by
of the Requirements
August 2008
Copyright © by Bryan Dale Mixon 2008
ii
ACKNOWLEDGEMENTS
I would first like to thank my supervising Professor, Dr. Kent Lawrence for all
of his guidance throughout the duration of this project. I would also like to thank Dr.
Wen Chan and Dr. Don Wilson for taking the time to serve on my committee. Also, I
would like to thank the Department of Mechanical and Aerospace Engineering, Dr.
Wilson, Dr. Chan, and Dr. Chudoba for the financial support that they have provided to
I would like to thank the members of the AVD Lab, Gary Coleman, Amit Oza,
and Nauman Mhaskar. Amit and Gary: You guys brought a little bit of OU to UTA. I
would also like to thank JuneHyun Kim and his family: thanks for making me feel like a
I wish to thank my family for all of their support over the last few years. I could
not have done this without you. To my brother, Brad, for having as bad of taste in Sci-
Fi movies as I do and for being a technical sounding board throughout my project. And
to Kristen Roberts for your insight and advice over the years.
August 8, 2008
iii
ABSTRACT
Cross section properties are crucial in the analysis and design of beams as they
play a key role in the determination of the stress state encountered. The development of
an FEM tool in MATLAB for the calculation of beam cross section properties will be
discussed. Properties of interest include general cross section properties as well as the
non-trivial warping-dependent properties for both the unrestrained and restrained cross
section conditions for solid and hollow, open and closed sections.
iv
TABLE OF CONTENTS
ACKNOWLEDGEMENTS....................................................................................... iii
ABSTRACT .............................................................................................................. iv
LIST OF ILLUSTRATIONS..................................................................................... ix
Chapter
1. INTRODUCTION ......................................................................................... 1
v
3.1.1. Weak Solution for Poisson and Laplace Equations................. 35
vii
5.7 Stress Function vs. Warping Function ................................................ 98
Appendix
viii
LIST OF ILLUSTRATIONS
Figure Page
ix
4.8 Global stiffness matrix development for 3-Node and
6-Node triangles (BCSP.m)............................................................................. 77
5.5 I-Beam............................................................................................................. 96
x
5.6 3-Node straight-sided and 6-Node
curved-sided elements ..................................................................................... 97
xi
LIST OF TABLES
Table Page
xii
LIST OF SYMBOLS AND ABBREVIATIONS
A Area
Qy First Moment of the Area about the y-axis
Qz First Moment of the Area about the z-axis
yc Coordinate of Centroid in y-direction
zc Coordinate of Centroid in z-direction
Iy Area Moment of Inertia about the y-axis
Iz Area Moment of Inertia about the z-axis
Iyz Product of Inertia
ω Warping Function
J Torsional Constant
ω* Restrained Warping Function
Γ Warping Constant
ux Displacement in the x-direction
uy Displacement in the y-direction
uz Displacement in the z-direction
θ Angle of Twist Per Unit Length
λ Lame’s Constant
G Shear Modulus
γ xy Shear Strain Component on the x-Plane in the y-Direction
xiii
τ yz Shear Stress Component on the y-Plane in the z-Direction
xiv
u *y Warping Displacement in y-Direction about Shear Center
xv
CHAPTER 1
INTRODUCTION
1.1 Introduction
Beam cross sections come in many shapes and sizes, being of either open or
possibilities, there are several fundamental properties associated with the cross section
These fundamental properties appear in their most basic form as general cross
section properties such as area, moments of inertia, and centroidal coordinates. More
torsion and transverse shear. For the purpose of this thesis, only general cross section
circular shape experience an out-of-plane displacement called warping (see Figure 1.1),
a contrary result to the previous assumptions made by C.A. Coulomb who suggested
that cross sections remain plane under torsion [1]. Building upon this result, Barrie de
analysis, describing the axial displacement due to torsion in the form of a function [1].
1
Figure 1.1: Warping of an I-Beam cross section [2]
Analytical solutions for these cross section properties exist for simple cross
section shapes but become increasingly difficult for multiply connected cross sections
and cross sections of arbitrary shape. This is especially true for the properties obtained
in the problem of torsion. Specifically, the fact that the warping function is determined
from the solution of a second order partial differential Laplace equation for which
explicit solutions are rare makes the problem complicated [3,4]. Thus, in order to
obtain the fundamental cross section properties for a torsional analysis, it is prudent to
numerically approximate these properties over a given cross section. To this end, such
a task is accomplished proficiently through the use of the finite element method.
capability for the calculation of fundamental cross section properties. It must handle
open or closed cross sections of arbitrary shape that are singly or multiply connected.
2
Examples of such cross sections are the square section (singly connected, closed
section), the hollow cylinder (multiply connected, closed section), and the I-Beam
The cross section properties of interest range of typical cross section properties
to those due to warping on a cross section. The following listing briefly summarizes
Area (A), first moments of the area (Qy and Qz), centroid coordinates (yc and zc),
area moments of inertia (Iy and Iz), and product of inertia (Iyz).
Warping function (ω), torsional constant (J), restrained warping function (ω∗),
The following figure defines the coordinate system used for these properties.
3
Figure 1.3: Cross section coordinate system [5]
Finite element tools capable of the requirements set forth in the previous section
have been developed previously. Perhaps most notably is the powerful finite element
modeling tool known as ANSYS. Within the program is a built-in module called Beam
Tool. The Beam Tool module has 11 common cross sections and the ability to
manually create uncommon cross sections. In both cases, all of the necessary cross
section properties are calculated based on the solution of the warping function [6].
Another example of an FEM tool is that developed by Walter Pilkey, created in the
FORTRAN programming language. First, the user provides an input file describing the
geometric data defining the cross section. After meshing the geometry, the mesh
generator applies nodes and elements to the geometry, and then the FEM code solves
for the desired properties and stresses generated by a given external force and/or
moment [5].
The remaining chapters of this paper detail the theory and development of a
finite element tool to calculate beam cross section properties under torsional loading
4
conditions. Chapter 2 provides a brief background to the theoretical framework. Chapter
3 outlines the finite element formulation of the theory discussed in the previous chapter.
Chapter 4 offers an overview of the finite element code developed in MatLab. Chapter
5 presents validation and application studies, focusing on key cross section types,
followed by their results in order to verify the performance of the tool developed in
Chapter 4. Chapter 6 discusses conclusions and offers recommendations for future study
in this area.
5
CHAPTER 2
Before a finite element tool can be created to calculate the cross section
theory for the problem at hand. As outlined in Chapter 1, there are several cross section
properties of interest. These properties break down into two sections: warping
independent and warping dependent. The warping independent cross section properties
are more common in everyday use such as area, moments of inertia, and the like. The
warping dependent cross section properties break down further into two sub-sections,
presented as the Saint-Venant torsion problem and represents the foundation for the
warping dependent properties. The restrained warping case will demonstrate the
adjustments necessary to augment the Saint-Venant theory for open cross sections.
For any cross section, there are several cross section properties that are
independent of warping. These properties are most often defined by the following
equations; these have been obtained from References 5 and 7. See Figure 2.1 below.
A = ∫ dA (2.1)
A
6
where the operator, dA, represents a differential area on the body and has coordinates y
and z.
The first moments of the area, Qy and Qz, are defined as:
Q y = ∫ z dA (2.2)
A
Q z = ∫ y dA (2.3)
A
in which y and z are the coordinates of the differential area dA [5,7]. The first moments
of the area are essential in determining the location of centroid coordinates yc and zc, as
Qz
yc = (2.4)
A
Qy
zc = (2.5)
A
The moments of inertia for a plane area with respect to the y and z axes are
I y = ∫ z 2 dA (2.6)
A
I z = ∫ y 2 dA (2.7)
A
respectively. Due to this, the moments of inertia are sometimes referred to as the
I y z = ∫ y z dA (2.8)
A
7
2.2 Warping Dependent Cross Section Properties
The warping dependent cross section properties outlined in this section are
classified by either the unrestrained warping case or the restrained warping case.
Initially, the problem of torsion was approached by C.A. Coulomb based on the
assumption that cross sections under torsion will remain plane after deformation, thus
only experiencing a rotation about the axis of twist [1]. In other words, the displacement
perpendicular to the cross section is zero. However this method proved to be accurate
only in beams of circular cross sections [5]. Building upon this idea, Barrie de Saint-
Venant made a slightly different assumption. He proposed that the cross section of a
beam, when acted upon by pure torsion about the axis of twist, will not only rotate
about the axis of twist, as theorized by Coulomb, but will also experience a
displacement in the direction perpendicular to the beam cross section. The out of plane
warping experienced during pure torsion would also, when viewed as a projection back
onto the original plane, retain the original shape of the cross section [5].
8
Figure 2.1: Cross section under torsion about the x-axis with displacements uy and uz. [5]
The Figure 2.1 shows a cross section under torsion about the x-axis, which is
also the axis of twist. For a specific point on the cross section, the displacement in the y
u y ≈ − z xθ (2.9)
u z ≈ y xθ (2.10)
where θ is the angle of twist per unit length and xθ is the rotation of a section of the
Coulomb’s work shows that beams of circular cross section under pure torsion
will only rotate, while having no displacement axially (warping). For all non-circular
cross sections, experimental results demonstrate that an axial displacement occurs and
that this displacement is relatively constant throughout the length of the beam [5]. This
implies that axial displacement caused by torsion is governed by some function that is
dependent on the y and z directions but independent of the axial (x) direction.
9
Consequently, Saint-Venant proposed a relation describing axial displacement, ux,
u x = θ ω ( y, z ) (2.11)
This warping displacement is proportional to the angle of twist per unit length, θ, and a
function describing the out of plane warping of the cross section, ω(y,z). The unknown
formulation and eventual solution of the unknown warping function, ω(y,z). Saint-
The basis of Saint-Venant’s approach from the case of linear elasticity for the
∂u x ∂u y ∂u
εx = , εy = , εz = z (2.12)
∂x ∂y ∂z
∂u y ∂u x ∂u ∂u ∂u ∂u y
γ xy = + , γ xz = z + x , γ yz = z + (2.13)
∂x ∂y ∂x ∂z ∂y ∂z
Stress-Strain Relations:
10
τ x y = Gγ x y , τ y z = Gγ yz , τ x z = Gγ x z (2.15)
∆v
e = εx +εy +εz = .
v
∂σ x ∂τ x y ∂τ x z
+ + + pv x = 0 (2.16)
∂x ∂y ∂z
∂τ x y ∂σ y ∂τ y z
+ + + pv y = 0 (2.17)
∂x ∂y ∂z
∂τ x z ∂τ y z ∂σ z
+ + + pv z = 0 (2.18)
∂x ∂y ∂z
where pvx, pvy, and pvz are the body forces per unit volume.
The equations governing the surface forces on a body are given by:
p x = σ x nx + τ x y n y + τ x z nz (2.19)
p y = τ x y nx + σ y n y + τ y z nz (2.20)
p z = τ x z nx + τ y z n y + σ z nz (2.21)
where nx, ny, and nz are the unit vectors normal to the surface and px, py, and pz are the
To this point, the strain-displacement relations for linear elasticity have been
shown in terms of the displacements. Next, the strains are related to the stresses through
11
the stress-strain relations. Now it is possible, using these relations, to represent the
stresses in terms of the displacements ux, uy, and uz. In other words:
ε = Du , σ = E ε , σ = E Du (2.22)
where ε, σ, and u are vectors containing the strains, stresses, and displacements
operators. Introducing Equation 2.22 into the equations of equilibrium provides a set of
D T σ + pv = 0 (2.23)
Substituting the results of Equation 2.22 into the equilibrium equations above will
(λ + G ) ∂e + G ∇ 2 u x + pv x =0 (2.24)
∂x
(λ + G ) ∂e + G ∇ 2 u y + pv y =0 (2.25)
∂y
(λ + G ) ∂e + G ∇ 2 u z + pv z =0 (2.26)
∂z
∂2 ∂2 ∂2 ∆v
with ∇ = 2 + 2 + 2 , e = ε x + ε y + ε z =
2
∂x ∂y ∂z v
where ∇ 2 is the Laplacian operator and e is the change in unit volume per unit volume,
or dilatation.
12
The next step is to apply the background outlined in the previous section to the
Equations 2.9 thru 2.11. Beginning again with the strain-displacement relations
(Equations 2.12 and 2.13) and applying the assumed displacement conditions, the
∂u x
εx = = θ 'ω ( y , z ) = 0 (2.27)
∂x
∂u y
εy = =0 (2.28)
∂y
∂u z
εz = =0 (2.29)
∂z
∂u y ∂u x ∂ω
γ xy = + = θ − z (2.30)
∂x ∂y ∂y
∂u z ∂u x ∂ω
γ xz = + = θ + y (2.31)
∂x ∂z ∂z
∂u z ∂u y
γ yz = + =0 (2.32)
∂y ∂z
With σx, σy, σz, and τyz equal to zero, the stress-strain relations for the case of
pure torsion (Equations 2.14 and 2.15) are greatly simplified, leaving only shear stress
τ x y = Gγ x y , τ x z = Gγ x z (2.33)
13
Substituting the shear strain equations from Equation 2.30 and 2.31 above provides the
∂ω
τ x y = Gγ x y = Gθ − z (2.34)
∂y
∂ω
τ x z = Gγ x z = Gθ + y (2.35)
∂z
Finally, returning to the equations of equilibrium (Equations 2.24 thru 2.26), the
effects of the displacement conditions with the unknown warping function must be
considered. Note that for the pure torsion problem, all body forces will be zero [5].
Equations 2.36 and 2.37 below restate the equations of equilibrium for the y and z
components.
(λ + G ) ∂e + G ∇ 2 u y + pv y =0 (2.36)
∂y
(λ + G ) ∂e + G ∇ 2 u z + pv z =0 (2.37)
∂z
Realizing that,
∂u x ∂u y ∂u z
e = εx +εy +εz = + + =0 (2.38)
∂x ∂y ∂z
and that the Laplacian of uy and uz are likewise equal to zero, Equations 2.36 and 2.37
equations for the pure torsion problem with the displacement condition in the axial
14
u x = θ ω ( y, z ) (2.39)
∂e x
(λ + G ) + G∇ 2 (θ ω ( y, z )) = 0
∂x
∇ 2ω ( y, z ) = 0
∂ 2ω ∂ 2ω
∇ 2ω ( y , z ) = + =0 (2.40)
∂y 2 ∂z 2
With the formulation of Equation 2.40, the unknown warping function for the
problem of torsion is now defined as the solution to this partial differential equation [1,
5, 8]. Specifically, a PDE of this nature is described as Laplace’s equation, the solution
surface conditions defined in Equations 2.19 thru 2.21, with respect to the problem at
by first writing the surface equations in terms of strains using the material laws, and
then using the strain-displacement relations to express the strains in terms of the
−
− −∂u
p x = λ e n x + G n ⋅ ∇u x + G n ⋅ (2.41)
∂x
−
− −∂u
p y = λ e n y + G n ⋅ ∇u y + G n ⋅ (2.42)
∂y
15
−
− − ∂u
p z = λ e n z + G n ⋅ ∇u z + G n ⋅ (2.43)
∂z
− −
where e is again equal to zero, n is the normal unit vector, and u is the displacement
vector.
−
For the condition that the x component of the unit normal vector, n , is equal to
zero, and with no applied forces on the body in question (i.e. px, py, and pz = 0),
Equation 2.42 and Equation 2.43 vanish. The x-direction of the surface condition, under
∂u y ∂u x ∂u y ∂u
ny + nz + ny + nz z = 0 (2.44)
∂y ∂z ∂x ∂x
∂ω ∂ω
n y − z + nz + y = 0
∂y ∂z
∂ω ∂ω
ny + n z = zn y − yn z
∂y ∂z
∂ω ∂ω ∂ω
where ny + nz =
∂y ∂z ∂n
∂ω
= zn y − yn z (2.45)
∂n
The boundary condition of Equation 2.45 is of the Neumann type since it is in the form
of a normal derivative.
16
Having now devised the boundary condition requirement necessary for the
Laplace boundary value problem (BVP) for the warping function for pure torsion, it is
now prudent to inspect the effect of torsion on the body. The unit normal vectors found
px = 0 (2.46)
py =τ xy (2.47)
pz = τ xz (2.48)
Moment equilibrium requirements for the cross section depicted in Figure 2.2
M x = ∫∫ (τ x z y − τ x y z )dy dz (2.49)
A
17
∂ω ∂ω
M x = Gθ ∫ + y y − − z z dA (2.50)
A
∂z ∂y
which defines the moment about the x-axis for a cross section of arbitrary shape.
The moment equation about the axis of twist for a circular cross section, which
M x = GJθ (2.51)
where J is the polar moment of inertia, (Iy + Iz) [1, 5, 8]. As expected, when zero
warping is assumed in Equation 2.50, the equation for a circular cross section, Equation
2.51, is obtained. Based off this discovery, the integral portion of Equation 2.50 is now
identified as one of the desired cross section properties known as the torsional constant,
J.
∂ω ∂ω
J = ∫ + y y − − z z dA (2.52)
A
∂z ∂y
resistance to torsion.
The Saint-Venant torsion problem is now fully formulated. Solving the partial
differential Laplace equation (Equation 2.40) with the boundary condition given by
Equation 2.45, the warping function is obtained. With the warping function solved, the
torsional constant, J, of Equation 2.52 can be found. Section 2.2.2 will later supplement
this formulation with the introduction of another cross section property: the warping
constant Γ.
18
It is important to note that the warping function approach, outlined by the Saint-
Venant torsion problem, is only valid for certain types of cross sections, under certain
torsional loading scenarios. Saint-Venant made the assumption that cross sections are
free to warp without restraint (i.e. unrestrained warping). If, for example, the beam has
external supports, is not prismatic, or has a distributed torsional load, the previously
made assumption is not satisfied [5]. Even so, it is seen that for closed, solid or hollow
cross sections, the warping function approach yields acceptable, approximate results
[5]. Unfortunately, for open thin-walled cross sections, it is necessary to take some
additional steps. Section 2.2.2 explains these steps and the additional properties that
formulated. The resulting equations formed a boundary value problem of the Laplace
equation in terms of the warping function ω(y,z) with a Neumann type boundary
condition. According to reference [4], “the nature and form of the boundary
conditions” relating to this solution are inconvenient in analysis for cross sections of
irregular shape. From the theory of analytic functions, an alternative approach was
force formulation. The main advantage given here is the existence of greatly simplified
19
∂ψ ∂ψ
τ xy = , τ xz = − (2.53)
∂z ∂y
with σx, σy, σz, τyz, and body forces equal to zero for the case of pure torsion about the
∂τ x y ∂τ x z
+ =0 (2.54)
∂y ∂z
∂τ y x ∂τ zx
= =0 (2.55)
∂x ∂x
The goal now is to determine the value of the Prandtl stress function over the cross
section. From the compatibility equations for linear elasticity and the shear strain
relations for the problem of torsion, Prandtl obtained the following two relations for the
∂ 2
∇ψ =0 (2.56)
∂y
∂ 2
∇ψ =0 (2.57)
∂z
∂2 ∂2 ∂2
with ∇ 2 = + +
∂x 2 ∂y 2 ∂z 2
Relating the above equations to the original stress function relation for the shear stresses
∂ 2ψ ∂τ ∂ 2ω
= − = − Gθ + 1 (2.58)
∂y 2
∂y ∂ y∂ z
20
∂ 2ψ ∂τ ∂ 2ω
= = Gθ − 1 (2.59)
∂z 2
∂y ∂z ∂ y
Summing these two results provides the following partial differential equation.
∇ 2ψ = −2Gθ (2.60)
Boundary conditions for the stress function formulation are derived from the
same equation as for the warping function approach, Equation 2.45, resulting in the next
equation.
∂ψ d y ∂ψ d z ∂ψ
+ = =0 (2.61)
∂ y d s ∂z d s ∂ s
From this equation it follows that ψ must be a constant along the boundary of the cross
section. For simplicity the value of the stress function is taken to be zero. In a similar
manner to the Saint-Venant torsion problem in the previous subsection, the equation
defining the moment about the axis of twist in terms of the stress function is
∂ψ ∂ψ
M x = ∫∫ − y− z dy dz (2.62)
A
∂y ∂z
M x = 2∫∫ψ dy dz (2.63)
A
Finally, the torsional constant may be defined for Prandtl’s Stress function approach as
Mx 2
Gθ Gθ ∫A
J= = ψ dA (2.64)
21
2.2.2. Restrained Warping Case
torsion is based largely on the assumption that cross sections are left free to warp, or in
other words, encounter unrestrained warping [5]. This free warping assumption
provides valid results for solid and closed thin-walled cross sections, but poses a
problem for open thin-walled cross sections [5, 16]. When open thin-walled cross
sections experience a torsional load, the deflections encountered are much higher in
comparison, due to a much smaller torsional rigidity, GJ. As a result, the effects of
axial stress, σx, cannot be ignored as was done previously in Saint-Venant’s theory [5].
twist is not important, since it has no direct affect on torsional shear stress calculations
[5]. In restrained warping, however, the location of the axis of twist is not arbitrary and
must be determined. The process used to formulate restrained warping proceeds much
the same as Saint-Venant’s unrestrained warping, but with the inclusion of the unknown
Firstly, let Figure 2.3 define a centroidal coordinate axis Cyz and also a
22
Figure 2.3: Coordinate translation for restrained warping [5]
y* = y − b (2.65)
z* = z − c (2.66)
The goal of unrestrained warping is to arrive at a value for the warping function,
ω(y,z), about the defined coordinate axis. In the case of restrained warping, let us find
the warping function about the Oy*z* coordinate axis resulting in the following notation.
ω * ( y* , z* ) (2.67)
must be expressed for the restrained warping case. Using the coordinate equations for
y* and z* and Equations 2.9-2.11, the displacement relations for restrained warping
appear as
u *y = − z * xθ = − xθ ( z − c) (2.68)
23
u *z = y * x θ = x θ ( y − b) (2.69)
u x* = θ ω * ( y * , z * ) (2.70)
where θ is the angle of twist per unit length and xθ is the rotation of the cross section a
Based on these displacement conditions and the fact that for pure torsion, in
restrained warping, σy, σz, and τyz are equivalently zero (σx is non-zero, unlike in the
previous sections, and will be addressed soon), the stress state for the cross section can
∂ω *
τ x y = Gγ x y = Gθ − z * (2.71)
∂y
∂ω *
τ x z = Gγ x z = Gθ + y * (2.72)
∂z
Through examination of the equilibrium equations (Equations 2.24 thru 2.26) for
the given displacement conditions set in this subsection, it is noticed that both the y* and
z* components are satisfied, leaving only the x* component. Like in the unrestrained
∇ 2ω * = 0 (2.73)
This time, however, the equation is referenced to the unknown location of the
for a cross section with zero body forces acting on its surface. Taking into account the
24
displacement conditions y* and z*, found in Equation 2.68 and Equation 2.69, the
∂ω * ∂ω *
n y − z * + n z + y * = 0 (2.74)
∂y ∂z
Inserting the equations for y* and z*, the boundary condition is expanded as follows
∂ω * ∂ω *
n y + c + n z − b = zn y − yn z (2.75)
∂y ∂z
Equation 2.75 may be conveniently simplified and rewritten as the following normal
derivative
∂ *
∂n
( )
ω + cy − bz = zn y − yn z (2.76)
Reference [5], suggests that since ω* satisfies the Laplace equation of Equation
( ) ( )
2.73, so too must ω * + cy − bz . In addition, ω * + cy − bz satisfies the boundary
condition of Equation 2.76, which, in comparison, shares the same form as the
boundary condition relation for unrestrained warping (Equation 2.45). Implied from
this observation is that ω* and ω at most only differ by a constant [5]. In equation form
ω * ( y, z ) + cy − bz = ω ( y, z ) + C (2.77)
ω * ( y, z ) = ω ( y, z ) − cy + bz + C (2.78)
25
u x* = θ ( x) ω * ( y , z ) (2.79)
When compared to the warping displacement assumed for unrestrained warping, the
difference here is that the angle of twist per unit length, θ, now varies as a function of
axial location and the restrained warping function, ω*, is incorporated. The restrained
develop that can no longer be ignored as they once were in unrestrained warping. For
this reason, Equation 2.79 includes the rate of angle of twist that varies as a function of
x. This inclusion allows for the development of axial stress or warping normal stress, in
σ x = σ ω = Eε x (2.80)
with
∂u x ∂θ ( x)
εx = = (ω * ) (2.81)
∂x ∂x
Inserting εx from Equation 2.81 into Equation 2.80 yields the warping normal stress,
Equation 2.82.
∂θ ( x)
σ x = E (ω * ) (2.82)
∂x
Since the beam is in an assumed state of pure torsion, the only normal stresses
encountered are those generated by the warping of the cross section. Therefore, the
26
normal stresses that develop must be equivalent to a zero force-couple system where the
∫σ
A
x dA = 0 M z = − ∫ yσ x dA = 0
A
M y = ∫ zσ x dA = 0
A
(2.83)
Substitution of Equation 2.82 for σx into each of the above integrals will provide the
∫ ω dA = 0
*
(2.84)
A
M z = ∫ yω * dA = 0 (2.85)
A
M y = ∫ zω * dA = 0 (2.86)
A
following:
∫ ω dA = ∫ ω dA − c ∫ y dA + b ∫ z dA + CA = 0
*
(2.87)
A A A A
The first integral term on the right hand side of the equation is defined as the warping
dependent cross section property called the first moment of warping, Qω, and presented
here as Equation 2.88 [5]. The second and third integral terms on the right hand side
Qω = ∫ (ω dA) (2.88)
A
Equation 2.87 may now be written as the combination of this geometric property and
27
∫ ω dA = Qω + CA = 0
*
(2.89)
A
Qω
C=− (2.90)
A
With the relation for the constant C above, Equation 2.78 now becomes
Qω
ω * ( y, z ) = ω ( y, z ) − cy + bz − (2.91)
A
where the second and third integral terms represent the familiar moment of inertia, Iy,
and product of inertia, Iyz. The first integral in Equation 2.92 is defined as the warping
dependent cross section property called the sectorial product of area, Iyω, about the y-
I yω = ∫ y ω dA (2.93)
A
The moment condition, My, of Equation 2.86, when submitted to the same
treatment as Equation 2.85 provides the definition of another warping dependent cross
28
where the second and third integrals of this relation represent the familiar moment of
inertia, Iz, and product of inertia, Iyz. The first integral of Equation 2.94 is defined as the
warping dependent cross section property called the sectorial product of area, Izω, about
I zω = ∫ z ω dA (2.95)
A
Inserting the defined warping dependent cross section properties (Q∗, Iy∗, Iz∗)
into Equations 2.92 and 2.94 accordingly, yields relations for the unknown constants b
M z = I yω − cI z + bI yz = 0 (2.96)
M y = I zω − cI yz + bI y = 0 (2.97)
I yz I yω − I z I zω
b = ys = (2.98)
I y I z − I yz2
I y I yω − I yz I zω
c = zs = (2.99)
I y I z − I yz2
which are equivalent to the Trefftz shear center approximation equations [5].
Incorporating the shear center constants back into the restrained warping
equation, Equation 2.91, the final development of the restrained warping function
equation is reached.
Qω
ω * ( y * , z * ) = ω ( y, z ) − z s y + y s z − (2.100)
A
29
In Section 2.2.1.1, it was seen that the Saint-Venant torque, developed in terms
of the warping function, contained what was defined as the torsional constant, J.
∂ω ∂ω
M x = Gθ ∫ + y y − − z z dA (2.101)
A
∂z ∂y
∂ω ∂ω
J = ∫ + y y − − z z dA (2.102)
A
∂z ∂y
In much the same fashion, another important cross section property is defined
from the investigation of shear forces on the cross section. Specifically, when a shear
force acts at any given point on the cross section it is appropriate to replace it with a
force-couple about the shear center which is equivalent to the original shearing force.
The moment produced by the force-couple is referred to as the warping torque, Tω, and
is added to any pure torque that might also be applied to the cross section. The
complete derivation of the warping torque may be found in reference [5] and is defined
Tω = − Eθ '' ∫ (ω * ) 2 dA (2.103)
A
where E is Young’s modulus, θ ' ' is the second derivative of the angle of twist per unit
length. The integral of the square of the restrained warping function is identified as
another desired cross section property known as the warping constant, Γ [5].
Γ = ∫ (ω * ) 2 dA (2.104)
A
30
The warping constant Γ of Equation 2.104 is crucial in calculations for restrained
warping. It is useful when calculating the warping torque, Tω, the warping normal
stress, σω, and other important parameters of the restrained warping analysis [5].
Inserting the restrained warping function, ω∗, from Equation 2.100 into Equation 2.104
one obtains:
2
Q
Γ = ∫ ω ( y, z ) − z s y + y s z − ω dA (2.105)
A
A
2 Qω2
Γ = ∫ ω dA − − 2 z s I yω + 2 y s I zω + z s2 I z − 2 y s z s I yz + y s2 I y (2.106)
A A
where the integrated term represents the final warping dependent cross section property
of interest, Iω.
I ω = ∫ ω 2 dA (2.107)
A
Adding this newly defined property to Equation 2.106 and applying some simplification
Qω2
Γ = Iω − − y s2 I y + 2 y s z s I yz − z s2 I z (2.108)
A
dealing with cross sections of an open nature. In such instances, a component of axial
stress, called the warping normal stress, is present. Thus an alternative formulation of
the warping function is created upon which six important warping dependent cross
section properties were defined. For convenience, these results are listed again below:
31
Qω
ω * ( y * , z * ) = ω ( y, z ) − z s y + y s z − (2.109)
A
Qω2
Γ = Iω − − y s2 I y + 2 y s z s I yz − z s2 I z (2.110)
A
Qω = ∫ (ω dA) (2.111)
A
I yω = ∫ y ω dA I zω = ∫ z ω dA I ω = ∫ ω 2 dA (2.112)
A A A
32
CHAPTER 3
warping function and the stress function approach to the problem of torsion. The
warping function approach is of the form of Laplace’s equation and the stress function
approach is of the form of Poisson’s equation. These partial differential equations are
classified as elliptical boundary value problems, and in both cases an explicit solution is
often difficult or nearly impossible to obtain for many cross sections [3]. With this in
This approach is the finite element method. The finite element method is one of
taking a typically analytical problem and discretizing over the domain into much
problem over a set of basis functions, or shape functions, which define the nodes of a
given element on the subspace. In this vein, the following chapter presents the
theoretical foundation upon which a finite element approach may be applied to the
cover the domain are developed. Finally, the finite element representation for the
problem of torsion and each of the cross section properties are given.
33
3.1 Elliptical Boundary Value Problem Formulation
The first step toward producing a finite element formulation to the problem of
torsion is to define a general model of the problem. This translates into a general model
for both Laplace’s equation, whose solution provides the warping function, and
Poisson’s equation, whose solution yields the stress function. These two may be
− ∇ 2 u = f in Ω (3.1)
u = u D on ΓD (3.2)
∂u
= g on ΓN (3.3)
∂w
where Ω denotes the entire domain of interest, while ΓD and ΓN denote the Dirichlet
the warping function of Chapter 2. Equations 3.2 and 3.3 are Dirichlet and Neumann
of Equation 3.3 is assumed to vanish, the set of equations would then represent
With this general problem definition set forth, a variational or weak formulation
labeled due to the lessened smoothness requirement that is necessary within the
34
existence of a solution [10]. In addition, weak formulations for elliptical boundary
value problems, such as the problem of torsion, naturally lend themselves to the
3.1.2 [10].
The general model for the prescribed problem of torsion is defined by Equations
3.1 thru 3.3. In order to produce a weak formulation for this case, begin by making the
− ∇ 2 u = f in Ω (3.4)
u = 0 on ΓD (3.5)
∂u
= g on ΓN (3.6)
∂w
integrate over the entire domain using Green’s Theorem resulting in the following
equation [10]:
∫ (∇u ⋅ ∇w ) dΩ = ∫ f w dΩ + ∫ g w dΓ
Ω Ω ΓN
N (3.7)
For this new relation to make sense, it is necessary that u ∈ H 1 (Ω) and w ∈ H 10 (Ω) ,
where H represents a vector space [10]. Also, remember that the dirichlet condition of
u ∈ H 1 (Ω) , u = u D on ΓD
35
∫ (∇u ⋅ ∇w ) dΩ = ∫ f w dΩ + ∫ g w dΓ
Ω Ω ΓN
N (3.8)
It is important to note that the trial function, u, and the test function, w, do not
lie in the same space. In order to ensure the existence of a unique solution to the
alternative formulation. Reference [10] proposes the following method for augmenting
the problem space. Since u D ∈ H 1 / 2 (ΓD ) , and given the relation γ ( H 1 (Ω) = H 1 / 2 (ΓD ) ,
then there exists a function U D such that γ U D = u D . Thus setting u equal to the
following relation
u = v +UD (3.9)
and substituting this back into Equation 3.8, the problem now becomes one of finding
v ∈ H 10 (Ω)
∫ (∇v ⋅ ∇w ) dΩ = ∫ f w dΩ + ∫ g w dΓ − ∫ (∇U
Ω Ω ΓN
N
Ω
D ⋅ ∇w) dΩ (3.10)
Since both v and w lie in the same space (i.e. V = H 01 (Ω) ), the Lax-Milgram lemma
proves that a unique solution exists to the problem such that [10, 12]:
36
a (v, w) = ∫ (∇v ⋅ ∇w ) dΩ (3.12)
Ω
represents the weak solution for the general problem of torsion. The next step is to
apply a Galerkin Discretization to Equation 3.11 to fully compose the finite element
formulation of the second order, elliptical, partial differential Laplace and Poisson
equations.
Instead of minimizing the test function in the weak solution described above
over the vector space V = H 01 (Ω) , it is useful to define a finite dimensional subspace
upon which to minimize the problem. Choose the subspace S D ∈ V , which represents a
smaller or discretized space [13]. The problem now becomes one of seeking
vD ∈ S D
∫ (∇v
Ω
D ⋅ ∇wD ) dΩ = ∫ f wD dΩ + ∫ g wD dΓN − ∫ (∇U D ⋅ ∇wD ) dΩ
Ω ΓN Ω
(3.14)
for all wD ∈ S D
T
wD = N j w e = w e N Tj (3.15)
T T
seek v D ∈ S D a (v D , w e N Tj ) = b ( w e N Tj ) j =1,2, ..., n (3.16)
37
or
vD ∈ S D
Ω
∫ (∇v D ⋅ ∇N Tj ) dΩ = ∫ f N Tj dΩ + ∫ g N Tj dΓN − ∫ (∇U D ⋅ ∇N Tj ) dΩ
Ω ΓN Ω
(3.17)
Next, from the Ritz-Galerkin method, v D and U D are assumed to take on the
form of [13]:
n
vD = ∑ xk N k (3.18)
k =1
n
U D = ∑U k N k (3.19)
k =1
Substitution of Equations 3.18 and 3.19 into Equation 3.17 yields the following final
n
[ ∫ (∇N k ⋅ ∇N Tj ) dΩ] xk = ∫ f N Tj dΩ + ∫ g N Tj dΓN − ∑U k ∫ (∇N k ⋅ ∇N Tj ) dΩ (3.20)
Ω Ω ΓN k =1 Ω
A( N k , N Tj ) xk = b ( N Tj ) (3.21)
where
A = ∫ (∇N k ⋅ ∇N Tj ) dΩ (3.22)
Ω
and
n
b = ∫ f N Tj dΩ + ∫ g N Tj dΓN − ∑U k ∫ (∇N k ⋅ ∇N Tj ) dΩ (3.23)
Ω ΓN k =1 Ω
38
Section 3.3 will discuss what Equations 3.22 and 3.23 represent for the problem
of torsion. Before this is done, it is necessary to develop the basis functions used to
describe the subspace of this section. These basis or shape functions will describe the
behavior of small triangles and the nodes contained therein for the entire domain.
basis functions have been introduced. These functions, called shape functions, given
as {N1 , N 2 ,..., N n } over the discretized domain, individually define small elements, but
together represent a mesh covering the region [13]. Each element is composed of three
or more node points which are connected to one another by straight or curved lines.
Each node’s displacement is governed by the shape function of the element it resides in
[14].
For the development of a finite element tool to perform the goals set forth in this
project it is necessary to define the elements that will be used. Two different element
types are developed. The first is a straight sided three node triangle and the second is a
The three node triangle, also called the linear triangle, is the first element to be
developed here. It is composed of three nodes connected by three straight edges [14].
Figure 3.1 illustrates a three node triangle, one node at each vertice, numbered
in a counter-clockwise manner starting from the lower left node, with a single interior
39
point, P(y,z). Point P is used to divide this triangle into three sub-triangles. Area one is
located opposite of node one. Areas two and three are located opposite of their
respective nodes.
Figure 3.1: Representative three node triangle divided into area coordinate relations
Three functions are described using the ratio of each sub-triangle to the total area of the
triangle. These functions are the shape functions which define each node of the linear
A1 A A
η1 = , η 2 = 2 , η3 = 3 (3.24)
Ae Ae Ae
The areas A1, A2, and A3 above are easily obtained as half the determinant of the
1 y z
A1 = det 1 y 2 z 2
1
(3.25)
2
1 y 3 z 3
1 y z
A2 = det 1 y1 z1
1
(3.26)
2
1 y 3 z 3
40
1 y z
1
A3 = det 1 y1 z1 (3.27)
2
1 y 2 z 2
The shape function relations of Equation 3.24 may now be rewritten in terms of
the area equations above after, the determinant has been taken and terms have been
A1 1 A2 1
η1 = = [a1 + b1 y + c1 z ] η2 = = [a2 + b2 y + c2 z ]
Ae 2 Ae Ae 2 Ae
A3 1
η3 = = [a3 + b3 y + c3 z ] (3.28)
Ae 2 Ae
where the variables a, b, and c are given by the following relationships in Equation
3.29.
ai = y j z k − y k z j bi = z j − z k ci = − ( y j − y k ) (3.29)
In the above equation i, j, and k represent a specific node of the element and should be
applied as 1-2-3 consecutively. In practice this means that if a1 is sought, then j=2 and
The area for the element may be represented similarly to Equations 3.25-3.27
except the matrix should contain all three of the nodal coordinates as:
1 y1 z1
Ae = det 1 y 2 z 2
1
(3.30)
2
1 y 3 z 3
41
Each element may now be characterized by the shape functions of Equation
3.28. In vector form this equates to the following row vector of shape functions for a
single element.
N e = [ N1 N 2 N 3 ] (3.31)
The y and z coordinates of the problem space may now be represented in terms of the
3
y1
y = ∑ N i yi = [ N1 N 2 N 3 ] y 2 (3.32)
i =1
y3
3
z1
z =∑ N i z i = [ N1 N 2 N 3 ] z 2 (3.33)
i =1
z3
a ! b ! c!
∫N N 2b N 3c dAe = 2 Ae
a
1 (3.34)
Ae
(a + b + c + 2) !
This equation will be used in Section 3.3 during the finite element formulation of the
problem of torsion and cross section properties for the 3-node triangle.
application towards cross sections having curved boundaries. In this case, fewer
42
elements will be needed to accurately capture the curvature of the surface. See Figure
3.2.
Figure 3.2: Representative six node triangle divided with curved sides
The curved-sided 6-node element is developed from Reference [20] and [21].
The shape function vector describing the displacement behavior of all six nodes is given
as:
N e = [ N1 N 2 N 3 N 4 N 5 N 6 ] (3.35)
where
N1 = (1 − (ξ + η )) (1 − 2(ξ + η )) (3.36)
N 2 = ξ (2ξ − 1) (3.37)
N 3 =η (2η − 1) (3.38)
N 4 = 4ξ (1 − (ξ + η )) (3.39)
N 5 = 4ξη (3.40)
N 6 = 4η (1 − (ξ + η )) (3.41)
43
With regard to the actual element, N1, N2, and N3 are the shape functions for the three
corner nodes, moving in counter-clockwise order. N4, N5, and N6 are the shape
functions for the midside nodes such that N4 defines the midside node between N1 and
N2, N5 defines the midside node between N2 and N3, and N6 defines the midside node
The y and z coordinates of the problem space may now be represented in terms
y1
y
2
6 y
y = ∑ N i yi = [ N1 N 2 N 3 N 4 N5 N6 ] 3 (3.42)
i =1 y4
y5
y 6
z1
z
2
6 z
z =∑ N i z i = [ N1 N 2 N 3 N 4 N5 N6 ] 3 (3.43)
i =1 z4
z5
z 6
1 1−ξ
1 n
∫ ∫ I (ξ ,η ) J e dη dξ = ∑ wi I (ξ i ,η i ) J ei
2 i
(3.44)
0 0
44
where n is the number of Gauss points used, wi, ξ i , and η i are the weighting function
and shape functions at each Gauss point. J e is the determent of the Jacobian matrix,
For a 6-node triangle with n equals 6 Gauss points, the following table of values
for weighting functions and shape functions at each point is used in Equation 3.44.
Table 3.1: Weighting function and shape function values for each Gauss point [20, 21]
Equation 3.44 will be used in Section 3.3 during the finite element formulation of the
problem of torsion and cross section properties for the 3-node triangle.
Equation 3.22 illustrates one such contribution: the element stiffness matrix.
K e = ∫ ∇N k ⋅ ∇N Tj dA (3.45)
Ae
∂ ^ ∂ ^
and ∇ = j+ k .
∂y ∂z
45
3.3.1.1. Three Node Triangles
In order to develop the stiffness matrix representation for the three node element
defined in Section 3.2.1, it is first necessary to describe the partial derivatives of the
shape function over the element as seen in Equation 3.44. For a three node triangle this
∂N 1 ∂N1 ∂N 2 ∂N 3 1
= = [b1 b2 b3 ] (3.46)
∂y 2 Ae ∂y ∂y
∂y 2 Ae
∂N 1 ∂N1 ∂N 2 ∂N 3 1
= = [c1 c2 c3 ] (3.47)
∂z 2 Ae ∂z ∂z
∂z 2 Ae
∂N T ∂N T ∂N ∂N
Ke = ∫ j+ k j+ k dAe (3.48)
∂y
Ae
∂z ∂y ∂z
∂N T ∂N ∂N T ∂N
Ke = ∫ + dAe (3.49)
∂y ∂y ∂z ∂z
Ae
∂N ∂N ∂N T ∂N T
where and are row vectors (1 x 3) and the transposed and are
∂y ∂z ∂y ∂z
b1 c1
1 b [b b 1
Ke = ∫ 2 2 1 2 b3 ] + c2 [c1 c2 c3 ] dAe (3.50)
Ae 4 Ae 4 Ae2
b3 c3
46
Since everything in this equation is known via Equations 3.29 and 3.30 for the given y
and z coordinates of each node for each three node element, the area integral applies
only to the differential operator, dAe. The result, after some minor simplification, takes
the form:
Finally, Equation 3.51 represents the finite element formulation for the element stiffness
matrix of a three node triangle. The result is a symmetric matrix of dimension [3 x 3].
A global stiffness matrix is then assembled using the direct stiffness method [14].
In order to develop the stiffness matrix representation for the 6-node curved-
sided element defined in Section 3.2.2, it is first necessary to expand Equation 3.45 as
before:
∂N T ∂N ∂N T ∂N
K e = ∫ ∇N k ⋅ ∇N dA = ∫
T + dAe (3.52)
j ∂y ∂y ∂z ∂z
Ae Ae
where N represents the row vector of the shape function given by Equation 3.34.
Rewriting Equation 3.52 by factoring out the non-transposed vectors leads to the
∂N
∂N T ∂N T ∂y
Ke = ∫ dAe (3.53)
∂y
Ae
∂z ∂N
∂z
47
Reference [5] defines the follow with respect to the above equation:
Ae = ∫
Ae
dAe = ∫ J e dη dξ (3.54)
∂y ∂z
∂ξ ∂ξ
with J e = (3.55)
∂y ∂z
∂η ∂η
∂N ∂N ∂η ∂N ∂ξ ∂η ∂ξ ∂N ∂N
∂η +
∂y ∂ξ ∂y ∂y ∂y ∂η = J −1 ∂η = B
∂y = =
∂η ∂N ∂ξ ∂η
and (3.56)
∂N ∂N ∂ξ ∂N e
∂N
+
∂z ∂η ∂z ∂ξ ∂z ∂z ∂z ∂ξ ∂ξ
matrix of derivatives of y and z with respect to the shape function variables, ξ i and η i
called the Jacobian matrix, and a new variable is introduced; the B matrix. The B
matrix, as seen above, is a (2x6) matrix composed of the product of the shape function
∂N T ∂N T
=B
T
(3.57)
∂y ∂z
These relations provide a new form for element stiffness matrix of Equation 3.53:
1 1−ξ
Ke = ∫ ∫ Be Be J e dη dξ
T
(3.58)
0 0
Quadrature methods outlined in Subsection 3.2.2 by Equation 3.44 and Table 3.1. If
48
I (ξ ,η ) = Be Be
T
(3.59)
1 1−ξ
1 n
ke = ∫ ∫ Be Be J e dη dξ = ∑
T T
wi Bei Bei J ei (3.60)
0 0
2 i
where n is the number of Gauss points used, wi, ξ i , and η i are the weighting function
and shape functions at each Gauss point given by Table 3.1. The element stiffness
matrix has dimensions of (6x6). A global stiffness matrix is then assembled using the
Now that the left hand side of the Ritz-Galerkin discretization of Equation 3.17
has been addressed, attention may be turned to the right hand side which contains three
important contributions to the finite element formulation for the problem of torsion.
n
b= ∫ f N T dAe + ∫ g N T dΓN − ∑U k ∫ ∇N T ∇N dAe (3.61)
Ae ΓN k =1 Ae
The first integral of Equation 3.61 relates to the internal volume force for a
given cross section. The second integral accounts for the necessary Neumann boundary
conditions placed on the cross section, while the third integral accounts for the Dirichlet
boundary conditions which are required to “pin down” the stiffness matrix to obtain a
vanishes in the warping function formulation of the torsion problem because the
49
function f is taken to be zero. This can be seen by examining Equations 2.40 and 3.1.
However, in the case of the stress function formulation, f is equal to 2; see Equation
For the case of the stress function formulation, as developed in Chapter 2, the
with N T = [N1 N 3 ] is solved using the following area integral relation [9, 14]:
T
N2
a ! b ! c!
∫N N 2b N 3c dAe = 2 Ae
a
1 (3.63)
Ae
(a + b + c + 2) !
Applying this to Equation 3.62 provides a numerical solution for the integral of the
N1 1
N dA = f Ae 1
∫fN dAe = f ∫
T
2 e (3.64)
3
Ae Ae
N 3 1
where f depends on the problem; warping function, f=0, or stress function, f=2.
The second integral of Equation 3.61 defines how the Neumann boundary
50
∂ω
Here the function g represents the warping normal derivative, , found in the warping
∂n
For the line integral of Equation 3.65, it is useful to employ the following line
∫N ( s) N ( s) ds = ∫ l1a l 2b Le dl2
a b
1 2 (3.66)
L 0
1! L
∫gN d s = g Le = eg
T
(3.67)
L
(1 + 1) ! 2
where Le is the length of the line defined by the node pair of an element that is touching
∂ω dz − dy
= zn y − ynz = z − y (3.68)
∂n ds ds
where z and y correspond to the y- and z-coordinate location of the boundary condition
dz − dy
and and represent the boundary normal vector components ny and nz.
ds ds
The third and final portion of Equation 3.59 defines how the Dirichlet boundary
conditions are applied to the problem of torsion for both the warping and stress function
51
n
b = ... − ∑ U k ∫ ∇N T ∇N dAe (3.69)
k =1 Ae
There are two distinct pieces that can be seen in the above equation. The first portion is
a column vector containing all of the nodal values of U that are known explicitly
through Dirichlet boundary conditions for the given cross section. These values will be
called UD. The second portion of Equation 3.69 is exactly the same as the stiffness
K 11 K 12 xu bD
= (3.70)
K
21 K 22 x D bu
The subscript D refers to known values and the subscript u refers to unknown values.
The above example represents two linear equations, the first of which, after
K 11 ⋅ xu + K 12 ⋅ x D = bD (3.71)
Ignoring the second equation resulting from Equation 3.48 and subtracting the known
values of Equation 3.71 over to the other side of the equals sign gives:
K 11 ⋅ xu = bD − K 12 ⋅ x D (3.72)
which translates to
52
such that known values of the displacement vector, x, are subtracted from the volume
force.
It is necessary only to address the first integral of Equation 3.61 representing the
element volume force when applied to the 6-node triangle. The second and third
integrals relating the Neumann and Dirichlet boundary conditions back to the volume
force remain the same for the 6-node triangle as they were for the 3-node triangle in
b= ∫fN
T
dAe (3.75)
Ae
This integral must now be integrated using the Gauss Quadrature integration as
used for the element stiffness matrix. By applying Equation 3.75 to the Gauss
Quadrature integration equation, the following approximate relation is obtained for the
1 1−ξ n
1
be = f ∫ ∫ N T (ξ ,η ) J e dη dξ =
2
f ∑w N
i
i
T
(ξ i ,η i ) J ei (3.76)
0 0
where n is the number of Gauss points used, wi, ξ i , and η i are the weighting function
and shape functions at each Gauss point given by Table 3.1. Here again, f is a function
of the problem being solved; either the stress function or warping function.
53
3.4 Beam Cross Section Properties
This section presents the finite element formulations for the cross section
The warping independent cross section properties provided in Section 2.1 are
rederived for finite element application in the following subsections for both three node
All of the following warping independent cross section properties that require
integration, with the exception of area, were integrated using Equation 3.34 and is
a ! b ! c!
∫N N 2b N 3c dAe = 2 Ae
a
1 (3.77)
Ae
(a + b + c + 2) !
Total area for a given cross section is obtained through the summation of the
following elemental area equation that was previously developed in the three node
1 y1 z2
1
Ae = det 1 y 2 z 2 (3.78)
2
1 y3 z3
The first moments of the area, Qy and Qz, for a given cross section are defined
54
Q y = ∫ z dA (3.79)
A
Q z = ∫ y dA (3.80)
A
where y and z may be replaced with the following nodal representations for an element:
y1 z1
y = [ N ][ y ] = [ N1 N 2 N 3 ] y 2 , z = [ N ][ z ] = [ N1 N 2 N 3 ] z 2 (3.81)
y3 z3
Insertion of these relations into Equations 3.79 and 3.80 along with the application of
the integral relation for area coordinates given in Equation 3.77 supplies the following:
Ae
Qy e = [z1 + z 2 + z 3 ] (3.82)
3
Ae
Qz e = [ y1 + y2 + y3 ] (3.83)
3
where total values may be obtained for the entire cross section by summing these
Qz
yc = (3.84)
A
Qy
zc = (3.85)
A
The finite element formulation of the moments of inertia and product of inertia
become:
55
I ye =
Ae 2
6
[
z1 + z 22 + z 32 + z1 z 2 + z 2 z 3 + z1 z 3 ] (3.86)
I ze =
Ae 2
6
[
y1 + y 22 + y32 + y1 y 2 + y 2 y3 + y1 y3 ] (3.87)
Ae
I y ze = [ y1 z1 + y 2 z 2 + y3 z 3 ] + Ae [ y1 z 2 + y2 z1 + y1 z 3 + y3 z1 + y 2 z3 + y3 z 2 ] (3.88)
6 12
with total values for these three properties being, again, the sum of each over every
All of the following warping independent cross section properties that require
integration were integrated using Gauss Quadrature numerical integration and Table
3.1. For the Gauss Quadrature integration of a given function I (ξ ,η ) , the following
1 1−ξ
1 n
∫ ∫ I (ξ ,η ) J e dη dξ = ∑ wi I (ξ i ,η i ) J ei
2 i
(3.89)
0 0
where n is the number of Gauss points used for the element, wi, ξ i , and η i are the
weighting function and shape functions at each Gauss point. J e is the determent of the
Jacobian matrix, Je. The Jacobian matrix as given in Equation 3.55, repeated below:
∂y ∂z
∂ξ ∂ξ
Je = (3.90)
∂y ∂z
∂η ∂η
Element area for a six node triangle is obtained using Equation 3.73 [5].
56
1 1−ξ
Ae = ∫
Ae
dAe = ∫
0
∫
0
J e dη dξ (3.91)
1 n
Ae = ∑ wi J ei
2 i
(3.92)
When summed over all six Gauss points, the element solution is reached. Through
summation of Equation 3.92, the total area for a cross section may be obtained.
The first moments of the area, Qy and Qz, for a given cross section are defined
Q y = ∫ z dA (3.93)
A
Q z = ∫ y dA (3.94)
A
where y and z may be replaced with the following nodal representations for an element
6
ye = ∑ N i yi = [N ][ y ] (3.95)
i =1
6
z e = ∑ N i z i = [N ][z ] (3.96)
i =1
[N] is the (1x6) row vector of shape functions and [y], [z] are the (6x1) column vectors
Inserting Equations 3.95 and 3.96 into Equations 3.93 and 3.94 respectively, and
57
1 1−ξ
1 n
Q ye = ∫
0
∫ [N ][z ] J e
0
dη dξ = ∑ wi [N i ][zi ] J ei
2 i
(3.97)
1 1−ξ n
1
Q ze = ∫
0
∫ [N ][y ] J e dη dξ = 2 ∑i wi [N i ][ yi ] J ei
0
(3.98)
Qz
yc = (3.99)
A
Qy
zc = (3.100)
A
The finite element formulation of the moments of inertia and product of inertia
become:
1 1−ξ
1 n
I ye = ∫ ∫ ( [N ][z ] ) J e dη dξ = ∑ wi ( [N i ][z i ] ) J ei
2 2
(3.101)
0 0
2 i
1 1−ξ
1 n
I ze = ∫ ∫ ( [N ][ y] ) J e dη dξ = ∑ wi ( [N i ][ y i ] ) J ei
2 2
(3.102)
0 0
2 i
1 1−ξ
1 n
I y ze = ∫
0
∫ ( [N ][y] )( [N ][z ] ) J e
0
dη dξ = ∑ wi ( [N i ][yi ] )( [N i ][z i ] ) J ei (3.103)
2 i
The warping dependent cross section properties provided in Section 2.2 are
rederived toward a finite element application in the following subsections for both three
58
3.4.2.1. Three Node Triangles
All of the following warping dependent cross section properties were integrated
using the same integral equation that was used for the warping independent cross
section properties for three node triangles and is repeated here in Equation 3.104 [14].
a ! b ! c!
∫N N 2b N 3c dAe = 2 Ae
a
1 (3.104)
Ae
( a + b + c + 2) !
Qω e = ∫ ω dA (3.105)
Ae
where the warping function, ω, may be represented as below for each element.
3
ω1
ω = ∑ N i ω i = [ N 1 N 2 N 3 ] ω 2 (3.106)
i =1
ω 3
Applying Equation 3.104 as previously, following the insertion of Equation 3.73, the
Ae
Qω e = [ω1 + ω2 + ω3 ] (3.107)
3
Iω e =
Ae 2
6
[
ω1 + ω 22 + ω32 + ω1ω 2 + ω 2ω3 + ω1ω3 ] (3.108)
59
Ae
I yω e = [ y1ω1 + y 2ω2 + y3ω3 ] + Ae [ y1ω 2 + y 2ω1 + y1ω3 + y3ω1 + y 2ω3 + y3ω 2 ] (3.109)
6 12
Ae
I zω e = [z1ω1 + z 2ω2 + z3ω3 ] + Ae [z1ω2 + z 2ω1 + z1ω3 + z3ω1 + z 2ω3 + z3ω2 ] (3.110)
6 12
Equations 3.108 – 3.110 may be summed across every element on the cross
The finite element formulation of the torsional constant, J, for the warping
J = I y + I z − bT ω (3.111)
Where bT is the row vector, (1 x n), representing the global volume force vector and ω is
the column vector, (n x 1), representing the global warping function vector
The finite element formulation of the torsional constant, J, for the stress
J e = 2 ∫ψ dA (3.112)
Ae
3
2 Ae
Je =
3
∑ [ψ
1
1 +ψ 2 + ψ 3 ] (3.113)
2
Q
Γ = I ω − ω − y s2 I y + 2 y s z s I yz − z s2 I z (3.114)
A
where ys and zs are the coordinates for the shear center whose equations are found in
Section 2.2.2.
60
3.4.2.2. Six Node Triangles
The warping dependent cross section properties for the 6-node triangle after
Qω e = ∫ ω dA (3.115)
Ae
where the warping function, ω, may be represented as below for each element.
6
ω = ∑ N i ω i = [ N ][ω ] (3.116)
i =1
[N] is the (1x6) row vector of shape functions and [�] is the (6x1) column vectors of
1 1−ξ n
1
Qωe = ∫
0
∫ [N ][ω ] J e dη dξ = 2 ∑i wi [N i ][ωi ] J ei
0
(3.117)
1 1−ξ
1 n
I ωe = ∫ ∫ ( [N ][ω ] ) J e dη dξ = ∑ wi ( [N i ][ω i ] ) J ei
2 2
(3.118)
0 0
2 i
1 1−ξ
1 n
I yωe = ∫
0
∫ ( [N ][y] )( [N ][ω ] ) J e
0
dη dξ = ∑ wi ( [N i ][yi ] )( [N i ][ωi ] ) J ei (3.119)
2 i
1 1−ξ
1 n
I zωe = ∫
0
∫ ( [N ][z ] )( [N ][ω ] ) J e
0
dη dξ = ∑ wi ( [N i ][zi ] )( [N i ][ωi ] ) J ei
2 i
(3.120)
61
Equations 3.118 – 3.120 may be summed across every element on the cross
The finite element formulation of the torsional constant, J, for the warping
function approach stays the same that used for the 3-node triangles [5].
J = I y + I z − bT ω (3.121)
The finite element formulation of the torsional constant, J, for the stress
1 1−ξ
1 n
Je = ∫
0
∫ [N ][ψ ] J e
0
dη dξ = ∑ wi [N i ][ψ i ] J ei
2 i
(3.122)
2
Qω
Γ = Iω − − y s2 I y + 2 y s z s I yz − z s2 I z (3.123)
A
where ys and zs are the coordinates for the shear center whose equations are found in
Section 2.2.2.
62
CHAPTER 4
Having now established the finite element formulation for the problem of
torsion and all of the desired cross section properties, it is necessary to translate these
aspects into an appropriate programming environment. For this project, the MATLAB
programming environment has been chosen. MATLAB, which stands for matrix
is necessary with the array element in this system, which makes it very usefully for
solving systems of matrices and vectors [17, 18]. Other programming languages such
as FORTRAN, while incredibly powerful, only use scalar elements requiring more
matrices [17].
tool for the determination of beam cross section properties in the case of pure torsion
using the MATLAB programming environment. The objective of this FEM tool is to
provide a capability for the calculation of these properties for cross sections of arbitrary
Chapter 3 have been implemented in the MATLAB programming environment with the
end result being a user friendly, intuitive tool for calculating section properties.
63
4.1 Overview of FEM Code- Basic Structure
The finite element tool for calculating cross section properties presented here is
based in part on the finite element code, FEM_50, which provides a generic framework
for the solution of the Laplace and Poisson equations using 3-node triangles and 4-node
quadrilateral elements [11]. The FEM_50 code has been modified and expanded in
order to solve the problem of torsion for the warping function and each of the desired
cross section properties using both 3-node, straight edged triangles and 6-Node, curved-
sided triangles.
The main body of the code may be divided generically into three sections: input,
analysis, and output; each of which is discussed in detail in the remaining sub-sections
of this chapter. A brief overview of the tool is presented here. See Figure 4.1.
Before analysis begins, it is first necessary to construct proper input files that
define the geometry of the cross section. This is accomplished by first meshing the area
of interest to produce a collection of nodes and elements over given surface and then
linking this data to the analysis tool in the proper format. There are three types of input
files required for this FEM tool, and all may be constructed using the meshing software,
EasyMesh [19], or in the case of the 6-Node triangles, by hand. One input file contains
nodal y- and z-coordinates, one contains element-node data, and the input file type
dirichlet.dat).
64
Figure 4.1: Basic logic flow for cross section FEM Tool
The analysis portion of the tool consists of one main driver program (BCSP.m)
and five external functions (f.m, u_d.m, stima3.m, stima6.m, and volume6.m).
Depending on the type of problem being solved, warping function or stress function,
each of the external functions listed may or may not be used. The input data is read into
the main program and used to solve for either the warping function or the stress
function. If the warping function approach is chosen then both warping independent
and warping dependent cross section properties are calculated (Sections 2.1 and 2.2). If
the stress function is used, then only the warping independent cross section properties
and the torsional constant, J, are calculated. Figure 4.1 illustrates the project flow as
described above.
65
4.2 Required Inputs
The beam cross section property finite element tool, or BCSP, requires data
given by three types of input files in order to operate. As mention in Section 4.1 these
To obtain the data above, the cross section of interest is discretized, or meshed,
into many small triangular elements composed of three or six nodes, depending on the
user’s preference. Meshing a cross section with 3-Node triangles is performed using
EasyMesh, a GNU meshing tool developed by Reference 19. Meshing a cross section
with 6-Node curved-sided triangles, however, has been performed by hand since access
to a meshing tool for this type of element is not readily available for this current project.
EasyMesh version 1.4 has been used to mesh 3-Node straight edge triangles for
this project. The meshing tool, when given an input file that defines the boundaries of a
certain cross section, provides three output files containing nodal coordinates, element
node numbers, and side node numbers. Instructions and examples for input file creation
After running the mesher for a specific cross section, EasyMesh provides the
three outputs listed above. The output containing nodal coordinate data has the format
66
• First line: (number of nodes)
The marker entry in the output above is used to identify nodes lying in regions or on
boundaries of interest that are set by the user in the EasyMesh input file. The nodal
The second output from EasyMesh contains the element node data and has the
• Following lines:
(element #) (i) (j) (k) (ei) (ej) (ek) (si) (sj) (sk) (xv) (yv) (marker)
Here i, j, and k represent the nodes that belong to each element and are listed in counter-
clockwise order. For the element output file, (marker) is used to denote locations
composed of different materials. The rest of the data listed in this output file provides
circumcenter coordinates. The element output file is saved as filename.E when meshing
is complete.
The third output from EasyMesh contains the element side node data and has the
where (c) and (d) indicate the start and end points of the side, (ea) and (eb) represent the
element to right or left of the side, and (marker) can be used to identify sides lying in
67
regions or on boundaries of interest that are set by the user in the EasyMesh input file.
When each of the three EasyMesh output files have been obtained for a specific
cross section it is next important to extract only the necessary data required to run the
finite element tool, BCSP. Since a given problem has the potential to have a large
number of nodes, elements, and sides, it is convenient to automate the sorting process in
order to gather only the needed data, in the proper format, in a short amount of time.
This has been done through the creation of a FORTRAN code and is discussed below in
Section 4.2.1.
As mentioned in the introduction to Section 4.3 above, the BCSP code requires
three types of input files to provide the needed data defining a cross section. The first
input file contains only the nodal y- and z-coordinates for each node. The input file
The node coordinate input file must be a data file (.dat) to be accepted by MATLAB
The second input file type contains the element node data for each element on
the cross section. The element input file format for each row of data is as follows for
68
The 3-Node triangle nodes must be listed in a counter-clockwise fashion to insure
proper results as dictated by the area coordinate development of Chapter 3. The 6-Node
triangle format requires that node 1, node 2, and node 3 be comprised of the nodes
occupying the vertices of the triangle and are listed in a counter-clockwise direction.
The midside nodes must then be defined as node 4, node 5, and node 6 listed in a
counter-clockwise direction as well, with node 4 being the midside node between nodes
1 and 2, node 5 being the midside node between nodes 2 and 3, and finally node 6 being
the midside node between nodes 3 and 1. The 3-node and 6-node element input files
must be data files (.dat) to be accepted by MATLAB and must be named elements3.dat
The third input file type contains the data required for application of boundary
conditions. The two possible boundary conditions for this project are of the Neumann
and Dirichlet type. In each case, the input file should contain the node pairs that define
every location where boundary conditions will be applied. The boundary condition
The input files containing data corresponding to Neumann boundary conditions should
The number and type of the above inputs depends on the problem being solved
in the finite element tool. If the user chooses the stress function approach, only
Dirichlet boundary conditions will be used. If the warping function approach is used,
69
then Neumann boundary conditions must also be applied. The type of element used can
imperative that every input files be present in the folder with the main driver
program BCSP, whether they are used or not, to ensure functionality (i.e.
There is an obvious difference between the format of the output files obtained
from EasyMesh and the required input format for the BCSP code. To remedy this a
FORTRAN code was produced that can quickly collect only the pertinent data from the
EasyMesh output files for the case of the 3-node triangle, returning four files of the
proper format, file name, and file type required by MATLAB and the BCSP code. A
Given the format of the EasyMesh output files filename.N, filename.E, and
filename.S which represent the nodal, elemental, and side data of the meshed area, the
FORTRAN input file creator must read only the data corresponding to the BCSP input
file format.
only the y- and z-coordinates on each row are selected by the FORTRAN code and
70
• First line: (number of elements)
• Following lines:
(element #) (i) (j) (k) (ei) (ej) (ek) (si) (sj) (sk) (xv) (yv) (marker)
only the i, j, and k node numbers on each row are selected by the FORTRAN code and
the FORTRAN input file creator must identify and select only the sides (node pairs) that
the user has marked via the marker option, in EasyMesh, as lying on boundaries where
B.C.’s are to be applied. For the sides located along the “marked” boundaries, only the
c and d node numbers representing the side (node pair) on each row are selected by the
FORTRAN code and written into a data file (.dat) named neumann.dat and
In general, the finite element tool, BCSP.m, presented here solves firstly the set
Au=b (4.1)
for a given cross section where the left side of the equation contains A, the stiffness
matrix, and u, the stress or warping function. The right side of the equation contains b,
71
the volume force and the related Neumann and Dirichlet boundary condition relations
that apply. When the stress function or warping function has been found, BCSP.m then
calculates the warping independent and warping dependent cross section properties for
the cross section. Throughout this process, as illustrated in Figure 4.2, six external
functions are called by the driver program to aid in calculations for the stiffness matrix,
A and the volume force, b. These functions are detailed in Subsection 4.3.1 during a
discussion of BCSP.m. Subsection 4.3.2 presents a brief description on how the code
may be conditioned to solve for either the stress function or the warping function.
A walkthrough of the driver program, BCSP.m, and the six external functions
seen in Figure 4.2 above, is presented here. The BCSP code is divided among the
The first action performed by the driver program is to read each of the input
files necessary for analysis using the load command. This action is performed on all
72
five input files (coordinates.dat, elements3.dat, elemtents6.dat, neumann.dat, and
Next, in Figure 4.4, a symmetry condition is introduced allowing the user to take
advantage of symmetry, modeling only that portion of the geometry. When BCSP is
executed, the user is prompted to input a value for the symmetry condition being used.
If full geometry of the cross section is defined by the input files, zero is selected
indicating zero symmetry. If ½ symmetry exists about the y-axis for the cross section, 1
is selected. Similarly, if ½ symmetry exists about the z-axis for the cross section, 2 is
selected. Finally, if the geometry is symmetric about the y- and z-axes, the user may
73
Figure 4.4: Cross section symmetry factor
Having loaded all node, element, and boundary condition data from each
input file, the BCSP code’s next objective is to develop the element stiffness and then
global stiffness matrices for the problem of torsion. Figure 4.5 is the main portion of
the external function stima3.m whose purpose is to calculate the element stiffness
matrix for every 3-node element given in the elements3.dat input file. The logic for the
stima3.m function was developed by Reference [11] and is equivalent to the 3-node
Figure 4.6 and 4.7 show the external function stima6.m, whose purpose is to
calculate the element stiffness matrix for every 6-node element given in the
elements6.dat input file. As discussed in Subsection 3.3.1.2, the integral required for
the determination of the element stiffness matrix is performed using Gauss Quadrature.
Figure 4.6 contains the portion of the function that defines the y- and z- coordinates for
74
all 6-nodes of the element as received from the driver program. The necessary Gauss
points for a 6-node element are also specified here as found in Table 3.1.
In Figure 4.7, the Jacobian matrix, J, and the shape functions are used to obtain
the B matrix and the product BTB. This data is obtained for each Gauss point for the
stiffness matrix is given by the summation of the product of BTB and the determinant of
1 n
ke = ∑
2 i
T
wi Bei Bei J ei (4.2)
This result is sent to the main driver BCSP.m after all six Gauss points have been
evaluated.
75
Figure 4.7: 6-Node element stiffness matrix (stima6.m) (continued)
Figure 4.8 depicts the global stiffness matrix development for the 3-Node
triangle and then the 6-Node triangle. Initially a sparse matrix A is created whose size
is (# of nodes by # of nodes). BCSP sends element node data to the stima3.m and/or the
stima6.m functions. Upon return from the external functions, the sparse, empty, matrix
A is filled with element values for the stiffness matrix. The global matrix is filled by
way of the direct stiffness method [14] and is illustrated in Figure 4.8. The for loop
sends every element in turn to the one of the stima functions and upon return, adds the
76
Figure 4.8: Global stiffness matrix development for 3-Node and 6-Node triangles (BCSP.m)
of the element volume force. BCSP first creates a sparse, empty, vector, b whose
dimension is (# of nodes by 1). Next, in the case of volume force for 3-node triangles
(line 70-76, Figure 4.9), element data is used to calculate the volume force of each
element according to Equation 3.62 of Chapter 3 and stored in b according to the node
numbers to obtain the global volume force vector for 3-node triangles adding nodes
together when appropriate. The logic for the element volume force for the 3-node
triangle was developed by Reference [11] according to Equation 3.62 of Chapter 3 and
is conditioned by the external function f.m whose value is zero for the warping function
In the case of 6-node triangles (lines 70-83, Figure 4.9), the main driver must
call the volume6.m function to perform a Gauss Quadrature integral for each element,
77
which is fed back into the main program for addition into the global b vector according
to node number.
formed global volume force vector, b according to Equations 3.59 and 3.65 of Chapter
3. Line 90 of BCSP checks to see if the neumann.dat boundary condition input file is
empty. If so, the global volume force is left unchanged. When the neumann.dat input
has data within, the global volume force will be augmented by subtracting the warping
normal derivative of Equation 2.45, for each neumann.dat node pair defining the
boundary. Lines 98-100 of Figure 4.10 serve to access every volume force entry that
has a node in common with any neumann.dat nodes and subtract the value of the
warping normal derivative boundary constraint at that node via the external function
78
Figure 4.10: Neumann boundary condition application
The external function warp.m provides the value of the warping normal
derivative for each pair of nodes sent to it from BCSP. The warping normal derivative
∂ω dz − dy
= zn y − ynz = z Ave − y Ave (4.3)
∂n ds ds
where zAve and yAve correspond to the midpoint between each node pair given in
dz − dy
neumann.dat and and represent the normal vectors ny and nz.
ds ds
This boundary condition is required for all internal and external boundaries when
solving for the warping function, and is zero when dealing with the stress function.
79
The final type of boundary condition is the Dirichlet boundary condition. It is
applied in BCSP to augment the global volume force and to identify the nodes in the
stiffness matrix that can be disregarded in the solution process of the next section,
Subsection 4.3.1.4. After applying the Dirichlet boundary condition, the global stiffness
matrix will no longer be singular. The theory was presented in Equations 3.68-3.70 of
Chapter 3. Here a sparse, empty vector u is created and filled only the nodes
corresponding to the Dirichlet boundary node pairs given in dirichlet.dat and are each
given a value of zero by the external function u_d.m. The values of u given here
represent either known values of the warping function, ω, or known values of the stress
function, ψ, depending on the problem. The u vector is then multiplied by the cells of
the global stiffness matrix, A, that correspond to the node number and subtracted from
the global volume force vector, b, based on Equation 3.70, as seen in Figure 4.12, line
115 [11].
To this point the main driver program, BCSP.m, and its external functions have
developed a global stiffness matrix and a global volume force vector. Boundary on the
problem of torsion have been applied also been applied, resulting in an augmented
80
global volume force and a non-singular global stiffness matrix. Now, the linear system
of equation can be solved for the unknown values of the vector u for either the stress
function or the warping function depending on the problem. Figure 4.13 displays the
manner in which this is achieved in the finite element tool. In the Figure on line 121, a
vector containing all of the free nodes is identified. This vector consists of every node
unknown nodes. Next, a solution for the u vector is obtained by calling only the
coefficients of the stiffness matrix and volume force vector that are attributed to the free
nodes and performing the left inverse of A and b as seen in line 124 of Figure 4.13 [11].
Having solved for the u vector for the stress function or the warping function,
the beam cross section properties may be now calculated. As presented in Chapter 3,
these properties are identified as warping independent and warping dependent. The
BCSP code calculates warping independent and dependent cross section properties first
for 3-node triangles and then for 6-node triangles. Figure 14.14 presents the element
level calculations for the 3-node triangle element. Each cross section property is
implemented here using the area coordinate integral results of Subsection 3.4.1.1.
81
When a value is obtained for each element, it is then summed with the previous
elemental result, producing an answer for the total cross section for each property.
Figure 4.14: 3-Node triangle warping independent and dependent cross section properties
Figure 4.15 illustrate how the y- and z-coordinate of the centroid is found. In
addition to this the parallel axis theorem is applied to the moment of inertia calculations
and the product of inertia to translate them from being defined about the origin to the
centroid location.
calculate the warping independent and dependent beam cross section properties for 6-
Node triangles. It is necessary once again to utilize Gauss Quadrature to calculate the
82
integrals governing the cross section properties for application with 6-Node triangles.
Figure 4.16 begins by defining the Gauss points and shape functions for a 6-Node
triangle.
Figure 4.17: 6-Node triangle warping independent and dependent cross section properties
Figure 4.17 obtains from BCSP the y- and z- coordinates, as well as the warping
function for each node on the element from previous sections section of the code. With
this data, it is important calculate the Jacobian matrix, Je, and the shape function vector,
N, for each gauss point on the element (Figure 4.18). From here the finite element
formulations of Subsection 3.4.1.2 and 3.4.2.2 the warping independent and dependent
83
Figure 4.18: 6-Node warping independent and dependent cross section properties (continued)
84
Figure 4.20: Unrestrained and Restrained Warping Cross Section Property Calculations
Figure 4.20 highlights the portion of the of the cross section property tool were
the warping independent and dependent properties are put to use in order to calculate
the torsion constant, J, shear center location, ys and zs, and the warping constant, Γ, if
the warping function approach has been taken. If the stress function approach is taken
however, only the torsional constant is obtained from this portion of the code.
Since the BCSP code is designed for application to both the stress function and
the warping function approach to the problem of torsion, it is prudent to discuss how
85
4.3.2.1. Stress Function
Given:
∇ 2ψ = −2Gθ (4.4)
ψ =0 (4.5)
For this problem, external function f.m must be set equal to 2 and external function
u_d.m must be set to zero. For the stress function approach only Dirichlet boundary
As a result the expected outputs from BCSP for the solution of the torsion
problem with the stress function will be the warping independent cross section
Given:
∇ 2ω = 0 (4.4)
ω=0 (4.5)
∂ω
=g (4.6)
∂n
The Dirichlet boundary condition of Equation 4.5 is necessary to mark areas of zero
warping on the cross section so that the stiffness matrix may be fixed in place [16] at a
86
placed on all external and internal boundaries of the given cross section. For this
As a result, the expected outputs from BCSP for the solution of the torsion
problem with the warping function, ω, will be the warping independent and warping
dependent cross section properties and the torsion constant, J, and the warping constant,
Γ.
87
CHAPTER 5
The development of a finite element tool for the calculation of beam cross
section properties has been presented from theoretical background to the finite element
applications of the BCSP FEM code; a circular cylinder, two square, multiply-
connected cross sections, a C-Beam, and an I-Beam. Following these brief applications
of the BCSP code, the circular cross section once again is presented, but this time to
Node straight-sided triangular element. A comparison of the stress function and the
warping function is also performed. Results from the validation problems are compared
The first application of the BCSP FEM code is toward a circular cylinder. As
mentioned previously, the circular cross section is the only cross section which does not
Figure 5.1 provides the proper dimensions for the given problem, which is
solved using the warping function approach only, since the geometry is multiply
connected. For use in BCSP, this problem is modeled taking advantage of symmetry.
88
Problem Setup:
external surfaces.
Table 5.1 contains results as obtained by ANSYS Beam Tool and BCSP for the
89
Table 5.1: Circular Cylinder Results: ANSYS and BCSP
A small discrepancy can be seen between ANSYS results and BCSP results.
The meshed cross section used to produce inputs files for BCSP was not fine enough,
however with only 338 elements still managed to remain accurate having an error of
only 1.41% on the torsional constant value. As expected, the warping constant is
The next application is toward a square 2x2 cross section with a square 1x1 hole
cut out of its center as seen in Figure 5.2. Results from the BCSP code are compared in
90
Figure 5.2: Square cross section with square hollow center
Problem Setup:
external surfaces.
A mesh of 700 3-Node elements was placed on the cross section providing very
good agreement with the results obtained by ANSYS. There is only a 0.54% difference
between the values seen here for the warping constant, Γ, and 0.048% difference in the
torsional constant, J.
91
Table 5.2: Square cross section with square cut-out results
The next application is toward a square 10x10 square cross section with two
square 3x8 spaces cut out. The two cut-out sections are centered symmetrically about
the centroid of the cross section with coordinates as given in Figure 5.3. Results from
the BCSP code are compared in Table 5.3 to those found using the ANSYS Beam Tool.
Problem Setup:
external surfaces.
92
Figure 5.3: Square cross section with two hollow rectangular cut-outs
A mesh of 3,100 3-Node elements was placed on the cross section providing
very good agreement with the results obtained by ANSYS. There is only a 0.42%
difference between the values seen here for the warping constant, Γ, and 0.087%
93
5.4 C-Beam Application
with height of 19 inches. and top and bottom flange length of 8.5 inches. See Figure 5.4
below. Results from the BCSP code are compared in Table 5.4 to those found using
the ANSYS Beam Tool and the Pilkey cross section tool, see Reference [5].
Problem Setup:
of symmetry.
94
A mesh of 1,848 3-Node elements was placed on the cross section providing
very good agreement with the results obtained by ANSYS. There is almost exact
agreement between BCSP and ANSYS for this application, with the Pilkey code
differing only slightly. Any difference is so small that almost no effect is made.
The next application is toward a symmetric I-Beam, an open section, with height
of 20 inches. and top and bottom flange length of 10 inches. See Figure 5.5 below.
Results from the BCSP code are compared in Table 5.5 to those found using the
95
Figure 5.5: I-Beam
A mesh of 2,278 3-Node elements was placed on the cross section providing
very good agreement with the results obtained by ANSYS. There is a 0.04% difference
between the values seen here for the warping constant, Γ, and 0.002% difference in the
torsional constant, J.
96
5.6 Circular Cylinder Application: 3-Node vs. 6-Node Triangles
The same circular cylinder discussed in Section 5.1 is repeated here in Figure
5.7 to be solved using first 3-Node straight-edged triangular elements and then 6-Node
requires the user to generate a very fine mesh of the cross section to better approximate
curvature. The advantage of the 6-Node curved-sided element is that it can easily form
to a curved shape allowing for a better approximation with much fewer elements. See
Figure 5.6.
97
The circular cylinder was fit with a mesh of 136 triangular straight-sided
elements in one case and only 5 6-Node curved-sided elements in another. In spite of
the low number of 6-Node elements used, the results obtained are as good as or better
than the 3-Node triangles because the element could form to the curved surface. See
Table 5.6.
From the table above it can be seen that there is only a 0.05% difference
between BCSP with the 6-Node element and ANSYS in the value of the torsional
constant. The 3-Node element however has a slightly less accurate answer in
to the warping function approach for the calculation of the torsional constant. For the
rectangular cross section in Figure 5.8, both the stress function and warping function
were employed to solve for the torsional constant using a varying amount of 3-Node
triangular elements.
98
Figure 5.8: Rectangular C-S: Stress Function vs. Warping Function
Figure 5.9 shows the convergence of both cases toward the real value of the
torsional constant as calculated in ANSYS. The green curve represents the warping
function and is seen to converge quickly in comparison to the stress function approach;
red curve. After approximately 1000 elements the warping function converges to
within 0.054% of the value given by ANSYS. The stress function takes over 4000
elements to arrive at an appropriate value of the torsional constant. The reason for the
99
Figure 5.9: Convergence of Stress Function vs. Warping Function on Torsional Constant
Results of the stress function and the warping function solution compared to ANSYS
100
CHAPTER 6
The purpose of this project has been accomplished through the creation of a
finite element tool that calculates fundamental warping independent and warping
dependent cross section properties for the case of pure torsion. The tool has been
created in MATLAB and is able to handle open or closed cross sections for which at
least one point of zero warping is known. Though the ability to determine the cross
section properties of a cross section with an arbitrary shape was desired, the restrictions
placed on the problem at hand by the stiffness matrix prevented this. To prevent the
stiffness matrix from being singular, at least one point of zero warping is needed in
order to apply the Dirichlet boundary condition. By securing the stiffness matrix in this
determined.
The code has been tested and validated through the use of five cross sections for
which the properties determined by the code were compared to the same results
determined from ANSYS. The results match extremely well, differing by less than one
percent with a fine mesh. Also, a study has been conducted which compared the results
obtained using the code with 3-Node straight-edge triangles and with 6-Node curve-
sided triangles to the results obtained from ANSYS. The 6-Node triangles are able to fit
a curved surface, therefore, for the same level of accuracy, the 6-Node triangle mesh
101
required less elements (5) than the 3-Node triangles (136) to reach the result. Finally, a
study has been completed which compared the results obtained with the stress function
versus the results obtained with the warping function. The results show that the stress
function requires significantly more elements than the warping function to arrive at the
same level of accuracy in comparison to the results returned by ANSYS, 4000 versus
In conclusion, it has been determined that the warping function is the best
approach for a project of this type, as it allows for the calculation of warping dependent
cross section properties that include the warping constant as well as the torsional
constant. In contrast, the stress function only allows for the calculation of the torsional
constant. Additionally, the case studies of this project show that 6-Node, curve-sided
triangular element is more efficient for applications with curved surfaces than the 3-
At the conclusion of this project, there are a few recommendations for further
how the stiffness matrix can be non-singular naturally as Reference [5] suggests; that is,
how the stiffness matrix can be used to solve the warping function without the
restriction of a known location of zero warping for Dirichlet boundary conditions. With
this ability, a more in-depth comparison of the results returned through the warping
102
Secondly, it would be beneficial to modify the code to allow input for an
applied moment or torque, as it would allow for further calculation of the angle of twist,
the shear stresses, and the warping normal stress of a cross section.
Lastly, the theory of transverse shear should be incorporated into the code to further
calculate cross section properties of shear deformation coefficients and the shear center
103
APPENDIX A
104
%% BCSP applies the finite element method to Laplace or Poisson's
Equation.
%
%
clear
%
% Read and load the nodal coordinate input file.
%
load coordinates.dat;
%
% Read and load the 3-Node and 6-Node triangle input files.
%
load elements3.dat;
load elements6.dat;
%
% Read and load the Neumann boundary condition input file.
% EVAL command is to create an empty NEUMANN array if no neumann file
% is found.
%
eval ( 'load neumann.dat;', 'neumann=[];' );
%
% Read and load the Dirichlet boundary condition input file.
%
load dirichlet.dat;
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%
% Set problem symmetry type:
% =0: Zero Symmetry
% =1: Symmetric about Y-Axis
% =2: Symmetric about Z-Axis
% =4: Symmetric about Y and Z-Axis
% Ex: If value=4 then the problem can be modeled as a single
Quadrant
% with final answers being adjusted accordingly.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%
sym=input('Set symmetry value: [No sym->0; sym y->1; sym z->2; y & z-
>4] ');
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%
% Stiffness Matrix Assembly
%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%
105
% 3-Node Triangles
for j = 1 : size(elements3,1)
A(elements3(j,:),elements3(j,:)) =
A(elements3(j,:),elements3(j,:)) ...
+ stima3(coordinates(elements3(j,:),:));
end
% 6-Node Triangles
for j = 1 : size(elements6,1)
A(elements6(j,:),elements6(j,:)) =
A(elements6(j,:),elements6(j,:)) ...
+ stima6(coordinates(elements6(j,:),:));
end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%
% Assemble Volume Forces
%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%
b = sparse ( size(coordinates,1), 1 );
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%
% Neumann Boundary Conditions
%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%
for j=1:size(neumann,1)
106
% Application of normal warping derivative 'g' toward the Right Hand
Side
% through the addition of (E/2)*g where E is the length of the side
% represented by each neumann.dat node pair
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%
% Apply Dirichlet Boundary Conditions
%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%
%
% Determine which nodes are associated with Dirichlet conditions.
% Assign the corresponding entries of u, and adjust the right hand
side.
%
u = sparse ( size(coordinates,1), 1 );
BoundNodes = unique ( dirichlet );
u(BoundNodes) = u_d ( coordinates(BoundNodes,:) );
b = b - A * u;
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%
% Solve A * u = b for the remaining unknown values of u.
%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%
% Cross Section Properties
%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%
u; % u= nodal warping function value if f=0
107
% u= nodal stress function value if f=2
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%
% Warping Independent and Dependent Cross Section Properties
%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%
%%%%%%%%%%%%%%%%%%%% 3-Node Triangle Calculations
%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%
for j=1:size(elements3,1)
y1=coordinates(elements3(j,1),1);
y2=coordinates(elements3(j,2),1);
y3=coordinates(elements3(j,3),1);
z1=coordinates(elements3(j,1),2);
z2=coordinates(elements3(j,2),2);
z3=coordinates(elements3(j,3),2);
u1=u(elements3(j,1));
u2=u(elements3(j,2));
u3=u(elements3(j,3));
Ae=0.5*((y2*z3-y3*z2)+(y1*(z2-z3))+(z1*(y3-y2)));
Area=Area+Ae;
Qyi=Qyi+((Ae/3)*(z1+z2+z3));
Qzi=Qzi+((Ae/3)*(y1+y2+y3));
Qwi=Qwi+((Ae/3)*(u1+ u2 + u3));
Iyi=Iyi+((Ae/6)*((z1^2)+(z2^2)+(z3^2)+(z1*z2)+(z2*z3)+(z1*z3)));
Izi=Izi+((Ae/6)*((y1^2)+(y2^2)+(y3^2)+(y1*y2)+(y2*y3)+(y1*y3)));
Iyzi=Iyzi+((Ae/6)*((y1*z1)+(y2*z2)+(y3*z3)))+((Ae/12)*((y1*z2)+(y2*z1)
...
+(y1*z3)+(y3*z1)+(y2*z3)+(y3*z2)));
Iwi=Iwi + ((Ae/6)*((u1^2)+(u2^2)+(u3^2)+(u1*u2)+(u2*u3)+(u1*u3)));
108
Iywi=Iywi +
((Ae/6)*((y1*u1)+(y2*u2)+(y3*u3)))+((Ae/12)*((y1*u2)+(y2*u1)...
+(y1*u3)+(y3*u1)+(y2*u3)+(y3*u2)));
Izwi=Izwi +
((Ae/6)*((z1*u1)+(z2*u2)+(z3*u3)))+((Ae/12)*((z1*u2)+(z2*u1)...
+(z1*u3)+(z3*u1)+(z2*u3)+(z3*u2)));
if (f == 2)
Jstress = Jstress + (2/3)*Ae*sum(u(elements3(j,:)));% Torsion
Constant for stress function approach
end
end
Area; Qyi; Qzi; Qwi; Iyi; Izi; Iyzi; Iwi; Iywi; Izwi;
Yci=Qzi/Area;
Zci=Qyi/Area;
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%
%%%%%%%%%%%%%%%%%%%% 6-Node Triangle Calculations
%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%
for j=1:size(elements6,1)
109
y(1,1)=coordinates(elements6(j,1),1);
y(2,1)=coordinates(elements6(j,2),1);
y(3,1)=coordinates(elements6(j,3),1);
y(4,1)=coordinates(elements6(j,4),1);
y(5,1)=coordinates(elements6(j,5),1);
y(6,1)=coordinates(elements6(j,6),1);
z(1,1)=coordinates(elements6(j,1),2);
z(2,1)=coordinates(elements6(j,2),2);
z(3,1)=coordinates(elements6(j,3),2);
z(4,1)=coordinates(elements6(j,4),2);
z(5,1)=coordinates(elements6(j,5),2);
z(6,1)=coordinates(elements6(j,6),2);
u1(1,1)=u(elements6(j,1)); u1(2,1)=u(elements6(j,2));
u1(3,1)=u(elements6(j,3)); u1(4,1)=u(elements6(j,4));
u1(5,1)=u(elements6(j,5)); u1(6,1)=u(elements6(j,6));
J(1,1)= (-3+4*Z(1,i)+4*E(1,i))*y(1,1)+(4*Z(1,i)-
1)*y(2,1)+0*y(3,1)+...
(4-8*Z(1,i)-4*E(1,i))*y(4,1)+4*E(1,i)*y(5,1)+(-
4*E(1,i))*y(6,1);
J(1,2)= (-3+4*Z(1,i)+4*E(1,i))*z(1,1)+(4*Z(1,i)-
1)*z(2,1)+0*z(3,1)+...
(4-8*Z(1,i)-4*E(1,i))*z(4,1)+4*E(1,i)*z(5,1)+(-
4*E(1,i))*z(6,1);
J(2,1)= ((-3+4*Z(1,i)+4*E(1,i))*y(1,1)+ 0*y(2,1)+(4*E(1,i)-
1)*y(3,1)+...
(-4*Z(1,i))*y(4,1)+(4*Z(1,i))*y(5,1)+(4-4*Z(1,i)-
8*E(1,i))*y(6,1));
J(2,2)= ((-3+4*Z(1,i)+4*E(1,i))*z(1,1)+ 0*z(2,1)+(4*E(1,i)-
1)*z(3,1)+...
(-4*Z(1,i))*z(4,1)+(4*Z(1,i))*z(5,1)+(4-4*Z(1,i)-
8*E(1,i))*z(6,1));
Jdet=det(J);
Ae = temp + 0.5*(Jdet*W(1,i));
Qye = temp1 + 0.5*(Jdet*W(1,i)*N*z);
Qze = temp2 + 0.5*(Jdet*W(1,i)*N*y);
Iye = temp3 + 0.5*(Jdet*W(1,i)*((N*z)^2));
110
Ize = temp4 + 0.5*(Jdet*W(1,i)*((N*y)^2));
Iyze = temp5 + 0.5*(Jdet*W(1,i)*((N*y)*(N*z)));
Qwe = temp6 + 0.5*(Jdet*W(1,i)*(N*u1));
Area;
Qyi;
Qzi;
Qwi;
Iywi;
Izwi;
Yci=Qzi/Area;
Zci=Qyi/Area;
Iyi;
Izi;
Iyzi;
Qwi;
if (sym == 0) % No Symmetry
symmetry=sym+1
Qy=Qyi, Qz=Qzi, Qw=Qwi
Yc=Yci, Zc=Zci
Iyc=Iyi-(Area*(Zci^2)), Izc=Izi-(Area*(Yci^2))
Iyzc=Iyzi-(Area*Yci*Zci)
Areatotal=Area
111
Qy=Qyi-Qyi, Qz=Qzi*symmetry;
Yc=Qz/Areatotal
Zc=Qy/Areatotal
Qw=Qwi-Qwi;
Iy=Iyi*symmetry;
Iz=Izi*symmetry;
Iyz=Iyzi-Iyzi;
Iyc=Iy-(Areatotal*(Zc^2))
Izc=Iz-(Areatotal*(Yc^2))
Iyzc=Iyz-Iyz
Iw=Iwi*symmetry
Iyw=Iywi-Iywi
Izw=Izwi*symmetry
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%
% Unrestrained and Restrained Warping Cross Section Properties
%
112
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%
end
if (f == 2)
Jstress=Jstress*symmetry
end
113
REFERENCES
114
[21] Burnett, David S. Finite Element Analysis : From Concepts to Applications. New
York: Addison-Wesley Longman, Limited, 1987.
115
BIOGRAPHICAL INFORMATION
Bryan Mixon was born in Oklahoma in 1981. Bryan graduated from Mustang
High School in 2000 and went on to pursue an undergraduate degree at the University
member of the Phi Kappa Phi National Honor Society, and attended two AIAA Annual
Aerospace Sciences Meeting and Exhibit and Author and Co-Authored three conference
116