AN APPROXIMATE SOLUTION TO BUCKLING OF

PLATES BY THE GALERKIN METHOD

by

YATTENDER RISHI DUBEY

Presented to the Faculty of the Graduate School of

The University of Texas at Arlington in Partial Fulfillment

of the Requirements

for the Degree of

MASTER OF SCIENCE IN MECHANICAL ENGINEERING

THE UNIVERSITY OF TEXAS AT ARLINGTON

August 2005
 ACKNOWLEDGEMENTS

I would like to acknowledge with great pleasure and appreciation the support

and encouragement by Dr. Seiichi Nomura throughout the thesis work. I am also

thankful to Dr. Wen S. Chan and Dr. Dereje Agonafer for serving on my thesis

committee.

Special appreciation is extended to my parents Dr. Chandra Prakash Dubey,

Mrs. Sarita Dubey, and my sibling Ms. Swati Dubey. I appreciate the encouragement

and help that I have received from my roommates and all my friends.

July 25 , 2005

ii
 ABSTRACT

AN APPROXIMATE SOLUTION TO BUCKLING OF

PLATES BY THE GALERKIN METHOD

Publication No. ______

Yattender Rishi Dubey, M.S.

The University of Texas at Arlington, 2005

Supervising Professor: Dr. Seiichi Nomura

This thesis presents a semi-analytical method to solve the governing differential

equation of plates using the Galerkin method. Symbolic algebra software is extensively

used to perform necessary calculations.

In this thesis, the lateral deflection of a plate is expressed in a series of

polynomials each of which satisfies the given homogeneous boundary conditions. The

coefficients of these polynomials are found out by the Galerkin method. Symbolic

algebra software works best while handling necessary algebra to generate admissible

polynomials and build required matrices for solving for the coefficients.

This thesis demonstrates the calculation of the lateral load for different plates

under different homogeneous boundary conditions and initial condition. As this analysis

iii
involves very complicated computation, it is almost impossible to handle all the

calculations without the help of symbolic algebra software.

Numerical examples are presented and the results are compared with the known

analytical solutions. It was shown that a reasonable level of convergence is achieved

with the present method.

iv
 TABLE OF CONTENTS

ACKNOWLEDGEMENTS....................................................................................... ii

ABSTRACT .............................................................................................................. iii

LIST OF ILLUSTRATIONS..................................................................................... vii

LIST OF TABLES..................................................................................................... viii

Chapter

1. INTRODUCTION……… ............................................................................. 1

1.1 Overview…………….............................................................................. 1

1.2 Galerkin Method…….............................................................................. 3

1.3 Symbolic algebra software ...................................................................... 4

2. BUCKLING THEORY…………………………………………………….. 7

2.1 Historical Background ............................................................................. 7

2.2 What is Buckling? .................................................................................... 8

2.3 Derivation of the Governing equation………………………………….. 9

2.4 Galerkin method…..……………………………………………………… 22

3. METHOD OF WEIGHTED RESIDUALS................................................... 25

3.1 Introduction………….............................................................................. 25

3.2 Application of Galerkin method .............................................................. 27

v
 4. NUMERICAL RESULTS….. ........................................................................ 46

4.1Simply supported boundary condition………………………………….. 46

4.2 Mixed boundary condition....................................................................... 49

4.3 Weinstein’s Theory…………………………………………………….. 51

4.4 Rectangular plate axially compressed in one direction…………………. 54

5. CONCLUSIONS AND RECOMMENDATIONS…………………………… 59

Appendix

A. MATHEMATICA PROGRAMS TO SOLVE THE

BUCKLING LOAD EQUATION OF A PLATE
FOR DIFFERENT CASES……………………………………………….. 61

REFERENCES .......................................................................................................... 71

BIOGRAPHICAL INFORMATION......................................................................... 73

vi
 LIST OF ILLUSTRATIONS

Figure Page

2.1 Differential plate element with stress resultants............................................... 9

2.2 Rectangular plate under concentrated load ..................................................... ..16

3.1 Rectangular plate............................................................................................ ...28

4.1 Simply supported plate……………………………………………………… .. 47

4.2 Plate with mixed boundary conditions………..……………………………….. 49

vii
 LIST OF TABLES

Table Page

3.1 Coefficients of 8th order polynomial for

simply supported plate with aspect ratio 1.0 ..................................................... 38

3.2 Coefficients of 9th order polynomial for

simply supported plate with aspect ratio 1.0 ..................................................... 39

3.3 Coefficients of 6th order polynomial for isotropic

plate with mixed boundary conditions .............................................................. 43

3.4 Coefficients of 7th order polynomial for isotropic

plate with mixed boundary conditions .............................................................. 44

4.1 Comparison of eigenvalues for simply supported plate........................................48

4.2 Comparison of eigenvalues for plate with mixed

boundary condition..……………………………………………………………50

4.3 Comparison of Eigenvalues for a simply supported

plate loaded axially……………………………………………………………..57

viii
 CHAPTER 1

INTRODUCTION

1.1 Overview

The first significant analysis of plates began in the early 1800s, with much of

the work attributed to Cauchy, Poisson, Navier, Lagrange, and Kirchoff. The first

mathematical approach to the membrane theory of plates was formulated by Euler

(1766), who solved the problem of free vibrations of rectangular and circular elastic

membranes using the analogy of two systems of stretched strings perpendicular to each

other [17]. The great engineer, Navier (1785-1836), can be considered the real

originator of the modern theory of elasticity. His numerous scientific activities included

the solution of various plate problems. Navier derived the correct differential equation

of rectangular plates with flexural resistance. Navier’s method is based on the use of

trigonometric series introduced by Fourier in the same decade. Poisson extended (1829)

the use of the governing plate equation, derived by Navier, to the lateral vibration of

circular plates. The boundary conditions of the problem, as formulated by Poisson,

however, are applicable only to thick plates. Kirchoff (1824-1887) [18] is considered

the founder of the extended plate theory which takes into account the combined bending

and stretching. The research done by these early “Engineers” was extremely significant,

and many of the techniques which they developed are still used in engineering analysis

today.

1
 However, in the last 30 years, with the advent of the digital computer, other

methods such as finite differences and finite elements have become practical. In 1956

Turner, Clough, Martin and Topp [15] introduced the finite element method, which

permits the numerical solution of complex plate and shell problems in an economical

way. The use of the finite element method, however, anticipates the availability of high-

speed computers with considerable storage capacity. Another computer-oriented

method for the static and dynamic analysis of plates of arbitrary shape subjected to

arbitrary loads is based on improved finite difference techniques developed by Stussi

and Collatz.

The equation defining small lateral deflections of the middle surface of a thin

plate subjected to lateral loads may be formulated in different ways. The most general

method involves the elimination of the unimportant terms from the equations of the

three-dimensional elasticity as the thickness is made small compared with other

dimensions. A second method, which requires less mathematical rigor but more

physical interpretation, utilizes the basic assumptions made in the theory of beams to

generate directly the equations for thin plates. Both methods lead to the same results.

The development of the governing partial differential equation defining small

lateral deflections of the middle surface of thin plates, as well as the development of the

companion relationships, is based on certain assumptions adopted because of prior

knowledge about the behavior of beams. Because of the limitation of small deflections,

1. The middle surface is assumed to remain unstrained. Because of the limitation of

thinness,

2
2. Normals to the middle surface before deformation remain normal after deformation,

and

3. Normal stresses in the direction transverse to the plate are neglected. A fourth

assumption based on the material properties is that

4. The material is homogeneous, isotropic, continuous, and linearly elastic.

1.2 Galerkin Method

Prior to the development of the Finite Element Method, there existed an

approximation technique for solving differential equations called the Method of

Weighted Residuals (MWR).The basic idea of MWR is to use a trial function with a

number of unknown parameters to approximate the solution. Then a weighted average

over the interior and boundary is set to zero. The idea is to approximate the solution

with a polynomial involving a set of parameters. The polynomial is made to satisfy both

the differential equation and the associated boundary conditions.

The Method of Weighted Residuals is broadly classified into:

1.2.1 Collocation Method

1.2.2 Least Squares Method

1.2.3 Galerkin Method

The Galerkin method has been used to solve problems in mechanical

engineering such as structural mechanics, dynamics, fluid flow, heat and mass transfer,

acoustics, neutron transport and others. The Galerkin method can be used to

approximate the solution to ordinary differential equations, partial differential equations

and integral equations.

3
 The origin of the method is generally associated with a paper published by

Galerkin in 1915 on the elastic equilibrium of rods and thin plates. The use of Galerkin

method increased rapidly during the 1950’s when it was used for analyzing dynamics of

aeronautical structures. The link with the Galerkin method permitted the finite element

techniques to be extended into areas where no variational principle was available for

example, such as many aspects of fluid mechanics, solid mechanics and heat transfer.

The finite element and Galerkin methods are currently the standard numerical

technique in use to solve various nonlinear problems. The methods retain the

advantages of weak formulations, which lower the continuity requirements of matching

elements and permits to use simple basis functions. However, these methods demand a

great amount of numerical integration effort in updating Jacobian matrix of each

Newton-Raphson iteration.

1.3 Symbolic algebra software

Since the 1960's, research work on the design of algorithms and systems for performing

symbolic mathematics (now commonly known as ''computer algebra'') was underway at

institutions such as MIT. The systems which were used during the 1970's (such as

MACSYMA and REDUCE) were very large LISP-based systems requiring many

megabytes of RAM and extensive CPU time to perform routine mathematical

computations. Consequently, only a tiny number of researchers with access to large

mainframe computers (and only if they didn't have to time-share with many other users)

were able to exploit this technology. With the emergence of modern symbolic algebra

software packages such as MATHEMATICA, MATLAB and MAPLE, which were

4
written in the C language and its variations, the megabytes of RAM and CPU running

time was considerably reduced.

Mathematica is the one of the most useful general computation system. Ever

since the 1960s individual packages had existed for specific numerical, algebraic,

graphical, and other tasks. But the visionary concept of Mathematica was to create once

and for all a single system that could handle all the various aspects of technical

computing--and beyond--in a coherent and unified way. The generality of symbolic

expressions allows Mathematica to cover a wide variety of applications. The symbolic

function is built-in and is very simple and robust and does not require any additional

commands to activate it. Programs written in Mathematica have better readability.

This thesis demonstrates how to find the critical buckling load value of square

plates under different constrained conditions using Galerkin method. It takes advantage

of both numerical techniques and symbolic algebra software. In this thesis, it is assumed

that the lateral deflection is expressed by a series of polynomials each of which satisfies

the given boundary conditions. The Galerkin method is used to determine the

coefficients of these polynomials.

Chapter 2 focuses on the buckling theory of plates and derivation of the governing

equations for buckling of square plates with associated boundary conditions. The

application of Galerkin method is also illustrated in this chapter.

5
Chapter 3 discusses solving the buckling load equation for a square plate subjected to

uniform compressive load with different boundary conditions. There are two types of

boundary conditions that are being discussed .Examples include –

a) Simply supported boundary conditions

b) Mixed boundary conditions

Chapter 3 will also illustrate the use of Galerkin method to solve for the buckling load

using Mathematica. Since Mathematica handles all the expressions symbolically, a

major advantage it has is the ability to carry out calculations and substitutions with ease.

After finding the buckling load using the same approximating function as the square

plate, a higher order approximating function will be used to show the convergence of

this method.

In Chapter 4 numerical results obtained from Mathematica programs will be compared

with standard numerical examples.

Chapter 5 discusses conclusion and future recommendations to the existing work.

6
 CHAPTER 2

BUCKLING THEORY

2.1 Historical Background

The initial theoretical research into elastic flexural-torsional buckling was preceded by

Euler’s 1759 treatise[12] on column flexural buckling, which gave the first analytical

method of predicting the reduced strengths of slender columns, and by Saint-Venant’s

1855 memoir [13]on uniform torsion, which gave the first reliable description of the

twisting response of members to torsion.

However, it was not until 1899 that the first treatments were published of

flexural-torsional buckling by Michell and Prandtl, who considered the lateral buckling

of beams of narrow rectangular cross-section. Their work was extended by

Timoshenko[14] to include the effects of warping torsion in I-section beams. Most

recently the invention of high-speed electronic computers (1950) exerted a considerable

influence on the static and dynamic analysis of plates. Although in 1941 Hrennikoff

had already developed an equivalent grid work system for the static analysis of complex

plate problem, his fundamental work in discretization of continua could not be fully

utilized due to the lack of high-speed computers. In 1956 Turner, Clough, Martin and

Topp [15] introduced the finite element method, which permits the numerical solution

of computers with considerable storage capacity.

7
 The extension of the general finite element method of structural analysis to flexural-

torsional buckling studies, since almost any particular situation can now be analyzed

using a general purpose computer program. This development is similar to that which

occurred in the in-plane analysis of plane rigid-jointed frames, in which the tabulations

of solutions used in the 1930s were replaced by general purpose frame computer

analysis programs.

2.2 What is Buckling?

When a slender structure is loaded in compression, for small loads it deforms with

hardly any noticeable change in geometry and load carrying ability. On reaching a

critical load value, the structure suddenly experiences a large deformation and it may

lose its ability to carry the load. At this stage, the structure is considered to have

buckled.

Buckling, also known as structural instability may be classified into two

categories:

a) Bifurcation buckling

b) Limit load buckling

In bifurcation buckling, the deflection under compressive load changes from one

direction to a different direction (e.g., from axial shortening to lateral deflection). In

limit load buckling, the structure attains a maximum load without any previous

bifurcation, i.e., with only a single mode of deflection.

8
 2.3 Derivation of the Governing equation

We consider the classical thin plate theory (CPT) in deriving the governing equation,

which is based on the Kirchhoff hypothesis:

(a) Straight lines perpendicular to the mid-surface (i.e., transverse normals) before

deformation remain straight after deformation

(b) The transverse normals do not experience elongation (i.e., they are inextensible).

(c)The transverse normals rotate such that they remain perpendicular to the mid-surface

after deformation.

The condition of equilibrium is met by considering a rectangular differential element of

dimensions dx, dy and h as shown in figure. For simplicity, only the middle surface of

the plate is shown. The stress resultants, which are internal reactions per unit of length,

exist because of the transverse distributed load of intensity p(x,y) on the upper surface

of the plate.

Fig 2.1 Differential plate element with stress resultants

9
 2 2
w w (2.1)
M x = D( + )
x2 y2
2 2
w w
M y = D( 2
+ ) (2.2)
x y2
2
w (2.3)
M xy = M yx = D(1 )
x y

Expressions for the stress resultants acting on the transverse faces at (x+ dx, y) and at

(x, y +dy) are obtained by expanding each into a Taylor’s series about (x, y). The higher

order terms are considered negligible since dx and dy are quantities of infinitesimal

value.

Q yz
Q yz * = Q yz + dy (2.4)
y

Q xz
Q xz * = Q xz + dx
x
My
M y* = M y + dy
y
Mx
M x* = M x + dx (2.5)
x
M yz
M yx * = M yx + dy
y
M xy
M xy * = M xy + dx
x

The condition of a vanishing resultant force in the z-direction results in the equation

Q yz Q xz
+ + p = 0. (2.6)
y x

10
If the resultant moment about an edge parallel to the x-axis is set equal to zero while

neglecting higher order terms, the resulting equation is

My M xy
Q yz = 0. (2.7)
y x

Similarly, the equilibrium equation with respect to rotation about an edge parallel to the

y-axis is

Mx M yx
Q xz = 0. (2.8)
x y

if the expression for Qyz from Eq.(2.7) and the expression for Qxz from Eq. (2.8) is

substituted into Eq. (2.6), the resulting equation is

2 2
2
Mx M xy My
2 + = p (2.9)
x2 x y y2

Substitution of Eqs. (2.1), (2.2), and (2.3) into Eqs. (2.7) and (2.8) gives the expressions

for Qxz and Qyz in terms of the deflection of the middle surface

3 3
w w
Q xz = D( 3 + )= D ( 2
w) (2.10)
x x y2 x

and

3 3
w w
Q yz = D( 3
+ 2
)= D ( 2
w) (2.11)
y x y y

2 2
where 2
= + (2.12)
x2 y2

11
 2.3.1 Governing Equation

The governing partial differential equation defining the lateral deflection of the

middle surface of the plate in terms of the applied transverse load is obtained by direct

substitution of Eqs. (2.1), (2.2), and (2.3) into the equilibrium Eq.(2.9). The result of

this is

4 4 4
w w w p 2w
+2 2 2 + 4 = (2.13)
x4 x y y D x2

p 2w
or 4
w= (2.14)
D x2

p 2w
or 2 2
w= (2.15)
D x2

4 4 4
Where 4
= +2 + (2.16)
x4 x2 y2 y4

The fourth-order partial differential Eq. (2.13) can be reduced to two second-order

partial differential equations which are sometimes preferred, depending upon the

method of solution to be used. This reduction is accomplished as follows. The addition

of Eqs. (2.1) and (2.2) gives

2 2
w w
M x + M y = D(1 + )( 2
+ ) (2.17)
x y2

2
w 2
w Mx + My
or + = (2.18)
x 2
y2 D(1 + )

M
or 2
w= (2.19)
D

12
 Mx + My
where M= (2.20)
1+

Substitution of Eq. (2.19) into (2.15) gives

M p
2
( )= (2.21)
D D

or 2
M=p (2.22)

Thus, the fourth order Eq. (2.13) has been reduced to the two second-order Eqs, (2.19)

and (2.22). If boundary conditions and the transverse load p are known, Eq. (2.22) can

be solved for M(x,y). Then Eq. (2.19) can be solved for w(x,y).

A complete solution of the governing Eq. (2.13) depends upon the knowledge

of the conditions of the plate at the boundaries in terms of the lateral deflection of the

middle surface w(x,y). Thus, expressions for these conditions must be developed. In

general, there are four types of mathematically “exact” solutions available for plate

problems:

The rigorous solution of plate problems is essentially a boundary value

problem of mathematical physics. Since the fulfillment of the boundary conditions

usually presents considerable mathematical difficulties, in general, rigorous solutions of

plate problems are rare. The most general form of the rigorous solution of the governing

differential equation can be written as

w( x, y ) = wH ( x, y ) + wP ( x, y ) (2.23)

where wH represents the solution of the homogeneous equation 2 2

w = 0 , and wP is a

particular solution of the non-homogeneous differential equation of the plate. There are

13
few cases, however, when the solution can be obtained directly, without employing the

above mentioned superposition principle.

Certain boundary conditions permit the use of special solutions, such as the

Navier solution. In the Navier solution wH = 0; thus

w(x, y) = wP (x, y). (2.24)

2.3.1.1 Solution of the Homogeneous Equation.

The physical interpretations of the solution of the biharmonic equation

( 4
w = 0) is to obtain the deflection of the plate wH (x, y) when only edge forces are

acting. Consequently, the solution of the homogeneous equation fulfills the prescribed

boundary conditions and maintains the equilibrium with the external boundary forces.

The fundamental difficulty of the rigorous solution is the proper choice of functions

wH(x, y) for a given problem.

The biharmonic equation ( 4

w = 0) permits the use of the solutions of the

Laplace equation 2
w = 0 , which are

x, xy, cos x , cosh y , x2 – y2, and x3 – 3xy2 , (2.25)

Where , represents an arbitrary constant. In Eq. (2.25) we may interchange x

and y and replace cos by sin and cosh by sinh, respectively. If w1(x, y) and w2(x, y) are

solutions of the Laplace equation, then

w1 + xw2 , w1 + yw2, and w1 + (x2 + y2 )w2 (2.26)

are solutions of the biharmonic equation 2 2

w = 0.

14
 2.3.1.2 Particular Solution

Although the solution wH of the biharmonic equation effectively describes the

equilibrium conditions of the plate subjected to edge forces, the expression of the

deflection w(x, y) is not complete without also considering the equilibrium of the lateral

forces Pz. for this purpose, a particular solution wP of the nonhomogeneous differential

equation (2.13) must also be determined. We require from the particular solution that it

satisfy the differential equation of the plate (2.13), but the fulfillment of the boundary

conditions is not mandatory. In the case of an infinite series solution, however, a more

rapid convergence can be obtained if the particular solution at least fulfills the boundary

conditions of two opposite edges of the plate.

For rectangular plates with fixed, or partially fixed, boundary conditions at

certain or all edges, the expression of deflections of the simply supported plate can be

used as a particular solution. This can be obtained with relative ease using Navier’s

method. Finally, the possibility of expressing the true singularity of the problem, in the

case of a concentrated lateral load, must be mentioned briefly. If such a force acts at the

center of a rectangular plate, for instance, the particular solution can be written as

Pz x2 + y2
wP(x, y) = ( x + y ) ln
2 2
. (2.27)
16 D a2

15
 Fig. 2.2 Rectangular plate under concentrated load

These types of solutions coupled with Maxwell’s law of reciprocity are used for

obtaining influence surfaces for plates.

2.3.1.3 Fourier Series

The Fourier series are indispensable instruments in the analytical treatment of

many problems in the field of applied mechanics, such as the solution of partial

differential equations of the theory of elasticity, vibrations, flow of heat, transmission of

electricity, and electromagnetic waves. The extension of the Fourier series leads to the

Fourier integrals and to the Fourier transforms.

2.3.1.4 Single Fourier series

Fourier’s theorem states that any arbitrary function y = f(x) can be expressed by

an infinite series, consisting of sine and cosine terms. Thus the original function is

replaced by the superposition of numerous sine and cosine waves. If f(x) is a periodic

function, the Fourier’s theorem states that

16
 1 2 x 4 x 2n x
f(x) = A0 + A1 cos + A2 cos + ......... + An cos + ......... (2.28)
2 T T T
2 x 4 x 2n x
+ B1 sin + B2 sin + ............. + Bn sin + ........
T T T

or, in more concise form,

1
f(x) = A0 + An cos n x + Bn sin n x, (2.29)
2 1 1

where A0 , An , and Bn (n=1,2,3….) are the coefficients of Fourier expansion;

represents

2
= (2.30)
T

and T is the period of the function

Equation (2.29) is valid for any piecewise regular periodic function, which

might also have discontinuities, and represents the arbitrary periodic function f(x) in the

full range from x = to x = + ; thus it is called full-range expansion.

The coefficients A0 , An , and Bn are obtained from (2.28)(2.29)(2.30)

T
2 (2.31)
A0 = f ( x) dx
T 0
T
2 (2.32)
An = f ( x) cos n xdx (n = 1,2,3,……)
T 0

2 (2.33)
Bn = f ( x) sin n xdx,
T

17
 When the function f(x) is not given in analytical form or it is too complicated to

perform the integrations prescribed, then the so-called harmonic analysis, which

replaces the integrals by summations, can be advantageously utilized. Dividing the

period T into 2m equals intervals, the Fourier coefficients are determined from

1 2m 1
A0 = yk (2.34)
m k =0
1 2m 1 kn
An = y k cos (2.35)
m k =0 m
kn
1 2 m 1 y k sin , (2.36)
Bn = m
m k =0

(k=0, 1, 2… 2m and n = 1, 2, 3…m)

Another approximate method for evaluating the constant of the Fourier

expansion consists of plotting f ( x), f ( x) cos(2n x / T )andf ( x) sin(2n x / T ) curves and

determining the area of the respective curves by planimeter.

2.3.1.5 Double Fourier series

In the static and dynamic analysis of plates, as discussed later, a given function

f(x, y) is often expanded into sine series of two variables, x and y, using the following

expression:

m x n y
f ( x, y ) = Fmn sin sin . (2.37)
m =1 n =1 a b

The above equation represents a half range sine expansion in x, multiplied by a

half-range sine expansion in y, using for the period of expansion T=2a and T=2b,

18
respectively. To obtain the coefficient Fmn, we first multiply the equation by

sin(k y / b)dy and then integrate between the limits 0 and b. Thus, we can write,

b b
k y m x n y k y
f ( x, y ) sin dy = Fmn sin sin sin dy. (2.38)
0
b m =1 n =1 a 0 b b

if n k , then

b
n y k y
sin sin dy = 0. (2.39)
0
b b

if n= k, then

b
n y b
sin 2 dy = . (2.40)
0
b 2

Utilizing a similar approach for the variable x, we obtain

a
m x a
sin 2 dx = . (2.41)
0
a 2

Thus Eq. (2.37) becomes

a b
ab m x n y
Fmn = f ( x, y ) sin sin dxdy. (2.42)
22 0 0
a b

Hence the coefficient of the double Fourier expansion is

a b
4 m x n y
Fmn = f ( x, y ) sin sin dxdy. (2.43)
ab 0 0 a b

Three types of boundary are considered at this time: simply supported, clamped, and

mixed.

19
2.3.2 Boundary Conditions

A complete solution of the governing Eq. (2.13) depends upon the knowledge of the

conditions of the plate at the boundaries in terms of the lateral deflection of the middle

surface w(x, y). Thus, expressions for these conditions must be developed. Three types

of boundary are considered at this time: simply supported, clamped, and free.

2.3.2.1 Simply Supported Edge Conditions

A plate boundary that is prevented from deflecting but free to rotate about a

line along the boundary edge, such as a hinge, is defined as a simply supported edge.

The conditions on a simply supported edge paralled to the y-axis at x =a, are.

w | X =a = 0 (2.44)

2 2
w w
M x | X = a = D( + ) X =a = 0 . (2.45)
x2 y2

Since the change of w with respect to the y coordinate vanishes along this edge, these

conditions become

w | X =a = 0 (2.46)

2
w
| X =a = 0 . (2.47)
x2

On a simply supported edge parallel to the x-axis at y = b, the change of w with respect

to the x-coordinate vanishes; thus, the condition along this boundary are

w |Y = b = 0
(2.48)
2 2
w w
M y |Y =b = D( 2 + ) Y =b = 0
y x2 (2.49)

20
 2
w
= D |Y =b = 0 (2.50)
y2

2.3.2.2 Clamped Edge Conditions

If a plate is clamped, the deflection and the slope of the middle surface must

vanish at the boundary. On a clamped edge parallel to the y-axis at x=a, the boundary

conditions are

w | x =a = 0
(2.51)

w
| x =a = 0 (2.52)
x

The boundary conditions on a clamped edge parallel to the x- axis at y=b are

w | y =b = 0 (2.53)

w
| y =b = 0. (2.54)
y

2.3.2.3 Mixed Edge Conditions

If a plate is simply supported along two opposite sides and clamped on the other

two sides at y=0, and y=b. The boundary conditions are

w = 0, (2.55)

w
=0 at y=0, b (2.56)
y

also, on the simply supported edge parallel to the y-axis , the boundary

conditions at x=0, and x=a are

w=0, (2.57)

21
 2 2
w w
and M | x = a = D( + ) x = a = 0. (2.58)
x2 y2

According to Kirchoff, only two boundary conditions are sufficient for the

complete determination of the deflection w satisfying the governing equation of the

plate and the three conditions derived from physical reasoning are too many. This

inconsistency is due to the assumption that the normals of the middle plane before

bending are deformed into the normals of the middle plane after bending. Without using

an assumption, a sixth-order differential equation can be obtained for which all the three

boundary conditions can be satisfied. It can be shown, however, that, except in the

immediate region of the boundary, the stress distribution given by this new equation is

substantially the same as that given by the governing equation. If this plate is thin, the

sixth-order terms can be neglected and this new equation reduces to governing

equation. This justifies the use of governing equation in the study of the bending of thin

plates.

2.4 Galerkin method

The Galerkin method is a member of the methods of weighted residuals

(MWR). This concept of weighted residuals was introduced by Crandall [16].

Consider a differential equation that can be represented in the following form,

L (T) = f, (2.59)

in a domain D(x, y), with boundary conditions T = 0 on D , the boundary of D. Where

L is a differential operator in the Sturm-Liouville system defined as

Lf = (k (x)f’)’. (2.60)

22
The Galerkin method assumes that T can be approximately represented as

N
Ta ( x, y ) = aj j ( x, y ), (2.61)
j =1

Where,

j = Known analytic functions (trial functions),

a j = Coefficients yet to be determined,

If we have a partial differential equation with a non-homogeneous boundary condition,

then we can rewrite eq. (2.16) as

N
Ta ( x, y ) = T0 ( x, y ) + aj j ( x, y ), (2.62)
j =1

where T0 was introduced to satisfy the inhomogeneous boundary conditions.

The residual, R, is defined as,

R(a 0 , a1 ,............, a n , x, y ) = L(u a ) f

N
= L(u 0 ) + a j L( j ) f. (2.63)
j =1

if an inner product,(f,g) , between two functions, f(x, y) and g(x, y), can be defined as

( f , g) = fgdxdy (2.64)
D

then the unknown coefficients, aj ,are obtained by solving the following systems of

equations,

( R, k )=0 k=1,……,N (2.65)

where k ' s are the same trial functions. The above equation can be written as a matrix

equation for the coefficient aj as,

23
 N
a j ( L( j ), k ) = ( L(u 0 ), k ) (2.66)
j =1

So, by solving the above equation, we can find out aj. we can substitute, aj in eq (2.15)

to obtain Ta(x, y).

The trial functions, j ( x) s , should be chosen from a complete set of polynomials of the

Nth order. It is necessary condition for convergence to the exact solution as N .

The conditions required in applying the traditional Galerkin method are as

follows:

(1) The functions k , are chosen from the same family as the trial functions j.

(2) The trial functions must be linearly independent.

(3) The trial functions should satisfy the homogeneous boundary conditions exactly.

The accuracy of the Galerkin method is influenced by the order of polynomials

and the choice of trial functions. The efficiency of the Galerkin method can be defined

in terms of the solution accuracy per unit of computer execution time.

There exists a close connection between the Rayleigh-Ritz method and the

Galerkin method in connection with the finite-element method. The importance of the

equivalence of the Galerkin method and the Rayleigh-Ritz method is that the

convergence of the Rayleigh-Ritz solution to the exact solution as N, the number of

terms, approaches to infinity is well established as long as the trial functions are

members of a complete set of functions. The convergence properties associated with the

Rayleigh-Ritz method carry over to the Galerkin method and the equivalence is also

exploited in obtaining error estimates.

24
 CHAPTER 3

METHOD OF WEIGHTED RESIDUALS

3.1 Introduction

Prior to the development of the finite element method, there existed an

approximation technique for solving differential equations called the method of

weighted residuals (MWR).The basic idea of MWR is to use a trial function with a

number of unknown parameters to approximate the solution. Then a weighted average

over the interior and boundary is set to zero. The idea is to approximate the solution

with a polynomial involving a set of parameters. The polynomial is made to satisfy both

the differential equation and the associated boundary conditions.

The main objective is to solve general linear equations in the form

Lu=c (3.1)

Where L is a linear operator (differential operators, matrices etc.), u is the unknown

function and c is a given function.

An approximate solution to eq. (3.1) is sought by a linear combination of N base

vectors in the linear space as

~ N
U= u i ei , (3.2)
i =1

where ui is the unknown coefficient and ei is the base vector in the linear space.

25
The residual (error), R, between the approximate solution and the exact solution is

defined as

RVL9–c (3.3)

N
=L u i ei c (3.4)
i =1

N
= u i Lei ( x) c( x) (3.5)
i =1

Note that for a function space, R is a function of the position, i.e.

N
R(x) = W ui L ei(x) – c(x). (3.6)
i=1

The Method of Weighted Residuals is broadly classified into the following methods

3.1.1 Collocation method

Choose ui so that the residual (error) vanishes at N selected points, i.e.

R (xi) = 0, i = 1,……N (3.7)

Although this method gives the exact values at the selected points, there is no guarantee

that the approximation behaves nicely between the selected points.

3.1.2 Least Square method

Choose ui so that the magnitude of residual (error) becomes the minimum, i.e.

|| R(x) || X min. (3.8)

This method is expected to give an overall well-behaved approximation.

26
3.1.3 Galerkin method

Choose ui so that R is orthogonal to N base functions (ei), i.e.

(R, ei) = 0 i = 1, ….., N . (3.9)

The idea of the Galerkin’s method is that if ei’s span the entire linear space, a vector that

is perpendicular to all the base vectors must be a zero vector.

The Galerkin method has been used to solve problems in mechanical

engineering such as structural mechanics, dynamics, fluid flow, heat and mass transfer,

acoustics and other related fields. The Galerkin method can be used to approximate the

solution to ordinary differential equations, partial differential equations and integral

equations.

3.2 Application of Galerkin method

Initially the finite-element method was used for constructing matrix solutions to

stress and displacement calculations in structural analysis. The finite-element method is

a special case of the Galerkin method in which the base functions are chosen such that

each base function becomes 1 at the corresponding nodes but otherwise 0 at other

nodes. The link with the Galerkin method permitted finite-element techniques to be

extended into areas such as fluid mechanics and heat transfer. At the present time, the

Galerkin finite element formulation is the most popular finite-element method.

27
Consider a rectangular plate as shown in figure with dimensions of a and b.

Fig 3.1 Rectangular plate

The governing differential equation for a plate can be written as:

4 4 4
w w w p z ( x, y ) 2 w
+ 2 + = (3.10)
x4 x2 y2 y4 D x2

Where pz is the external load acting on the plate surface and D is the bending or flexural

rigidity of the plate. The exact solution of the governing plate equation (3.10) must

simultaneously satisfy the differential equation and the boundary conditions of any

given plate problem. Since Eq (3.10) is a fourth-order differential equation, two

boundary conditions, either for the displacements or for the internal forces, are required

at each boundary.

The displacement components to be used in formulating the boundary

conditions are lateral deflections and slope. At fixed edges, for instance, the deflection

and the slope of the deflected plate surface are zero.

28
 w
( w) x = 0, ( )x = 0 (x=0 or x=a) (3.11)
x

w
( w) y = 0, ( )y = 0 (y=0 or y=b) (3.12)
y

Equation (3.10) can be rewritten using the two-dimensional Laplacian operator

2 2
( 2
= + ) as :
x2 y2

2
pz w
2 2
w= (3.13)
D x2

Consider a differential operator L, defined as

4 4 4
L= + + .
x4 x2 y2 y4

So, eq.(3.10) becomes,

pz
Lw = (3.14)
D

We can express the solution to the above equation in terms of the eigenfunction and

eigenvalues, which are defined as:

Lenm ( x, y ) = nm Lenm ( x, y ) (3.15)

Where enm(x, y) are the eigenfunctions and nm are the corresponding eigenvalues. Once

the eigenfunctions and eigenvalues are known, it is possible to express w(x, y) as,

w( x, y ) = wnm e nm
enm ( x, y ) (3.16)
n =1 m =1

Where, wnm are unknown coefficients

29
The eigenfunctions, enm(x, y), are expressed as a linear combination of base functions,

fi(x, y), as

N
nm
enm ( x, y ) = ci f i ( x, y ) (3.17)
i

We have to select the base functions, fi(x, y), which satisfy the boundary conditions. So

substituting Eq. (3.17) in Eq. (3.15) gives,

N N
nm nm
ci Lf i ( x, y ) = nm i c Lf i ( x, y ). (3.18)
i i

now, multiplying Eq.(3.18) with another base function fj(x, y) which also satisfies the

boundary conditions as

N N N N
nm nm
ci Lf i ( x, y ) f j ( x, y ) = c
nm i Lf i ( x, y ) f j ( x, y ) (3.19)
i j i j

We can write the above equation in the following form,

Ac = nm B c, (3.20)

Where

a b
a ij = LLf i ( x, y ) f j ( x, y )dydx
0 0
a b
| (3.21)
bij = Lf i ( x, y ) f j ( x, y )dydx
0 0

The quantities A and B are N x N square matrices as shown below:

a11 ...............a1N b11 ...............b1N

......................... .........................
A= B=
......................... .........................
a NN ................a NN bNN ...................bNN

30
Where nm , is the eigenvalue and c will be the corresponding eigenvectors.

For calculating the eigenvalues and corresponding eigenvectors, we used the

Cholesky decomposition.

Applying the Cholesky decomposition on matrix B as

B = UT U, (3.22)

where, U is the upper triangular matrix and UT is the transpose of matrix U.

So Eq. (3.20) becomes,

Ac = U TU c .
nm (3.23)

Substituting,

Uc=x

c =U 1
x

Eq. (3.22) then becomes,

AU 1
x= UT x
nm (3.24)

1 1
U T AU 1
x= UT UT x
nm (3.25)

Finally Eq. (3.25) becomes,

x= nm x,

Where,
1
= U T AU 1 . (3.26)

31
 We can calculate the eigenvalues of nm and corresponding eigenvectors c

with the help of symbolic algebra software Mathematica. The matrix B is positive

definite and symmetric. Cholesky decomposition is a faster and stable method than any

other alternative methods for solving linear equations.

We can also calculate the values of unknown coefficients wnm used in the

calculation of approximate solution as follows:

a b
wnm = f ( x, y )enm ( x, y )dydx (3.27)
0 0

Where f(x, y), is the initial condition.

So substituting the calculated values in Eq. (3.16) we can get the lateral deflection of

the rectangular plate.

It was mentioned in Chapter 2 that two types of boundary conditions are being

considered here.

3.2.1 Simply Supported Boundary Condition

To study each of the boundary conditions stated before, we need to begin with

the governing differential equation of the plate. At first, considering the simply

supported boundary condition for a square plate subjected to lateral loads.

The governing differential equation of the plate subjected to lateral loads is:

4
w 2 4w 4
w p z ( x, y ) 2 w
+ + = (3.28)
x4 x2 y2 y4 D x2

32
 where,

pz is the lateral load that is being applied

D is the bending or flexural rigidity of the plate

A plate boundary that is prevented from deflecting but free to rotate about a line along

the boundary edge, such as a hinge, is defined as a simply supported edge. The

conditions on a simply supported edge paralled to the y-axis at x =a, are.

w | X =a = 0 (3.29)

2 2
w w
M x | X = a = D( + ) X =a = 0 (3.30)
x2 y2

Since the change of w with respect to the y coordinate vanishes along this edge, these

conditions become

w | X =a = 0 (3.31)

2
w
| X =a = 0 . (3.32)
x2

On a simply supported edge parallel to the x-axis at y =a, the change of w with respect

to the x-coordinate vanishes; thus, the conditions along this boundary are

w |Y = a = 0 (3.33)
2 2
w w (3.34)
M y |Y = a = D( + ) Y =a = 0
y2 x2

By assuming a solution in the form

w( x, y ) = c i ei ( x , y ) (3.35)
i

33
Eq. (3.28) can be solved using the standard Galerkin method. Substituting Eq. (3.35)

into Eq. (3.28) and integrating over the entire plate produces the following eigenvalue

problem.

[A]c = [ B] c (3.36)

by introducing the notation

= , (3.37)
x y

also, = . (3.38)

Using the Galerkin method, the elements of [A] and [B] matrices are represented as:

b a
aij = ei e j dxdy (3.39)
0 0

b a
bij = ei e j dxdy (3.40)
0 0

These expressions can be simplified by integration by parts, thus distributing the

operator over both ei and ej.

b a
a ij = ei e j dxdy (3.41)
0 0

b a
= ( . ) . ei e j dxdy (3.42)
0 0

When applying surface integral over the entire surface of the plate and homogeneous

boundary conditions aij becomes

34
 b a
aij = ( . )ei e j dxdy (3.43)
0 0

Using another homogeneous boundary condition gives the expression:

b a
aij = ( . )ei ( . )e j dxdy (3.44)
0 0

By substituting equation (3.37) into equation (3.44), the final expression for members aij

is revealed as.

2 2
b a 2
ei 2
e ej ej
aij = 2
+ 2i . + dxdy (3.45)
0 0 x y x2 y2

Deriving the expression for bij in the similar way by starting with equation (3.40) and

using integration by parts, we get.

b a
bij = ei e j dxdy (3.46)
0 0

The first step in solving this problem is to systematically choose a trial function e that

satisfies the plate’s boundary conditions. Polynomial approximating functions will be

used to represent the lateral displacement of the plate. In this discussion, the trial

function ei (x, y) will be represented as:

N N

i ( x, y ) = a[i, j ]u j ( x, y ) (3.47)
i =1 j =1

Where

L Mj
u j ( x, y ) = x j . y (3.48)

and, Lj and Mj are positive integers and i are coefficients to be determined.

35
 In the simply supported boundary condition example, it is found that an eight

order polynomial is the lowest order possible to satisfy the boundary conditions.

i = a[1,1] + a[2,1]x + a[3,1] y + a[4,1] xy

+ a[5,1]x 2 + a[6,1] y 2 + a[7,1]xy 2 + .....

(3.49)
+ a[42,1]x 3 y 5 + a[43,1]x 2 y 6 + a[44,1]xy 7 + a[45,1] y 8

where a[1,1] through a[45,1] are unknown coefficients of i.

Once a general Nth order polynomial is defined as mentioned in Eq.(3.49), the

next step is to apply the boundary conditions and solve for the unknown coefficients.

This is carried out by developing eight new equations using Mathematica.

Each equation represents a different boundary condition.

For Deflection

BC (1): (0, y ) = 0 (3.50)

BC (2): ( a, y ) = 0 (3.51)

BC (3): ( x,0) = 0 (3.52)

BC (4): ( x, b ) = 0 (3.53)

For Moments

2 2
BC (5): + =0 (3.54)
x2 y2 x =0
2 2
BC (6): + =0 (3.55)
x2 y2 x=a
2 2
BC (7): + =0 (3.56)
x2 y2 y =0

36
 2 2
BC (8): + =0 (3.57)
x2 y2 y =b

This results in an undetermined system of one hundred forty five equations with forty-

five unknowns. Flattening out all the nested eight equations above using the Flatten

[Table]. command in Mathematica. Tabulating all the eight equations and solving, we

get values of coefficients a [i,j].

Substituting these values of the coefficients back in the generalized polynomial function

we end up with an eighth order polynomial (3.49), and in this case, one independent

equation results in the following polynomial.

1 ( x, y ) = {x(1 2 x 2 + x 3 ) y (1 2 y 2 + y 3 )} (3.58)

Using Mathematica to evaluate the elements of matrices [A] and [B] wherein

integration operations are carried over the polynomial. The eigenvalue is evaluated

using the Eigenvalue[] command in Mathematica, which matches with the value found

from Eq. (3.36).

The general eighth order polynomial that is generated from the generalized

polynomial Eq. [3.49] is as follows:

1 ( x, y ) = a[1] + xa[2] + ya[3] + x 2 a[4] + xya[5] + y 2 a[6] + (3.59)

x 3 a[7] + x 2 ya[8] + xy 2 a[9] + y 3 a[10] + x 4 a[11] +

x 3 ya[12] + x 2 y 2 a[13] + xy 3 a[14] + y 4 a[11] + x 3 ya[12]

+ x 2 y 2 a[13] + xy 3 a[14] + y 4 a[15] + ................. + x 3 y 3 a[25]

+ x 2 y 4 a[26] + xy 5 a[27] + y 6 a[28] + ................. + x 3 y 5 a[42]

+ x 2 y 6 a[43] + xy 7 a[44] + y 8 a[45].

37
The coefficients a[i,j] which are obtained by solving the equations of boundary

conditions are as follows :

Table 3.1 – Coefficients of 8th order polynomial for simply supported

plate with aspect ratio 1.0
a[5,1] a[12,1] a[14,1] a[17,1] a[20,1] a[25,1] a[32,1] a[33,1]

1.00 -2.00 -2.00 1.00 1.00 4.00 -2.00 -2.00

Once the eigenvalue problem is solved using the generalized method, the entire

procedure is carried out by calculating the eigenvalues and corresponding eigenvectors

using the Cholesky decomposition. Wherein the Cholesky decomposition is applied to

matrix [B] as

B =UT U,

where, U is the upper triangular matix and UT is the transpose of U.

The result obtained using the Cholesky decomposition was very accurate and

compatible with the direct eigenvalue method.

In order to improve on the percent error and get good convergence, a higher

order approximating polynomial must be considered. A ninth order polynomial is

formed by adding terms to equation (3.49).

i = a[1,1] + a[2,1]x + a[3,1] y + a[4,1] xy

+ a[5,1]x 2 + a[6,1] y 2 + a[7,1]xy 2 + .....

+ a[42,1]x 3 y 5 + a[43,1]x 2 y 6 + a[44,1]xy 7 + a[45,1] y 8
(3.60)
+ a[46]x 9 + a[47]x 8 y + a[48]x 7 y 2 + a[49]x 6 y 3 + .....
+ a[52]x 3 y 6 + a[53]x 2 y 7 + a[54]xy 8 + a[55] y 9

38
 In order to evaluate the result for a ninth order polynomial, it is necessary to

change the eight order polynomial to a ninth order polynomial. By applying the

boundary conditions to this new approximating function and solving for all the

coefficients of the ninth order using the Mathematica program, we get a system of three

trial functions 1 , 2 and 3 .

The corresponding coefficients to these three trial functions are:

Table 3.2 – Coefficients of 9th order polynomial for simply supported

plate with aspect ratio 1.0
a[5,1] a[12,1] a[14,1] a[17,1] a[20,1] a[25,1] a[32,1] a[33,1]
1.00 -2.00 -2.00 1.00 1.00 4.00 -2.00 -2.00

a[5,2] a[12,2] a[14,2] a[20,2] a[23,2] a[25,2] a[33,2] a[40,2]

2.33 -3.33 -4.66 2.33 1.00 6.66 -3.33 -2.00

a[5,3] a[12,3] a[14,3] a[17,1] a[25,1] a[27,1] a[32,3] a[42,3]

2.33 -4.66 -3.33 2.33 6.66 1.00 -3.33 -2.00

Substituting these values into Eq. (3.60) we get the following trial functions:

1 ( x, y ) = {x(1 2 x 2 + x 3 ) y (1 2 y 2 + y 3 )} (3.61)

x
2 ( x, y ) = { (7 10 x 2 + 3 x 4 ) y (1 2 y 2 + y 3 )} (3.62)
3

x
3 ( x, y ) = { (1 2 x 2 + x 3 ) y (7 10 y 2 + 3 y 4 )} (3.63)
3

While the first order eigenvalue system was solved by simply multiplying the

single element matrix [A] by the single element inverse matrix [B], this third order

system is solved using the Mathematica eigenvalue function. The elements of matrices

39
[A] and [B] found by the definite integral command in Mathematica are then operated

upon using the eigenvalue function in Mathematica.

The convergence of the critical buckling load from as overestimated value

toward the exact values is a property of the Galerkin method. Such a property occurs

when the order of the approximating function is increased. It was found that when the

order of the approximating function was increased, the lower order polynomial was

kept. In the current example 1 is the same for the eighth and ninth order polynomials.

If this did not occur, the Galerkin system would not be complete and the calculated

value would be incorrect.

3.2.2 Mixed Boundary Conditions

In the mixed boundary conditions, we consider two opposite edges to

be simply supported and the other two edges of the plate to be fixed. If a plate is simply

supported along two opposite sides and clamped on the other two sides at y=0, and

y=b. The boundary conditions are

w = 0, (3.64)

w
=0 at y=0, b (3.65)
y

Also, on the simply supported edge parallel to the y-axis, the boundary

conditions at x=0, and x=a are

w=0, (3.66)

2 2
w w
and M | x = a = D( + ) x = a = 0. (3.67)
x2 y2

40
 In order to find the buckling load of a plate subjected to mixed

boundary conditions as mentioned above, we need to begin with the governing

differential equation of a plate i.e. Eq. (3.28).

We start with assuming the solution to be in the form:

w( x, y ) = c i ei ( x , y )
i

Using the standard Galerkin method to solve the governing differential Eq.

(3.28) and then generating an eigenvalue problem by integrating over the entire plate.

The elements of matrices [A] and [B] are evaluated in the same way as shown for the

case of simply supported boundary condition.

Once the expression for the elements of matrices [A] and [B] are obtained, the

next step is to choose a trial function that satisfies the plate’s boundary conditions. As

before, the trial function is represented as in Eq. (3.47)

N N

i ( x, y ) = a[i, j ]u j ( x, y )
i =1 j =1

L Mj
Where u j ( x, y ) = x j . y

For the Mixed boundary condition example, it is found that a sixth order

polynomial is the lowest order possible to satisfy the boundary conditions.

i = a[1,1] + a[2,1]x + a[3,1] y + a[4,1] x 2

+ a[5,1]xy + a[6,1] y 2 + a[7,1]x 3 + .....

+ a[14,1]xy 3 + a[15,1] y 4 + a[16,1]x 5 + ....... (3.68)
+ a[25,1]x 3 y 3 + a[26,1]x 2 y 4 + a[27,1]xy 5 + a[28,1] y 6 .

Where a[1,1] to a [28, 1] are unknown coefficients of i .

41
 Once a general Nth order polynomial is defined as mentioned in equation

(3.68), the next step is to apply the boundary conditions and solve for the unknown

coefficients. This is carried out by developing eight new equations using Mathematica.

Each equation represents a different boundary condition.

For Deflection

BC (1): (0, y ) = 0 (3.69)

BC (2): ( a, y ) = 0 (3.70)

BC (3): ( x,0) = 0 (3.71)

BC (4): ( x, b ) = 0 (3.72)

For Moments

2 2
BC (5): + =0 (3.73)
x2 y2 x =0

2 2
BC (6): + =0 (3.74)
x2 y2 x=a

For Slope

BC (7): =0 (3.75)
x y =0

BC (8): =0 (3.76)
x y =a

Having solved all the eight boundary condition equations and tabulating them

together using the Table[] command in Mathematica, we get values of coefficients

a[i,j]. Substituting these values of the coefficients back into the generalized polynomial

42
function we end up at a sixth order polynomial which results in one independent

equation in the following polynomial.

1 = x(1 2 x 2 + x 3 )( 1 + y ) y (3.77)

Using Mathematica, we evaluate the elements of matrices [A] and [B]. The

eigenvalue is evaluated using the EigenValue[] command in Mathematica, which

matches with the value found from Eq. (3.36).

The general sixth order polynomial that is generated from the generalized

polynomial Eq. (3.68) is as follows:

1 ( x, y ) = a[1,1] + a[2,1]x + a[3,1] y + a[4,1] x 2 +

+ a[5,1]xy + a[6,1] y 2 + a[7,1] x 3 + a[8,1] x 2 y + .........

+ a[14,1] xy 3 + a[15,1] y 4 + a[16,1]x 5 + a[17,1]x 4 y + ...... (3.78)
+ a[ 25,1] x y + a[26,1] x y + a[ 27,1] xy + a[ 28,1] y .
3 3 2 4 5 6

The coefficients a [i,j] which are obtained by solving the equations of boundary

conditions are as follows :

Table 3.3 – Coefficients of 6th order polynomial for isotropic plate with mixed
boundary conditions.
a[5,1] a[9,1] a[12,1] a[17,1] a[18,1]
-1.00 1.00 2.00 -1.00 -2.00

Once again, following the same procedure as in simply supported boundary

condition, the eigenvalue problem is solved using the generalized method, the entire

procedure is carried out by calculating the eigenvalues and corresponding eigenvectors

using the Cholesky decomposition. Wherein, the Cholesky decomposition is applied to

matrix [B].

43
 As before, the result obtained using the Cholesky decomposition was pretty

accurate and compatible with the direct eigenvalue method. In order to improve on the

percent error and get good convergence, a higher order approximating polynomial must

be considered. A seventh order polynomial is formed by adding terms to Eq. (3.78).

1 ( x, y ) = a[1,1] + a[2,1]x + a[3,1] y + a[4,1]x 2 + a[5,1]xy + ......

+ a[11,1]x 4 + a[12,1]x 3 y + a[13,1]x 2 y 2 + ........

+ a[18,1]x 3 y 2 + a[19,1]x 2 y 3 + a[20,1]xy 4 + ..
(3.79)
+ a[25,1]x 3 y 3 + a[26,1]x 2 y 4 + a[27,1]xy 5 + .....
+ a[34,1]x 2 y 5 + a[35,1]xy 6 + a[36,1] y 7 .

In order to evaluate the result for a seventh order polynomial, it is necessary to

change the sixth order polynomial to a seventh order polynomial. By applying the

boundary conditions to this new approximating function and solving for all the

coefficients of the seventh order using the Mathematica program, we get a system of

three trial functions 1 , 2 and 3 .

The corresponding coefficients to these three trial functions are:

Table 3.4 – Coefficients of 7th order polynomial for isotropic

plate with mixed boundary conditions.
a[5,1] a[9,1] a[12,1] a[17,1] a[18,1]
-1.00 1.00 2.00 -1.00 -2.00

a[5,1] a[9,1] a[12,1] a[18,1] a[23,1]

-2.33 2.33 3.33 -3.33 -1.00

a[5,1] a[12,1] a[14,1] a[17,1] a[25,1]

-1.00 2.00 1.00 -1.00 -2.00

44
 Substituting these values into Eq. (3.79) we get the following trial functions:

1 ( x, y ) = {x(1 2 x 2 + x 3 )( 1 + y ) y} (3.80)

x
2 ( x, y ) = { (7 10 x 2 + 3 x 4 )( 1 + y ) y} (3.81)
3

3 ( x, y ) = {x(1 2 x 2 + x 3 ) y ( 1 + y 2 )} (3.82)

As incase of simply supported boundary condition, While the first order

eigenvalue system was solved by simply multiplying the single element matrix [A] by

the single element inverse matrix [B], this third order system is solved using the

Mathematica eigenvalue function. The elements of matrices [A] and [B] found by the

definite integral command in Mathematica are then operated upon using the eigenvalue

function in Mathematica.

This convergence of the critical buckling load from an over estimated value

toward the exact value is a property of the Galerkin method. Such a property occurs

when the order of the approximating functions is increased.

45
 CHAPTER 4

NUMERICAL RESULTS

This chapter presents the numerical results achieved for the problems

discussed in chapter 3. All of the computation was carried out with the help of a

symbolic algebra software system, Mathematica [11].

4.1 Simply supported boundary condition

In our example, we analyze the buckling load of a square plate for the given

simply supported boundary condition. In this type of boundary condition wherein all

sides are simply supported, the plate is prevented from deflecting but free to rotate

about a line along the boundary edge, such as a hinge.

The conditions on a simply supported edge paralled to the y-axis at x =a, are.

w | X =a = 0 (4.1)

2 2
w w
M x | X = a = D( 2 + ) X =a = 0 (4.2)
x y2

On a simply supported edge parallel to the x-axis at y =a, the change of w with respect

to the x-coordinate vanishes; thus, the condition along this boundary are

w |Y = a = 0 (4.3)
2 2
w w
M y |Y = a = D( 2 + ) Y =a = 0
y x2 (4.4)

46
 Fig 4.1 Simply supported plate

Once the boundary conditions have been applied, the next step is to

choose the suitable trial functions which satisfy the boundary conditions completely.

For the current boundary conditions, the trial function can be chosen as

N N

i ( x, y ) = a[i, j ]u j ( x, y ) (4.5)
i =1 j =1

L Mj
where , u j ( x, y ) = x j . y (4.6)

In the simply supported boundary condition example, it is found that an eight

order polynomial is the lowest order possible to satisfy the boundary conditions.

i = a[1,1] + a[2,1]x + a[3,1] y + a[4,1] xy

+ a[5,1]x 2 + a[6,1] y 2 + a[7,1]xy 2 + .....

(4.7)
+ a[42,1]x 3 y 5 + a[43,1]x 2 y 6 + a[44,1]xy 7 + a[45,1] y 8

Where a[1,1] through a[45,1] are unknown coefficients of 1.

47
 Table 4.1 Comparison of eigenvalues for simply supported plate
Order Direct Approach Cholesky’s Decomposition
8 389.969 389.969
9 389.69, 2463.33 389.969,401401.4
10 389.636, 2463.33, 6304.32 389.969, 401401.4, 6.34*10^7
11 389.637, 2435.33, 6304.32 389.969, 401401.4
12 389.63, 6234.5 389.96, 20443.4, 401401.4
13 383.26, 2435.07, 6234.56 389.969, 401401.39
14 831.41, 2373.52, 6388.73, 389.969, 401401.39
complex
15 670.524,1147.44, 4740.78, 389.969, 401401.39
complex

Once the eigenvalue problem is solved using the generalized method, the entire

procedure is carried out by calculating the eigenvalues and corresponding eigenvectors

using the Cholesky decomposition. Wherein the Cholesky decomposition is applied to

matrix [B] as

B =UT U, (4.8)

Where, U is the upper triangular matix and UT is the transpose of U.

The result obtained using the Cholesky decomposition was very accurate and

compatible with the direct eigenvalue method.

It was seen that the values obtained for the eigenvalue for the least order of the

polynomial matched with that of a simple one-dimensional problem. The published

values for higher orders were not to be found and so it was not possible for any

comparison to be done for them.

48
 4.2 Mixed boundary condition

In example 2, we are going to find the buckling load of a square plate

with mixed boundary conditions. we consider two opposite edges to be simply

supported and the other two edges of the plate to be fixed. If a plate is simply supported

along two opposite sides and clamped on the other two sides at y=0, and y=b. The

boundary conditions are

Fig 4.2 Plate with mixed boundary conditions

w = 0, (4.9)

w
=0 at y=0, b (4.10)
y

Also, on the simply supported edge parallel to the y-axis, the boundary

conditions at x=0, and x=a are

w=0, (4.11)

2 2
w w
and M | x = a = D( + ) x = a = 0. (4.12)
x2 y2

49
 For the Mixed boundary condition example, it is found that a sixth

order polynomial is the lowest order possible to satisfy the boundary conditions.

i = a[1,1] + a[2,1]x + a[3,1] y + a[4,1] x 2

+ a[5,1]xy + a[6,1] y 2 + a[7,1]x 3 + .....

+ a[14,1]xy 3 + a[15,1] y 4 + a[16,1]x 5 + ....... (4.13)
+ a[25,1]x 3 y 3 + a[26,1]x 2 y 4 + a[27,1]xy 5 + a[28,1] y 6 .

Where a[1,1] to a [28, 1] are unknown coefficients of i .

Table 4.2 Comparison of eigenvalues for plate with mixed boundary condition
Order Direct Approach Cholesky’s Decomposition
6 294.96 294.96
7 294.96,926.71,2376.34 294.96,19509.7
8 926.71,2376.34,4910.4,cmplx 294.96,19509.7,304128.4
9 2646.6,4910.4,complex 294.96,19509.7,304128.4
10 236.23,364.25,2464.6,complex 294.96,19509.7,304128.4
11 134.067,236.48,364.43,complex 294.96,19509.7
12 139.19,1233.01,2441.74,complex 294.96,19509.7,304128.4
13 5317.25,10442.9,complex 294.96,19509.7
14 6615.23, 11432.9,complex 294.96,19509.7,304128.4

15 88.491,440.52,1192.64,complex 294.96,19509.7,304128.4

Once a general Nth order polynomial is defined as mentioned in equation

(4.13), the next step is to apply the boundary conditions and solve for the unknown

coefficients. Once again following on the lines of example 1 and using both eigenvalue

function and Cholesky decomposition we see that the results obtained were accurate and

compatible with each other.

50
 4.3 Weinstein’s Theory

Weinstein’s theory is related to the stability of a square plate with four clamped

edges under uniform thrust in all directions in its plane. Both Weinstein and Trefftz [19]

have obtained approximate solutions by the Rayleigh method and have shown how

lower limits to the buckling load may be obtained by considering the cases in which the

edge conditions are less stringent. Using the exact method, they considered the stability

of the rectangular plate loaded by compressive forces distributed over two opposite

edges and varying linearly from zero at one corner.

According to Weinstein,

Consider the class of all even functions w(x, y) which have continuous

derivatives of the fourth order in the square S : x ! ,y ! ,

2 2

and which satisfy the conditions

(4.14)
w=0
w (4.15)
| x =0 = 0.
x

On the boundary, C ( x = ,y = ) of the square plate, it is required to find the

2 2

least value of the expression

I ( w)
,
D( w)

where I(w) and D(w) denote the following integrals taken over the square plate S:

2 2
I ( w) = ( w xx + w yy ) 2 dxdy; D( w) = ( w x + w y )dxdy.
S S

51
also, w = wxx + w yy ; (grad w)2 = wx2 + wy2

The function w which renders (I / D) a minimum satisfies the differential equation

w+ w = 0, (4.16)

with (4.14) and (4.15) as boundary condtions.

Solving the equation (4.16) they obtained the inequality < 5.33….., this

number being the value of ( I/ D) for the function

w* = cos2 x cos2 y (4.17)

Taking a set of functions which, like w*, satisfy the boundary conditions (4.14) and

(4.15), Weinstein could obtain in a similar way by the Ritz method a non-increasing

sequence of upper limits for the true value of . According to Weinstein, such a

sequence would always converge but not necessarily to , except when the chosen set is

complete. For instance, Weinstein computed the value of (I / D) for

w** = a cos2 x cos2 y + b cos3 x cos3 y, (4.18)

it was found that, with a suitable choice of a:b, the inequality < 5.31173.

Weinstein has obtained solutions both for the case when the two edges parallel

to the load are simply supported and for the case when they are clamped. In both the

cases, the loaded edges were always simply supported.

When the Mathematica program was modified according to the parameters in

Weinstein’s theory it was seen that the results were very accurate and matched with the

results obtained by Weinstein’s theory. The minimum order of polynomial for the given

set of boundary conditions came out to be eighth.

52
 i = a[1,1] + a[2,1]x + a[3,1] y + a[4,1] xy

+ a[5,1]x 2 + a[6,1] y 2 + a[7,1]xy 2 + .....

(4.19)
+ a[42,1]x 3 y 5 + a[43,1]x 2 y 6 + a[44,1]xy 7 + a[45,1] y 8

Where a[1,1] through a[45,1] are unknown coefficients of 1.

Applying the set of boundary conditions wherein slope and deflection on each

side is zero.

w | x =0 ,a = 0 w | y = 0, a = 0
w w (4.20)
| x =0 ,a = 0 | y = 0, a = 0
x y

Using the set of equations for the boundary condition, we can evaluate the

coefficients of the polynomial. Solving these equations we get a single trial function for

the eighth order polynomial. Applying the Laplacian operator L over the entire area of

the plate, we evaluate the elements of matrices [A] and [B]. Then applying the Cholesky

decomposition on matrix [B], the eigenvalues that are calculated come out to be

= 5.4713, which is very close to the desired value of = 5.33.

To get better convergence of the results, a higher order polynomial was

considered and Cholesky decomposition was applied again to achieve more stable

results.

53
 4.4 Rectangular plate axially compressed in one direction

Consider a simply supported rectangular plate subjected to a uniform axially

compressive force Nx per unit length along the edges x = 0, a. putting Ny and Nxy equal

to zero, we get the governing equation as

4 4 4 2
w w w N w
4
+2 2 2 + 4 + x = 0. (4.21)
x x y y D x2

In terms of the non-dimensional parameters " and# , defined as

x = a " , y = b# , p = a/b (4.22)

rewriting Eq. (4.21) as

4 4 4 2
w w w w
+ 2 p2 + p4 + 2
= 0. (4.23)
" 4
" #
2 2
# 4
"2

Since all the edges are simply supported, the boundary conditions are given by

The conditions on a simply supported edge paralled to the y-axis at x =a, are.

w | X =a = 0 (4.24)

2 2
w w
M x | X = a = D( + ) X =a = 0 (4.25)
x2 y2

On a simply supported edge parallel to the x-axis at y =a,

w |Y = a = 0 (4.26)
2 2
w w (4.27)
M y |Y = a = D( + ) Y =a = 0
y2 x2

In a non-dimensional form, these conditions can be written as

2 2
w w
w= + p2 =0 ( " = 0, 1), (4.28)
" 2
#2

54
 2 2
w w
w= + p2 =0 (# = 0, 1) (4.29)
" 2
#2

Since w = 0 along " = 0, 1 and # = 0, 1, we have

2
w
=0 ( # = 0, 1),
"2

2
w
=0 ( " = 0, 1)
#2

Making use of these relations, we find the boundary conditions from Eqs. (4.28) and

(4.29) reduce to

2
w
w= = 0 ( " = 0, 1), (4.30)
"2

2
w
w= = 0 ( # = 0, 1). (4.31)
#2

The differential equation (4.23) and the boundary conditions (4.30) and (4.31) contain

terms which are of even order in " and# ; hence, a variable separable type of solution is

sought. Thus, for Eq. (4.23), assume a solution of the form

w(" ,# ) = Amn sin m " sin n # (m = 1,2,…., n = 1,2,……), (4.32)

n =1 m =1

Where m and n define the number of half waves that the plate buckles in the X-

and Y- direction, respectively, and Amn represents the amplitudes of the shape function

or the mode shapes. Eq (4.32) satisfies the boundary conditions exactly.

Since we need to calculate the minimum value for Nx, minimizing the equation

to get least value of Nx. we get,

55
 2
4D
( N x ) cr = (4.33)
b2

D – Flexural rigidity

b – Plate width

Thus, a simply supported plate buckles with one half wave in the Y – direction and p

half waves in the X – direction, i.e.; p must be an integer. This implies that the plate

buckles into square plates.

Assuming D and b as unity, then from Eq. (4.33) the minimum critical load Nx

is calculated to be 39.478.

The minimum order of polynomial for the given set of boundary conditions

came out to be of eighth order. This eighth order polynomial has 45 coefficients to be

evaluated to calculate the buckling load.

= a[1] + xa[2] + ya[3] + x 2 a[4] + xy a[5] + a[6] + x 3a[7] + x 2ya[8] + xy 2a[9] + y 3a[10] +
i

x4a[11] + x 3ya[12] + x 2y 2a[13] + xy 3a[14] + y 4a[15] + x5a[16] + x4ya[17] + x 3y 2a[18] + x 2y 3a[19] + (4.34)
xy4a[20] + y 5a[21] + x6a[22] + x 5ya[23] + x4 y 2a[24] + x 3 y 3a[25] + x 2y 4a[26] + xy5a[27] + y 6a[28] + x 7a[29]

+ x 6ya[30] + x5y 2a[31] + x4y 3a[32] + x 3y 4a[33] + x 2y 5a[34] + xy 6a[35] + y 7a[36] + x8a[37] + x 7 ya[38] +

x6y 2a[39] + x 5y 3a[40] + x 4y 4a[41] + x 3y 5a[42] + x 2y 6a[43] + xy 7a[44] + y 8a[45]

Using the set of equations for the boundary condition, we can evaluate the coefficients

of the polynomial.

w | x =0 ,a = 0 w | y = 0, a = 0
2 2
w (4.35)
w
| x = 0, a = 0 | y =0 ,a = 0
x2 y2

56
 Once the boundary conditions are applied, the next step is to solve all the

equations involving the coefficients of the Nth order polynomial. The procedure is

carried out using the Galerkin method to arrive at the eigenvalues for each order of the

polynomial.

Table 4.3 - Comparison of Eigenvalues for a simply supported plate loaded axially
Order Direct Approach Cholesky’s Decomposition

8 39.506 39.506

9 39.5066,62.205,249.55 39.5066,10136.4

10 39.478,62.205,113.41,159.2 39.5066, 10136.4, 40664.8

11 39.478,61.687,113.41,159.2 39.5066, 10136.4, 40664.8

Similarly evaluating for 10th order polynomial, we end up with 66 coefficients

that are to be determined in order to arrive at the minimum buckling load value.

Applying the given boundary conditions and solving the problem using the Galerkin

method.

= a[1] + xa[2] + ya[3] + x 2 a[4] + xy a[5] + a[6] + x 3a[7] + x 2ya[8] + xy 2a[9] + y 3a[10] +
i

x4a[11] + x 3ya[12] + x 2y 2a[13] + xy 3a[14] + y 4a[15] + x 5a[16] + x4ya[17] + x 3 y 2a[18] + x 2 y 3a[19] + (4.36)
xy4a[20] + y 5a[21] + x6a[22] + x5 ya[23] + x 4y 2a[24] + x 3y 3a[25] + ...................... +

x5 y 3a[40] + x4 y 4a[41] + x 3y 5a[42] + x 2y 6a[43] + xy 7a[44] + y 8a[45] + .............. + y 9 a[55] + x10a[56] + .....................

x5 y 5a[61] + x 4y 6a[62] + x 3y 7a[63] + x 2y 8a[64] + x y 9a[65] + y10a[66]

57
 Solving for the coefficients of the trial functions, we get a single trial function

for the eighth order polynomial. Galerkin method is used to evaluate the eigenvalues.

The minimum eigenvalue calculated using the Mathematica program is = 39.506.

On further increasing the order of the polynomial to 10th order, and following the

procedure of applying the boundary conditions and following the Galerkin method, we

get the value of = 39.478 which is exactly the same as theoretical value.

Hence we can conclude that the lowest eigenvalue calculated using the

Mathematica program is very close to the value obtained from the theoretical solution

of Nx. To achieve better convergence and accuracy of the results, a higher order

polynomial is required.

58
 CHAPTER 5

CONCLUSIONS AND RECOMMENDATIONS

The Galerkin method was used to solve the governing differential equation of

the plate for different boundary conditions. The procedure used for solving for the

coefficients of the approximating polynomials and the Galerkin method would be

inconceivable if it were not for the symbolic software. By successfully manipulating the

given expressions and symbolically retaining the variables, the usage of symbolic

algebra software is suited for parametric study.

One particular advantage symbolic software has over present numerical

methods is the ability to retain parameters throughout the calculations and to input them

at appropriate times determined by the user. The method of selecting polynomials that

satisfy all boundary conditions, geometric and natural, provides for better accuracy and

faster convergence. This allows for lower order polynomials to be used as

approximating function.

Examples of the Mathematica programs used in this method are included in the

Appendices. The procedures explained in chapter 3 along with the examples provided in

the Appendix are sufficient to show the steps necessary to reach a solution. The results

obtained throughout this thesis were close to existing data.

59
 Weinstein’s theory was one of the first works done on plate analysis and it

really formed a basis of understanding the plate behavior under different conditions. It

formed the yardstick for future work that was carried out in this field.

The method of solving for the buckling load presented here takes advantage of

the speed and convenience of symbolic software. With the availability of routines such

as Mathematica and Matlab, classical analytical procedures can be considered when

analyzing engineering problems. These routines provide for a way of solving a problem

other than numerical techniques and finite element methods. Future applications of

symbolic software are encouraged. These applications will hopefully lead to a better

understanding of classical analytical procedures and also make ways for faster and more

accurate methods in solving such types of problems.

60
 APPENDIX A

MATHEMATICA PROGRAMS TO SOLVE THE

BUCKLING LOAD EQUATION OF A PLATE
FOR DIFFERENT CASES

61
 Program 1

This program calculates the buckling load for a simply supported plate using

the Galerkin method. The region considered for the plate in this case is,

x=0, x=1, and y=0, y=1

(* Definition of trial function which satisfies the given boundary conditions. The least

order defined for this case.*)

poly[n_] : = Sum [a(i + j) (i + j + 1) / 2 + i + 1] x^ iy^ j, {i, 0, n}, {j, 0, n - i}]

order = 8;

(*Generating the trial function using the above function definition*)

poly[order]

a[ 1 ] + xa[ 2 ] + ya[ 3 ] + x 2 a [4] + xy a [5] + a[6] + x 3a[7] + x 2ya[8] + xy 2a[9] + y 3a[10] +

x4a[11] + x 3ya[12] + x 2y 2a[13] + xy 3a[14] + y 4a[15] + x5a[16] + x4ya[17] + x 3y 2a[18] + x 2y 3a[19] +

xy4a[20] + y 5a[21] + x6a[22] + x5ya[23] + x4y 2a[24] + x 3y 3a[25] + x 2y 4a[26] + xy5a[27] + y 6a[28] + x7a[29]

+ x6ya[30] + x5y 2a[31] + x4y 3a[32] + x 3y 4a[33] + x 2y 5a[34] + xy6a[35] + y 7a[36] + x8a[37] + x7 ya[38] +

x6y 2a[39] + x5y 3a[40] + x4y 4a[41] + x 3y 5a[42] + x 2y 6a[43] + xy7a[44] + y 8a[45]

(*Applying the boundary conditions *)

eq1 = CoefficientList[poly[order] / . x 0 , y]

eq2 = CoefficientList[poly [order] / . x 1 , y]

eq3 = CoefficientList[poly [order] / . y 0 , x]

eq4 = CoefficientList[poly [order] / . y 1 , x]

62
eq5 = CoefficientList[D[poly[order] , {x, 2}] + ( D[poly[order], {y, 2}] ) / . x 0 , y]

eq6 = CoefficientList[D[poly[order] , {x, 2}] + ( D[poly[order], {y, 2}] ) / . x 1 , y]

eq7 = CoefficientList[D[poly[order] , {y, 2}] + ( D[poly[order], {x, 2}] ) / . y 0 , x]

eq8 = CoefficientList[D[poly[order] , {y, 2}] + ( D[poly[order], {x, 2}] ) / . y 1 , x]

(*Flatten command used for grouping all the eight equations together *)

Eq = Flatten [ { eq1, eq2, eq3, eq4, eq5, eq6, eq7, eq8} ]

(* Tabulating all the eight equations using the Table command *)

Equn = Table[ Eq[ [ i ] ] == 0, {i , 1, Length[Eq] } ]

(* Solving all the eight equations and generating the coefficients *)

sol = Solve[ Equn, Table[a[i], {i, 1, 60} ] ] [ [ 1] ]

(* Substituting the values of coefficients back into the equations *)

poly1 = poly[order] / . sol

(* Simplifying the equation by taking out common terms *)

m = Table[Coefficient[ poly1, a[i] ] , {i, 1, Length[poly[order] ] } ] // Simplify

(* Deleting all the integers to separate out all the derived base functions *)

e = DeleteCases[ m, _Integer]

(* Calculation of Matrix A*)

A = Table[ Integrate[( D[e[[i]], {x, 4}] + D[e[[i]], {y, 4}] + (2 * D[D[e[[i]], {y, 2}], {x, 2}])) *
e[[j]], {x, 0, 1} , {y, 0, 1}], {i, 1, 1}, {j, 1, 1}] // N; MatrixForm[A]

(* Calculation of Matrix B*)

B = Table[Inte grate[ e[[i]] * e[[j]], {x, 0, 1}, {y, 0, 1}], {i, 1, 1}, {j, 1, 1}] // N; MatrixForm [B]

63
(* Calculating inverse of Matrix B*)

B1 = Inverse [B ]

(* Calculation of the product of Matrices A and B1*)

A2 = %.A; MatrixForm[ A2 ]

(* Calculation of Eigenvalues for the Matrix A2*)

E1 = Eigenvalues [A2 ]

When the Cholesky decomposition is used, the entire procedure is carried out

till evaluating the matrices [A] and [B]. After calculating [A] and [B] the following

steps are followed for Cholesky decomposition.

(* To load the Cholesky decomposition routine *)

<< LinearAlgebra ' Cholesky'

(* Calculating the Cholesky decomposition for matrix [B]*)

( u = CholeskyDecompositio n[ B ] ) // MatrixForm

(* Calculation of transpose for matrix u*)

ut = Transpose[ u ]

(* Calculation of inverse of ut*)

uti = Inverse[ ut ]

(* Calculating the product of uti , A and ui*)

A2 = (uti) * A * (ui)

(* Calculating the Eigenvalues for A2*)

E1 = Eigenvalues [ A2]

64
 Program 2

This program calculates the buckling load for mixed boundary conditions using

the Galerkin method. The region considered for the plate in this case is

x = 0, x =1 and y = 0, y = 1

(* Definition of trial function which satisfies the given boundary conditions. The least

order defined for this case.*)

poly[n_] : = Sum [a(i + j) (i + j + 1) / 2 + i + 1] x^ iy^ j, {i, 0, n}, {j, 0, n - i}]

order = 6;

(*Generating the trial function using the above function definition*)

poly[order]

a[ 1 ] + xa[ 2 ] + ya[ 3 ] + x 2 a [4] + xy a [5] + a[6] + x 3a[7] + x 2ya[8] + xy 2a[9] + y 3a[10] +

x4a[11] + x 3ya[12] + x 2y 2a[13] + xy 3a[14] + y 4a[15] + x5a[16] + x4ya[17] + x 3y 2a[18] + x 2y 3a[19] +

xy4a[20] + y 5a[21] + x6a[22] + x5ya[23] + x4y 2a[24] + x 3y 3a[25] + x 2y 4a[26] + xy5a[27] + y 6a[28]

(*Applying the boundary conditions *)

eq1 = CoefficientList[poly[order] / . x 0 , y]

eq2 = CoefficientList[poly [order] / . x 1 , y]

eq3 = CoefficientList[poly [order] / . y 0 , x]

eq4 = CoefficientList[poly [order] / . y 1 , x]

eq5 = CoefficientList[D[poly[order] , {x, 2}] + ( D[poly[order], {y, 2}] ) / . x 0 , y]

eq6 = CoefficientList[D[poly[order] , {x, 2}] + ( D[poly[order], {y, 2}] ) / . x 1 , y]

65
eq7 = CoefficientList[D[poly[order] , {x, 1}] / . y 0 , x]

eq8 = CoefficientList[D[poly[order] , {x, 1}] / . y 1 , x]

(*Flatten command used for grouping all the eight equations together *)

Eq = Flatten [ { eq1, eq2, eq3, eq4, eq5, eq6, eq7, eq8} ]

(* Tabulating all the eight equations using the Table command *)

Equn = Table[ Eq[ [ i ] ] == 0, {i , 1, Length[Eq] } ]

(* Solving all the eight equations and generating the coefficients *)

sol = Solve[ Equn, Table[a[i], {i, 1, 60} ] ] [ [ 1] ]

(* Substituting the values of coefficients back into the equations *)

poly1 = poly[order] / . sol

(* Simplifying the equation by taking out common terms *)

m = Table[Coefficient[ poly1, a[i] ] , {i, 1, Length[poly[order] ] } ] // Simplify

(* Deleting all the integers to separate out all the derived base functions *)

e = DeleteCases[ m, _Integer]

(* Calculation of Matrix A*)

A = Table[ Integrate[( D[e[[i]], {x, 4}] + D[e[[i]], {y, 4}] + (2 * D[D[e[[i]], {y, 2}], {x, 2}])) *

e[[j]], {x, 0, 1} , {y, 0, 1}], {i, 1, 1}, {j, 1, 1}] // N; MatrixForm[A]

(* Calculation of Matrix B*)

B = Table[Inte grate[ e[[i]] * e[[j]], {x, 0, 1}, {y, 0, 1}], {i, 1, 1}, {j, 1, 1}] // N; MatrixForm [B]

(* To load the Cholesky decomposition routine *)

<< LinearAlgebra ' Cholesky'

66
(* Calculating the Cholesky decomposition for matrix [B]*)

( u = CholeskyDecompositio n[ B ] ) // MatrixForm

(* Calculation of transpose for matrix u*)

ut = Transpose[ u ]

(* Calculation of inverse of ut*)

uti = Inverse[ ut ]

(* Calculating the product of uti , A and ui*)

A2 = (uti) * A * (ui)

(* Calculating the Eigenvalues for A2*)

E1 = Eigenvalues [ A2]

67
 Program 3

This program calculates the buckling load using Weinstein’s theory. The

region considered for the square plate according to Weinstein’s theory is

x = 0, x = , y = 0, y =

(* Definition of trial function which satisfies the given boundary conditions. The least

order defined for this case.*)

poly[n_] : = Sum [a(i + j) (i + j + 1) / 2 + i + 1] x^ iy^ j, {i, 0, n}, {j, 0, n - i}]

order = 8;

(*Generating the trial function using the above function definition*)

poly[order]

a[ 1 ] + xa[ 2 ] + ya[ 3 ] + x 2 a[4] + x y a[5] + y 2a[6] + x 3a[7] + x 2ya[8] + xy 2a[9] + y 3a[10] +

x4a[11] + x 3ya[12] + x 2y 2a[13] + xy 3a[14] + y 4a[15] + x5a[16] + x4ya[17] + x 3y 2a[18] + x 2y 3a[19] +

xy4a[20] + y 5a[21] + x6a[22] + x5ya[23] + x4y 2a[24] + x 3y 3a[25] + x 2y 4a[26] + xy5a[27] + y 6a[28] + x7a[29]

+ x6ya[30] + x5y 2a[31] + x4y 3a[32] + x 3y 4a[33] + x 2y 5a[34] + xy6a[35] + y 7a[36] + x8a[37] + x7 ya[38] +

x6y 2a[39] + x5y 3a[40] + x4y 4a[41] + x 3y 5a[42] + x 2y 6a[43] + xy7a[44] + y 8a[45]

(*Applying the boundary conditions *)

eq1 = CoefficientList[poly[order] / . x 0 , y]

eq2 = CoefficientList[poly [order] / . x Pi , y]

eq3 = CoefficientList[poly [order] / . y 0 , x]

eq4 = CoefficientList[poly [order] / . y Pi , x]

68
eq5 = CoefficientList[D[poly[order] , {y, 1}] / . y 0 , x]

eq6 = CoefficientList[D[poly[order] , {y, 1}] / . y Pi , x]

eq7 = CoefficientList[D[poly[order] , {x, 1}] / . x 0 , y]

eq8 = CoefficientList[D[poly[order] , {x, 1}] / . x Pi , y]

(*Flatten command used for grouping all the eight equations together *)

Eq = Flatten [ { eq1, eq2, eq3, eq4, eq5, eq6, eq7, eq8} ]

(* Tabulating all the eight equations using the Table command *)

Equn = Table[ Eq[ [ i ] ] == 0, {i , 1, Length[Eq] } ]

(* Solving all the eight equations and generating the coefficients *)

sol = Solve[ Equn, Table[a[i], {i, 1, 60} ] ] [ [ 1] ]

(* Substituting the values of coefficients back into the equations *)

poly1 = poly[order] / . sol

(* Simplifying the equation by taking out common terms *)

m = Table[Coefficient[ poly1, a[i] ] , {i, 1, Length[poly[order] ] } ] // Simplify

(* Deleting all the integers to separate out all the derived base functions *)

e = DeleteCases[ m, _Integer]

(* Calculation of Matrix A*)

A = Table[ Integrate[( D[e[[i]], {x, 4}] + D[e[[i]], {y, 4}] + (2 * D[D[e[[i]], {y, 2}], {x, 2}])) *
e[[j]], {x, 0, Pi} , {y, 0, Pi}], {i, 1, 1}, {j, 1, 1}] // N; MatrixForm[A]

69
(* Calculation of Matrix B*)

B = Table[Inte grate[ (D[ e[[i]] , {x, 2}] + (D[e[ [i] ], {y, 2}] ) ) * e[[j]],
{x, 0, Pi}, {y, 0, Pi}, {i, 1, 1}, {j, 1, 1}] // N; MatrixForm [B]

When the Cholesky decomposition is used, the entire procedure is carried out

till evaluating the matrices [A] and [B]. After calculating [A] and [B] the following

steps are followed for Cholesky decomposition.

(* To load the Cholesky decomposition routine *)

<< LinearAlgebra ' Cholesky'

(* Calculating the Cholesky decomposition for matrix [B]*)

( u = CholeskyDecompositio n[ B ] ) // MatrixForm

(* Calculation of transpose for matrix u*)

ut = Transpose[ u ]

(* Calculation of inverse of ut*)

uti = Inverse[ ut ]

(* Calculating the product of uti , A and ui*)

A2 = (uti) * A * (ui)

(* Calculating the Eigenvalues for A2*)

E1 = Eigenvalues [ A2]

70
 REFERENCES

[1] Timoshenko, Stephen, 1878-1972. Theory of plates and shells [by] S. Timoshenko
[and] S. Woinowsky-Krieger.
[2] Szilard, Rudolph, 1921- Theory and analysis of plates: classical and numerical
methods.
[3] R D MINDLIN 0840 Journal of Applied Mechanics 07\ 20_27 Influence of rotary

inertia and shear on flexural motions of isotropic elastic plates.

[4] S. Nomura and B. P. Wang, Free vibration of plate by integral method

Computers & Structures, Volume 32, Issue 1, 1989, Pages 245-247

[5] Trahair, N. S. Flexural-torsional buckling of structures / N.S. Trahair.

[6] Wang, C. M. Exact solutions for buckling of structural members / C.M. Wang, C.Y.

Wang, J.N. Reddy.

[7] Jawad, Maan H. Theory and design of plate and shell structures / Maan H. Jawad.

[8] Leipholz, H. H. E. (Horst H. E.), 1919- Theory of elasticity [by] H. Leipholz.

[9] Rashed, Youssef F. Boundary element formulations for thick plates / Youssef F.

Rashed.

[10] Maeder, Roman. Programming in Mathematica / Roman E. Maeder.

[11] Wolfram, S. (1989) Mathematica Addison-Wesley Publishing Company.

[12] Euler, L (1759) Surla la force des colonnes. Memoires Academic Royale des

Sciences et Belle Lettres, Berlin, 13.

71
[13] Saint-Venant, B. [1855] Memoir sur la torsion des prismes. Memoires des Savants

Estrangers, XIV.

[14] Timoshenko, S.P (1953) sur la stabilite des systemes elastiques, in Collected

Papers of Stephen P. Timoshenko.

[15] Turner, M.J; Clough, R.W; Martin, G.C; Topp, L.J; “Stiffness and deflection

analysis of Complex Structures”.

[16] Galerkin, B.G (1915), Series solution of some problems in elastic equilibrium of

rods and plates. Vestnik inzhenerov i tekhnikov 19(7), 897-908

[17] EULER, L; “De motu vibratorio tympanorum,” Novi Commentari Acad.

Petropolit; 10 (1766), 243-260.

[18] KIRCHOFF, G; Vorlesungen uber mathematische Physik, Vol 1, B.G Teubner,

Leipzig, 1876

[19] WEINSTEIN, A; Journal of London Mathematical Society 10, 184- 192 (1935)

[20] IYENGAR, N.G.R; Structural stability of columns and Plates (1988)

72
 BIOGRAPHICAL INFORMATION

Yattender Rishi Dubey received his Bachelor in Engineering in Mechanical

Engineering from Visveswaraiah Technological University, Belgaum, India in 2002.

Immediately after his graduation, he worked as a Graduate Trainee at Indian Institute of

Chemical Technology, Hyderabad, India from August 2002 to June 2003. His

responsibilities involved process equipment design and process control for the pilot

plant used for the manufacturing of benzaldehyde from the oxidation of toluene.

Later in the fall of 2003, he joined The University of Texas at Arlington to

pursue his Masters of Science in Mechanical Engineering. His master’s thesis dealt with

the study of buckling load for an isotropic plate using the Galerkin method. His research

interests include finite element methods, engineering design and optimization using

numerical techniques and extending the Galerkin theory to more complex structural

analysis.

73