SE F N -C NF RING DES I FINI E ELE E AN [SIS F PL ES SHELLS
Transcript of SE F N -C NF RING DES I FINI E ELE E AN [SIS F PL ES SHELLS
-CIVIL ENGINEERING STUDIES STRUCTURAL RESEARCH SERIES NO. 401
SE F N -C NF FINI E ELE E
F PL ES
UllU-ENG-73 .. 2019
RING DES I AN [SIS
SHELLS
By CHANG-KOON CHOI
W. C. SCHNOBRICH
Issued as
a Technical Report
of a Research Program
Sponsored by
THE OFFICE OF NAVAL RESEARCH
DEPARTMENT OF THE NAVY
Contract NOO0l4-67-A-0305-00l0
UNIVERSITY OF ILLINOIS
at URBANA-CHAMPAIGN
URBANA, ILLINOIS
SEPTEMBER 1973
UIlU-ENG ... 73",,2019
USE OF NONCONFORMING MODES IN FINITE ELEMENT
ANALYSIS OF PLATES AND SHELLS
By
Chang-Koon Choi
and
William C. Schnobrich
Issued as a Technical Report of a Research
Program Sponsored
by
The Office of Naval Research Department of the Navy
Contract N 00014-67-A-0305-00l0
University of Illinois Urbana, Illinois
September, 1973
iii
ACKNOWLEDGMENT
This report was prepared as a doctoral dissertation by Mr. Chang
Koon Choi submitted to the Graduate College, University of Illinois at Urbana
Champaign, in partial fulfillment of the requirements for the degree of Doctor
of Philosophy in Civil Engineering.
The study was conducted under the supervision of Dr. W. C. Schnobrich,
Professor of Civil Engineering, as part of a research study of numerical analysis
supported by the Office of Naval Research under contract N00014-67-A-0305-00l0.
Also the support of Mr. Choi's graduate study by a University of Illinois Fellow
ship is gratefully acknowledged.
The numerical calculations were performed on the IBM 360/75 system
of the Computing Service Office of the University of Illinois.
CHAPTER
2
3
iv
TABLE OF CONTENTS
INTRODUCTION .
1 . 1 Genera 1 . 1.2 Literature Survey. 1.3 Object and Scope. 1.4 Notations
ISOPARAMETRIC ELEMENTS.
Page
1 2
10 11
15
2.1 General. 15 2.2 Isoparametric Shell Element 16 2.3 Ahmad-Irons Modification 17 2.4 Reduced Integration Technique. 20 2.5 Addition of Nonconforming Displacement Modes. 22
ELEMENT FORMULATION.
3.1 Variational Principle 3.2 The Basic Geometry and Displacement Field. 3.3 Nonconforming Displacement Modes. 3.4 Strain-Displacement Relationships 3.5 Stress-Strain Relationships 3.6 Element Stiffness Matrix 3.7 Selection of Nonconforming Modes. 3.8 Forced Interelement Boundary Compatibility
26
26 28 29 31 33 34 37 39
4 ASSEMBLY AND SOLUTION OF THE LOADDISPLACEMENT EQUATIONS. 42
5
4.1 General. 4.2 Assemblage of Stiffness Matrix and
Boundary Conditions . 4.3 Load Vector. 4.4 Solutions of Load-Deflection Equations. 4.5 Strains. 4.6 Stresses and Forces .
NUMERICAL RESULTS
5.1 General . 5.2 Deep Beams . 5.3 Plates 5.4 Cylindrical Shell. 5.5 Elliptic Paraboloid.
42
42 43 44 45 45
47
47 48 51 54 56
6
v
5.6 Spherical Shell . 5.7 Hyperbolic Paraboloid. 5.8 Conoidal Shell
CONCLUSIONS AND SUGGESTIONS FOR FURTHER STUDIES .
6.1 Conclusions 6.2 Suggestions for Further Studies.
LIST OF REFERENCES .
APPENDIX
A SHAPE FUNCTIONS.
B LOCAL COORDINATE SYSTEM .
C LOCAL STRAINS AND TRANSFORMATIONS.
D NUMERICAL INTEGRATIONS .
E TYPICAL COMPONENTS OF ELEMENT STIFFNESS MATRIX
Page
57 58 59
62
62 63
65
103
105
107
110
. 112
Table
2
3
vi
LIST OF TABLES
Element Designations.
Comparison of Results of Cantilevered Beam Using Various Elements
Abscissae and Weight Coefficients of the Gaussian Quadrature Formula.
Page
70
71
72
Figure
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
vii
LIST OF FIGURES
3-DIMENSIONAL ISOPARAMETRIC ELEMENTS .
LINEAR ISOPARAMETRIC ELEMENT UNDER PURE BENDING
RESTORING THE ACTUAL BEHAVIOR OF THE ELEMENT BY ADDING NONCONFORMING MODE.
DEFINITIONS OF GLOBAL AND LOCAL COORDINATES.
NONCONFORMING DISPLACEMENT (w) MODES FOR LINEAR ELEMENT
NONCONFORMING DISPLACEMENT (w) MODES FOR QUADRATIC ELEMENT
INTEGRATION POINTS FOR PLATE (n = 3) .
GAUSS INTEGRATIONS .
TYPICAL UNIFORM MESH (2x2) OF CLAMPED SQUARE PLATE UNDER CONCENTRATED LOAD AT THE CENTER.
CENTRAL DEFLECTION OF CLAMPED SQUARE PLATE UNDER CONCENTRATED LOAD AT THE CENTER.
CENTRAL DEFLECTION OF CLAMPED SQUARE PLATE .
COMPARISON OF CENTRAL DEFLECTION FROM VARIOUS PLATE BENDING ELEMENTS.
CLAMPED EDGE CIRCULAR PLATE UNDER UNIFORM LOAD.
GEOMETRY AND MATERIAL PROPERTIES OF CIRCULAR CYLINDRICAL SHELL
DISPLACEMENTS OF CIRCULAR CYLINDRICAL SHELL UNDER UNIFORM LOAD .
STRESS RESULTANTS OF CIRCULAR CYLINDRICAL SHELL UNDER UNIFORM LOAD .
GEOMETRY AND MATERIAL PROPERTIES OF A SQUARE ELLIPTIC PARABOLOID
Page
73
74
75
76
77
78
79
79
80
80
81
82
83
84
84
86
89
viii
Page
18 VERTICAL DISPLACEMENT OF SIMPLY SUPPORTED SQUARE ELLIPTIC PARABOLOID UNDER UNIFORM LOAD . 89
19 STRESS RESULTANTS OF SIMPLY SUPPORTED SQUARE ELLIPTIC PARABOLOID UNDER UNIFORM LOAD . 90
20 GEOMETRY AND MATERIAL PROPERTIES OF SPHERICAL SHELL. 91
21 STRESS RESULTANTS OF SPHERICAL SHELL UNDER UNIFORM LOAD. 92
22 GEOMETRY AND MATERIAL PROPERTIES OF HYPERBOLIC PARABOLOID 93
23 VERTICAL DEFLECTION ACROSS MIDSPAN OF CLAMPED HYPERBOLIC PARABOLOID UNDER UNIFORM NORMAL LOAD . 93
24 STRESS RESULTANTS OF CLAMPED HYPERBOLIC PARABOLOID UNDER UNIFORM NORMAL LOAD 94
25 GEOMETRY AND MATERIAL PROPERTIES OF CONOIDAL SHELL . 95
26 DISPLACEMENTS OF SIMPLY SUPPORTED CONOIDAL SHELL UNDER UNIFORM LOAD. 95
27 STRESS RESULTANTS OF SIMPLY SUPPORTED CONOIDAL SHELL UNDER UNIFORM LOAD. 97
28 DEFLECTIONS OF CLAMPED CONOIDAL SHELL UNDER UNIFORM LOAD. 100
29 STRESS RESULTANTS OF CLAMPED CONOIDAL SHELL UNDER UNIFORM PRESSURE 101
Chapter 1
INTRODUCTION
1.1 General
Various types of shell structures have been built for various pur
poses in a number of countries in the past. With the structural efficiency
of this type of structure, it is obvious that more such structures will be
built in the future. Architectural or roof shells are one form of this
structural shape which has attracted a good deal of engineering research ef
fort and designers· attentions as well. Aesthetics, long column-free span
ning capacities and economy, both in the sense of a weight-to-space ratio
as well as in material quantities used, have made this type of structure very
popular.
In recent days, however, it has been observed that while architec
tural or roof shell construction has somewhat tapered off, particularly in
countries where the wages of the construction worker have risen sharply,
other types of shells, such as aeronautical structures, nuclear reactor
vessels and marine structures, have become more widely used. Consequently,
recent research efforts have been directed in these latter areas. The shells
to be analyzed have also become more complicated.
Solutions by classical methods of shell analysis require the solving
of complicated governing differential equations. Therefore, such solutions are
extremely rare and limited in scope. Exact analytical solutions have been
possible only for those shells of simple regular geometry, ideal boundary
conditions and simple loadings. This is in great measure due to mathematical
2
complexity of the resulting equations. Simplification of the analysis pro
cedures, such as membrane theory, although of great value, cannot provide the
detail necessary. Experimental studies as an alternative are either not widely
accepted or are too expensive and time consuming to be used for other than
research studies into a specific configuration.
The finite element method provides a powerful tool for shell analysis
as it does for many other structural analysis problems. The structure may be
approximated by an assemblage of a finite number of discrete elements inter
connected at a finite number of nodal points. Then, piecewise continuous
displacement and/or stress fields are assumed in each element. The applica
tion of variational principles yield a system of simultaneous algebraic equa
tions [lJ. Once the final matrix equations have been established, the solutions
are straight-forward. Thus the unknown variables, displacements and/or stress
es, are obtained.
It is possible to obtain numerous different finite element models
by introducing different equilibrium or compatibilit~ conditions at the inter
element boundaries and modifying these conditions by the use of Lagrangian
multipliers [2J. According to the variables assumed in the element, the re
sulting finite element models may be classified as 1) DispJacement (Compatible)
Models, 2) Force (Equilibrium) Models, 3) Mixed Models and 4) Hybrid Models.
A brief discussion of these alternative models will be presented later.
1.2 Literature Survey
The development and the application of the finite element method
followed, in general, in the order of plane stress problems, plate bending
3
problems and curved shell problems. Key and Krieg [3J summarized the early
stages of finite element development by different groups, i .e., mathematicians~
engineers and physicists.
Courant [4J has been frequently acknowledged as one of the first to
apply the finite element method, as known today, to structural mechanics prob
lems. He used this method in analyzing the St. Venant torsion problem of a
prismatic bar by the means of an assemblage of triangular elements.
The first formulations of the finite element method to be used in
structural analysis were of the "Direct Stiffness Matrix·· type approach as
applied to in-plane stress problems [5,6J. This approach has the conceptual
limitation of applying to the simple elements. Later, Melosh [7J extended the
definition of the direct stiffness method to make it a variationally based
approach, expressing the finite element formulation in terms of the Minimum
Potential Energy Principle. Thus, the variational method, as a generalized
form of the Ritz method, began to be most frequently u~ed as the basis for
the element formulation.
Although the majority of the early efforts using the finite element
method belong to the classification of the displacement model, several other
forms or variations of the variational method with various displacement and/or
stress assumptions are possible. A goodly number of these have been investi -
gated. After the successes with plane stress problems [5,6,8J, plate bending
applications began to appear.
Before proceeding further with the discussion of plate bending prob-
lems, the convergence requirements for the displacement model are discussed
briefly. If the following conditions are met, the results of finite element
will converge toward the true solutions as the mesh size is reduced:
4
1. Interelement boundary compatibility,
2. Inclusion of rigid body modes,
3. Inclusion of all pertinent constant strain states.
The compatibility requirement for plate bending problems becomes
more demanding than for plane stress problems since both normal and rotational
displacements compatibilities should be satisfied. A detailed discussion of
compatibility requirements can be found in Reference 20. Melosh [9J and
Zienkiewicz [lOJ used 12 term polynomials to represent the normal displace
ment for rectangle and quadrilateral element shapes, respectively. In both
cases normal displacement compatibility is satisfied but not the rotational
or normal slope compatibility.
It has been proved to be difficult to ~chieve intere1ement compati
bility when using deflection shapes for the element established from conven
tional displacement components defined only at vertices. This difficulty is
due in part to the disparity between the nUUlber of terms in a complete poly
nomial used to define the displacements and the number of terms needed to
express the displacement field by variables defined at the vertices. Dropping
some terms from a complete polynomial may result in geometric anisotropy and
even destroy convergence [llJ.
One of the alternatives is the adoption of smooth surface interpo
lation functions. Bogner, Fox and Schmit [12J used complete third order
Hermitian polynomial interpolation formulas for the rectangular plate ..
Fraeijs de Veubeke [13J achieved the conforming quadrilateral element by
subdividing it into 4 triangles, assuming a complete cubic polynomial dis
placement field in each triangle and finally condensing out the interior
degrees of freedom.
5
For triangular elements the task of achieving interelement compati
bility was found to be even more difficult. The use of higher order poly
nomial representations of displacement functions and its derivatives is one
solution. In this case, however, additional nodes placed along the sides
and/or reference points located inside of the elements have to be utilized.
For the triangle it is impossible to achieve complete compatibility with any
thing lower than a complete quintic (21 terms) representation. Bell [14J and
Argyris [15J adopted the complete quintic representation with the associated
21 degrees of freedom (normal displacement, and its first and second deriva
tives at each vertex and angular displacement at each mid-side point). Bell
[14J later eliminated the mid-side node to reduce the element to one with
18 degrees of freedom.
Argyris [15~16J adopted even higher order representations, i.e.,
sextic (28 term) and septic (36 term) representations. It is obvious that
the higher the order of the representation of the displacement that is used,
the more costly the solutions become. Thus while some were adhering to
achieving full interelement compatibility by using higher order polynomials,
others sought an adequate low order incompatible element [8,17J.
The use of a cubic polynomial coupled with the concept of area
coordinates, which simplifies the development and is effective in avoiding
geometric anisotropy, has been successfully employed by Bazeley [17J and
others [18J even though it is an incomplete polynomial. The approach using
subdomain triangles was used by Clough and Tocher G19J and Clough and Felippa
[20J. The resulting elements are 9 D.D.F. and 12 D.D.F. compatible elements,
respectively.
6
Thus far, the discussion has been limited to formulations within
the scope of a minimum potential energy principle or displacement models.
Fraeijs de Veubeke [21J worked on one of the early stress models basing his
work on a minimum complementary energy principle. He showed that upper and
lower bound convergences on the solutions are obtained by the stress and by
the displacement models, respectively.
The mixed model which involves simultaneous assumptions on dis
placements and stresses for each element was used first by Herrman [22J.
The compatibility and equilibrium are easily achieved by expanding the vari
ables into relatively low order polynomials. Herrman used linear variations
for the triangle. With this assumption, however, the results do not appear
to demonstrate any marked superiority over other models. This mediocre be
havior seems to be due to too simple an assumed variation of the selected
variables. Adoption of a higher order representation of variables by Visser
[23J showed some improvement in the solutions.
Herrman [22,24J did not take Reissner1s Principle as a starting
point of formulation, but later works in mixed model by Prato [25J, Pian and
Tong [2J and Will and Connor [27J are based on Reissner1s Principle. There
are several versions of element methods based on Reissner1s principle depend
ing on the selection of the continuity requirements to be imposed along the
boundary [28J.
Formulations utilizing the hYbrid model, a terminology introduced
by Pian [26J, are based on either 1) assumed equilibrating stress fields in
side of each element and compatible displacements along the interelement
boundary (hybrid stress model) or 2) assumed continuous displacements inside
of each element with equilibrating surface traction along the interelement
7
boundary (hybrid displacement model). Again, there may be several modified
versions of these models. The detailed classification of the variations of
this model appears in Ref. 28.
Tong [29J pointed out the advantages of the hybrid displacement
model over the hybrid stress model. In a hybrid displacement model the as
sumed displacements need not be compatible at the beginning. The compatibil
ity along the boundary can be produced by introducing a Lagrangian multiplier
which is physically interpreted as the self-equilibrating boundary tractions.
Harvey and Kelsey [30J followed this approach for flat triangular elements.
The first, and perhaps simplest, finite element approach applied
in shell analysis is the approximation of the surface by an assemblage of
flat elements. This can be done relatively easily by combining the uncoupled
membrane and plate bending behavior through coordinate transformations as
necessary. In this way virtually all of the plate bending elements, in proper
combination with plane stress elements, can be used in shell analysis. Green
et al. [31J first used this approach in shell analysis. Johnson and Clough
[32J and many others successfully applied flat elements in shell analysis.
Chu and Schnobrich [18J used flat triangular plate elements with higher order
displacement functions, resulting in an element with 27 generalized displace
ments, in an attempt to provide compatibility between shell and eccentrically
placed supporting beams.
It is noted here that even with the discrete errors caused by ap
proximating the curved shell surface by flat elements and the lack of inter
element compatibility of some displacement components, in particular, the
discontinuity of slopes, the flat element solutions, generally, converged
to the real solutions within a reasonable limit.
8
Finally curved shell elements began to appear in finite element
shell analysis in the late 60's. Some of the difficulties experienced in
curved shell formulations are: 1) the inplane and bending actions are now
coupled and can no longer be calculated separately, 2) the continuity of dis
placements becomes more difficult to achieve because this now requires the
three displacement components (u, v, w) to be of equal order when the dis
placements are expressed in the global coordinate system, 3) the inclusion
of rigid body modes is much more difficult to achieve, especially when the
displacement is described in curvilinear coordinates.
Connor and Brebbia [33J expressed the radial displacement (w) by
a 12 term polynomial and the membrane displacements (u, v) with linear ex
pansions for a shallow shell element whose geometry was defined by a quadratic
function of the coordinates. For this element, the rigid body modes are not
included nor is the compatibility completely satisfied due to the disparity
in the order of polynomials used for radial and membrane displacements.
Pecknold and Schnobrich [34J enforced rigid body motion in one of their
curved elements by introducing additional unknowns and writing the additional
constraint equations.
Bogner et al [35J used 16 term polynomial for all tnree displace-
ment components resulting in 48 D.O.F. circular cylindrical shell element.
Rigid body modes are not included in this element so it must be used only
with small arc dimensions. Compatibility requirements are however satisfied.
Cantin and Clough [36J developed a curved cylindrical element including rigid
body motion. Utku [37J achieved a relatively simple algebraic representation
for curved triangular elements by using linear variations for the three
9
displacement and two angular displacement components. Argyris and Scharpf
[16J proposed higher order variations, i.e., complete quintic polynomial for
all three displacement fields.
While the efforts to achieve complete satisfaction of convergence
conditions, as discussed earlier, have continued, it has been shown that even
with elements which do not satisfy the above conditions, the satisfactory
results can be obtained and the justification of using higher order formula
tions essential to achieving the strict convergence conditions must be ques
tioned.
Bazeley et al. [17J dramatized the above question by showing that
a nonconforming triangular element can produce superior results over conform
ing elements in plate bending. Pecknold and Schobrich [34J also found that
the conforming and nonconforming elements used in investigating skewed shallow
shell problems produced almost exactly the same numerical results. The non
essentialness of inclusion of complete rigid body modes has also been reported
in Ref, 38. It has already been shown that many of the previously discussed
elements [8,18,33J produced satisfactory convergence characteristics with one
or more of the conditions relaxed.
Recently the isoparametric element and its modifications for shell
analysis have received much attention. Ever since the first introduction of
the isoparametric element [39J, many attempts have been made to improve the
bas i c accuracy of ttli s element. Among the illOi'"e s i gni fi cant deve 1 opments in
the use of the isoparametric element for shell analysis are the modification
of a three dimensional isoparametric element to ordinary shell element by Ahmad
and Irons [40J, the reduced integration technique by Too and Irons [41J,
and addition of nonconforming displacement modes by Wilson et al. [42J.
10
The latest development in the finite element field is the extension
of the linear analysis to nonlinear problems. The major areas of interests
include elastoplastic analysis [43J, cracking problems [44J, combination of
material and geometric nonlinearities [45J and even dynamic response of non
linear structures [46J.
1.3 Object and Scope
The behavior of the isoparametric element for three-dimensional
problems is well established and documented and the efficiency of this type
of element is also well known. The application of this element to thin shell
problems, however, is quite limited. Only the cubic elements produce reason-
able accuracy. Due to the many degrees-of-freedom involved, the economy of
the use of this element is questionable. There still exists the need for
developing a new economical shell element based on the isoparametric element.
The main objective of this study is to develop such a versatile,
yet economical, shell element based mainly on isoparametric concepts. The
weaknesses which the original isoparametric element has are corrected by
means of the addition of nonconforming displacement modes. The most beneficial
combinations of these additional modes are investigated for various orders of
elements, i.e., linear, quadratic, and cubic elements.
A computer ,program which is capable of handling problems of various
order elements, various shapes of elements and various types of shells, is
developed.
Some selected numerical results for plate and shell problems are
presented and compared with those from classical solutions, from other well
established fin~te element models and from experimental studies, whenever
11
available, in order to evaluate the efficiency of the elements presented in
this study.
The major findings and the suggestions for future studies based on
the findings of this study are presented at the conclusion of this study.
1.4 Notations
The symbols used in this study are defined where they first appear.
Following is a summary of the symbols which are used frequently.
A
B, B
c
I
0, 0
E
e
F
i ,j
J, I J I
area of an element
matrix relating nodal displacement and strains, con
forming and nonconforming part, respectively
constraint matrix
material property matrix in global and local coordinate
system
Young's modulus
superscript indicating lIelement"
body force components
subscripts indicating node i and nonconforming mode j
Jacobian and determinant of Jacobian
stiffness matrix of structure and stiffness matrix of
an element
M , M , M )( Y xy
N, N
N , N , N x y xy
p~ 1
Q
R
r
S
s
T
to' t. 1
U, U. 1
IT
12
submatrices of Ke
condensed element matrix
moments in x and y directions and twisting moment
shape functions in curvilinear coordinates, conforming and nonconforming, respectively
forces in x and y directions and shear force
generalized load at node i in an element
interpolation functions for surface tractions
intensity of uniformly distributed load and intensity
of load at node i
load vector
boundary tractions
surface
corrective factor for shear
surface traction
uniform thickness of shell and thickness at node
nodal displacement vector and displacement at node i
amplitude of nonconforming displacement modes
u, v, w
I
U , V , W
v
cSw
X, X. 1
x, y, z
, I
X , Y , z
Ct, B
E, E
Ex,Ey'Yxy '
Yxz,Yyz
8
KX I 'K Y I 'K)( I Y
13
components of displacement in the x, y, z directions
components of displacement in the x , y , z directions
volume
J I
unit direction vectors in the x , y , z directions,
respectively
virtual work
global coordinates and coordinates at node i
global coordinates (components of X)
local coordinates
J
rotations about the x ,y axes
strain vector in global and local coordinate systems, respectively
components of E
curvilinear coordinates
direction cosine matrix = [v l ' v2' v3 J
curvature changes of middle surface of shell
Poissonis ratio
potential energy function
0, 0
cp. 1
14
stress vector in global and local coordinate system, respectively
=
15
Chapter 2
ISOPARAMETRIC ELEMENTS
2.1 General
One of the approaches to obtain the desired displacement field is
to represent it by Hermitian interpolation (or shape) functions [llJ as has
been discussed in the previous chapter. The similar approach using Lagrangian
interpolation functions may be used to begin the development of the isopara-
metric element, so designated because it uses the same shape functions for
representing both the global coordinates and the displacements. Zienkiewicz,
Irons and their co-workers at the University of Wales, Swansea [39-41,47,48J
are the most prolific contributors to the development and application of this
type of element.
The basic concept and derivation of this element has been well
publicized and one need only refer to Zienkiewicz's textbook [47J or the
references cited above for complete details of the basic approach. Therefore,
a detailed development of the basic element will not be presented here. How-
ever, the significant characteristics of the isoparametric element are sum-
marized below:
1 . The same shape functions are used for defining the geometry
and the unknown displacement variables.
X I N. 1
x. (2. 1 ) 1
U L N. 1
U. 1
(2.2)
16
where X = global coordinates
N. = shape function at node i (Appendix A) 1
x. = coordinates at node i 1
U = global displacements of any point in the element
U. = displacement at node i 1
2. Gausian quadrature is the only practical method to calculate
the element properties.
3. Geometrical conformity and displacement continuity are satisfied
after deformation.
4. The constant derivative condition is satisfied for all iso-
parametric elements. This means that rigid motions do not
cause strain.
5. In order to calculate the stiffness matrices~ transformations
are necessary to evaluate the global derivatives from curvi-
linear derivatives and to perform volume integration.
The isoparametric element is most advantageous when applied to three
dimensional structures, but it can be extended to plate and shell problems as
well. In this study the investigation is limited to plate and shell elements
derived from the three dimensional isoparametric element.
2.2 Isoparametric Shell Element
A three dimensional "Brick Type fl isoparametric element (Fig. 1)
may be used directly for plates and shells. If the thickness of the ele-
ment is reduced enough in comparison with the element side length it will
actually function as a plate or shell element. This type of element, however,
17
has two weaknesses or faults which restrict its practical application. These
deficiencies are:
1. The element is too stiff against flexure. This is the result
of IIparasitic shears ll which develop because of the constraints
placed on the transverse displacement mode. The phenomenon is
particularly severe for low order elements. Figures 2b and 2c
show the constrained mode and the true mode of the linear ele
ment when subjected to pure bending.
2. Too large a number of variables have to be included as com
pared with conventional shell elements. For instance, the
cubic isoparametric element (Fig. lc) has 32 nodal points which
translates into a total of 96 unknowns while most conventional
quadrilateral shell elements have only 20 unknowns. This mul
tiplication of unknowns results from not taking into account
such things as normals remaining normal.
So far, no specific assumptions which a conventional shell element
utilizes have been made. Therefore, transverse shear deformation is taken into
account for this element. The retention of the shear effect is important
particularly when the element is used for thick shell problems where the trans
verse shear effect is no longer negligible.
The correction of the deficiencies of the isoparametric element given
above is the subject of the following sections.
2.3 Ahmad-Irons Modification
The need to require an excessive number of unknowns, listed as
second undesirable feature of original isoparametric shell element may be
18
corrected in part by the application of the well known Kirchoff's hypothesis,
i.e., the normal to the mid-surface remains straight and normal after defor-
mation with the corresponding strain energy ignored. Thus, the unnecessary
nodes thru the thickness of the element can be eliminated (Figs. ld and l~).
With this linear expansion in the normal direction ~ there remains now a
total of six degrees of freedom for each set of top and bottom nodes (Fig. 4a).
Ignoring the elongation in the normal direction ~ in addition to
the above assumption, Ahmad et ale [40J replaced the six degrees of freedom
by three mid-surface displacements (u, v, and w) in the global directions
(x, y, and z) and two rotational displacements (a and S) around the two
1 ()r~ 11 \I rlo-f"i nor! :::l voe (v I ::> nrl \I I \ • ' .. n ......... 'J '-"'- I • "'-,"" UA'-..) \ A UIIU.J I. Thus the element has five basic degrees
of freedom at each node just as a conventional shell element would have. The
definitions of the curvilinear coordinate system (~, n and ~), the global I i
coordinate system ~,y, and z) and the localized coordinate system (x , y , i
and z ) are""given in Appendix B.
Now the Cartesian coordinates (x, y, and z) of any point in the
shell element is expressed thru its curvilinear coordinates (~, n and ~) and
the mid-surface global coordinates. Thus Eq. 2.1 of the original isopara
metric element is changed [40J to the form
x X. 1
Y L N. y. + L N ~ \/3i (2.3) 1 1 i 2
z Z. 1
where the vector \/3" " 1 1.s given as
19
x. X. 1 1
V3i = y. y.
1 1
Z. z· 1 J top 1 bottom
For a shell with constant thickness, t , o
and
N. = shape functions in curvilinear coordinates 1
to = constant thickness of shell
v3i unit vector normal to the shell surface at node i
(2.3.1)
(2.3.2)
The corresponding displacement components (u, v and w) in the
directians of global axes (x, y and z) are expressed in the form
where
u u. 1
-:: ] 1
L N. L N. £ t o cP i (2.4) v = v.
I +
1 1 1 2
w W. 1 J
I I
a.,S· = rotations around the axes (x ,y ) in the directions of 1 . 1
vl and v2 respectively at node i
cP . 1
=
One may refer to Appendix B of this thesis for the reference axes I I
(x ,y and z ) and their direction vectors (vl ' v2 and v3) and Appendix A
for the definition of the shape functions. It is noted here that the shape
20
functions in Eqs. 2.3 and 2.4 are defined by only two coordinates (~,n) while
those of Eqs. 2.1 and 2.2 are defined by three (~, nand s).
This element substantially reduces the number of unknowns from the
original isoparametric element, but it still retains most of the character
istics of the original element. Therefore, the first mentioned deficiency
of the original element (that it is much too stiff in bending) has been left
virtually untouched.
Some numerical results using linear, quadrilateral and cubic ele
ments are given in Fig. 10 to demonstrate the behavior of this type of ele
ment. The figure shows the central deflection of a clamped square plate
under a concentrated load placed at the center plotted against the mesh size
for the various elements. As expected, only cubic element shows fairly good
convergence to the real solution while quadratic element is rather slow in
converging and the linear element behaves so badly that, unless the sub
division is reduced to uneconomical limits, the element is not applicable
to a plate or shell problem.
, Schemes for the improvement of this weakness are discussed in the
following two sections where a reduced integration technique and the addition
of nonconforming displacement modes are described.
2.4 Reduced Integration Technique
It has been noted in the previous section that the low order ele
ment is far too stiff to be an economical element to use when the depth of
the member is small compared to its. plan dimensions. Figure 2 explains this
source of error present in the use of the element. The constraints in the
21
normal direction n require the strains to be as shown in Fig. 2b. Associated
with this configuration is the tendency to absorb too much of the deformation
energy by shear energy.
One of the ways to overcome this difficulty is the elimination of
the excessive shear energy by the means of a selective reduction of the
order of numerical integration. This technique was first introduced by
Doherty et al. [49J and later elaborated on by Pawsey [50J. This procedure
was also investigated independently by Zienkiewicz et al. [41J. Marked im-
provement was shown in the problems these groups investigated.
The.basic reasons why the simple reduction in integration order
produced such improvement are:
1. The reduction in the order of integration tends to reduce
the value of stiffness of an element below the value exact
integration produces. Since the element starts too stiff,
the softening has desirable effects.
2. The reduced integration can exclude the regions where the dis
placement may be over constrained. For the linear element (Fig.
2a), for instance, a one point numerical integration at the center
of the element where the shear strain is zero results in zero
shear energy. Similar reductions occur with the quadrilateral
element when a 2 x 2 'integration is used.
Obtaining the stiffness matrix of an element involves evaluating
an integral of the general form
r r r T J J J B d B IJI d ~ d n d ~ (2.5)
where
22
/JI = determinant of Jacobian
D = material property matrix
B = matrix relating nodal variables and strains
The integration is most readily carried out numerically using Gaussian
quadrative with an integration order of 2 x 2, 3 x 3, or 4 x 4 mesh of Gauss
points for the linear, quadratic or cubic elements respectively [50J. These
integration points are necessary to evaluate the volume of an element ade
quately.
As pointed out in the beginning of this section, the source of the
errors (Fig. 2) is excessive shear energy. Therefore, the reduced integra
tion order needs to be applied only to the transverse shear component while
maintaining the usual complete integration on the other strain components.
The integration procedure, however, should be divided up so that each com
ponent of strain energy is integrated over its own grid of integration points.
Although the selectively reduced integration improves the perfor
mance of the element [41,50J, a uniform or all-round reduction of integration
order, for instance, 2 x 2 integration on all the strain components for a
quadratic element, produces even more rapid convergence and better ac-
curacy [41J. The reasons of such an improvement, however, are not readily
explained mathematically.
Two-fold benefits accrue from this technique. These are the im
provement in accuracy but also coupled with a saving in the time required
to calculate the element stiffness.
2.5 Addi.tion of Nonconforming Displacement Modes
Another general approach to improve the basic accuracy of the
23
isoparametric element by eliminating the excessive shear energy is thru the
addition of nonconforming modes. Instead of disregarding the shear energy
during the integrating procedure, Wilson et al. [42J adopted a new approach,
that of eliminating the transverse displacement constraints which cause the
excessive shear strain in the element. This is achieved by adding extra dis-
placement modes. The additional displacement modes are of the same form as
the errors in the original displacement approximation.
A two-dimensional element under pure bending, for instance, is
shown in Fig. 2a. The exact displacement field for this type of loading
(Fig. 2c) is of the form
where
u
v 1 2 '2 (Xl (1 - E;, )
2 + (X2 (1 - Tl )
(Xl amplitude of the selected displacement shape
= a function of material properties, for w = 0, cx2
(2.6)
(2.7)
o
The linear isoparametric element is capable of representing the
displacement component u (Eq. 2.6) but not capable of including the trans
verse displacement component v (Eq. 2.7). This constraint in the transverse
direction results in extraneous shear strains (Fig. 2b). What was lacking
in the representation of the displacement field by the isoparametric finite
element is the very source of the error. The addition of extra displacement
modes in the same form as Eq. 2.7,
v + 2 13 2 (l - Tl ) (2.8)
24
results in the elimination of the extraneous shear strain and the restoration
of the actual behavior of an element (Fig. 3). These additional modes gen-
erally destroy the compatibility of the isoparametric element.
Now the total displacement field of any element can be expressed
by adding the additional modes to the original isoparametric element displace-
ment field, Eq. 2.2. Thus,
where
u \ N. U. + L\ N. U. L 1 1 J J
N. additional nonconforming displacement modes J
U. = additional unknowns corresponding to the additional J
displacement modes
(2.9)
These additional unknowns can be taken as the displacements of a
selected set of internal modes of the element simply as a set of generalized
coordinates associated with the respective nonconforming modes. The addi-
tional unknowns are also determined by minimizing the strain energy in the
element. The resulting stiffness matrix, using the displacement field ex-
pressed by Eq. 2.9, has been enlarged over the usual isoparametric element
matrix due to the additional modes and the corresponding unknowns. This
enlarged matri~ however, can be partitioned then condensed back to the order
of the original matrix, i.e., without the additional modes, by means of
static condensation. This is possible since the additional unknowns occur
only in the equations applicable to the specific element into which the non-
conforming modes have been applied.
25
The examples given by Wilson et al. [42J demonstrate marked im
provement over the low order element. Convergence to the true solution is
obtained. Their work~ however, was limited to the cases of plane strain and
thick shell problems. The necessity of developing a new general thin shell
element still exists.
The following chapter deals with the development of elements based
on the approaches discussed in this chapter.
26
Chapter 3
ELEMENT FORMULATION
3.1 Variational Principle
As seen in Section 1.2, the variational principle has been fre-
quently used as the basis of element formulation [2J. For a typical displace-
ment model, without any boundary constraints, the total potential energy
functional TIp may be expressed, utilizing tensor notations, as
where
- f T.U. d5 50 1 1
TIe = potential energy functional for element e p
s. . = strain tensor components lJ
Eijkl = elastic constants
v
F. 1
U. 1
5
T. 1
=
=
=
=
=
volume
body force component
displacement
surface
surface traction
(3. 1 )
50 -" portion of 5 over which the surface tractions are pre-
scribed
27
The displacement U is defined by suitable functions expressed in
terms of variables at nodes, i.e.,
U = [N] U. 1
E = [B] u. 1
where subscript i indicates node i, not tensor notation, and
[N] interpolation functions
[B] matrix relating strains to nodal displacements
Substituting Eqs. 3.2 and 3.3 into Eq. 3.1, one gets
It p r
= J v
(-21 U~ BT EBU. - U~ NT F) dV -1 1 1 J
r U ~ NT T dS So 1
(3.2)
(3.3)
(3.4)
where the subscripts for tensor notation have been dropped for reasons of
.simplicity. This equation can be written in a more compact form as
where
Iv BT E B dV
J NT F dV +
v NT T dS
(3.5)
(3.5.1 )
(3.5.2)
e e K and R are referred to as element "stiffness matrix" and "load vector,'1
respectively.
Taking variations with respect to the nodal variables U., then 1
Eq. 3.5 yields
o (3.6)
28
Assembling the element stiffness matrix and load vector into whole structure,
KU R = 0 . (3.7)
Thus a system of linear algebraic equations, which is known as the
IIload-displacement equation,1I is derived.
3.2 The Basic Geometry and Displacement Field
The isoparametric shell element and some improvements and modifi-
cations to it have been described in the preceding chapter. Some of the con-
cepts in the element development as described are directly utilized in this
study. The Ahmad-Irons definitions for the shell geometry and displacement
field, Eq. 2.3 and Eq. 2.4, respectively, are taken as the starting point of
this study, The equations expressing these definitions are rewritten here
as
and
r : L
Z
u
v
w
X. 1
= L N. y. 1 1
Z. 1
u. 1
= L N. v· 1 1
w· 1
(3.8)
t. [:',.' 1 + I Ni -+ L, <Pi JJ J (3.9)
It is noted again that if the shape functions N. are such that no 1
volume discontinuity or gap or overlay is produced along interelement bound-
aries by unique coordinates at the nodes, the displacement compatibility is
also maintained at the interelement boundaries.
29
The expression defining the mapping of the element as indicated by
Eq. 3.8 is used throughout this study without any modifications to represent
the global coordinates of any point of the shell. However, the equations ex-
pressing the displacement such as Eq. 3.9 are used only as a starting point
in this study and subsequent modifications are made to these equations.
3.3 Nonconforming Displacement Modes
As discussed in Section 2.5, one of the techniques to get rid of the
excessive stiffness of the isoparametric shell element is thru the addition
of nonconforming displacement modes [42J. Thus, by sacrificing the inter-
element displacement compatibility present in the original isoparametric
element, it becomes possible to eliminate or at least minimize the built-in
constraints that the element has against the transverse displacements. It
was shown in Section 1.2 that for displacement models strict interelement
boundary displacement compatibility is not essential in order to assure
convergence. This fact is used in this study as the basis for relaxing the
strict compatibility requirement.
Thus, in addition to the basic shape functions defined in Appendix
A, the following nonconforming displacement modes are introduced. For a
linear element, the nonconforming shapes to be appended are
Nl = ~2
N2 2 = n
N3 = n (1 ~2) (3.10)
N4 = E,; ( 1 n2)
N5 ( 1 - E,;2) (1 ? ,
= n-)
30
while for a quadratic element~ the shapes are expressed by
Nl = ~ (1 ~2)
N2 nO 2 = n )
N3 ~n(l ~2) (3.11)
N4 = n~(l n2
)
N5 ( 1 n2) (1 _ ~2)
As seen in Section 2.3 (Fig. 10), since the cubic element behaves in
an adequate manner and there is already a sizeable number of degrees of freedom
associated with that element, no additional displacement modes are or should
be added to improve its behavior. The cubic element is used primarily to
evaluate the accuracy of the new nonconforming elements by comparison.
The additional modes in Eqs. 3.10 and 3.11 are so selected to have
zero values at each node and to eliminate the undesirable constraints present
in the original isoparametric shapes. For instance, the first two modes in
Eqs. 3.10 are designed to eliminate the transverse displacement constraints
in the same way as was discussed in Section 2.5. The 3rd and 4th modes in
Eqs. 3.10 contribute to the softening of the twisting constraints while the
5th mode adds the bubble shape that provides for displacements totally within
the region.
The additional modes for the quadratic element~ as listed in (Eq.
3.11), are selected along the same lines as for the linear element. These
modes are illustrated in FigS. 5 and 6 for linear and quadratic elements,
respecti ve ly.
Now the general form of the displacement field with nonconforming
31
modes included can be formed by adding the extra modes of Eq. 3.10 or Eq.
3. 11 to Eq. 3.9. Thus,
'\ N. L 1
u. 1
v· 1
W. 1
+
u. J
v. J
+
w. J
(3.12)
Equation 3.12 is the general case where all five additional modes
(Eq. 3.10 or 3.11) are applied to all five variab1es (u,v,w, aand G) at each
node. It may be unnecessary, even excessive, to carryall nonconforming
modes for all displacement components. Slower convergence may result or the
highly nonconforming element may prove uneconomical. The optimum combination
of additional modes for each nodal variable is investigated in Section 3.7.
3.4 Strain-Displacement Relationships
Once the displacement field is defined, the strain-displacement relation-
ship is established by combining the proper derivatives of the displacement com-
ponents. If the basic shell assumptions are adopted, it is most convenient to
erect a local coordinate system (Xl, yl, Zl) as defined in Appendix B and to take
the displacement derivatives with respect to this coordinate system.
The components of strains in the local coordinate system are
32
I
au E: I
D
X ax I
av E: I
I Y ay I B
Y I I ~+~ { 11 x Y I I
(3.13) E: J = = ay ax I I
Yx'z I aw + ~ I I
ax az I I
Y I I ~+~ y.. Z I I
ay az
where E: I is neglected in order to be consistent with the assumptions made . z in Section 2.3.
Before obtaining the strain vector (Eq. 3.13), it is convenient to
establish first a matrix consisting of the local derivatives of all displace-
ment components (Eq. 3. 14) . The terms needed for the strain vector can then
be selected and assembled.
r~ I I
1
av aw I I
ax ax ax I I !
au av aw (3.14) I I I
ay ay ay I I I
au av aw J I I
az az az
Since the displacements are defined in a global coordinate system
(Eq. 3.9) and the shape functions, N. and N. in Eq. 3.8 and ·Eq. 3.12, are 1 J
expressed in terms of curvilinear coordinates (~,n,~), two elaborate trans-
formations are necessary in order to obtain the strains in a local coordinate
33
system. The operations of these transformations are given in Appendix C.
The resulting expressions for strain components in the local coordinate sys-
tem with the global displacements at the nodes are
u.
cis\[:;j u.
[ ~~ J 1 t. J t.
{c:} B.8 T 1 B.8T J - T v. + 2 + v. + -2 C.8 cpo
1 1 J J J J
w. W. 1 J
(3.15)
where B., C., B. and C. are defined in Appendix C. 1 1 J J
For simplicity the strains in Eq. 3.15 are expressed with two sep-
arate parts, i.e., conforming (ordinary isoparametric element) and noncon-
forming parts; thus
{c} [8, -B ] l ~ 1 (3.16)
3.5 Stress-Strain Relationships
The stress components corresponding to the strain components are
given by
a i -1
X
a I y
I
{a } = a I I
x y (3.17)
o "j x z
a I I
Y z
I . I
where the stress vector {a } and material property matrix [0 ] are expressed
34
again in the local coordinate system (Appendix B). It is also noted that
is neglected to be consistent with shell assumptions.
o I
Z
For simplicity, a homogeneous linearly elastic material whose pro-
perties are the same in all the directions is considered. For this material I
the matrix [0 J is of the form
where
[O'J E 2
- ]J
l sym~ E Young's modulus
]J = Poisson's ratio
- ]J
2 (3.18)
A numerical corrective factor s is needed since the actual shear
distribution across the thickness is quadratic whereas a constant (or aver-
age) shear distribution is assumed as a consequence of the shell assumptions.
The ratio of actual shear strain energy to the assumed shear strain energy is
approximately 6/5 for rectangular beams [51J and this corrective factor is
adopted in this study.
3.6 Element Stiffness Matrix
I I
Once the material property matrix [0 J and the strain vector {€ }
are obtained, the element stiffness matrix is obtained by the direct application
35
of variational principles as shown in Section 3.1. The thus obtained stiff-
ness matrix is given in Eq. 3.5.1, or
(3.19)
where the superscript e denotes the individual element and the volume dV has
to be transformed for the integration process as
dV = dx dy dz IJI dt;: dn de: (3.19.1)
where IJI = determinant of the Jacobian matrix.
Equation 3.19 is now of the form
(3.20)
The explicit integration of this equation is so complicated that
numerical integration is the only practical way to achieve the integration.
An outline of the Gaussian numerical integration method is given in Appendix
D. More details can also be found in References 47 and 50.
With the displacement vector U and the [S] matrix subdivided into
conforming and nonconforming parts, then Eq. 3.7 may be rewritten in the form
::: 1 [~l = [: 1 (3.21 )
where
r ST E S dV Jv
(3.21.1)
f ST E B dV (3.21.2) v
36
K22 ~ I v B T E B dV (3.21.3)
The submatrix Kll is the ordinary isoparametric element stiffness
matrix and Kl2 and K22 have been formed from the nonconforming displacement
modes. Thus the element stiffness matrix has been enlarged. The expressions
for typical components of the element stiffness matrix are given in Appendix
E.
Separating Eq. 3.21 into two matrix equations
Kl1U + Kl2U = R (3.22)
T -K12~ + K22U = 0 (3.23)
The nonconforming modes are condensed out first. Solving Eq. 3.23 for U,
the amplitude of the additional modes are found to be,
U = (3.24)
On substituting IT into Eq. 3.22, the variables for the nonconforming displace-
ment modes are eliminated. The final set of equations to be used is
i
K U R (3.25)
where I
K = (3.25.1)
I
Thus, the newly formed element stiffness matrix K has been con-
densed back to the same' order as an ordinary isoparametric element stiffness
matrix Kll"
37
It should be noted, however, that the advantages which the additional
nonconforming modes provide may not be worth the cost if the operations in-
vo1ved in forming matrices K12 and K22 and in subsequently condensing them I
back out to form matrix K (Eq. 3.25.1) are too expensive. Since the cost of
calculations depends on the size of the additional matrices, K12 and K22 , it
is desirable to reduce the order of such matrices as much as possible without
hurting the accuracy of the nonconforming element.
This can be achieved in part by applying the nonconforming modes
selectively to the various variables in much the same fashion as selective
reduction of integration points (Section 2.4) has been used.
3.7 Selection of Nonconforming Modes
Thus far it has been assumed that all five additional modes in Eq.
3.10 or 3.11 are applied to all five variables at each node, Bakhrebah [52J.
Examining the behavior of the isoparametric elements (Fig. 2), however, it
can be observed that not all the displacement components are constrained to
the same degree. While the transverse displacement in Fig. 2 is severely
constrained, the longitudinal displacement may be adequately represented by
the linear variation present in the isoparametric element.
This indicates the strong desirability of using a selective addition
of nonconforming modes. The reduction of additional modes applied to the
conforming modes is particularly important because the numerical integration
and matrix condensation procedures in forming the final stiffness matrix are
expensive and maybe more so than is necessary.
For a linear element, inclusion of the fifth mode of Eq. 3.10 re-
sults in a singular matrix for K22 . Therefore, that mode is deleted from
38
further consideration. The next question raised is the validity of applying
the nonconforming modes to the rotations. As seen in Section 2.3 the rota
tions at the nodes (a and S) are approximated as the average difference, in
inplane displacements (u and v) across the thickness. Thus, the rotations
at node i are expressed by
(v . v. ) ltop lbottom
a :::::
t. 1 (3.26)
(u. u. ) S ::::: 'top 1 bottom
t. 1
According to the approximations in Eqs. 3.26, the rotations are
established solely from the inplane displacements. Therefore, if the inplane
displacements are satisfactory with the original element, no additional ro-
tational nonconforming modes are necessary or even desirable for the new
element.
It has been seen that the parity of orders of polynomial represen-
tations for the displacement components u, v and w is not essential for a dis-
placement model [34J. A linear expansion for displacement components u and v
and cubic expansion for w has been used with a great deal of success [32,33J.
This should also hold true for nonconforming elements. The linear represen-
tation of the u and v displacement component is exact for the case of constant
moment and quadratic representation should be sufficient for linearly varying
moment. Therefore, the addition of the first two nonconforming modes of Eq. )
3.10 to the displacement components u and v should be sufficient approximation
for a linear element. For a quadratic element which already has second order
39
terms in representing u and v displacements, the addition of the nonconforming
modes to u and v displacement components is less important than in the case of
a linear element.
Finally, the vertical displacement component w can be approximated
adequately by a cubic representation. For both the linear and the quadratic
elements, the inclusion of the 3rd and 4th modes from Eq. 3.10 or Eq. 3.11
(twisting modes) are essential to eliminate the twisting constraints. The
decision as to whether the 5th mode of Eq. 3.11 is essential to improve the
accuracy of the quadratic element is difficult. Further discussion of this
matter is delayed to Chapter 5 where the numerical results are available to
form the basis of a judgement.
From among the possible nonconforming modes those considered to be
most beneficial have been predicted to be NC4-4.3R for the linear element and
NC8-4.3R or NC8-5R for the quadratic element (see Table 1). For both the
linear and the quadratic cases, the reduction of size of matrices due to the
addition of only selected nonconforming modes is significant. For instance,
the order of matrix K22 in Eq. 3.21 is 25 x 25 for both linear and quadratic
elements using all nonconforming modes and before the reduction. This matrix
K22 is reduced to order 8 x 8 after selected reduction.
3.8 Forced Interelement Boundary Compatibility
Due to the introduction of nonconforming displacement modes (Sec
tion 3.3), the element used in this study starts with a violation of compati
bility along the interelement boundaries. This may be in the form of gaps,
overlaps and/or kinks along these boundaries. Initially, it was considered
40
desirable to re-establish compatibility along these boundaries. One of the
ways to re-establish this interelement boundary compatibility is by the means
of self-equilibrating boundary tractions which push, pull and/or rotate the
boundary back to the compatible position.
For this case, the element described in this study becomes equiva-
lent to that known as.a hybrid displacement model. For a hybrid displacement
model, the equations equivalent to Eq. 3.7 are given as
KU R + T C r = o
c U = '0
or in matrix form
where
cT = I N Q dS as
Q interpolation function for boundary tractions
r = nodal tractions
as = portion of boundary where the tractions are applied.
(3.27)
(3.28)
The matrix C is known as a constraint matrix. It is simply the statements
of compatibility along the interelement boundaries.
The second equation of Eqs. 3.27, which is a system of homogeneous
linear equations, generally does not have unique solutions except zeros for
all unknowns U. Of course, if the matrix C has a rank (r) less than the
number (n) of unknowns U, some r of the unknowns can be solved in terms of
41
the remaining n - r unknowns. In this case, there are infinite number of
solutions. Substituting these solutions into the first equation of Eq. 3.27~
the final solutions can be obtained. The procedure, however, is not as simple
as setting up the equations themselves.
Another alternative approach is to eliminate U before r in the
equation solving process. Solving the first equation of Eqs. 3.27 for U,
(3.29)
and substituting U into the second equation of Eq. 3.27
= o (3.30)
Solving the Eq. 3.30 for r and substituting it into Eq. 3.29, one gets nodal
displacement U and nodal traction r. The solution is costly, however, mainly
because of the necessity to invert the stiffness matrix K and in the process
loose the banded character of the enlarged matrix to be solved. This could
be avoided if the equations were properly ordered in the machine with the
constraint equations first and the variables in proper order.
In this study it was originally intended to re-establish complete
interelement compatibility by the means of boundary tractions thru the process des
cribed above. This operation being applied after having added the nonconforming
modes which destroyed the compatibility to begin with. The performance of the
nonconforming modes, even 10111 order element, however, was so satisfactory that
it has been found unnecessary to apply the constraint equations.
For the above two reasons, i.e., cost of the equation solving pro-
cess and the already adequate nature of the solution, the idea of re-establishing
compatibility of the interelement boundaries has been abandoned in this study.
42
Chapter 4
ASSEMBLY AND SOLUTION OF THE LOAD-DISPLACEMENT EQUATIONS
4.1 General
By the application of one of the variational principles (Mi~imum
Potential Energy Principle in this study) the general form of load-deflection
equation was obtained in Section 3.1.
KU R = 0 (4. 1 )
In this systems of lfnear simultaneous algebraic equations, the
stiffness matrix of the whole structure K and the load vector R are known.
Therefore Eq. 4.1 is solved for unknown displacement vector U.
Once the displacement U is obtained, the corresponding strains c
and stress 0 and/or forces can be obtained by the relations of Eq. 3.16 and
Eq. 3.17, respectively.
4.2 Assemblage of Stiffness t-.1atrixand Boundary Conditions
The stiffness matrix for the total structure K is formed by summing
the contributions of all the individual element stiffness matrices. This
initial stiffness matrix would be singular. The final assembled stiffness
matrix K must reflect the boundary conditions. Then Eq. 4.1 becomes solvable.
Boundary conditions are incorporated by deleting or modifying the
appropriate rows and columns from total stiffness matrix K corresponding to
the dis~lacementS·constrained or specified at appropriate nodes. The deletion
can be performed effectively during the stiffness matrix assembling process.
43
One effective method of assembling the stiffness matrix is the code number
system discussed by Tezcan [64J.
4.3 Load Vector
In the finite element formulations, the applied load must be trans-
formed to the nodal loads. Though the generalized load is the theoretically
correct and more acceptable load to use the statically equivalent load can
be used satisfactorily [34J.
The generalized load is defined as the equivalent nodal loads which
produces the same amount of work done as the actual loading would when moved
through the corresponding virtual displacement. For distributed loads, the
virtual work is expressed by
where
p~ 6u 1 f q. N. dA 6u
All
p~ = generalized load at node 1
6u = virtual displacement
q. intensity of load at nodes 1
N. shape function 1
in an element
cancelling the virtual displacement 6u in Eq. 4.2.
p~ 1
r
J q. N. dA
All
For the uniformly distributed load q , o
p~ 1
r qu- J N. dA A 1
(4.2)
(4.3)
(4.4)
44
Again the integration is carried out by the numerical integration method just
as with the element stiffness matrix.
The assemblage of load vector R can be performed in the same way
as the stifness matrix assemblage. The loads corresponding to the constrained
displacements should be omitted during assemblage.
4.4 Solutions of Load-Deflection Equations
Solving the simultaneous algebraic equations Eq. 4.1 following the
formation of structural stiffness matrix, is the central part of finite ele
ment method. Therefore, in order to achieve the economy of the method, it is
essential to take maximum advantage of well-known special characteristics of
the stiffness matrix, i.e., symmetric, banded and positive definite.
Gauss elimination procedure takes fewer operations than other
method using rational operations for a dense matrix of coefficients [54J.
One of the most efficient modifications based on Gauss elimination is one
known as the "frontal solution technique lf [55J.
Instead of using a band algorithm, the frontal method uses the
specific IIfrontl! concept. As the front advances, the element stiffness co
efficients are assembled and the variables which appear for the last time
at that front are separated and eliminated immediately. This elimination
process is the same way as the Gauss elimination. The size of the Iffront,"
however, is smaller or, at worst, same as the band width of the equation.
The frontal solution is one of the more effective methods for
element~ with nodal points at more than just the corners, or three dimensional
elements whose band width becomes very large. Branching structures is another
45
class of problem where the frontal method posses advantages. -For the element
with nodes at corners only this approach may not be as effective as for the
elements mentioned above due to the large input-output requirements of the
method. The frontal solution is adopted as the equation solver in this study.
4.5 Strains
The strain-displacement relationships have been established in
Section 3.4 (Eqs. 3.15 and 3.16).
(4.5)
Thus the local strains are calculated with the global displacements.
If only the values at the nodes are desired, then Eq. 4.5 can be
simplified by omitting the amplitudes of the nonconforming nodes IT and the
corresponding B. Since the nonconforming displacement modes have little
direct effect on strain values computed at the nodes, as defined in Section
3.3, U'and B need not be calculated. It achieves substantial savings in
calculation effort. Thus the nodal strains are now obtained by
I
{s} = [B] {U} (4.6)
It should be noted, however, that if the values are desired at
points other than the nodes, such as the Gaussian integration points, then
Eq. 4.5 with the nonconforming contribution rather than Eq. 4.6 is used in
order to obtain the best values at those points.
4.6 Stresses and Forces
Stresses in local coordinate systems are obtained by Eq. 3.17, i.e.,
46
I I •
{o } [0 ] {E } (4.7)
For the thin shell problems, it is convenient to obtain the forces in the
shell coordinate system. The axial forces per unit shell width are obtained
by the relations
And the moments are obtained by
where
M I = H(K I + 1-1K I) X X Y
M I = H(K I + 1-1K I) Y Y x
M I i = H( 1 ;1-1)(2K I.) X Y X Y
H = 2 12(1 - 1-1 )
K I, etc. x =
(4.8) ,
(4.9)
The approximations N ' I = N I I and M 8 I =- MIl are applied in x y y x x y y x
accorda~ce with thin shell assumptions.
47
Chapter 5
NUMERICAL RESULTS
5. 1 Genera 1
The series of numerical examples presented in this chapter are
1) to provide a numerical background for selecting the most beneficial COI11-
binations of nonconforming modes (Section 3.7), 2) to demonstrate the improve
ment of accuracy obtainable by the addition of these nonconforming displacement
modes and 3) to demonstrate the applicability of such a procedure to practical
problems.
In order to discuss the various combinations of nonconforming modes,
a special notational system was established. The designations for some of the
elements under consideration are tabulated in Table 1 using the notational
system described below. Taking NC4-4.3R, for instance, the NC4- indicates
the element is a 4- noded element (linear element) with ~on~onforming dis
placement modes. The figures _~~_ indicate that ± nonconforming modes (Eq.
3.10) are applied to 1 variables (u, v, w). Finally R indicates that a Re
duced number of nonconforming modes, in this case 2 out of possible 4 modes,
are actually applied to the inplane displacement components while all 4 non
conforming modes are applied to the normal deflection w (Table 1). For another
example, C4 indicates that the element is a fonforming ~-node element, i.e.,
original isoparametric element with no additional nonconforming modes.
A number of different types of problems were studied to assess the
selective nonconforming approach to the element development. In order to
evaluate inplane behavior and simple bending behavior of the various nonconforming
48
elements, a deep beam was investigated for various load combinations. Next
clamped rectangular plates were investigated for a concentrated load applied
to the center. These tests evaluate the plate bending behavior of noncon-
forming elements. All major elements listed in Table 1 have been used in
the above listed problem series, the results are compared and the most bene-
ficial elements chosen. Only those elements passing this preliminary screen-
ing were used in subsequent investigations.
A cylindrical shell and an elliptic paraboloid were investigated
to evaluate the basic behavior of nonconforming elements for single and double
curvature shells, respectively. A hyperbolic paraboloid, a spherical shell
and a conoidal shell were also investigated in order to demonstrate the versa-
tility of newly developed nonconforming elements.
5.2 Deep Beams
5.2.1 Cantilevered Beam
A deep cantilevered beam whose dimensions and material properties I
are shown in Table 2 has been examined for two different types of loadings.
The results are tabulated in Table 2 and compared with the "exact" solution
by Popov [51J in which the shear effect has been taken into account.
It is obvious that it is unfair to compare the linear (4-node),
quadratic (8-node) and cubic (12-node) elements in the terms of number of
elements. Such a comparison is grossly biased to the higher order elements.
An alternative is to use the number of unknowns involved. This comparison
is also given in Table 2. An argument can also be made for the basis of com-
parison to be total computer time, programming effort, etc.
49
Due to the shape of the element and type of loading, any element
including N5 (Eq. 3.11) was not considered here so evaluation of them was
delayed until the next section where the bending behavior of plates was in
vestigated. The mesh sizes used are indicated in parenthesis in Table 2.
For this first loading, the structure is subjected to constant
bending. Except element C4, all the elements investigated gave very good
solutions, as expected. Element C12 gave a little higher values for both de
flection and stress but these differences are negligible.
For the linear element, the improvement obtained by adding two non
conforming modes is the most impressive. (Compare the results from NC4-2.3,
NC4-4.3R and C4 in Table 2.) Addition of another two nodes to inplane dis
placement components (NC4-4.3) had no noticeable effect for the type of ap
plied loading. Because the additional nonconforming modes involve additional
computation they actually have a harmful or degrading effect.
Since the eight-node element (C8) has a quadratic representation of
displacement in its development, it produces good solutions. The addition
of nonconforming modes (NC8-2.3, NC8-4.3R and NC8-4.3) or utilizing higher
order element (C12) produced the same accuracy as NC4-4.3R or C8. Going to
this additional expense then produced no net return.
Loading Case (B)
The structure is now under a linearly varying moment. Only the
element designated as C12 gave accurate results while other elements produced
50
slightly different results. This is mainly because only the element C12 is
capable of representing the parabolic shear distribution across the depth of
the beam.
The linear elements (NC4-2.3~ NC4-4.3R, NC4-4.3) lack the capability
of expressing the shear distribution thru the depth and approximate the verti
cal displacement by a quadratic variation while the real variation is cubic.
However, this does not mean the other elements are to be discarded because
the improvement achieved by NC4-4.3R (NC4-2.3 is actually the same element
under inplane loading) can be seen to be almost as good as C12 element. Con
sidering the fewer unknowns involved, the result obtained by NC4-4.3R is very
satisfactory.
It has been seen that both NC4-4.3R and NC8-4.3R produced the same
accuracy as NC4-4.3 and NC8-4.3, respectively. In other words, for inplane
displacements, the addition of two nonconforming modes should be sufficient
and the addition of another two nonconforming modes is either unnecessary or
undesirable.
5.2.2 Summary of Beam Investigations
The major findings form the limited beam investigations are sum
marized as follows:
1. The addition of nonconforming displacement modes to the original
isoparametric elements gives generally markedly improved results,
i.e.~ approximately th~ same accuracy as that from a one degree
higher order element but at less expense. For instance, just
for inplane, the element C8 has 16 degrees of freedom while
51
NC4-4.3R has 12 degrees of freedom which is then condensed to
8 while NC4-4.3 has 16 degrees of freedom which is condensed to
8. The results from NC4-4.3R are close to those achieved by C8.
With these savings the difference of accuracy was 1 percent or so.
2 .. For both linear and quadratic elements, the addition of two non
conforming modes (Nl , N2) is sufficient to improve the inplane
behavior of elements. Addition of another two nonconforming
modes (N 3, N4) is either unnecessary or undesirable. Therefore
the elements NC4-4.3 and NC8-4.3 are deleted from further con-
siderations.
3. The improvement of the linear element has been found to be most
significant. When smaller grid size, equivalent to the distance
between nodes in higher order elements, is used the element
NC4-4.3R gives as accurate a result as the quadratic or cubic
element.
5.3 Plates
In the preceding section, the inplane behavior of nonconforming modes
was examined. In order to examine plate bending the addition of nonconforming
modes to deflection wand rotations a and S was investigated.
5.3.1 Square Plate
A clamped edge square plate under a concentrated load applied at
the center was investigated. The geometry and material properties are shown
in Fig. 9. Since the exact solution to this problem is available [56J, a
S2
number of authors have selected this case to evaluate the basic accuracy of
their element.
The central vertical deflection as computed by the linear and the
quadratic elements along with various combinations of nonconforming displace
ment modes is plotted versus the mesh sizes in Figs. 10 to 11.
For the linear element, the inclusion of the NS mode (Eq. 3.10) re
sults in a singular matrix of K22 (Eq. 3.21.3). Therefore, the element which
includes this nonconforming mode, NC4-S.S, is excluded from further consider
ation. The elements NC4-4.S and NC4-2.S, in which the nonconforming modes are'
applied to rotations, not only do not converge to the right solutions, but
also the magnitude of error is far too large (Fig. lla). As discussed earlier
in Section 3.7, rotations are established solely from the inplane displacements
and are not directly related to out-of-plane displacements. If the inplane
displacements are satisfactory, rotations should also be satisfactory (Fig.
2). Therefore, the addition of rotational nonconforming modes may result in
a more flexible structure than actual one.
The results obtained by the elements NC4-4.3R and NC4-2.3 are not
very good when a coarse mesh size is used. The ratio of convergence, however,
is very fast in both cases and very accurate results are obtained when the
grid has been refined to reasonable sizes. The element NC4-4.3R produced
superior results to the element NC4-2.3 because the latter does not include
the nonconforming twisting mode.
For the quadratic elements, the element which includes nonconforming
modes to the rotational degrees-of-freedom (NC8-S.S) behaves again much worse
than the elements which do not include the rotational nonconforming modes
{NC8-4.3R and NC8-S.3R}. Unlike the linear element, however, the magnitude
53
of the difference between the two, with and without, results is not nearly
as large as in the linear element case. The results indicate that the ele-
ment containing these nonconforming modes applied to the rotations (NC8-5.5)
has some tendency toward convergence to the real solution.
The results obtained by elements NC8-5.3R and NC8-4.3R are nearly
identical (Fig. llb). The former includes N5 mode while the latter does not.
The mode N5 has virtually no effect for plate bending problems.
The major findings of this section are
1. The application of nonconforming modes to the rotations may
results in either erroneous convergence (linear element) or at
least markedly slower convergence (quadratic element);
2. The fifth mode N5 is not appliable to low order elements due
to the singular nature of the matrix (linear element) or else
it has little effect and can be neglected for the higher order
elements (quadratic element);
3. With both linear and quadratic elements, the inclusion of twisting
modes (N3
and N4 for w) has significant effects and should not
be omitted.
As a conclusion the most rnmhin;:)+innc: \""'VIlIU I IIU v I VlloJ
of nonconforming modes are NC4-4.3R and NC8-4.3R for linear and quadratic
elements, respectively. Thus from here on only elements NC4-4.3R, NC8-4.3R
and C12 are considered. Although based on quite limited data, since the ele-
ment choice represents the intuitively more reasonable elements, and the cost
of continuing the examine the other elements was too high, the decision to re-
strict the study to just these elements is not premature. The designation of
these elements can now be simplified to NC4 and NC8, respectively.
54
The results of these elements compare favorably with those obtained
by other finite elements as shown in Fig. 12. It may be noted that results
from all the three elements (NC4, NC8 and C12) appear to converge to a value
slightly higher than the real solution. It is due to the fact that the use
of a minimum order of numerical integration in claculating the stiffness matrix
tends to under-estimate the stiffness of element (Appendix D). Since the dif
ference is negligible and the use of reduced integration represents a signifi
cant saving in time, its use is continued.
5.3.2 Circular Plate
The next test involves a circular clamped plate under a uniform
load. The main purpose of investigating this problem is to test the appli
cability of nonconforming modes when used in conjunction with a nonrectangular
element arrangement.
As seen in Fig. 13, the solutions indicate good convergence char
acteristics. Thus it is concluded that the nonconforming elements can be
used for both a rectangular element arrangement and for a nonrectangular ele
ment arrangement. Marked departures from rectangularity should however be
avoided if at all possible.
5.4 Cylindrical Shell
A circular cylindrical shell supported by a rigid diaphgram and with
no longitudinal edge beam has become a standard test problem frequently inves
tigated by a number of authors [32,40,57,58J. Thus the behavior of this type
of shell is well established. The purpose of investigating this shell in this
55
study is to evaluate the accuracy achievable with the newly developed ele-
ments when seeking solutions for singly curved shells.
The dimensions and material properties of the shell considered are
shown in Fig. l4a. Two loading cases as shown in Fig. l4b are considered, i.e.,
1) pressure load (47 psf on shell surface) plus live load (25 psf) and 2) dead
weight (150 pCf) plus live load (25 psf). The elements NC4 (10x8) and NC8
(5.4) are used with the mesh sizes indicated in the parentheses. Each was used
for both loading cases. The costs of computer usage by these two different
elements were approximately same.
The r,esults obtained are presented in Fig. 15 (displacements) and
Fig. 16 (stress resultants). These results are compared with a solution com
puted by numerical integration of the Donnell-Jenkin1s equation [57J. Although
some differences are distinguishable in the results for the two loading cases, !
particularly in wand M I,for all intensive purposes the two cases yield the x
same result. Thus the approximation of the gravity load by a normal-to-surface
load which has been used previously by many investigators can be justified.
The agreement between results from Ref. 57 and those from NC4 and
NCB for loading case II is excellent with the exception of the hoop stress
N I and hoop moment M I. These stress resultants are rather sensitive for this x x
geometry and boundary conditions. The magnitude of the disagreement is more-
over relatively small and thus thought to be of small consequence.
The results obtained by NC4 and NC8 are generally in close agree-
ment for each loading case. However, the stress resultants obtained by NC8
at mid-side nodes, particularly N I in Fig. l6b and N I I in Fig. 16e are not x x y
so good as those at corner nodes. This results in a somewhat zigzaged nature
of plotted curve.
56
Another interesting~ yet useful, comparison to make is with the re-
sults obtained by Ahmad et al. [40J. They used 24 conforming cubic elements
(designated as C12 in this study, 60 OOF) and 24 conforming parabolic elements
(designated as C8 in this study, 40 OOF) in solving the same problem. It is
noted that the comparison with Ref. 40 is somewhat approximate rather than
exact because of a slightly different geometry (thickness to span ratio) and
loading used in Ref. 40. Whenever the difference between the exact solutions
cited in this study and in Ref. 40 is distinguisable in figures~ this differ-
ence in exact solutions is also shown.
The improvement in accuracy obtained by NC8 over C8 using a smaller
number of elements, 20 elements vs. 24, is very significant. The results ob-
tained by NC8 are as good as those obtained by C12. Considering the bad be-
havior of C4 in Fig. 10, the improvement achieved by NC4 is very impressive.
Thus the linear element which has been ignored by previous investigators has
been shown to be as useful as any other element.
5.5 Elliptic Paraboloid
A simp 1 y supported squa re ell i pti c pa raboloid whose geometry and ma-
terial properties are shown in Fig. 17 is investigated. Due to the square
plan form, the equation of the mid-surface of this elliptic paraboloid is
simplified to be
z 1 2 2 = - (x + y ) 2R
The shell is loaded by a uniformly distributed load.
(5. 1 )
The same shell was investigated previously by several authors [59,60J
57
using somewhat different approaches. This shell is analyzed in this study
in order to evaluate the basic accuracy of NC4 and NC8 elements for doubly
curved shells. Since the behavior of this type of shell, i.e., a shell with
positive Gaussian curvature, is predominately membrane action with bending
moments being mostly of a secondary stress nature and only significant near
boundaries, smaller mesh sizes are needed particularly near the edges of shell
in order to pick up the sharp stress gradient.
The results obtained by using NC4 and NC8 elements are shown in
Fig. 18 (displacements) and Fig. 19 (stress resultants). Reasonably good
agreement with the results from Gustafson and Schnobrich [60J and Chu and
Schnobrich [18J has been obtained with the exception of the shear N I I at x y
corner support. Since the plotted values of NC4 and NC8 are those at the
Gauss integration points, somewhat distant from the corner, these values may
not represent corner values adequately particularly where such a sharp stress
gradients exists.
The difference between the results obtained by using NC4 (8x8) or
by using NC8 (4x4) is again minimal. The mesh sizes are indicated in the
parenthses. Both NC4 and NC8 with above mesh sizes not only produced very
close accuracy but also they used nearly the same computer running time (ap
proximately 10 percent difference). Thus the nonconforming elements, both
NC4 and NC8 have been shown to be used satisfactorily for both single and
double curved shells.
5.6 Spherical Shell
One problem which gave a good bit of trouble in the early stages
58
of the development of the finite element procedure was the pressurized shell
of revolution. Therefore a spherical shell with a built in edge and subjected
to a uniform normal load is analyzed. The dimensions of the shell and the ma
terial properties are given in Fig. 20. In this analysis, nonrectangular ele
ments are used. Due to the symmetric shape only one quarter of the shell was
actually solved.
This shell has also been solved analytically by Timoshenko [56] and
by finite element method using 24 cubic type isoparametric elements [40J.
Those are a few of the many analyses done for this shell. The latter refer
ence is of particular interest because it contains the results obtained by
the element with no nonconforming modes, i.e., with the original fully com
patible element.
The results obtained in this study are shown in Fig. 21. The agree
ment with the analytical solution and with other finite element solutions are
very good for both meridional bending moment and hoop forces.
Considering the relatively coarse meshes used in this study and the
computational efforts involved in element formulation and solution for twenty
four 60-degree-of-freedom elements used in Ref. 40, the results obtained by
this study are extremely satisfying. The stress resultants obtained by NC4
and NC8 are almost identical except the peak poi,nt where NC4 is slightly un
favorable.
5.7 Hyperbolic Paraboloid
A hyperbolic paraboloid shell with all edges clamped and subjected
to a uniform normal load is considered next. The geometry and material pro
perties are shown in Fig. 22. The equation of mid-surface of the shell is
59
z k x y (5.2)
where
k f = a"b
The results obtained by using NC4 and NCB and with two different
corner rises are presented along with the results of other investigators in
Figs. 23 and 24. This shell with the inadvertent variation noted below has
been investigated by Chetty and Tottenham [61J, Pecknold and Schnobrich [34J,
Chu and Schnobrich [lBJ and many others. While the results from Ref. 61 and
Ref. 34 are in good agreement, the results from Ref. lB are quite different
from above two.
Due to a typographical error in the paper of Iyengar and Srinivasin
[63J two different shell geometries have developed for what was thought to
be the same shell. The results presented in Refs. 34 and 61 are for a shell
with corner rise of 1.304 while Ref. 1B used 1.034 inches as listed in Ref. 63.
The results from this study show good agreement to both Refs. 61
and 34 and to Ref. lB, depending on the corner rise used. From this it can
be concluded that all methods would give the same results when applied to the
same problem. Also the results show that while NC4 produces results slightly
higher than the values from the references cited, NCB produced results slightly
lower. The difference is not significant.
5.B Conoidal Shell
The conoidal shells, usually in the truncated form, are most commonly
used for the factory type buildings because of their natural lighting
60
characteristics. Only a limited amount of work has been done on this type
of shell. The more reliable methods of analysis are yet to be developed.
A truncated conoidal shell whose geometry and material properties
are shown in Fig. 25 and subjected to a uniform normal load is investigated.
Two different boundary conditions are considered, i.e., 1) all edges are
clamped and 2) all edges are simply supported. Due to the symmetry, one half
of the shell is actually solved. The equation of the mid-surface of the
shell is
z f = - x L
1 ) (5.3)
This shell was analyzed analytically and experimentally by Hadid
[62J. The results obtained for this study are presented in Figs. 26 thru 29.
5.B. 1 Simply Supported Edges
Figures 26 and 27 show a generally good agreement between the numer-
ical results from Ref. 62 and those of this study. The comparison with the
experimental results reported in Ref. 62 is of particular interest and is very
useful as well because if the analytical result is backed by experimental re-
sults this greatly increases confidence in the numerical solutions.
The difference between results from NC4 (BxB) and NCB (5x4) is
minimal. The mesh sizes are indicated in the parentheses. However, the nor
mal displacement Wi at X = ~ obtained by NCB is not in such a good agreement
as elsewhere (Fig. 26a). Both results from NC4 and NCB are in closer agreement
with Hadid!s [62].analytical solution than with his experimental solution.
61
5.8.2 Clamped Edges
As with simply supported case, the general agreement between the re-
sults with the nonconforming elements and those from Ref. 62 is reasonably
good. The largest discrepancy among the results is again seen for the normal
I - L ( ) displacement w at x = 2 Fig. 28b. The sensitivity of the structure may be
a major cause for such a discrepancy. For example when the support is clamped
the maximum deflection is reduced by a factor of over five from that value
obtained when the edge was simply supported. This is very unusual behavior
for a shell and is a possible explanation for the large difference between
experimental and numerical results for the simply supported case.
Comparing with the general deflection profile at x = ~. for a simply
supported shell in Fig. 26a, the results from this study seem to be more rea-
sonable than those from Ref. 62 (Fig. 28b). However, the differences shown
in Fig. 28b are not readily explanable in view of the other agreements.
Since relatively coarse grid sizes are used in this study, particular-
ly for NC4, the sharp stress gradients near clamped edge due to the edge dis-
turbance are not adequately represented. In the central region of the shell
where the behavior has settled down to essentially membrane conditions the
agreement is much better.
62
Chapter 6
CONCLUSIONS AND SUGGESTIONS FOR FURTHER STUDIES
6.1 Conclusions
The development of yet another new finite element was presented in
this study. When the original isoparametric element is used in shell analysis,
either the structural representation is too stiff (lower order element) or too
many unknowns are involved (higher order element) making the use of this type
of element uneconomical. Of the original isoparametric family of shell ele
ments only the sixty-degree-of-freedom elements, designated as C12 in this
study, can produce reasonable accuracy.
By adding nonconforming displacement modes selectively to the low
order isoparametric elements, highly improved new shell elements have been
established. Addition of various combinations of nonconforming modes to ~rig
inal elements have been investigated and the elements designated as NC4-4.3R
(linear element) and NC8-4.3R (quadratic element), later abreviated to NC4' and
NC8, respectively, have been found to be most effective in terms of the number
of nonconforming modes used versus the improvement obtained.
For both NC4 and NC8, only two (Nl , N2 in Eq. 3.10 or 3.11) noncon
forming modes are added to the inplane displacement and four (Nl , N2, N3, N4,
in Eq. 3.10 or 3.11) to out-of-plane displacement. No additional modes are
applied to rotations. These elements, NC4 and NC8 along with C12, have been
tested with deep beam, plate and various shell problems.
The results obtained from these new elements compare with the re
sults from well established analysis extremely satisfactorily. The improvement
63
incorporated into the four-node element, which has been ignored by virtually
all previous investigators, is most impressive.
The elements NC4 and NCB can be used equally successfully for any
type of shell problems. The element NC4 may be used with more flexibility
for the shells with irregular boundaries and/or irregular shapes. While ele
ments NCB and C12 can be used advantageously where the sharp stress gradients
exist.
The shells investigated in this study are just a few examples of
what these new elements can solve. The versatility of the elements is evi
dent. With the unknowns selected for these elements there are no difficulties
involved in adding rib or stiffening elements either concentric or eccentric
to the shell.
6.2 Suggestions for Further Studies
One of the most promising areas which this type of element can be
used effectively is with nonlinear material behavior problems. The extention
to nonlinear material behavior problems is not a difficult task. Unlike other
conventional finite elements, the state of stress or progress of plasticity
can be defined individually at each integrating point in an element. Thus
part of the element may be plastic while the remainder is elastic. Besides,
these types of elements do not have difficulties in obtaining a stiffness
matrix of an element which includes both elastic and plastic regions at the
same time. Therefore, either initial stress approach or updated stiffness
approach can be used in solving the nonlinear problems.
Another extension would be the application to problems of response
64
of shells to dynamic loading. In this case, both material and geometrical
nonlinearities should be considered in order to obtain more accurate solutions.
Very little has been published in this area to date.
Another topic of practical interest is the comparative economies of
reduced integration technique versus nonconforming mode approach. Those two
approaches have some characteristics in common. They take the isoparametric
element as a starting point. They have the same motivations, i.e., to get
rid of the excessive shear energy present in the original low order isopara
metric element. The comparisons, such as in terms of applicability, gener
ality and economy, would be of practical importance.
65
LIST OF REFERENCES
1. Reissner, E., "On a Variational Theorem in Elasticity," Journal of Mathematics and Physics, Vol. 29, pp. 90-95, 1950.
2. Pian, T.H.H. and Tong, P., "Basis of Finite Element Methods for Solid Continua," International Journal for Numerical Methods in Engineering, Vol. 1, pp. 3-28, 1969.
3. Key, S. W. and Krieg, R. D., "Comparison of Finite Element and Finite Difference Methods," Int. Symp. bn Num. and Computer ~1eth. in Struc. Mech., University of Illinois, Urbana, Illinois, 1971.
4. Courant, R., IIVariationa1 Methods for the Solutions of Problems of Equilibrium and Vibrations," Bulletin of American Mathematical Society, Vol. 49, pp. 1-23, 1943.
5. Argyris, J. H. and Kelsey, S., "Energy Theorems and Structural Analysis," Aircraft Engineering, Vol. 26, 1954, Vol. 27, 1955.
6 . T urn e r, r~. J., C lou g h, R . l~. and Top p, L. J., " S t iff n e s san d De f 1 e c t ion Analysis of Complex Structures," Journal of Aeronautical Science, Vol. 23, pp. 805-823, 1956.
7. Melosh, R. J., "Basis for Derivation of Matrices for the Direct Stiffness Method," AIAA Journal, Vol. 1, No.7, pp. 1631-1637, July 1963.
8. Holland, 1. and Bell, K., "Finite Element Methods in Stress Analysis," TAPIR (Tech. Univ. of Norway Press), Trondheim, 1969.
9. Melosh, R. J., "A Stiffness Matrix for the Analysis of Thin Plates in Bend i ng," Journa 1 of Aerospace Sci ence, Vo 1. 28, No.1, pp. 34-42, 1961.
10. Zienkiewicz, O. C. and Cheung, Y. K., "The Finite Element Method for Analysis of Elastic Isotropic and Orthotropic Slabs," Proc. Institution of Civil Engineers, August 1964.
11. Gallagher, R. H., "Analysis of Plate and Shell Structures,'1 Proc. Symp. on Application of Finite Element Methods in Civil Engineering, Vanderbilt University, 1969.
12. Bogner, F. K., Fox, R. L. and Schmit, L. A., liThe Generation of Interelement Compatible Stiffness and Mass Matrices by the Use of Interpolation Formulars," Proc. Conference on Matrix Methods in Structural Mechanics, AFFDL-TR-66-80, pp. 397-444, 1965.
13. Fraeijs de \teubeke, B., "A Conforming Finite Element for Plate Bending," International Journal of Solids and Structures, Vol. 4, pp. 95-108, 1968.
66
14. Bell., Ko .. "A Refined Triangular Plate Bending Finite Element," International Journal for Numerical Methods in Engineering, Vol. 1, ppm 101-122, 1969.
15. Argyris, J. H., Fried, I. and Scharpf, D., liThe TUBA Family of Plate Element for the Matrix Displacement Method," Aeronautical Journal, Vol. 72, August 1968.
16. Argyri s, J. H. and Scha rpf, D., liThe SHEBA Fami ly of Shell El ements for the Matrix Displacement Method," Aeronautical Journal., October 1968, March, May 1969.
17. Bazeley, G. P. Cheung, Y. K., Irons, B. M. and Zienkiewicz, O. C., IITriangular Elements in Plate Bending - Conforming and Non-Conforming Solutions," Proc. Conference on Matrix Methods in Structural Mechanics, AFFDL-TR-66-80, November 1965.
18. Chu, T. C. and Schnobrich, W. C., "Finite Element Analysis of Translational Shells,1J Civil Engineering Studies, Structural Research Series No. 368, University of Illinois, Urbana, Illinois, November 1970.
19. Clough, R. W. and Tocher, J. L., "Finite Element Stiffness Matrices for Analysis of Plate Bending," Proc. Conference on Matrix Methods in Structural Mechanics, AFFDL-TR-66-80, ppm 397-444, 1965.
20. Clough, R. W. and Felippa, C. A., "A Refined Quadrilateral Element for Analysis of Plate Bending," Proc. Conference (2nd) on Matrix Methods in Structural Mechanics, Wright-Patterson AFB, Ohio, 1968.
21. Faeijs de Veubeke, B . ., "Displacement and Equilibrium Models in Finite Element Method," Stress Analysis, Edited by Zienkiewicz, O. C. and Hollister, G. S., John Wiley & Sons, 1965.
22. Herrman, L. R., IIA Bending Analysis of Plates," Proc. Conference on Matrix Methods in Structural Mechanics, AFFDL-TR-66-80, November 1965.
23. Visser, W., "A Refined Mixed Type Plate 'Bending Element," Technical Note, AIAA Journal, Vol. 7, No.9, ppm 1801-1803, 1969.
24. Herrman, L. R., flFinite Element Bending Analysis for Thin Shells,iI AIAA Journal, Vol. 6, No. 10, pp. 1842-1847, 1968.
25. Prato, C. A., "Shell Finite Element Method via ReissnerDs Principle,1l International Journal of Solids and Structures, Vol. 5, ppm 1119-1133, 1969.
26. Pian, T.H.H .. and Tong, IlRationalizationin Deriving Element Stiffness Ma tri x by Assumed Stress Approach," Proc. Conference (2nd) on r~a tri x Methods in Structural Mechanics, AFFDL-TR-68-l50, Wright-Patterson AFB, Ohio, 1968.
67
27. Connor, J. J. and Will, G. T., IIA t~ixed Finite Element Shallow Shell and Formulation,1I Proc. US-Japan Seminar on Matrix r~ethods of Structural Analysis and Design, 1969.
28. Pian, T.H.H., "Formulation of Finite Element ~-1ethods for Solid Continua," Proc. US~Japan Seminar on Matrix Methods of Structural Analysis and Des i gn , 1969.
29. Tong, P., "New Displacement Hybrid Finite Element Models for Solid Continua,11 International Journal for Numerical Methods in Engineering, Vol. 2, pp. 73-83, 1970.
30. Harvey~ J. W. and Kelsey, S., IITriangular Plate Bending Elements with Enforced Compa t i bi li ty:1 AIAA Journa 1, Vo 1. 9, No.6, pp. 1023-1026, June 1971.
31. Green, B. E., Strome, D. R. and Weikel, R. C., IIApplication of the St i ffness Method to the Ana lys is of Shell Structures:1 ASME, Paper No. 61-Av-58, August 1961.
32. Johnson, C. P. and Clough, R. W., IIA Finite Element Approximation for the Analysis of Thin Shells," International Journal for Solids and Structures, Vol. 4, pp. 43-60, 1968.
33. Connor, J. and Brebbia, C., "Stiffness ~1atrix for Shallow Rectangular Shell Element:1 Proc. ASCE J. of EM Div., No. EM5, October 1967.
34. Pecknold, D. A. and Schnobrich, W. C., "Finite Element Analysis of Skewed Shallow Shells,1I Civil Engineering Studies, Structural Research Series No. 332, University of Illinois, Urbana, 1968.
35. Bogner, F. K.,_ Fox, R. L. and Schmit, L. A., IIA Curved Cylindrical Shell Discrete Element,1! AIAA Journal, Vol. 59 No.4, April 1967.
36. Cantin, G. and Clough, R. W., IIA Curved Cylindrical Shell Finite Element,1! AIAA Journal, Vol. 6, No.6, June 1968.
37. Utku, S., IIStiffness ~~atrices for Thin Triangular Elements of Non-Zero Gaussian Curvature, II AIAA Journal, Vol. 5, No.9, September 1967.
38. Haisler, W. E. and Stricklin, J. A.s IIRigid Body Displacements of Curved Elements in the Analysis of Shells by the Matrix Displacement Method, II Techni cal Note'l AIAA Journal, pp. 1525-1527, Augus t 1967.
39. Ergatoudis, 10:; Irons, B. M. and Zienkiewicz, O. C., "Curved Isoparametri c, IIQuadri latera 111 El ements for Fi n i te El ement Ana 1 ys is, II I nternational Journa of Solids and Structures; Vol. 4, pp. 31-42, 1968.
68
40. Ahmad, S., Irons, B. M. and Zienkiewicz, O. C., flAnalysis of Thick and Thin Shell Structures by Curved Finite Elements,fI International Journal for Numerical Methods in Engineering, Vol. 2, pp. 419-451, 1970.
41. Zienkiewicz, O. C., Taylor, R. L. and Too, J. M., "Reduced Integration Technique in General Analysis of Plates and Shells," International Journal for Numerical Methods in Engineering, Vol. 3, ppm 275-290, 1971.
42. Wilson, E. L., Taylor, R. L., Doherty, t~. P. ,and Ghaboussi, J., "Incompatible Displacement Models," International Symposium on Numerical and Computer t~ethods in Structural Mechanics, University of Illinois, Urbana, September 1971.
43. Gupta, A. K., Mohraz, B. and Schnobrich, W. C., "Elasto-Plastic Analysis of Three Dimensional Structures Using the Isoparametric Element'lfl Civil Engineering Studies, Structural Research Series No. 381, University of Illinois, Urbana, August 1971.
44. Yuzugullu, O. and Schnobrich, W. C., flFinite Element Approach for the Prediction of Inelastic Behavior of Shear Wall-Frame Systems," Civil Engineering Studies, Structural Research Series 386, University of Illinois, Urbana, May 1972.
45. Zienkiewicz, O. C. and Nayak, G. C., "A General Approach to Problems of Plasticity and Large Deformation Using Isoparametric Elements," Proc. Conference (3rd) on Matrix Method in Structural Mechanics, Wright-Patterson AFB, Ohio, October 1971.
46. McNamara, J. F. and ~1arcal, P. V., "Incremental Stiffness Method for Finite Element Analysis of the Nonlinear Dynamic Problems," Proc. Int. Symp. on Numerical and Computer Methods in Structural Mechanics, Uni vers i ty of III i noi s, Urbana, III i noi s, September 1971 .
47. Zienkiewicz, O. C., liThe Finite Element t~ethod in Engineering Science," McGraw-Hill, 1971.
48. Zienkiewicz, O. C., "Isoparametric and Allied Numerically Integrated Elements - A Review," Int. Symp. on Num. and Computer Meth. in Structural Mechanics, University of Illinois, Urbana, September 1971.
49. Doherty, W. P., Wilson, E. L. and Taylor, R. L., "Stress Analysis of Axisymmetric Solids Utilizing Higher_Order Quadril~tefalFinite Elements," SESM Report 69-3, Structural Engineering Laboratory, University of California, Berkeley, California, 1969.
50. Pawsey, S. F., liThe Analysis of Moderately Thick to Thin Shells by the Finite Element Method,1I SESM Report 70-12, Structural Engineering Laboratory, University of California, Berkeley, California, 1970.
51. Popov, E. P., "Introduction to fviechanics of Solids,!' Prentice-Hall, Inc., 1968.
69
52. Bakhrebah, S., IIFinite Element Analysis of Intersecting Cylinders," Ph.D. Thesis, Department of Civil Engineering, University of Illinois, Urbana, 1973.
53. Kaplan, W., IIAdvanced Calculus," Addison-ltJesley.
54. Melosh, R. J. and Banford, R. M., IIEfficient Solution of Load-Deflection Equations,1I Proc. ASCE, Journal of Structural Division, Vol. 95, No. ST4, pp. 661-676, April 1969.
55. Irons, B. r'~., "A Frontal Solution Program for Finite Element Analysis,1I International Journal for Numerical Methods in Engineering, Vol. 2, pp. 5-32, 1970.
56. Timoshenko, S. and L\loinowsky-Krieger, S., IITheory of Plates and Shells:' 2nd Edition, McGraw-Hill, 1959.
57. Scordelis, A. C. and Lo, K. S., IIComputer Analysis of Cylindrical Shells,1I Journal of American Concrete Institute, Vol. 61, No.5, May 1964.
58. Chuang, K. P. and Veletsos, A. S., "A Study of Two Approximate f\1ethods of Analyzing Cylindrical Shell Roofs,1I Civil Engineering Studies, Structural Research Series No. 258, University of Illinois, Urbana, Illinois, October 1962.
59. Abu-Sitta, S. H., "Finite Difference Solutions of the Bending Theory of the Elliptic Paraboloid,1I Bulletin of the International Association for Shell Structures, No. 20, December 1964.
60. Gustafson, W. C. and Schnobrich, W. C., !lMultiple Translatlonal Shells,1I IASS-IMCYC Congreso International Sobre La Applicacion de Estructuras Larninares en Arquitectura, Mexico City, September 1967.
61. Chetty, S.~tK. and Tottenham, H., 'IAn Investigation into the Bending Analysis of Hyperbolic Paraboiloid Shells,iI Indian Concrete Journal, July 1964.
62. Hadid, H. A., nAn Analytical and Experimental Investigation into the Bending Theory of Elastic Conoidal Shell,1I Ph.D. Dissertation University of Southampton, March 1964.
63. Iyengar., K.T.R. and Srinivasin, R. S., IIBending Analysis of Hyperbolic Paraboloid Shell,1I Structural Engineer, Vol. 46, No. 12, pp. 397-401, Decernoer 1968.
64. Tezcan, S. 5., IIComputer Analysis of Plane and Space Structures, II Journal of the Structural Division, Proc. ASCE, Vol. 92, No. ST2, pp. 143-173, April 1966.
Table 1
Element Designations
Nonconforming Modes Applied To DOF per Element Designations -
U,v w a,S Original Condensed
* C4 None None None 20 NC4-2.3 Nl ,N2 Nl ,N2 None 26 20 NC4-2.5 N1,N2 N1,N2 N1 ,N 2 30 20
Linear NC4-4.3 Nl ,N2,N3,N4 Til ,N2,N3,N4 None 32 20 Elements NC4-4.3R** N1,N2 Nl ,N2,N3,N4 None 28 20
NC4-4.5 Nl ,N2,N3,N4 Nl ,N2,N3,N5 Nl ,N2,N3,N4 40 20
-......J NC4-5.5 N1,N2,N3,N4,N5 Nl,N2,N3,N4,N5 N1 ,N2 ,N3 ,N4 ,N5 45 20 a
* C8 None None None 40 NC8-2.5 Nl ,N2 Nl ,N2 N1,N2 50 40
Quadratic NC8-4.3 Nl ,N2,N3,N4 Nl ,N2,N3,N4 None 52 40 Elements NC8-4.3R** Nl ,N2 N1,N2,N3,N4 None 48 40
NC8-5.3 N1,N2,f\f3,N4,N5 Nl,N2,N3,N4,N5 None 55 40 NC8-5.3R** N, ,N2 N1,N2,N3,N4,N5 None 50 40 NC8-4.5. N, ,N2,N3,N4 N, ,N2,N3,N4 N1,N2,N3,N
4 60 40
NC8-5.5 N, ,N2,N3,N4,N5 Nl ' N2 ,N3 ,N4 'I N5 N, ,N 2 ' N 3 ' N 4 ' N 5 65 40 Cubic Element C12* None None None 60 * -'---._---
Equivalent to original isoparametric elements.
**Elements predicted as the most beneficial ones in Section 3.7.
Element & Mesh
C4 (3xl)
NC4-2.3 (3xl)
NC4-4.3 (3x1 )
NC4-4.3R (3x1 ) II (6x2 )
C8 (3xl )
NC8-2.3 (3x1 )
NC8-4.3 (3x1 )
NC8-4.3R (3x1 )
C12 (3xl)
EXACT
---.--
71
Table 2
Comparison of Results of Cantilevered Beam Using Various Elements
(A) Ph ~
tY2 ....-.
Number 6 at A o at B 6 at A Unknowns x 10-5 x 10-1 x 10-5
(B) t P
o at B x 10-3
-----------
12 1 .20 4.00 4.90
12 1 .80 6.00 7.40 --
12 1 .80 6.00 7.80
12 1 .80 6.00 7.40
24 7.58
30 1 .80 6.00 7.59
30 1 .80 6.00 7.61
30 1 .80 6.00 7.61 30 1 .80 6.00 7.61
48 1 .81 6.02 7.68
1 .80 6.00 7.68
----------_ ........
A ~------~----~_r
V:/'
~---~ -IA
~ ____ 3014-----*-
p = 2 1 b
E = 3 x 106 ps i
~ = 0.0
2.02
3.60
3.60
3.60
3.60
3.60
3.60
3.60
3.60
3.65
3.60
72
Table 3
Abscissae and Weight Coefficients of the Gaussian Quadrature Formula
fl PI
f(x) dx = _L llJ~a), -1 J= 1
±a
0-57735 02691 89626
0-77459 66692 41483 0-00000 00000 00000
0-86113 63115 9-l053 0- 33998 10435 S4856
0-90617 98459 38664 0-53846 93101 05683 0-00000 00000 00000
0-93246 95142 03152 0-66120 93864 6()265 0-23861 91860 83197
0-94910 79123 42759 0-74153 11855 9939-l ~40584 51513 77397 0-00000 00000 00000
0-96028 98564 97536 ~79666 64774 13627 0-52553 -24099 16329 0-18343 46424 95650
0-96816 02395 07626 0-83603 11073 2663'6 0-61337 14327 00590 0-32425 34234 O~S09 0-00000 00000 00000
n=2
n=3
n=4
n = 5
n = 6
n = 7
n = 8
n=9
n = 10
H
\-00000 00000 00000
0-55555 55555 55556 0-88888 888~8 81::889
0-34785 48451 37454 0-65214 51548 62546
0-23692 63850 56189 0-47862 1\(1704 993()6 0-56888 88888 88889
0-17132 44923 79170 0- 36076 15730 48139 0-46791 39345 72691
0- J 2948 49661 6S870 0-27970 53914 't,9277 0-381~3 00505 05119 0041795 91836 7346Y-
0-10122 85362 90376 0-22238 1034-t 53374 0-31370 66-f58 77887 0-36268 37833 78362
~OS127 43883 61574 0-18064 8160(J 9-!~57 0-26061 069h4 02935 0- 3123-t 70770 -lO003 0-33023 93550 OI~W
0-97390 65285 17172 0-06667 13.+43 086SS 0-86506 336(16 SSl)S5 O-I-l9.t5 134\) I 5!l5S 1 0-67940 956S2 9t)()~-1- 0-21lJOS 63«25 15t)S2 0-43339 53\)41 29247 Q-2c1'J2() 671l)3 Ol)996 0-14SS7 43JS9 S 16~ I Q-2lJ552 42247 14753
73
(a) Linear Element
(b) Quadratic Element
(d) Quadratic Element--Nodes Across Thickness Eliminated
(c) Cubic Element
(e) Cubic Element--Nodes Across Thickness Eliminated
FIG. 1 3-DIMENSIONAL ISOPARAMETRIC ELEMENTS
74
n
M M
(a) Linear Element Under Pure Bending
or or
(b) Constrained Deformation
I L _________ .....J
(c) Actual Deformation
FIG. 2 LINEAR ISOPARAMETRIC ELEMENT UNDER PURE BENDING
75
Conforming Mode
Nonconforming Mode
II
Combined Mode ,. I
/ I I L.... ___________ J
FIG. 3 RESTORING THE ACTUAL BEHAVIOR OF THE ELEMENT BY ADDING NDNCONFORr~ING MODE
76
(a) Displacement Components at the Top and Bottom Nodes
z,w
Y,V
Node i
(b) Global Coordinates, Local Coordinates and Corresponding Displacement Components
FIG. 4 DEFINITIONS OF GLOBAL AND LOCAL COORDINATES
77
(a) N]
(b) ~
FIG. 5 NONCONFORMING DISPLACEMENT (w) MODES FOR LINEAR ELEMENT
78
(b) R2
79
FIG. 7 INTEGRATION POINTS FOR PLATE (n = 3)
.33998
0.86113
FIG. 8 GAUSS INTEGRATIONS
80
a/2 = t = E =
a/2 ]J =
x
FIG. 9 TYPICAL UNIFORM MESH (2x2) OF CLAMPED SQUARE PLATE UNDER CONCENTRATED LOAD AT THE CENTER
40"
1 "
3 x 0.3
~ 100 ...
c o .,.... +-> ~
..--o
C/)
..--ro ClJ
c.c::
4-o OJ en ro
+-> c OJ u sOJ
0..
80
60
40
20
a 1 xl
8-Node Element (C8)
". 4-Node El ement (C4).......""
?v? C-AC- 3x3
Mesh Size in 1/4 of plate
~
4x4 5x5
FIG. 10 CENTRAL DEFLECTION OF CLAMPED SQUARE PLATE UNDER CONCENTRATED LOAD AT THE CENTER
106 psi
81 260
NC4-4.5
~
In
120 c: 0
or-..j...J
:::s 100 0
(,/')
ro 80 OJ 0:::
4-0 NC4-4.3R OJ 60 O"l ro
+-l c: OJ u 40 $.... OJ
0...
20
0 2x2 4x4 6x6 8x8
(a) Linear Element
120 NC8-5.5
~ Y .. ... 100 c: ~~ NC8-4.3R 0
NC8-5.3R or-+-l :::s
r-- 80 0
(,/')
ro 60 OJ
0:::
4-0
OJ 40 O"l ro
+-l c: OJ u 20 $.... OJ
0...
0 1 xl 2x2 3x3 4x4 5x5
Mesh Size in 1/4 of Plate (b) Quadratic and Cubic Elements
FIG. 11 CENTRAL DEFLECTION OF CLAMPED SQUARE PLP,TE
r-ct5 OJ
cr:: 4-o OJ en ct5 +' s:: OJ u SOJ
0..
100
80
I I
I I
I I I
82
ACM
" " f1 ,
......... -- ........ '"",,-~ ........ '" / ...-..., ...............
/C12 I
I
NC4
2 4 8
Mesh Size in 1/4 of Plate 10
Meaning of Symbols
M, AC~l
LCCT-9 DVS Q-19
Incompatible but complete triangular elements; 12 external OOF (Ref. 9)
Two LCCT-9 triangles; 12 external DOF (Ref. 19) Compatible quadrilateral; 16 external DOF (Ref. 13) Four LCCT-11 triangles; 12 external and 7 internal DOF (Ref. 20)
FIG. 12 COMPARISON OF CENTRAL DEFLECTION FROM VARIOUS PLATE BENDING ELEMENTS
83
80 ~
... c 0 y +-' t :::::i .--0
(/) 60 r 10 il
.--rd OJ t = 111
0::::
4- E = 3 x 106 psi 0
OJ 11 0.0 0') rd 40 ~x +-' C
I 4 ~ I OJ u S- r OJ
0..
4 8 12 16
Number of Elements
FIG. 13 CLAMPED EDGE CIRCULAR PLATE UNDER UNIFORM LOAD
z
(a)
84
L = 372"
R = 372"
t = 3.75 11
a = 40° E = 3 x .06 psi
II == 0.0
\ Case I ,/
\ / \ Case II ,/
\ /
FIG. 14 GEOMETRY AND MATERIAL PROPERTIES OF CIRCULAR CYLINDRICAL SHELL
-10
10
VI 20 OJ
..c:: u c
3: 30
40
20
~ Scordelis & Lo (Ref. 57)
" NC4} Loadi ng Case I • NCB
t::. NC4} Loading Case II o NCB
o C12 II CB }
Ahmad (Ref. 40)
(a) Wi at Midspan
cp, degrees
30
FIG. 15- DISPLACEMENTS OF CIRCULAR CYLINDRICAL SHELL UNDER UNIFORM LOAD
40
Vl QJ
..c u I::
N i o
.. >
Vl QJ
..c u
12
2
1::4 ...-
I o
8
o
85
------ Chuang (Ref. 58)
6 NC4} loading Case I • NC8
6 NC4} loading Case II o NC8
~ ~~ 2 . }. 'Ahmad (Ref. 40) .
---- Exact
(b) v at End
10 20
<p, degrees
(e) u' at Midspan
FIG. 15 (CDr.JTH1UED)
30 40
20
-+-l '+-....... (/)
a. ...... ..;,t.
">, 40 :z
60
-4
" -+-l 4-.......
(/)
a. -2 .,... ..;,t.
.. -x z
0
" ~
86
-- Scordelis & Lo (Ref. 57)
A NC4 } Loading Case I /I NC8 6- NC4 } Loadi ng Case II 0 NC8 0 C12 } Ahmad (Ref. 40) II C8
(a) N at Midspan y
" ~ 0
(b) N I at Midspan x
<1>. degrees
FIG. 16 STRESS RESULTANTS OF CIRCULAR CYLINDRICAL SHELL UNDER UNIFORM LOAD
87
3.0 Scorde1is & Lo (Ref. 5?)
" NC4 Loading Case • " • NC8 •
" {j, NC4 Loading Case II 0 NC8
2. 0 C12
fIIB C8 } Ref. 40 +l 4- Exact .......... +-l 4-
I VI fIIB 0-
'r-.:>L.
X 1.0 ::E:
(c) M I x at Midspan
LO
+-l 4-.......... +-l 0.5 4-
I VI 0-
..:;;:
>, 0 ::E
-0.5 0 10 20 30 40
¢, degrees (d) r~ at r,l ids pan
y
FIG. 16 (CONTINUED)
10
8
;t: 6 ""til Cl..
>,
x z 4
2
o
+-> ~ ......... +-> ~
I til Cl..
'r-.::.t.
>,
X ~
88
Scorde1is & Lo (Ref. 57)
It. NC4 } Loading Case I
• NC8
tJ. NC4 } Loadi ng Case II 0 NC8
(e) N I I at End x y
2r------------r------------r-----------~-------------o C12 I C8 ;} Ref. 40
I
10 20 30 40
cp, degrees
(f). Mx'y' at End
FIG. 16 (CONTINUED)
(/)
QJ ...c u s::
N I o
89
6 E = 3 x 10 psi ]J = 0.0 t = 0.429 11
q = 104 lb/ft2
a = b = 48 11
FIG. 17 GEOMETRY AND MATERIAL PROPERTIES OF A SQUARE ELLIPTIC PARABOLOID
x a
--- Chu & Schnobrich (Ref. 18) Gustafson & Schnobrick (Ref. 60)
6 NC4} This study o NC8
o
w Along Center Line
FIG. 18 VERTICAL DISPLACEMENT OF SIMPLY SUPPORTED SQUARE ELLIPTIC PARABOLOID UNDER UNIFORM LOAD
1 .0
100
75 t::
'r-
'-..0 ,.......
>, 50 z;, x z
-150
-100 t:: 'r-
'-..0
>,
x -50 z
3/4 'L b
90
1/2
(a) Nx" Ny' Along the Center Line
-~-Chu & Schnobrich
__ Gustafson & Schnobrich
o N I
II N x I } NCB y } This study
tJ. NX' NC4 ,. N I }
Y
1/4 a
O~~ ____ , ____ L-__________ L-__________ ~ ________ ~
3/4 1.0 a 1/4 1/2
(b) N I I Along the Edge x y
x a
FIG. 19 STRESS RESULTANTS OF SIMPLY SUPPORTED SQUARE ELLIPTIC- PARABOLOID UNDER UNIFORM LOAD
~ I+--~ l+-
I VI 0.. ..... ~ .. -
>. :::E
0
2
4
1.0
91
3/4
(c) M I Along Center Line y
1/2
FIG. 19 (CONTHJIJED)
q
1/4
R = 90"
t = 3"
0
11 = 0.1333
E = 3 x 106 psi q = 1 psi
FIG. 20 GEm1ETRY MJO r~ATERIi\L PROPERTIES DF SPHERIC/\L SHELL
o
s::: .r-......... ..Cl ..-
I
s:::
-& :::E:
C 'r-......... ..Cl ..-
'" <D z:
92
40
Analytical (Ref. 56) 30 0 C12 (Ref. 47)
t:. NC4 } This study 0 NCB
20
10
0
(a) Meridional Bending Moment
O~----~------~------r-----~------r-------r-----i
10
20
30
40
o o
50 ~----~------~----~------~------~----~-------35 30 25 20 15 10
cp, degrees
(b) Hoop Force
FIG. 21 STRESS RESULTANTS OF SPHERICAL SHELL UNDER UNIFORM LOAD
5 o
0
-4.0
U1 OJ
.J:: u
-8.0 C or-
M I 0 ..-
~
-12.0
-16.0 o
93
z
a = b = 6.46"
t = 0.25 11
f = 1.304 11
or 1 .034 11
Jl = 0.39 E = 5 x 105 psi
q = 1.0 psi
FIG. 22 GEOMETRY AND MATERIAL PROPERTIES OF HYPERBOLIC PARABOLOID
Chetty & Tottenham -,,- Pecknold & Schnobrich
--- Chu & Schnobrich
6. NC4} w/f = 1.304 0 NC8
-::?" .l NC4} w/J = ~ ... II NC8 1 3 .. ~d' .. :;;--
e......-t:r'''--
/' /'
/'
1/4
~/
x a
1/2 3/4 1.0
FIG. 23 . VERTICAL DEFLECTION ACROSS MIDSPAN OF tLAMPED HYPERBOLIC PARABOLOID UNDER UNIFORM NORMAL LOAD
>,
x z
c:
'-. .0
..-: c:
'::£.>,
94
-2.0------------~----------~------------~----------~
-1. 0
0
1.0
2.0
3.0 o
(a) N I I Across Midspan x y
Chetty & Tottenham _ .. - Pecknold & Schnobrich
---t::. NC4
0 NCB
At. NC4
• NCB
Chu & Schnobrich
w/ f = 1.304
w/ f = 1.034
1/4 x a
(b) M I Across Midsection y
}
1/2
This Study
3/4 1.0
FIG. 24 STRESS RESULTANTS OF CLAMPED HYPERBOLIC PARABOLOID UNDER UNIFORM NORMAL LOAD
95
x
b = 49.5"
t = 0.5"
L = 72" h = 18" 1 h = 9"
2 -6 E = 5.62 x 10 psi
~ = 0.15 q = 60 pst
(pressure)
FIG. 25 GEOMETRY AND MATERIAL PROPERTIES OF CONOIDAL SHELL
-5
FIG. 26 DISPLACEMENTS OF SIMPLY SUPPORTED CONOIDAL SHELL UNDER UNIFORM LOAD
U'l QJ .c u s::
('Y) I 0
3:
U'l QJ .c u s::
('Y) I o
-10
-5
o
\
96
--- H-~leth } Ref. 62
- -0-- Exp
6 NC4 } This Study o NC8
I
(b) w at y = 0
1/4 1/2 3/4
\ / \ /
5 \ / \ ~
\ 6 I \ I \ I ~ I \ I \ ;1 \ I
\ / \ / b /
, n
10
15
" //"" " ' 0--""/ '0 __ -,
20 0 1/4 -
x 1/2 3/4 [
(c) I b
w at y = z
FIG. 26 (CONTINUED)
1.0
1.0
+-l 4--V')
0.. 'r-..::..:. .. -x z
+-l 4--V')
0.. or-~
'" -x z
-0.6
-0.4
-0.2
;..;0.4
-0.2
____ H-Meth
--0-- Exp
97
} Ref. 62
A NC4
o NCB } This Study
(a) N I at y = ~ x z
- l (b) Nx' at x = Z
Y.. b
FIG. 27. STRESS RESULTANTS OF SIMPLY SUPPORTED CONOIDAL SHELL UNDER UNIFORM LOAD
0.6
0.4
+-> 4--.. (/)
0.. .~
..:.t:.
>, 0.2
z
98
~o
-, '0""
....... -.0....... 0 -------0.... IJ ,
H-Meth } Ref. 62 - -cr- -- Exp
t:. NC4
o NC8 This Study
\ \ \ ~ \ \ \ \ \
-0.2~ __________ ~ __________ ~ __________ ~ __________ ~
o 1/4 -x 1/2 3/4 1.0 [
(c) Ny' at y = 0
0.4
.~
..:.t:.
" 0.2
0 0 1/4
Y 1/2 3/4
b
(d) N • - L at x = -y z
FIG. 27 (CONTINUED)
.,...
........
..0 ..-c: .....
c: .,... ........ ..0 ..-
0
c ..... .. -
:£:."»>
1/4
a
4.0
8.0
99
........ , " ,
"
H-Meth
" ~ \
6 \ \ \ \
\ } Ref. 62 --0-- Exp
6 NC4 } This Study o NCB
- 1/2 x l
3/4
12.0 L-.. ____ --L. _____ ...L-.. ____ .........;;:m:;;;...... ___ ----'!
o 1/4 1.. 1/2 3/4 1.0 b
(f) My' at x = ~
FIG: 27 (COnTINUED)
til Q)
..c u 1.0 s::
(V) I 0
~
2.0
Vl
~ 1.0 u s:: .....
(V) I o
.. ') n L.V
o
100
-x 1/4
[ 1/2 3/4
-- Ref. 62
t:. NC4 }This study 0 NC8 0 0
t:.
I
(a) w at y = 0
I - L (b) w at x = "2
FIG. 28 DEFLECTIONS OF CLAMPED CONOIDAL SHELL UNDER UNIFORM LOAD
1.0
1.0
t:. 0 o
-10.0
. -20.0 c:: '''' ........ ..a ,....
.. 40.0
101
KKTG } Ref. 62 --0--- Exp
t:. NC4} Thi s Study o NCB
o 0 / --a--:;!{1
I /0
-50.0 L ____ ~~::::......---~----__:~----~ o - 1/2 3/4 1.0 x
[
(b) N I at y = 0 y .
FIG. 29 STRESS RESULTA~~TS OF CLA~'1PED COrJOIDAL SHELL UnDER UN I FORr'1 PRESSURE
s:: or-......... ..0
-6.0
102
Ref. 62
A NC4 } This Study o NCB
1/4
(c) M I at y = 0 x
_ 1/2 x [
FIG. 29 (CONTINUED)
3/4 1.0
103
APPENDIX A
SHAPE FUNCTIONS
If a shell is defined by the mid-surface coordinates only, the
shape functions for the shell may also be represented, as two dimensional
elements, using ~ and n only. Here ~ and n are two curvilinear coordinates
in the mid-surface of shell which vary between -1 and 1 (Fig. 1).
A.l Linear Element
N. 1
1 = - (1
4 + ~) (1 o
+ n) o
with ~ = ~ E,. and n n n· o 1 0 1
A.2 Quadratic Element
Corner Nodes
Mid-side nodes (~. = ± 1, n. = 0) 1 1
AD3 Cubic Element
Corner nodes
+ ~) (1 o 2
- n)
N .. = -3'2 (' + ~ ) (1 + n ){-10 + 9(~2 + n2)} 1 0 0
(A.l )
(A.2)
(A.3)
(A.4)
104
Mid-side nodes (~ =
(A.5)
l05
APPENDIX B
LOCAL COORDINATE SYSTEM
Besides the basic two coordinate systems, the Cartesian and curvi-I I
linear coordinates systems, another orthogonal coordinate system, x , y , and I
Z , is defined locally (Fig. 4). The corresponding unit vectors, vl' v2' and
v3'" in the new directions of new coordinate systems are also shown in Fig. 4.
The unit vector normal to the shell surface can be established from
the cross product of two vectors tangential to the reference surface. This
has been done very efficiently by taking the first tltJO columns of the transpose
of the Jacobian which is obtained in the process of forming the stiffness
matrix.
The transpose of the J~cobian is
X'l; x, n X, I;:
[J] T = Y'l; y'n y., I;: (B. 1 )
z'l; Z, n z'l;:
Taking the first two columns
x'l; x, n
V3 = Y'l; x Y'n (B.2)
z'l; z, n
Then normalizing V':j ') one can get unit vector v') . ..J -.J
Once the vector v3 is defined it is possible to erect a infinite
number of mutua lly perpendicular vectors orthogonal to it. It is convenient
106
to choose axes so as to relate to the global axes, x or y, wherever possible.
First it is checked if the first or second term in vector v3 is zero. If tHe
second term is zero, a unit vector along the y-axis, denoted as (j), is es-
tablished. If the second term is non-zero, a unit vector along the x-axis,
(i), is established.
For the former case
j x (B.3)
Now, vl is a unit vector perpendicular to a plane defined by the y-axis and
v3° v2 is erected silTljJly by
x (B.4)
For the latter case the calcul~tion_begins with v2
instead of vl
in Eq. B.3 and follows the same procedures as the former case.
= x i (8.5)
The direction cosine matrix can be expressed by these three unit vectors.
[e] (B.6)
Thus the local I I
coordinates x , y , z and correcponding direction cosines
vl ' v2' v3 are obtained. It is noted here that even though new coordinates I
X ,y lie in the surface defined by ~, n they do not coincide with ~, n gen-I I
erally. For the plate, the directions of x ,y z are"coincide with the di-
rections of global axes, x, y, z.
107
APPENDIX C
LOCAL STRAINS AND TRANSFORMATIONS
At first, the global derivative matrix is transformed into the local
derivative matrix by the standard axis rotations.
I J I
U, I x v, I
x W, I
x U, x V, x w, X
I I
[eJT [OJ (C.l ) u, I v, I W, I = U 'y V'y W'y Y Y Y I I I U, I z V, I z W, J z U, z v, z W, z
3u I dU where I
etc. u, I, u, etc. --, , dX x x 3x
[eJ = direction cosine matrix [vl ' v2' v3 J
Since the shape functions, N. and N. in Eq. 3.8 and Eq. 3.12, are 1 J
defined with respect to the curvilinear coordinates ~, nand r;, (Appendix A)
another transformation from the curvilinear to global coordinates is needed
in order to evaluate the global derivatives of displacement with respect to
global coordinates. By applying the chain rule [53J and introducing the in-
verse of the Jacobian, the global derivative matrix is expressed by
u, x v, x w, x u,~ v'E, w,~
U'y v, w, = [J] -1 u, v, w, y Y n n n (C.2)
u, z v, z
w, z u'r;, v'(, w'r;,
where [JJ -1 is the inverse of Jacobian
108
x,(; y, E~ z'(;
[JJ = x, Y'n z, . (C.2.1 ) n n
x'L: Y , L: z'L:
Substituting Eq. C.2 into Eq. C. 1
I I , U, I x V, I x
W, I x u'l: v'l: w'l: I I I [e]T[J]-l [eJ u, I V, I W,I = U, v, W, Y Y Y n n n
I I I
U, I z V, : z
W, I z u'L: v'L: It! , L:
(C.3)
Now, it is possible to evaluate the local displacement derivatives numerically
for any given set of curvilinear coordinates (l:,n,L:).
Taking the'advantage of the normality condition of vectors in cal-
[ T -1 culating 8J [J] ,one may get a matrix [A] in a special form
o
[A] [e]T[J]-l o (C.4)
Introducing new expressions
B1 . = AllNi,r, + A12 N. 1
,., n
B2. = A21 N;,l: + A22 N. 1
1 ,n (~.5)
B3. = A33 Ni,l: 1
aN. where N. l:' etc.
_ 1 etc.
1 , -~ ,
109
Picking up proper terms from Eq. C.3 and utilizing Eq. 3.12, the strain vector
can be obtained
I T [: C. oT [:1 e T [~ + C. 8
T Gt {c } B. 8 + + B. 1 1 1 J
w . J
(C.6)
where
B. Bl 0 0 1
0 Bl 0
B2 81 0
0 0 Bl
0 0 . B2
C. = c;B l ;0 0
1
0 c;B2 0
c;B 2 cB :> 1 0
B3 0 c;B l
0 B3 c;B2
and B. and C. are defined in a similar way. J J
The approaches in this appendix are basically the same as given in
Ref. 40 except minor changes.
110
APPENDIX 0
NUMERICAL INTEGRATIONS
With isoparametric element analysis, complex calculations are in-
volved in determining both the element stiffness and the load vector. Numer-
ical integration is the only practical way to perform the calculations.
Gaussian integration is most suitably used for these computations.
For simplicity, one dimensional function f(~) is considered. First,
sampling points, so located to achieve best accuracy, should be determined
(Fig. 8). Then, with corresponding weighting coefficients H., the integrals 1
can be written for n sampling points as
I
1
J -1
f(C:) de: = n I H. f(e:.) 1 1 1
(D. 1 )
The locations of sampling points and the weighting coefficients are
given in Table 3 which was taken from Ref. 47.
For two and three dimensional functions, the integrals can be
written respectively, as
1 1 n m
J (
I J
F(e:,n) de: dn = L .L H.H.f(e:., n·) (0.2) i = 1 j=l 1 J 1 J
-1 -1
and 1 1 1 1 ( ( ( n m
I j j ) f(e:,n,~) d~ dn d~ = ·I L L H.H.H k (~i ,nj ,sk) i=l j=l . k=l • 1 J
-1 -1 -1 (0.3)
Frequently n, m arid 1 are taken as the same number for convenience.
The minimum size of integration mesh should be so determined that
111
the volume of the element is integrated adequately. The integration meshes
of 2x2, 3x3 and 4x4 are sufficient for linear, quadratic and cubic elements
respectively.
112
APPENDIX E
TYPICAL COMPONENTS OF ELEMENT STIFFNESS MATRIX
The element stiffness matrix Ke in Eq. 3.20 can be partitioned as
( E . 1 )
where Kll , K12 and K22 are defined in Eqs. 3.20.1 to 3.
The matrix Kll can also be subdivided into n x n submatrices for
an element with n nodes.
Kll r K K K .. K. 1 ----11 --12 --1 J -1 n
K ---21
j (E.2)
K' l K .. K. -1 -lJ -In
K K . K --n 1 nJ nn
whose typical submatrix, K .. , is of the order of 5 x 5 reflecting the 5 degrees -lJ
of freedom at each node.
where
The matrix K .. is further subdivided as -1 J
K .. K K -lJ -pp ---pr
(3x3) (3x2)
K K ---rp -rr (2x3) (2x2)
1 1 1
J r r T K J J
8 [B . DB . ] 8 I J I dE: dn dZ;; --pp 1 J -1 -1 -1
etc.
(E. 3)
113
If the thin shell assumptions are made, where the variation of [8J
with respect to ~ is negligible, an explicit integration with respect to C;, (1
i.e., in the form of J f(C;) dr::;, can be performed easily_ By doing -1
so, the
numerical integration is needed in only two directions (~,n) and the computa-
tional efforts for element stiffness matrix can be reduced substantially.
Thus,
where
K -pp
K -rp
K "~pr
1 (
= 2 J -1
ti {
-1 (1
tj J -1
t.t. --' ~
6
1
J -1
{ -1 (1
J -1
(1
J -1
t.t. J ' J + -2--
1
-1
T 8[B.08.J e IJI d~ dn 1 J
¢~ 8[C.OB.J 8T IJI d~ dn 1 1 J
e[B.DC.J 8T ¢. IJI d~ dn
1 J J
(1 T 'T j ¢. 8[B.OB.J 0 ¢. IJI d~ dn
'1 1 J J -1 1
J( ¢ ~ e [C. DC . ] 8 T ¢. I J I d~ dn
1 1 J J -1
[B.DB.J , J B1i011Blj + B2i023B2j Bli012~2j + B2i033Blj
(E.3.1 )
o
B2i012Blj + Bli033B2j B2i022B2j + Bli033Blj' 0
o
o
o
o Bli044Blj
+ B2iOSSB2j
(E.3.2)
114
[C.DB.] :::; 0 0 B3i D44Blj 1 J
0 0 B3i D55B2j
0 0 0
[C.DC.] r
B3iD44B3j 0 0 1 J
0 B3i D55B3j 0
0 0 0
The submatrices K12 and K22 in Eq. E.l are, respectively, in the forms of
K .. r " "'1 --1 J
l:PP j --rp
(E.4)
and
K .. [~p] --1 J (E. 5)
The terms in these matrices can be obtained in the similar way as Eq. E.3.1
and Eq. E.3.2 by simply changing B., etc. to ~. etc. 1 1
PART 1 - GOVERNMENT
Administrative & Liaison Activities
Chief of Naval Research Department of the Navy Arlington, Virginia 22217 Attn: Code 439
Code 468 Code 437
Director ONR Branch Office 495 Summer Street Boston, Massachusetts 02210
Director ONR Branch Office 536 South Clark Street Chicago, III inois 60605
Director Naval Research Laboratory Attn: Library Code 2029 (ONRL) Washington, D.C. 20390
U. S. Naval Research Laboratory Attn: Technical Information Div.
(2)
(6)
Washington, D.C. 20390 (6)
Commanding Officer ONR Branch Office 207 West 24th Street New York, New York 10011
Director ONR Branch Office 1030 E. Green Street
all (J 1 ..JI J V D
Defense Documentation Center Cameron Station Alexandria, Virginia 22314
Commanding Officer U. S. Army Research Off. Durham Attn: Mr. J. J. Murray CRD-AA-IP Box CM, Duke Station Durham, North Carolina 27706
( 12)
Redstone Scientific Info. Center Chief, Document Section U. S. Army Missile Command Redstone Arsenal, Alabama 35809
Army R&D Center Fort Belvoir, Virginia 22060
Commanding Officer and Director Naval Ship Res & Dev Center Washington, D. C. 20007 Attn: Code 042 (Tech. Lib. Br.)
Code 700 (Struc. Mech. Lab.) Code 720 Code 725 Code 727 Code 800 (Appl. Math. Lab.) Code 901 (Dr. M. Strassberg) Code 941 (Dr. R. Liebowitz) Code 945 (Dr. W. S. Cramer) Code 960 (Mr. E. F. Noonan) Code 962 (Dr. E. Buchmann)
Naval Research Laboratory Washington, D. C. 20390 Attn: Code 8400
Code 8410 Code 8430 Code 8440
Director, Aero Mechanics Dept. Naval Air Development Center Johnsville Warminister, Pennsylvania 18974
Undersea Explosion Research Div. Naval Ship R&D Center Norfolk Naval Shipyard Portsmouth, Virginia 23709 Attn: Dr. Schauer Code 780
Nav. Ship R&D Center Annapolis Division Annapolis, Maryland 21402 Attn: Code A125 Mr. Y. F. Wang
Navy (Contid)
U. S. Naval Weapons Center China Lake, Cal ifornia 93557 Attn: Code 4062 Mr. W. Werback
Code 4520 Mr. Ken Bischel
Commanding Officer U. S. Naval Civil Engr. Lab. Code L31 Port Hueneme, Cal ifornia 93041
Technical Director U. S. Naval Ordnance Laboratory Whi te Oak Silver Spring, Maryland 20910
Technical Director Naval Undersea Rand D Center San Diego, Cal ifornia 92132
Supervisor of Shipbui 1ding U. S. Navy Newport News, Virginia 23607
Technical Director Mare Island Naval Shipyard Vallejo, Cal ifornia 94592
U." S. Navy Underwater Sound Ref. Lab. Office of Naval Research P. O. Box 8337 Orlando, Florida 32806
Chief of Naval Operation Department of the Navy Washington, D. C. 2~35~
Attn: Code OP-07T
Deep Submergence Systems Naval Ship Systems Command Code 39522 Dept. of the Navy Washington, D. C. 20360 Attn: Chief Scientist
Eng i neer i ng Dept .. U. S. Naval Academy Annapol is, Maryland 21402
2
Naval Air Systems Command Dept. of the Navy Washington, D. C. 20360 Attn: NAJR 320 Aero. & Structures
5308 Structures 604 Tech. Library
Naval Facilities Engineering Command Dept. of the Navy Washington, D. C. 20360 Attn: NFAC 03 Res. & Development
04 Engineering & Design 14114 Tech. Library
Naval Ship Systems Command Dept. of the Navy Washington, D. C. 20360 Attn: NSHIP 031 Chief Scientists for R&D
03412 Hydromechanics 037 Ship Silencing Div. 2052 Tech. Library
Naval Ship Engineering Center Prince George Plaza Hyattsville, Maryland 20782 Attn: NSEC 6100 Ship Sys. Eng. & Des. Dept.
6l02C Computer-aided Ship Design 6105 Ship Safety 6110 Ship Concept Design 6120 Hull Division 6120D Hull Division 6128 Surface Ship Struct. 6129 Submarine Struct.
Naval Ordnance Systems Command Dept. of the Navy Washington, D. C. 20360 Attn: NORD 03 Res. & Technology
035 Weapons Dynamics 9132 Tech. Library
Air Force
Commander WADD Wright-Patterson Air Force Base Dayton, Ohio 45433 Attn: Code WWRMDD
AFFDL (FDDS) Structures Division AFLC (MCEEA)
Air Force (Cont'd)
Chief, Appl ied Mechanics Group U. S. Air Force Inst. of Tech. Wright-Patterson Air Force Base Dayton, Ohio 45433
Chief, Civi 1 Engineering Branch WLRC, Research Division Air Force Weapons Laboratory Kirtland AFB, New Mexico 87117
Air Force Office of Scientific Res. 1400 Wilson Blvd. Arlington, Virginia 22209 Attn: Mech. Div.
NASA
Structures Research Division NASA Langley Research Center Langley Station Hampton, Virginia 23365 Attn: Mr. R. R. Heldenfels, Chief
National Aeronautic & Space Admin. Associate Administrator for Advanced Research & Technology Washington, D. C. 20546
Scientific & Tech. Info. Facility NASA Representative (S-AK/DL) P. O. Box 5700 Bethesda, Maryland 20014
Other Government Activities
Commandant Chief, Testing & Development Div. U. S. Coast Guard 1300 E Street, N. W. Washington, D. C. 20226
Technical Director Marine Corps Dev. and Educ. Command Quantico, Virginia 22134
Engineering Division Washington, D. C. 20550
3
Science & Tech. Division Library of Congress Washington, D. C. 20540
Director STBS Defense Atomic Support Agency Washington, D. C. 20350
Commander Field Command Defense Atomic Support Agency Sandia Base Albuquerque, New Mexico 87115
Chief, Defense Atomic Support Agcy. Blast & Shock Division The Pentagon Washington, D. C. 20301
Director Defense Research & Engr. Technical Library Room 3C-128 The Pentagon Washington, D. C. 20301
Chief, Airframe & Equipment Branch FS-120 Office of Fl ight Standards Federal Aviation Agency Washington, D. C. 20553
Chief, Research and Development Maritime Administration Washington, D. C. 20235
Deputy Chief, Office of Ship Constr. Maritime Administration Washington, D. C. 20235 Attn: Mr. U. L. Russo
Mr. Milton Shaw, Director Div. of Reactor Deve!. & Technology Atomic Energy Commission Germantown, Maryland 20767
Ship Hull Research Committee National Research Council National Academy of Sciences 2101 Constitution Avenue Washington, D. C. 20418 Attn: Mr. A. R. Lytle
PART 2 - CONTRACTORS AND OTHER TECHNICAL COLLABORATORS
Universities
Dr. J. Tinsley Oden Dept. of Engr. Mechs. University of Alabama Huntsville, Alabama 35804
Professor Me. E. Gurtin Dept. of Mathematics Carnegie Institute of Technology Pittsburgh, Pennsylvania 15213
Professor Jul ius Miklowitz Division of Engr. & Appl ied Sciences Cal ifornia Institute of Technology Pasadena, Cal ifornia 91109
Dr. Harold Liebowitz, Dean School of Engr. & Appl ied Scinece George Washington University 725 23rd Street Washington, D. C. 20006
Professor Eli Sternberg Div. of Engr. & Appl ied Sciences Cal ifornia Institute of Technology Pasadena, California 91109
Professor Paul M. Naghdi Div. of Applied Mechanics Etcheverry Hall University of California Berkeley, Cal ifornia 94720
Professor W. Nachbar University of Cal ifornia Dept. of Aerospace & Mech. Engr. La Jolla, Cal ifornia 92037
Professor J. Baltrukonis Mechanics Division The Cathoiic Univ. of America Washington, D. C. 20017
Professor A. J. Durell i Mechanics Division The Cathol ic Univ. of America Washington, D. C. 20017
Professor R. B. Testa Department of Civil Engineering Columbia University S. W. Mudd Bui lding New York, New York 10027
4
Professor H. H. Bleich Department of Civi 1 Engr. Columbia University Amsterdam & 120th Street New York, New York 10027
Professor R. D. Mindlin Department of Civil Engr. Columbia University S. W. Mudd Bui lding New York, New York 10027
Professor F. L. DiMaggio Department of Civil Engr. Columbia University 616 Mudd Building New York, New York 10027
Professor A. M. Freudenthal George Washington University School of Engineering and Appl ied Science Washington, D. C. 20006
Professor Wm. Prager Division of Engineering Brown University Providence, Rhode Island 02912
Dr. Y. Weitsman Department of Engineering Sciences Tel-Aviv University Ramat ... Aviv Tel-Aviv, Isreal
Professor P. V. Marcal Brown University Division of Engineering Providence, Rhode Island 02912
Professor Norman Jones Massachusetts Institute of Technology Department of Naval Architecture and Marine Engineering Cambridge, Massachusetts 02139
Dr. S. Dhawan, Director Indian Institute of Science Bangalore, India
Dr. Charles E. Hutchinson . School of Engineering University of Massachusetts Amherst, Massachusetts 01002
Universities (Cont1d)
Professor Lambert Tall LeHigh University Department of Civil Engineering Bethlehem, Pennsylvania 18015
Professor B. A. Boley Department of Theor. & Appl. Mech. Cornell University Ithaca, New York 14850
Professor P. G. Hodge Department of Mechanics 111 inols Institute of Technology Chicago, Illinois 60616
Dr. D. C. Drucker Dean of Engineering University of Illinois Urbana, III inois 61801
Rofessor N. M. Newmark Dept. of Civil Engineering University of Illinois Urbana, Illinois 61801
Professor E. Reissner University of California, San Diego Department of Applied Mechanics La Jolla, California 92037
Professor Wi 11 iam A. Nash Dept. of Mechs. & Aerospace Engr. University of Mass. Amherst, Mass. 01002
Library (Code 0384) U. S. Naval Postgraduate School Monterey, California 93940
Professor Arnold Al1entuch Dept. of Mechanical Engineering Newark College of Engineering 323 High Street Newark, New Jersey 07102
Professor E. L. Reiss Courant Inst. of Math. Sciences New York University 4 Washington Place New York~ New York 10003
5
Dr. George Herrmann Stanford University Department of Appl ied Mechanics Stanford, California 94305
Professor J. D. Achenbach Technological Institute Northwestern University Evans ton, III i no i s 60201
Director, Ordnance Research Lab. Pennsylvania State University P. O. Box 30 State College, Pennsylvania 16801
Professor Eugene J. Skudrzyk Department of Physics Ordnance Research Lab. Penn State Univ. P. O. Box State College, Pennsylvania
Professor J. Kempner
30 16801
Dept. of Aero. Engr. & Applied Mech. Polytechnic Institute of Brooklyn 333 Jay Street Brooklyn, New York 11201
Professor J. Klosner Polytechnic Institute of Brooklyn 333 Jay Street Brooklyn, New York 11201
Professor R. A. Schapery Civil Engineering Department Texas A and M University College Station, Texas 77840
Dr. Nicholas J. Hoff Dept. of Aero. & Astro. Stanford University Stanford, California 94305
Professor Max Anl iker Dept. of Aero. & Astro. Stanford University Stanford, California 94305
Professor W. D. Pilkey Department of Aerospace Engineering University of Virginia Charlottesville, Virginia 22903
Univ~rsities (Cont'd)
Professor Chi-Chang Chao Division of Engineering Mechanics Stanford University Stanford, California 94305
Professor S. B. Dong University of California Department of Mechanics Los Angeles, California 90024
Professor Tsuyoshi Hayashi Department of Aeronautics University of Tokyo Bunkyo-Ku Tokyo, Japan
Professor J. E. Fitzgerald, Ch. Dept. of Civil Engineering University of Utah Salt Lake City, Utah 84112
Professor R. J. H. Bollard Chairman, Aeronautical Engr. Dept. 207 Guggenheim Hall University of Washington Seattle, Washington 98105
Professor Albert S. Kobayashi Dept. of Mechanical Engr. University of Washington Seattle, Washington 98105
Librarian Webb Institute of Naval Arch. Crescent Beach Road, Glen Cove Long island, New York iiS42
Dr. Daniel Frederick Dept. of Engr. Mechs. Virginia Polytechnic Inst. Blacksburgh, Virginia 24061
Industry and Research Institutes
Dr. Vito Salerno Applied Technology Assoc., Inc. 29 Church Street Ramsey, New Jersey 07446
Library Services Department Report Section, Bldg. 14-14 Argonne National Laboratory 9700 S. Cass Avenue Argonne, Illinois 60440
6
Dr. M. C. Junger Cambridge Acoustical Associates 129 Mount Auburn Street Cambridge, Massachusetts 02138
Supervisor of Shipbuilding, USN, and Naval Insp. of Ordnance Electric Boat Division General Dynamics Corporation Groton, Connecticut 06340
Dr. L. H. Chen Basic Engineering Electric Boat Division General Dynamics Corporation Groton, Connecticut 06340
Dr. Joshua E. Greenspon J. G. Engr. Research Associates 3831 Menlo Drive Baltimore, Maryland 21215
Library Newport News Shipbui lding & Dry Dock Company Newport News, Virginia 23607
Mr. J. I. Gonza 1 ez Engr. Mechs. Lab. Martin Marietta MP-233 P. O. Box 5837 Orlando, Florida 32805
Dr. M. L. Merritt Division 5412 Sandia Corporation Sandia Base Albuquerque, New Mexico 87115
Director Ship Research Institute Ministry of Transportation 700, Shinkawa Mi taka Tokyo, Japan
Dr. H. N. Abramson Southwest Research Institute 8500 Culebra Road San Antonio, Texas 78206
Dr. R. C. DeHa rt Southwest Research Institute 8500 Culebra Road San Antonio, Texas 78206
Dr. M. L. Ba ron Paul Weidlinger Associates 110 East 59th Street New York, New York 10022
Industry and Research Institutes (Cont1d)
Mr. William Caywood Code BBE Appl ied Physics Lab. 8621 Georgia Ave. Silver Springs, Maryland 20034
Mr. J. Hrzina Bldg. R-l, Room 1104A TRW Systems 1 Space Park Dedondo Beach, Cal ifornia 90278
Mr. D. Wilson Litton Systems, Inc. AMTD, Dept. 400 El Segundo 9920 West Jefferson Blvd. Culver City, Cal ifornia 90230
Mr. P. C. Durup Aeromechanics Department, 74-43 Lockheed-Cal ifornia Co. Burbank, California 91503
Dr. Kevin J. Forsberg, Head Sol id Mechanics Orgn. 52-20, Bldg. 205 Lockheed Palo Alto Research Lab. Palo Alto, California 94302
Dr. E. M. Q. Roren Head, Research Department Det Norske Veritas Post Box 6060 Oslo, Norway
Westinghouse Electric Corporation Bettis Atomic Power Laboratory P. O. Box 79 West Mifflin, Pennsylvania 15122
Dr. Andrews F. Conn Hydronautics, Incorporated Pindell School Road, Howard County Laurel, Maryland 20810
7
Spell rit \" Cl a s sifica tion
DOCUMENT CONTROL DATA - R&D I 5"clJrity cla!>sificatiotl of title, body of abstract ilnd indexing annotation must be erl'tered when the overall report is classified) t---~---....:..-......~-----.:.:....---""""""""T""----"':'-----"-;"-"""""""
C" ·.~IN"'TING ACTIVITY (Corporate allthor) 2<1. REPORT SECURITY CLASSIFICATION
Department of Civil Engineering Univers i ty of III inois Urbana, Illinois 61801
2h. GROUP
\ f-. L P 0 R TTl T L E
Use of Nonconforming Modes in Finite Element Analysis of Plates and Shells
·1 (("iCRIPTIVE NOTES (Type of report and. inclusive Jates)
:, AU THOR(S) (First name, middle initial, last name)
Chang-Koon Choi and W. C. Schnobrich
6 REPORT DATE 7B. TOTAL NO. OF PAGES 17b. NO. OF REFS
September 1973 120 Ila. CONTRACT OR GRANT NO. 9a. ORIGINATOR'S REPORT NUMBER(S)
N00014-67-A-0305-00l0 SRS 401 b. PROJECT NO.
c.
d.
9b. OTHER REPORT NO(S) (Any other numbers that may be Bssigned this report)
U,I LU-ENG-73-20l9 10 DISTRIBUTION STATEMENT
Release Unlimited
11. SUPPLEMENTARY NOTES
13. ABSTRACT
1~. SPONSORING MILITARY ACTIVITY
Office of Naval Research Structural Mechanics Branch Arlington, Va. 22217
The development of yet another finite element is presented in this study. By adding nonconforming displacement modes selectively to the low order isoparametri elements, highly improved new shell elements are established. Addition of various combinations of nonconforming modes to the original elements has been investigated and the elements, both linear and quadratic, which have two nonconforming modes added to the inplane displacements, four to the out-of-p1ane displacement and no additional modes to rotations have been found to be most effective in terms of the number of nonconforming modes used versus the improvement obtained. These elements have been tested with deep beam, plate, and various shell problems. Cylindrical, elliptic paraboloid, spherical and conoidal shells were investi~ated in order to demonstrate the versatil ity of newly developed nonconforming elements. The results obtained from these investigations compare favorably with the results from well-establ ished analyses. The improvement incorporated into the four-node element is most impressive.
D FORM -1 NOV651
(PAGE 1) .. '
73 SIN 0101-807-6811 Security Classification
A-3140B
j" LIN K A LIN K B LIN K C
KEY WORDS
ROLE WT ROLE WT ROLE WT
~ Structural She 11 s Numerical Analysis Finite Element
I
I
FORM 1 1 NOV 65
(BACK)
Security Classification ·'\-31409