in Figure 8 was modeled in ABAQUS. The dam consists of 16 complete arches,
each of 40-feet spans between buttresses centerlines. The length of the dam is 688 feet. The
maximum dam height is about 84 feet. The arch thicknesses range from 1.0 foot at the top
to 3.95 feet at the lowest point. The upstream face is inclined 50 degrees from the
horizontal. The buttresses range in thickness from 1.85 feet at the top to 4.25 feet at the
lowest point.

Figure 8 shows the dam and rock foundation, with contours of the hydrostatic pressure
applied plotted on the upstream surface of the arches. The analyses were performed using:
a) 3D solid displacement element (without any enhancement); b) 3D solid elements with
incompatible modes; c) 3D hybrid stress element; and d) 3D solid element with Reduced
Integration formulation.

Copyright © 2018 U.S. Society on Dams. All Rights Reserved. 11


The concrete and rock were modeled as isotropic linear elastic materials with E conc = 1.5E6
psi, υ conc =0.2, E rock = 3.0E6 psi, and υ rock =0.2.

Figure 8. Example multiple arch dam.

A 3D 8-node element formulation using incompatible modes, with element surface area of
2’x 2’ (fine mesh) is used as the reference solution. Two layers of elements have been
modeled through the thickness of the arches. Figures 9 and 10 show the maximum tensile
stress and displacement contours, which are used for comparison. The maximum tensile
stress at the base of the dam is approximately 300 psi, and the maximum displacement at
the crest in the canyon direction is 0.094-in.

Figure 9. 3D solid elements-incompatible modes (reference solution) - Maximum tensile


stress contour.

Copyright © 2018 U.S. Society on Dams. All Rights Reserved. 12


Figure 10. 3D solid elements- incompatible modes (reference solution) – Displacement
contour.

Figures 11 and 12 illustrate the maximum tensile and displacement contours of the dam
using the displacement based element (without any enhancement) with mesh surface area
of 10’x10’. Only one layer of elements has been used through the thickness of the arches.
The maximum displacement value is in agreement with the values obtained from the
reference solution, but the stress contours are dramatically different from the reference
solution. The maximum tensile stress value at the base is approximately 110 psi, which is
one-third of the stress value obtained from the reference solution. The stress results show
that displacement base elements with one layer across the thickness is not robust enough
to capture the bending behavior of the structure.

Figure 11. 3D solid elements without enhancement- Maximum tensile stress contour.

Copyright © 2018 U.S. Society on Dams. All Rights Reserved. 13


Figure 12. 3D solid elements without enhancement- Displacement contour.

Figures 13 and 14 illustrate the maximum tensile and displacement contours of the dam
using incompatible displacement elements with mesh surface area of 10’x10’. Only one
layer of elements has been used through the thickness of the arches. The maximum
displacement value is in agreement with the values obtained from the reference solution,
and the maximum tensile stress at the base of the dam is approximately 220 psi, which is
less than the reference solution.

Figure 13. 3D solid elements with incompatible modes- Maximum tensile stress contour.

Copyright © 2018 U.S. Society on Dams. All Rights Reserved. 14


Figure 14. 3D solid elements with incompatible modes- Displacement contour.

Figures 15 and 16 illustrate the maximum tensile and displacement contours of the dam
using Hybrid stress elements with mesh surface area of 10’x10’. Only one layer of elements
has been used through the thickness of the structure. The maximum displacement value is
in agreement with the values obtained from the reference solution, but the maximum tensile
stress value at the base of the dam, which is approximately 130 psi, is drastically less than
that of reference solution.

Figure 15. 3D solid elements with Hybrid stress formulation- Maximum tensile stress
contour.

Copyright © 2018 U.S. Society on Dams. All Rights Reserved. 15


Figure 16. 3D solid elements with Hybrid stress formulation- Displacement contour.

Figures 17 and 18 illustrate the maximum tensile and displacement contours of the dam
using Reduced Integration formulation with mesh surface area of 10’x10’. Only one layer
of elements has been used through the thickness of the arches. The maximum displacement
value does not match the reference solution, nor does the displacement contour match the
reference solution. The stress contours also show that using one layer of element through
the thickness with one integration point is not capable of capturing the bending response
of the structure against out-of-plane forces.

Figure 17. 3D solid elements with RI formulation- Maximum tensile stress contour

Copyright © 2018 U.S. Society on Dams. All Rights Reserved. 16


Figure 18. 3D solid elements with RI formulation- Displacement contour.

Following table summarizes the stress and displacement values for the above example:

Table 1- summary of stress and displacement values


Formulation Displacement at Maximum tensile
the crest stress at the base
(in) (psi)
Incompatible mode (fine mesh) 0.094 300
Displacement element 0.093 110
Incompatible mode element 0.094 220
Hybrid stress element 0.093 130
Reduced Integration element 0.12 Wrong stress
values

The above example shows that:

• The analysis results change by using different element formulations.

• The Reduced Integration element with one element across the thickness is not as
robust as full integration elements for capturing bending response of structures.

• Element formulation with incompatible displacement modes is more robust as


compared to other formulations that are available in element library of ABAQUS
software.

• Comparison of the displacement field is not sufficient to validate analysis results.


As observed above, the displacement contours are very similar for all formulations,

Copyright © 2018 U.S. Society on Dams. All Rights Reserved. 17


except for the RI element. For C0 elements (e.g., 3D 8-node solid elements with
tri-linear shape functions, or 2D 4-node plane elements with bi-linear shape
functions), the stress field, which is the derivative of the displacement field should
be reviewed in order to validate the analyses results.

• Sensitivity analyses of the results is always required to check how the stress and
displacement results change by varying the element sizes and formulations.

CONSIDERATIONS AND RECOMMENDATIONS FOR THE FE USER

In general, solid and shell elements that are used more frequently for modeling 3D
structures, including dam and appurtenances, suffer from deficiencies called “locking
problems”. As described, ANS elements are suitable for addressing shear and trapezoidal
locking problems, while RI and SRI elements are more applicable to circumvent the
volumetric locking issues. Hybrid stress elements are also useful for incompressible
materials. EAS elements are very robust in alleviating almost all types of deficiencies for
the elements; however, they are computationally more expensive than other elements
because of the additional DOFs that are included at the element level. In some cases mesh
refinement is warranted dependent on geometric complexity.

Regardless of the kinematic enforcements used to develop low-order elements, these


elements are C0 elements, i.e., the derivative of displacement (strains), and consequently,
the stresses are not continuous between the two adjacent elements. This issue will increase
the errors in computing the stresses when using a coarse mesh arrangement (stress jump).
This issue is exacerbated when the uniform reduced integration elements are used. In order
to have smooth stress distribution among the elements, it is recommended to use higher-
order elements whenever possible; however, higher-order elements are computationally
more expensive since they have additional nodes. Care should be taken when using higher-
order elements for an explicit analysis, since the size of time-steps will drop and the
analysis will take much longer than the analysis using low-order elements.

Solid-shell (continuum shell) elements are another class of finite elements that have been
implemented in commercial software. These are capable of modeling the exact 3D
geometry of structures with shell-like behavior. The only difference between the
continuum shell and solid element is that the continuum shell has a thickness direction that
has to be defined prior to the analysis. The advantage of these elements is that they can be
used to model thin structures with only one layer of element across the thickness (such as
thin arch dam). It is therefore recommended to use solid-shell elements for modeling 3D
thin structures. Since Solid-shell elements have translational DOFs, they could easily be
connected to solid elements without any constraint requirements.

When using uniform reduced integration elements, it is recommended to limit the


Hourglass energy to 10% of total energy to make sure that the hourglass spurious modes
will not impact the solutions dramatically. Full-integration C0 elements are preferred since
they have more Gauss points across the element and yield more realistic stress/strain
contours.

Copyright © 2018 U.S. Society on Dams. All Rights Reserved. 18


It is recommended to use SRI method if steel plasticity, metal forming or rubber-like
materials are under investigation. This method retains the full rank of the stiffness matrix
while alleviating the volumetric and shear locking problems.

Tetrahedral solid and triangular plane elements are susceptible to locking problems due to
their geometry. They should only be used when the hexahedral solid or quadrilateral plane
elements cannot be used. When modeling thin curved structures with quadrilateral shell
elements, warping of the element will cause errors in the analysis. It is recommended to
use quadrilateral elements with warping stiffness when modeling curved shells with coarse
mesh arrangement (for example, LS-Dyna carries quadrilateral shell formulations with
warping stiffness that alleviates the errors caused by warping of the shell surface). Using
triangular shell elements or using quadrilateral shell elements with a fine mesh will also
resolve the issues caused by warping of quadrilateral shell elements.

Finally, it is recommended that the user performs a sensitivity analysis of the stress results
in order to check how the results change by varying the element size and formulation. The
example problem presented above showed how different formulations provide different
stress results.
CONCLUSIONS

The theoretical formulations and examples provided in this paper demonstrate that
application of varied element technology for the same problem in an FE analysis can
produce a wide range of results. Recommendations are provided to the user to avoid
common problems with certain element types, as well as recommendations for application
of specific element types to specific structures and material types.

Having a deep knowledge of FE methods is necessary in order to make the best use of
commercial FE software. A number of element formulations exist in commercial software
that are used for wide range of engineering problems, yet each formulation is not applicable
to every problem. The engineer utilizing FE software is encouraged to become familiar
with the application of each element formulation, as well as always perform manual
calculations for use in checking overall validity of FE results, in order to obtain the most
accurate results for their problem.

If FE software is used without sufficient knowledge of the theory behind element


formulations, the modeling and simulation of the structure’s behavior could result in
inaccurate, or even incorrect results that users inadvertently rely upon. This has recently
become a more serious issue, as most commercial FE programs utilize many underlying
“program controlled” computations and methodologies not readily defined for the user,
which in turn results in a higher probability of incorrectly applied methodologies and
erroneous solutions. It is highly recommended that users are trained and understand the
principles of finite element methods prior to using any FE commercial software as an
analytical tool.

Copyright © 2018 U.S. Society on Dams. All Rights Reserved. 19


REFERENCES

1. MacNeal, Richard H., Faulkner, L.L. (Ed.) (1994). Finite Elements: their Design and
Performance. New York, Marcel Dekker, Inc.

2. Betsch P, Gruttmann F, Stein E. A 4-node finite shell element for the implementation
of general hyperelastic 3D-elasticity at finite strains. Computer Methods in Applied
Mechanics and Engineering 1996; 130(1–2):57–79.

3. Bischoff M, Ramm E. Shear deformable shell elements for large strains and rotations.
International Journal for Numerical Methods in Engineering 1997; 40(23):4427–4449.

4. Brank B, Korelc J, Ibrahimbegovic A. Nonlinear shell problem formulation accounting


for through-the-thickness stretching and its finite element implementation. Computers
and Structures 2002; 80(9-10):699–717.

5. Cardoso RPR, Yoon JW. One point quadrature shell element with through-thickness
stretch. Computer Methods in Applied Mechanics and Engineering 2005; 194(9–
11):1161–1199.

6. Lu H, Ito K, Kazama K, Namura S. Development of a new quadratic shell element


considering the normal stress in the thickness direction for simulating sheet metal
forming. Journal of Materials Processing Technology 2006; 171(3):341–347.

7. Belytschko T, Lin JI, Tsay CS. Explicit algorithms for the nonlinear dynamics of shells.
Computer Methods in Applied Mechanics and Engineering 1984; 42(2):225–251.

8. Bathe KJ, Dvorkin EN. A four-node plate bending element based on Mindlin/Reissner
plate theory and a mixed interpolation. International Journal for Numerical Methods in
Engineering 1985; 21(2):367–383.

9. Park KC, Stanley GM. A curved C0 shell element based on assumed natural-coordinate
strains. Journal of Applied Mechanics-Transactions of the ASME 1986; 53(2):278–
290.

10. Simo JC, Fox DD. On a stress resultant geometrically exact shell model. I. Formulation
and optimal parametrization. Computer Methods in Applied Mechanics and
Engineering 1989; 72(3):267–304.

11. Simo JC, Rifai MS, Fox DD. On a stress resultant geometrically exact shell model. IV.
Variable thickness shells with through-the-thickness stretching. Computer Methods in
Applied Mechanics and Engineering 1990; 81(1):91–126.

12. Simo JC, Kennedy JG. On a stress resultant geometrically exact shell model. V.
Nonlinear plasticity: formulation and integration algorithms. Computer Methods in
Applied Mechanics and Engineering 1992; 96(2):133–171.

Copyright © 2018 U.S. Society on Dams. All Rights Reserved. 20


13. César de Sá JMA, Natal Jorge RM, Fontes Valente RA, Almeida Areias PM.
Development of shear locking-free shell elements using an enhanced assumed strain
formulation. International Journal for Numerical Methods in Engineering 2002;
53(7):1721–1750.

14. Schwarze M, Reese S. A reduced integration solid-shell finite element based on the
EAS and the ANS concept-Geometrically linear problems. International Journal for
Numerical Methods in Engineering 2009; 80(10):1322–1355.

15. Mostafa M, Sivaselvan M, Felippa C. A solid-shell corotational element based on


ANDES, ANS and EAS for geometrically nonlinear structural analysis. International
Journal for Numerical Methods in Engineering 2013; 95(2):145–180.R. L.

16. Mostafa M, An improved solid-shell element based on ANS and EAS concepts.
International Journal for Numerical Methods in Engineering 2016; Volume 108, Issue
11, 14 December 2016, Pages 1362–1380

17. ANSYS, Finite Element commercial software, Version 18.2: http://www.ansys.com/

18. ABAQUS, Dassaults System’s Simulia Inc. Finite Element commercial software,
CAE 2016: https://www.3ds.com/products-services/simulia/

19. Flanagan DP, Belytschko T. A uniform strain hexahedron and quadrilateral with
orthogonal hourglass control. International Journal for Numerical Methods in
Engineering 1981; 17:679–706.

20. LS-Dyna, http://www.dynasupport.com/howtos/element/hourglass

21. Zienkiewicz O.C, Taylor R. L., Too J.M. Reduced integration technique in general
analysis of plates and shells. International journal for numerical methods in
engineering, Volume 3, Issue 2, April 1971, Pages 275-290.

22. Alves de Sousa RJ, Natal Jorge RM, Fontes Valente RA, César de Sá JMA. A new
volumetric and shear locking-free 3D enhanced strain element. Engineering
Computations 2003; 20(7):896–925.

23. MacNeal RH. Simple quadrilateral shell element. Computers and Structures 1978;
8(2):175–183.

24. Voigt notation- https://en.wikipedia.org/wiki/Voigt_notation

25. Wilson E.L, Taylor R.L, Doherty W.P. Ghabousi J. Incompatible displacement Models,
Numerical and computer methods in structural mechanics, Book

26. Taylor, P. J. Beresford and E. L. Wilson. A non-conforming element for stress analysis,
International Journal for Numerical Methods in Engineering 1976;Volume10, 121 1-
1219

Copyright © 2018 U.S. Society on Dams. All Rights Reserved. 21


27. Simo JC, Rifai MS. Class of mixed assumed strain methods and the method of
incompatible modes. International Journal for Numerical Methods in Engineering
1990; 29(8):1595–1638.

28. Simo JC, Armero F. Geometrically non-linear enhanced strain mixed methods and the
method of incompatible modes. International Journal for Numerical Methods in
Engineering 1992; 33:1413–1449.

29. Felippa C, Advanced Finite Element course notes, The Three-Field Mixed Principle
of
Elastostatics: https://www.colorado.edu/engineering/CAS/courses.d/AVMM.d/AVM
M.Ch12.d/AVMM.Ch12.pdf

30. Taylor RL, Simo JC, Zienkiewicz OC, Chan ACH. The patch test—a condition for
assessing FEM convergence. International Journal for Numerical Methods in
Engineering 1986; 22:39–62.

31. Felippa C. Advanced Finite Element course notes, The HR Variational Principle of
Elastostatics: https://www.colorado.edu/engineering/CAS/courses.d/AVMM.d/AVM
M.Ch11.d/AVMM.Ch11.pdf

32. Pian. T.H.H. (1964), “Derivation of element stiffness matrices by assumed stress
distributions”, AIAA J., 2, 333- 1376.

33. Sze KY, Yao LQ. A hybrid stress ANS solid-shell element and its generalization for
smart structure modelling. Part I—solid-shell element formulation. International
Journal for Numerical Methods in Engineering 2000; 48:545–564.

34. Felippa C. Advanced Finite Element course notes, Hybrid variational principles:
formulations https://www.colorado.edu/engineering/CAS/courses.d/AVMM.d/AVM
M.Ch13.d/AVMM.Ch13.pdf

35. MacNeal RH, Harder RL. Proposed standard set of problems to test finite element
accuracy. Finite Elements in Analysis and Design 1985; 1(1):3–20.

Copyright © 2018 U.S. Society on Dams. All Rights Reserved. 22

You might also like