Shear Locking: Shear Locking, Aspect Ratio Stiffening, and Qualitative Errors
Shear Locking: Shear Locking, Aspect Ratio Stiffening, and Qualitative Errors
Shear Locking: Shear Locking, Aspect Ratio Stiffening, and Qualitative Errors
Related terms:
Beam, Flexure, Integration, Aspect Ratio, Stiffness Matrix, Shear Strain, Beam Ele-
ment, Plate Element, Strain Gradient
The full stiffness matrix for a Timoshenko beam is developed in Appendix B where
its modeling characteristics are investigated. The Euler-Bernoulli beam is developed
in Appendix A so the modeling characteristics of the two beam elements can be
contrasted. The parameters contained in the elements of the stiffness matrix are
Young's modulus, E; the shear modulus, G; the area, A; the moment of inertia, I;
the shear shape factor, K; and the element length, L. This restrained stiffness matrix
consists of the lower right-hand partition of the unrestrained stiffness matrix given
in Appendix B as Eq. B.19. It is found by forcing the displacement and rotation of
the left end to be zero.
The stiffness element K22 of Eq. 12.1 is separated into three components. The first
element, KGAL/4, is associated with the shear strain. The second element, KAGL/12,
is associated with the erroneous parasitic shear term. The final element, EI/L, is
associated with the flexure deformation.
The overall stiffness and condition number can be determined by evaluating the
two eigenvalues, max and min, associated with this restrained stiffness matrix. The
overall stiffness, given as the sum of the two eigenvalues or as k11 + k22, varies with
the length as shown in Fig. 3. The condition number, computed as ( max – min)/ min,
varies with the length as shown in Fig. 4. These characteristics are computed for a
beam with a unit cross section and a shape factor of 5/6.
Figure 3. Overall stiffness vs length
The sharp rise in the overall stiffness with increasing length occurs as a result of
the increase in K22 as L increases. The two terms in K22 with L in the numerator
are directly proportional to L. The increase in condition number occurs because
K22 increases and k11 decreases. The element k11 decreases because L is in the
denominator.
The k22 term in the restrained stiffness matrix given by Eq. 12.1 contains the parasitic
shear term KAGL/12. When this term is removed and the overall stiffness and
condition number are plotted in Figs. 3 and 4, we can see the relation between the
overall stiffness and condition number for the corrected and uncorrected stiffness
matrices. The point at which the overall stiffness begins its ascent is slightly delayed.
The rise in the condition number occurs sooner. This illustrates that the shear
locking mechanism is independent of the parasitic shear phenomenon and that
shear locking is caused by a second modeling deficiency.
(21)
and
(22)
Shear Locking
A third type of elemental error known as shear locking is contained in out-of-plane
bending elements with independent shear strain models. This error is due to im-
proper modeling of the out-of-plane shear strains by the displacement polynomials.
This error begins to dominate the problem by making the individual element
overly stiff when the thickness becomes small with respect to the other dimensions
of the element. This error is found in Timoshenko beam elements and Mindlin
plate elements unless corrections are made. The cause of this modeling error was
identified through the use of strain gradient notation [see Ref. 2].
An approach based on St. Venant's principle has been developed for controlling
this modeling deficiency through insights provided by the use of strain gradient
notation. The St. Venant correction factor is similar to ad hoc penalty methods
previously developed to control shear locking. However, the strain gradient approach
enables these correction factors to be developed using physical arguments.
Although this book focuses on plane stress problems, a Timoshenko beam is devel-
oped in Lesson 12 to identify the source of shear locking and to demonstrate that
this error is indeed different from parasitic shear. The Timoshenko beam element
is corrected using a St. Venant factor to control this error. Strain-gradient-based
Mindlin plate elements containing the St. Venant correction factor are developed
and demonstrated in Ref. 3. The mechanism that causes shear locking is also
shown to cause plane elasticity elements with high length-to-width ratios to be
overly stiff. This modeling deficiency is usually controlled by disallowing the use of
high-aspect-ratio elements. The presentation is first made from the point of view of
the out-of-plane bending elements because the development is more direct. Then
the error due to aspect ratio stiffening is considered.
(14.264)
where
(14.265)
are the transverse shears evaluated at the midpoints of the element boundaries.
Making use of the bilinear element interpolation for the four-node quadrilateral
element, it follows that
(14.266)
where , for , are the mid-surface position vectors of the element nodes.
At this point the transverse shear treatment algorithm changes depending upon the
choice of Gauss point interpolation of the director field, see the discussion in Section
14.5.3. In the formulation of the linear shell theory in Chapter 10, the continuum
consistent interpolation (14.219) is used. While this interpolation is appropriate and
the most straightforward for the nonlinear case, we illustrate below the algorithm
based on the full update in (14.222). By making use of the assumed shear field along
with the update formulae for the director field in (14.222), the assumed transverse
shear field may be written concisely in matrix notation. Recall the director field
update equation and first variation of the director field:
(14.267)
It follows from the element interpolation that (dropping the iteration index )
(14.268)
Define for convenience and use only in this discussion the following vectors:
(14.269)
(14.270)
where
(14.271)
(14.272)
(14.273)
Constitutive relations. The constitutive relations for the curvilinear components of the
resultant transverse shear force are written in terms of the transverse shear strains
as
(14.274)
(14.275)
The matrix is the inverse of the metric , where metric components in the reference
configuration are defined by the inner product
(14.276)
The transverse shear force components in the shell orthonormal coordinate system
are calculated with the coordinate transformation as
(14.277)
Geometric tangent operator. The calculation of the geometric stiffness matrix requires
the second variation of the assumed transverse shear field (see (14.235)):
(14.278)
(14.280)
(14.281)
(14.282)
(14.283)
Many interpolation schemes have been proposed to avoid transverse shear locking.
Here we describe an assumed strain method based on the Hu-Washizu principle.
This scheme derives from that by MacNeal [24], subsequently extended and refor-
mulated in Hughes and Tezduyar [25], and MacNeal [15], and revisited in Dvorkin
and Bathe [26].
Figure 10.28. Notation for the assumed strain field on the standard isoparametric
element.
(10.381)
where
(10.382)
are the transverse shears evaluated at the midpoints of the element boundaries.
Making use of the bilinear element interpolation, it follows that
(10.383)
where , for , are the mid-surface position vectors of the element nodes.
By making use of the assumed shear field along with the linear displacement field,
the assumed transverse shear field may be written concisely in matrix notation.
Recall the unit director field update equation in (10.263):
(10.384)
At this point we introduce the element interpolations for the linear displacement
field. For illustration purposes, an alternative approach is introduced that does not
follow from the developments above is adopted. Nodal rotations are interpolated
instead of the cross product of the rotation with the unit director field. While this
choice is not followed in the implementation of the element, numerical tests show
that both formulations performs substantially similar.
(10.385)
(10.386)
(10.387)
where
(10.388)
(10.389)
(10.390)
Constitutive relations. The constitutive relations for the curvilinear components of the
resultant transverse shear force are written in terms of the transverse shear strains
as
(10.391)
(10.392)
The matrix is the inverse of the metric , where metric components in the reference
configuration are defined by the inner product
(10.393)
The transverse shear force components in the shell orthonormal coordinate system
are calculated with the coordinate transformation as
(10.394)
The internal force contribution from the transverse shear field in (10.346) is replaced
by the following. Define vectors of nodal displacement quantities and test functions:
(10.395)
(10.396)
where is the area measure in the reference configuration and is the (symmetric)
transverse shear contribution to the stiffness, as follows. Define the symmetric
matrices
(10.397)
(10.398)
(10.399)
Let us now present the isoparametric formulation procedure via an example. The
example chosen for this demonstration is the three-node bar element. This element
is simple enough to easily see the details of the isoparametric formulation and com-
plex enough to highlight the causes of the strain modeling deficiencies in distorted
isoparametric elements. These errors are demonstrated with sample calculations.
• shell elements;
• solid elements.
Usually, shell (or plate) elements or one layer of 20-node solid elements yield
reasonably accurate results in the assessment of Type ‘a’ hot spots. It should be noted
that the goal is to omit the non-linear stress peak. This implies that 20-node solid
elements with reduced 2-point integration in the thickness direction are sufficient.
One problem with simple solid elements is the tendency towards the so-called shear
locking, which also makes this type of solid element preferable.
If multi-layer modelling of a plate with solid elements is used, which may also be
the case when only half the plate thickness is modelled, the surface stress includes
part of the non-linear stress peak, up to about 0.4t from the weld toe. Therefore, the
stress results should be linearized across the thickness, or the stresses should be read
outside that area and extrapolated to the weld toe. Shell elements are used to model
the middle planes of plates. Plate thickness is given as a property of the element.
Attachments on plates should be extended towards the mid-plane of the plate, or
connected to it with rigid links. If the welds are not modelled at all, the stiffness
of the plate between adjacent discontinuities becomes too low. For example, in the
rectangular hollow section joint shown in Fig.4.1, the welds should be modelled in
order to obtain realistic stiffness in the gap, g’, or in the corner. Solid elements are
preferable in such details.
4.1. Details that are unsuitable for modelling with thin shell elements (from ref.4).
General Introduction
John O. Dow, in A Unified Approach to the Finite Element Method and Error Analysis
Procedures, 1999
Such a comparison has identified the causes of the following types of errors in indi-
vidual finite elements: (1) parasitic shear, (2) spurious zero-energy modes, (3) shear
locking, (4) aspect ratio stiffening, and (5) incorrect coupling in laminated composite
plates. Procedures for correcting or controlling these errors are formulated using
strain gradient notation.
In the second approach, the element is given a set of nodal displacements that would
impose a specific strain distribution on an accurate element or the continuum. The
strain components actually generated in the element being analyzed are compared
to the strain components that should exist in the continuum. This comparison is
done numerically on a point-by-point basis to identify the distribution of the strain
modeling errors in the element. The results of two such comparisons are shown in
Fig. 4. The strain energies are computed and compared to give a measure of the
aggregated errors in the strain representations.
This approach enables the evaluation of any type of element. The evaluation of a
single element can be viewed as an extension of the “patch test.” In this extension, we
evaluate all of the element capabilities, rather than just the ability to represent rigid
body motions and constant strain states. This approach has identified the effects of
initial element distortion on the strain representations produced in isoparametric
elements. The identification of these modeling errors motivated the development
of an alternative finite element formulation procedure that produces accurate el-
ements. The elements created with this new procedure do not exhibit any of the
modeling deficiencies contained in isoparametric elements.
The MTLED solution algorithm has been successfully applied (Miller et al., 2012;
Zhang et al., 2012) in a comprehensive patient-specific surgical simulation. All
anatomical components are represented by surfaces, generated by segmentation
of the image raw data (Fig. 18.3). The parenchyma, tumor, and ventricle volumes
were represented (filled) by a total of 31,753 nodes, having an average spacing of
3.5 mm. Representing this complex geometry by a point cloud is a straightforward
and trivial procedure, whereas creating a good-quality finite element mesh, for the
same geometry, is a tedious and time-consuming procedure that requires a lot of
manual intervention (Miller et al., 2012). Integration across the volume is applied by
using a regular Cartesian grid that consists of 53,672 integration points with 3 mm
spacing between them (a background mesh, but this requires no manual work to
create). A contact interface has been defined between the skull and the parenchyma
surface, with nodes on the parenchyma surface not being able to penetrate the skull,
but could slide without friction. The tumor and ventricle surfaces were tied (by shared
nodes) to the internal surfaces of the parenchyma. The results obtained using the
MTLED method are as useful and accurate as those obtained with the FE method.
The contours in Fig. 18.4 show the good agreement between the results obtained
by the MTLED and those using the FE method. Only in a few places are slightly
different results seen and the differences are not larger than the voxel size of the
intraoperative brain MRI.
Figure 18.3. Definition and surface visualizations of the problem geometry;
parenchyma (yellow (light gray in print versions)), tumor (blue (dark gray in print
versions)), and ventricle (red (gray in print versions)).
3.1 Introduction
The finite element models of the Euler–Bernoulli beam theory and the Timoshenko
beam theory are now standard (see Reddy 1993). A number of Timoshenko beam
finite elements have appeared in the literature. They differ from each other in the
choice of interpolation functions used for the transverse deflection 0 and rotation
. Some are based on equal interpolation and others on unequal interpolation of 0
and .
The Timoshenko beam finite element with linear interpolation of both 0 and is the
simplest element. However, it behaves in a very stiff manner in the thin beam limit,
i.e. as the length-to-thickness ratio becomes very large (say, 100). Such behaviour
is known as shear locking (see Nickell and Secor 1972, Tessler and Dong 1981,
Prathap and Bhashyam 1982, and Averill and Reddy 1990). The locking is due to
the inconsistency of the interpolation used for 0 and . To overcome the locking,
one may use equal interpolation for both 0 and but use a lower-order polynomial
for the shear strain, . This is often realized by using selective integration, in which
reduced-order integration is used to evaluate the stiffness coefficients associated
with the transverse shear strain, and all other coefficients of the stiffness matrix
are evaluated using full integration. The selective integration Timoshenko beam
element is known to exhibit spurious energy modes (see Prathap and Bhashyam
1982 and Averill and Reddy 1990). Prathap and Bhashyam (1982) used a consistent
interpolation of the variables to alleviate locking.
The transverse shear strain in the Timoshenko beam theory (Timoshenko 1921,
1922) is represented as a constant through the beam thickness, and a shear cor-
rection factor is thus introduced to calculate the transverse shear force that would
be equal in magnitude to the actual shear force. Since the actual shear stress
distribution through beam thickness is quadratic, Jemielita (1975), Levinson (1981),
Bickford (1982) and Reddy (1984a) developed third-order beam theories to capture
the true variation of the shear stress. The displacement field of these third-order
theories accommodates a quadratic variation of the transverse shear strain and
stresses, and there is no need to use shear correction factors in a third-order
theory. The Levinson third-order beam theory has the same equations of equilibrium
as the Timoshenko beam theory but the force and moment resultants contain
higher-order strain terms. Bickford (1982) used Levinson's displacement field and
developed variationally consistent equations of motion of isotropic beams while
Reddy (1984) developed a variationally consistent third-order theory of laminated
composite plates.
Heyliger and Reddy (1988) used the third-order laminate theory of Reddy to develop
a beam finite element and studied bending and vibrations of isotropic beams. The
element is based on Lagrange linear interpolation of the rotation and Hermite
cubic interpolation of 0, as they are the minimum requirements imposed by the
weak form of the third-order theory (also see Phan and Reddy 1985 and Reddy
1997a).
In this chapter, we present the development of a unified beam finite element that
contains the finite element models of the Euler–Bernoulli, Timoshenko and the
refined third-order beam theory. The derivation of the unified element is based on
the exact relationships between the various theories presented in Chapter 2. The
relationships allow interdependent interpolation of 0 and and the rank deficiency
is removed, resulting in an efficient and accurate locking-free finite element for
the analysis of beams according to classical as well as refined beam theories.