-
Thin-Walled Structures 00 (2014) 1??
Thin-Walled
Structures
Elasticity-Based Free Vibration ofAnisotropic Thin-Walled
Beams
Paul R. Heyliger
Department of Civil and Environmental Engineering, Colorado
State University, Fort Collins, CO 80523, 970-491-6685,
[email protected]
AbstractThe free-vibration of anisotropic open and closed
thin-walled beams is studied using approximate solutions to the
three-dimensionalequations of motion of a linear elastic solid. The
three displacement components are approximated using grouped
polynomialsover both the cross-section coordinates and independent
polynomial or trigonometric functions over the axial coordinate of
severalbeams with representative boundary conditions. This method
is applied to several sections with varying degrees of anisotropy
andshape with both isotropic and orthotropic material properties.
Results are compared with those computing using beam or
shelltheories, and several new results are presented for these
geometries.
Introduction
Three-dimensional solids that have one-dimension much larger
than the other two and a cross-sectioncomprised of elements of that
are generally long and thin are usually classified as thin-walled
beams. Theirbehavior, which can combine elements of motion usually
associated with axial, flexure, and torsional mo-tion, has been
extensively studied ever since the pioneering work of Wagner and
Pretschner [1], Bleichand Bleich [2], and Kappus [3]. As noted by
Gjelsvik [4], the comprehensive theory of open thin-walledsections
of Vlasov was also developed during the same period as these early
works but not translated intoEnglish until 1963 [5]. By this time
Goodier [6] and Timoshenko [7] had independently developed
generaltheories. Additional work by Gere and Lin [8] on coupled
vibration behavior outlined some of the moresignificant issues
related to these components. Numerous computational models have
been further devel-oped, including the dynamic stiffness matrix
method of Banerjee and co-workers [9] and the finite elementmodel
of Mei [10]. More recent applications to composite beams have also
been considered by Kollar [11],Cortinez and Piovan [12], Piovan and
Cortinez [13]. Vibration of thin-walled beams under initial stress
hasbeen studied by Machado and Cortinez [14].
Benscoter [15] developed an early theory for thin-walled beams
with closed cross-section, where thegeneral behavior becomes more
complicated. Significant progress has been made since this time,
includinga number of approaches for anisotropic beams, including
those of Hodges and co-workers [16] and Songand Librescu [17].
Summaries of much of this work have been given by Murray [18],
Librescu and Song[19], and Hodges [20] for beams of open and closed
section under both static and dynamic response.
1
-
Paul R. Heyliger / Thin-Walled Structures 00 (2014) 1?? 2
By far the most comprenensive studies of thin-walled beams have
used beam, plate, or shell theories inwhich displacement fields are
assumed over the domain of the plate that restrict the motion of
the displace-ment components [21]. By their own definition, the
displacement fields of all beam theories, thin-walledor not,
require some sort of kinematic restriction. This usually takes the
form of assumed componentsof representative displacement and/or
stress over the cross-section of the beam. This is intended to
allowa rapid analysis that captures the key mechanics of these
solids. Approximate beam models are usuallycompared with either
more detailed elasticity results or experimental evidence to assess
their accuracy.It is less typical to allow for a full
representation of the usual displacement components consistent
withthe three-dimensional theory of elasticity. In this study,
elasticity-based displacement fields are assumedover the
cross-section of both isotropic and anisotropic beams. This is
accomplished by applying the Ritzmethod directly to the
three-dimensional equations of elasticity using power-series
approximations in theform introduced by Visscher [22]. The intent
of this work is to determine the level of accuracy of
moresimplified models, provide an alternative means of analysis for
these solids, and present numerical resultsfor thin-walled beams to
provide a basis for future comparison.
Geometry and Governing Equations
We consider a three-dimensional solid whose cross-section
coordinates are defined in the (x1,x3) or (x-z) plane with a much
larger dimension in the x2 or y axis with a length of L. The beam
is assumed to becomposed of an anisotropic material whose principal
material axes are aligned with the (x,y,z) axes. Thismaterial is
assumed to be originally orthotropic and having gone through a
rotation in the x-y plane at anangle as measured from the positive
x axis (the horizontal component of the cross-section axis). The
beamis hollow with a uniform wall thickness of t.
The Weak Form
The governing equations used for this study are the
three-dimensional equations of linear elasticity withan anisotropic
constitutive tensor that can represent what amounts to a monoclinic
material. These are notsolved explicitly at each point in the
domain, but instead approximate solutions are sought for their
weakform as expressed within Hamiltons Principle [23]. This is
usually expressed as
0 = t
0
V{11 + 22 + 33 + 44 + 55 + 66}dVdt + 12
t0
V(u2 + v2 + w2)dVdt (1)
Here the conventional contracted notation has been used for
Cauchy stress components (11 = 1, 23 =4, 1 = 11, 4 = 223,
C1111=C11, C1123=C14, etc.) and it is understood that the 1, 2, and
3 directions are(x1 = x, x2 = y, and x3 = z).
The general constitutive law that links the Cauchy stress to the
components of infinitesimal strain i jcan be written as
i j = Ci jklkl (2)
where the strain components are linked with the three
displacement components u, v, and w in the x, y, andz directions
as
2
-
Paul R. Heyliger / Thin-Walled Structures 00 (2014) 1?? 3
1 = 11 =ux
2 = 22 =vy 3 = 33 =
wz
4 = 23 =vz
+wy
5 = 13 =uz +
wx 6 = 12 =
vx
+uy. (3)
The general constitutive relation for the materials used in this
study can be expressed in matrix form as
123456
=
C11 C12 C13 0 0 C16C12 C22 C23 0 0 C26C13 C23 C33 0 0 C360 0 0
C44 C45 00 0 0 C45 C55 0C16 C26 C36 0 0 C66
123456
(4)
These stiffness components are associated with either a
naturally occurring monoclinic material with 13independent elastic
constants or a rotated orthotropic material. In the latter case,
the final Ci j as listed in the6x6 matrix are a function of the
on-axis elastic constants and the angle of rotation of originally
orthotropicproperties about the 3 or z axis. Their explicit forms
are given in Reddy [24].
The Ritz Model
The computational approach used in this study relaxes many of
the assumptions used in most beamtheories. Rather than introducing
specific displacement fields over the beam domain, the three
point-wisedisplacement components are approximated directly using
the three-dimensional equations of elasticity.There is no need to
represent any rotational variables or their equivalent in this
theory, since each of thedisplacements is computed at every
location in the beam.
The general form for the displacements can be expressed as
ui(x j, t) = uio (x j) +
Np=1
Aip(t)uip (x j) (5)
Here the Aip represent unknown constants that depend on time.
The three values of i indicate that there arethree components of
displacement and that three independent sets of functions can be
used to approximateeach of those displacements. The o terms
represent the simplest functions that satisfy the essential
bound-ary conditions for that displacement direction. In the study
discussed in this paper, the initial boundaryconditions will be
assumed to be zero, with no initial displacement or velocity. The o
terms are there-fore all equal to zero for the considered case. In
these approximation equations, p represents a selectedapproximate
function for each respective direction. These functions must
satisfy general requirements ofindependence as outlined by Reddy
[23]. By using a large number of these approximation terms for
eachdisplacement component, very accurate solutions can be
determined for the mode shapes and frequenciesof vibration for a
beam.
In this study, polynomial series are used over the cross-section
of the beam and then combined with otherpolynomials or
transcendental functions over the axial length of the beam. This
class of approximation hasa long and rich history in the field of
mechanics. Among the first to use it for the case of vibrating
solidswas Demarest [25], who used Legendre polynomials over the
three Cartesian coordinate directions to find
3
-
Paul R. Heyliger / Thin-Walled Structures 00 (2014) 1?? 4
the natural modes of vibrating parallelepipeds. This method was
further studied by Ohno [26], who alsodiscussed the splitting of
the resulting eigenvalue problem into a smaller number of
equivalent eigenvalueproblems using group theory. Polynomial
expansions were also used in general cylinder [27] and sphere[28]
vibrations. A significant and powerful expansion of this approach
was introduced by Visscher [22],who expanded the three displacement
components in terms of power series regardless of the shape of
thesolid. This is generally written as
up(x, y, z) = xiy jzk (6)
This is a very attractive family of approximations in that it
frequently yields very simple integration resultsover the domain of
the solid. The nature of this integration will be discussed for
specific cases in the sectionsthat follow. This type of
approximation has seen extensive application in resonanting solids
[29], vibratingbeams [30], and related areas such as granular media
[31]. In terms of applications to solids with moregeneral beam
geometry, Carrera and co-workers have recently developed a unified
beam theory using amodification of this type of approximation over
the beam cross section and have extensively applied it to awide
array of problems [32, 33].
If the displacement constants cip are replaced with the letters
c, d, and e for the three displacements(1=u, 2=v, 3=w) for any
index p, it is possible to generalize the matrix form of the Ritz
model applied toHamiltons principle following the assumption of
general harmonic motion. Specifically, the Ritz coeffi-cients can
be written as
A1p(t) = cp sint (7)
A2p(t) = dp sint
A3p(t) = ep sint
(8)
Substitution of these conditions and the general Ritz
approximation into Hamiltons principle leads to thegeneralized
eigenvalue problem that can be represented in matrix form as
[K11] [K12] [K13]
[K21] [K22] [K23][K31] [K23] [K33]
{c}{d}{e}
= 2 [M
11] 0 00 [M22] 00 0 [M33]
{c}{d}{e}
(9)The elements of the stiffness ([K]) and mass ([M]) matrices
are given in the Appendix.
In the specific cases that follow, the (x,z) plane is taken as
the cross-section of the beam and the y-axis isthe longitudinal
axis of the beam. For the most part, the approximation functions in
the y-direction appearas power series. For certain types of support
conditions this may require some adjustment or replacement.This
will be discussed for beams under simple support studied in the
next section.
Computationally, using global power series over the
cross-section greatly simplifies the integration re-quired for the
relevant coefficient matrices. For hollow sections, for example,
the integration can be com-pleted first over the outer bounding
dimensions of the solid and then subtracting the same integral
evaluatedover the inner ounding surfaces of the cross-section. The
operations for rectangular sections are straightfor-ward. For the
elliptical section, the integration requires a trigonometric
substitution that can be evaluated inclosed form for any powers of
x and z [31].
4
-
Paul R. Heyliger / Thin-Walled Structures 00 (2014) 1?? 5
One additional advantage of using global approximation functions
over the beam cross-section andlength is the potential to group the
specific approximations according to the physical character of the
dis-placement components they are being used to represent [26, 34].
Various conditions of cross-sectionalsymmetry, axial wavelength,
and grouping of the various approximation functions can drastically
reducethe size of the matrix eigenvalue problem. Rather than
generalizing this discussion, these features are dis-cussed for the
specific geometries for which they are used in the examples that
follow.
Beams with Simple Support: Semi-Analytic FEM Model
In cases where the axial (or y) dependence of the displacement
functions are known, the form of theapproximation functions can be
modified to reduce the dimension of the problem and fix the axial
depen-dence as a given quantity. In other words, when the axial
dependence is fixed the problem becomes two-dimensional in the
cross-sectional variables (x,z). This allows for the introduction
of a second independentmethod of analysis with which results of the
present model can be compared. In the case of some beamsunder the
end conditions of simple support beam, most beam theories usually
require that the compponentsof transverse displacement are zero at
the end-points and also that the resultant moment quantities are
equalto zero. Using the elasticity theory of the present model,
identical displacement conditions are imposed atthe endpoints of
the beam for all points over the cross-section at all times. This
implies that u(x,0,z,t) =u(x,L,z,t) = w(x,y,0,t) = w(x,L,z,t) = 0.
In addition, the traction vector component along the axis ty = yyny
= 0. For the types of materials considered in this study, each of
these conditions is satisfied for beams upto and including
orthotropic symmetry (C16 = C26 = C36 = C45 = 0) using the
approximations
u(x, y, z, t) =nj=1
U j(t)uj(x, y) sin ky
v(x, y, z, t) =nj=1
V j(t)vj(x, y) cos ky
w(x, y, z, t) =nj=1
W j(t)wj (x, y) sin ky
(10)
Here k is the wave number and the capital letters for the field
variables are each spatial constants asso-ciate with the individual
subscript j, each of which correspond to independent shape
functions as denotedby and each of which can still vary with time.
When combined with conventional finite element approx-imations in
two dimensions, they represent the nodal values of the displacement
components. As for thegeneral Ritz method, the superscript on each
function indicates that the approximation functions for each ofthe
variables need not be the same. Depending on the geometry of the
problem each approximation functioncan take a specific form. The
variations of the primary field variables within Hamiltons
Principle are thentaken as
u = ui (x, y) v = vi (x, y) w =
wi (x, y) (11)
Although solutions can be determined for the general transient
case, the focus in this study is on the calcu-lation of periodic
frequency for a given wave number k. Hence it is further assumed
that the time dependent
5
-
Paul R. Heyliger / Thin-Walled Structures 00 (2014) 1?? 6
behavior for the spatial constants is in the form given by
general periodic motion as
U j(t) = U j sint V j(t) = V j sint W j(t) = W j sint (12)
Here t is time and is the natural frequency of free vibration.
Substitution of these functions and collectingterms allows the weak
form from Hamiltons principal to be expressed in matrix form in a
manner nearlyidentical to the original Ritz method:
[K11] [K12] [K13]
[K21] [K22] [K23][K31] [K23] [K33]
{U}{V}{W}
= 2 [M
11] 0 00 [M22] 00 0 [M33]
{U}{V}{W}
(13)There are two primary differences between this form and that
of the general Ritz model that is the focus ofthe present study: 1)
the integration need occur only over the beam cross-section, and 2)
the constants (in thiscase the nodal variables) exist only over a
single cross-section. This drastically reduces the
computationalsize of the problem. This model is used as a method of
comparison for the beams with simple support inthe sections that
follow.
Results and Discussion
In the results that follow, the present method is labeled 3DRE
to represent the three-dimensional Ritzelasticity structure of the
analysis. This is used in all tabulated results for the examples
considered. Severalgeneral shapes are considered. The first are
somewhat standard beam cross-sections that are not
necessarilythin-walled but are useful to study the accuracy of the
present 3DRE model.
General Sections: The Reduction to Basic Beam Theories
One feature of Ritz-based methods based on power-series
approxmations of the displacement compo-nents is that it allows a
simple reduction to capture the behavior of many elementary
theories. Han et at.[35] have published a complete assessment of
the Euler-Bernoulli, Rayleigh, and Timoshenko beam the-ories under
a variety of boundary conditions. These results can be then
directly compared to the currentelasticity-based Ritz models to
show the agreement between approaches.
In a power-series representation of the displacement vector, the
displacement dependence on the cross-section coordinates x and z
can be restricted to exactly match a number of beam theories for
sectionsthat are symmetric about either of these axes (it is
assumed for this section only that the cross-sectionsare symmetric
about the z-axis): The Rayleigh theory can be recovered by setting
the Poisson ratio usedto calculate the elastic stiffnesses Ci j to
zero, setting i = k = 0 as a single term approximation for
thedisplacement components u and w for all y, and keeping only the
approximation i = 0 and k = 1 as theapproximation for v in Equation
6. In addition, the shear modulii are increased by 5-6 orders of
magnitudeto reduce the influence of shear deformation. The
Timoshenko theory can be simulated by using the samerestrictions as
for the Rayleigh theory but using the exact shear modulii to
account for first-order sheardeformation and using an appropriate
shear correction coefficient.
The clamped-free beam is a very common configuration that has
been extensively studied for a varietyof geometries. Han and
co-workers [35] have given formulas for the characteristic
frequency equations.The frequencies can be computed for any range
of input parameters. For the Rayleigh theory, these includethe
modulus of elasticity E, the density , the length L, and the
cross-sectional area A. For the Timoshenko
6
-
Paul R. Heyliger / Thin-Walled Structures 00 (2014) 1?? 7
model these parameters are required in addition to the shear
modulus G and the value of the shear coefficientk. This latter
value has been the subject of intense study but there are fairly
well accepted values for typicalcross-sections and noted by Han
[35].
As an example, a cantilever beam was studied whose relative
dimensions were 1 x 1 x 10 in the cross-section and length
directions. Two shapes were considered: the square and the circle.
The Rayleigh andTimoshenko approximations were used over the
cross-section and so the Poisson ratio was assumed tobe zero. Along
the axis of the beam, power series in y (y1, y2, and so on) were
used for each of thedisplacement components to denote the
requirement that all displacements are equal to zero at the
clampedend where y=0. For each cross-sectional shape, two
additional refinements were considered: the solidsection and the
thin-walled closed section where the wall thickness is 1/100 of the
outer dimension. Theresults gave agreement within four or five
figures between the constrained elasticity theory developed hereand
the Rayleigh and Timoshenko theories. The values of the shear
coefficient [36, 37] were taken fromHan [35] who had summarized the
work of Cowper [38].
Open Sections
The specific open thin-walled sections considered are the
unsymmetric channel shown in Figure 1 andthe symmetric semicircle
of Figure 2. Both of these sections have seen extensive study, and
the goal of theirinclusion here is to evaluate the accuracy of
earlier theories and models within the context of the
presentelasticity solution.
The Doubly Unsymmetric Channel
The doubly symmetric section shown in Figure 1 was originally
studied by Yaman [39] and has beenextensively used by a number of
other researchers. It is an excellent example of a doubly symmetric
sectionas there are no mirror planes that intersect the x-y
cross-sectional plane. The material properties used werean elastic
modulus E of 70 GPa, a shear modulus G of 26 GPa, and a density of
2700 km/m3. Yamanoriginally gave only the integrated section
properties rather than the actual cross-sectional dimensions.
Thelatter are needed to integrate the power series expansions using
the present elasticity model. Since theseproperties were not given
by Yaman, the integrated quantities were used to compute the leg
dimensionsand thickness by minimizing the absolute difference
between the area and the elements of the inertia tensor.Using this
approach, the Yaman beam was determined to have a constant wall
thickness of t = t2 = 1.275 (allmm), d1 = 13.29675, d3 = 25.9616,
and d2 = 39.375. This gives A = 9.68 (9.68) x 105 m2, Ixx =
5073.64(5080) x 106 m4, Iyy = 22400 (22400) x 106 m4, and Ixy =
4256 (4250) x 106 m4 where the values usedby Yaman have been given
in the parentheses. This gives a total absolute difference of 13.23
x 106 m4 forthe 3 values of the inertia tensor [40].
The level of agreement between competing formulations in regards
to this cross-section have beensomewhat mixed. Ambrosini [41] has
provided a recent summary and discussion regarding the results
ofseveral researchers who have considered this cross-section,
including the work of Arpaci and Bozdag [42]and Tanaka and Bercin
[43]. Much of this discussion centers around the use of various
elements of theinertia tensor of the beam cross section and other
features related to the beam theories used to study thebeam
behavior. This discussion does not apply to the present method
since the actual cross-section is usedalong the the
three-dimensional elasticity tensor components. But Ambrosini [41]
notes the differences inthe lower frequencies among these various
studies.
To assess the overall accuracy of the present method, the lowest
four frequencies of the simply-supportedbeam were computed and
compared with the values of Ambrosini [41] and those of Yaman [39]
along withthe results from the two-dimensional semi-analytic finite
element model and a fully three-dimensional finiteelement model. In
the case of the two-dimensional finite element model, the wave
number k can be directly
7
-
Paul R. Heyliger / Thin-Walled Structures 00 (2014) 1?? 8
input to extract the appropriate natural frequencies under the
conditions of simple support. For the lowestfour modes, three of
the frequencies are associated with the wave number k=pi. This is
consistent with modalvibration patterns that possess variations
along the axial length that are symmetric about the midpoint ofthe
beam. The third lowest frequency is associated with the lowest
anti-symmetric mode where k=2pi. Byanti-symmetric we mean about the
mid-point of the beam. The mesh used to represent the cross-section
forboth the two and three-dimensional finite element models has,
along the three legs, 4, 12, and 8 elementdivisions and four
element divisions through the thickness. The two corner regions
therefore have 4 x 4meshes. A total of 128 elements with 165 nodes
are used to model the complete cross-section. The natureof the
semi-analytic approximation ensures that all boundary conditions
are explicitly satisfied at the end-points of the beam. For the
three-dimensional model, this same cross-section is used along with
40 elementdivisions along the length of the beam. Symmetry
conditions were used so that only one-half of the beamlength is
modeled. Hence the the boundary conditions at the simply-supported
end and the mid-point of thebeam for the symmetric modes of the
three-dimensional model are given by
u(x, y, 0) = v(x, y, 0) = zz(x, y, 0) = 0 (14)
w(x, y, L/2) = xz(x, y, L/2) = yz(x, y, L/2) = 0 (15)
As is usually the case in standard finite element formulations,
the boundary conditions on the stress compo-nents are not satisfied
pointwise but rather in an integral sense over the entire
cross-section of the beam.
The results of these analyses are given in Table 1. There is
good agreement between the various methodsfor the lowest frequency
and also good agreement between the present methods, but
significant variationsin the remaining frequencies. The modal
shapes for the half-beams are also shown in Figures 3a-d.
The Thin-walled Semi-Circle
Friberg [44] was the first to study the vibration of the
cantilevered beam shown in Figure 2 using Vlasovbeam theory. Since
that time numerous researchers have also studied this particular
cross-section, includingJun [45, 46], de Borbon and Ambrosini [47],
and Petrolo and co-workers [48]. This beam section has
someadvantages over the previous section in that the vibrational
modes for isotropic beams can be split intogroups that reflect
symmetry or asymmetry about the vertical axis that bisects the
section. These two modalgroups have displacement components that
are either even or odd about this axis. For symmetric modes, uis
odd while v and w are even. For antisymmetric modes, u is even
while v and w are odd about x. Theproperties of this section
include an inner diameter di = 45 mm, a wall thickness of 4 mm, a
length of 820mm, an elastic modulus of E = 68.9 GPa, a shear
modulus of 26.5 GPa, and a density of 2711 kg/m3.
Two types of support conditions were considered: the fixed-free
beam originally studied by Friberg [44]and the simply-supported
beam of the same cross-section considered by Jun [45, 46] and de
Borbon andAmbrosini [47]. In the present model, 3 terms were used
in the x and y directions for the three variables(these terms
include only those represented by the appropriate symmetry group)
and 8 axial terms for thefixed-free beam. For the simply-supported
beam, only a single axial term was used using the
trigonometricapproximations from the semi-analytic model. This
explicitly represents the appropriate boundary condi-tions and
reduces the problem to being purely two-dimensional. The axial
response is identified by the wavenumber k. Six in-plane terms were
used for each of the displacement variables. For both support
conditions,the frequencies were computed using both the full set of
elastic constants and then repeated using the elastic
8
-
Paul R. Heyliger / Thin-Walled Structures 00 (2014) 1?? 9
constants that result with the assumption of zero Poisson
ratio.For the fixed-free beam, the results are shown in Table 2. As
a simple comparison, the bending frequen-
cies were also computed using the Rayleigh beam theory using the
approach of Han [35]. These numericalvalues are the same to four
significant figures with those computed by Friberg using the Vlasov
beam theory[44]. The frequencies computed using the Timoshenko
model of Jun [45] are in very good agreement withthe values
obtained using the assumption of zero Poisson ratio. There is some
discrepancy with the results ofPetrolo and co-workers [48], which
is mildly surprising given that their approch uses similar
power-seriesapproximations over the cross-section, albeit via forms
incorporated into finite element approximations.The largest
difference in frequency comes in comparison with the bending
frequencies of the present modelusing the two constitutive
theories. Those for zero Poisson ratio, which is a common
assumption in manybeam theories, are consistently five to ten
percent lower than those computed using the true elastic
stiffnesstensor. The torsional frequencies are also influenced but
to a much lesser extent.
The results for the simply-supported beam are given in Table 3,
for which the main comparitive studyis the work of Jun [45, 46]. As
before, the Rayleigh theory results can be computed using the
methodologyof Han for the simply-supported beam [35] and these
results are in good agreement with the Timoshenkoresults of Jun and
the results of de Borbon and Ambrosini [47] with somewhat less
agreement with thework of Petrolo and co-workers [48]. The present
model gives frequencies that are close to those of Jun butare
consistently lower by up to a single percentage. Additionally,
there is little change in results using thepresent approach when
the full elastic constants are used in the calculations.
Plots of the modal shapes along the length of the beam as
measured at the outer diameter/thickness of thesection are shown in
Figures 4-9 for the first three bending and torsional modes. The
bending frequenciesare those for which Rayleigh frequencies are
computer and hence give modal patterns dominated by thevertical or
transverse displacement component v. The torsional frequencies are
dominated by the in-planedisplacement components u and v but with a
smaller warping motion out of the plane.
Closed Sections
In this section, two representative examples are studied that
have been analyzed using other methods ofapproximation: the hollow
monoclinic rectangular cantilevered beam and a homogeneous hollow
ellipticalbeam under simple support. Both of these solids have
beam-like behavior and have been studied by at leastone kinematic
model that restricts the displacement field. The rectangular
section has relatively thick wallswith anisotropic constitutive
behavior. The elliptical section has isotropic properties with
relatively thinwalls. Hence they provide a good range of comparison
with the present model that does not restrict thenature of the
displacement field other then truncating the number of terms in the
series solution.
The Hollow Rectangle
Song and Librescu [17] considered the free vibrations of a
hollow orthotropic thin-walled beam com-posed of an orthotropic
solid that has been rotated about a single axis and is fixed at one
end. In thenomenclature of this paper, the fibers always lie in the
x-y plane with the rotation angle being measuredbetween the +x and
+y axis. When = 0, the fibers are perpendicular to the long axis of
the beam, andwhen = pi/2, the fibers are then along the axis of the
beam. The material properties used are those of Songand Librescu
and are taken as E1 = 206.843 GPa (30 x 106 psi), E2 = E3 = 5.171
GPa (0.75 x 106), 12 = 13= 23 = 0.25, G13 = G23 = 2.551 GPa (0.37 x
106 psi), and G12 = 3.1026 GPa (0.45 x 106. Here both SI andUS
units have been given as the latter were used in the original study
[17]. The original outer dimensionsof the rectangular section were
given as 0.026416 m (1.04 inches) in the x-direction by 0.006096 m
(0.24inches) in the y-direction with a thickness of 0.001016 m
(0.04 inches) and a length of 0.254 m (10 inches)
9
tiganiHighlight
-
Paul R. Heyliger / Thin-Walled Structures 00 (2014) 1?? 10
in the y-direction. The beam is fixed at y=0. This implies that
each of the displacement components is zerofor all (x,z) at y=0.
The density was taken to be 1528.15 kg/m3 (14.3 x 105
lb/s2/in4).
Song and Leissa incorporated a kinematic model that enforces
zero deformation in the cross-sectionalplane, allows for both
primary and secondary warping, and assumes constant transverse
shear strain. Sucha model is very useful in terms of computational
effort but is likely to result in an overly stiff system
thatoverestimates the bending frequencies. Part of the objective of
this work is to assess such limits.
For the present model, the accuracy of the results is linked to
the number of terms in the series usedto represent each of the
three displacement components. As the number of terms increases,
the computedfrequencies get lower and more accurate until an
increase in the number of terms has little influence on theresults.
For this geometry, an appropriate number of terms was assessed by
computing the values of thelowest six frequencies for the case of =
0. It was found that a combination of nine axial terms
(beginningwith y1 since the displacements must be zero at the fixed
support) and five in-plane terms provides sufficientaccuracy for
all of the results that follow.
For the extreme cases of = 0 and = pi/2, this beam configuration
has axial, torsional, and flexuralmodes that completely uncouple.
Group theory could have been used to drastically reduce the size of
theunknowns for this problem by exploiting the symmetry of the
resulting displacements. However, this wasnot used to maintain the
same number of degrees of freedom for random angles of fiber
orientation.
The axial frequencies of a bar that correspond to zero Poisson
ratio approximations and a simplifieddisplacement field are given
by [49]
n =(2n + 1)pic
2L(16)
Here c is the one-dimensional longitudinal wave speed given
byE/. E is used to denote the elastic
modulus in the direction of the longitudinal wave and L is the
length of the bar. For = 0, E = E2 and for = 90, E = E1. Similarly,
the bending frequencies in the transverse (w-dominant) and lateral
(u-dominant)directions can be approximated by the Rayleigh beam
theory [35]. These assume axial displacements thatare linear in
either z or x, respectively, and also incorporate the rotary
inertia of the cross-section in thekinetic energy term. However,
they do not include any shear deformation. These calculations are
moreinvolved than using a single formula, but the frequencies can
be easily calculated using the procedureoutlined by Han and
co-workers [35]. Once again, the values of modulus of elasticity
and second momentof area required by the Rayleigh theory are
obvious depending on the direction of bending and can becomputed
for the two extremes for the angle of fiber orientation.
The axial frequencies are given in Table 4 for the cases of = 0
and 90 degrees for the one-dimensionaland Song and Librescu
theories, and for 15 degree increments for the present model for
both the first andsecond modes. For all angles other than 0 or 90,
the axial modes are not uncoupled from the rest of themotion.
However, the modal displacements are the largest of all components
and are generally constant overthe beam cross-section making them
easy to identify. Two results are notable. First, there is a
surprisingfour percent disagreement between the result of Song and
Librescu for the first axial mode when = 0. Thismode is dominated
by the elastic modulus of the matrix material, and it is odd that
the theory of Song andLibrescu gives such a noticeable difference.
The second result of note is the strongly nonlinear
relationshipbetween the axial frequency and the angle of
orientation of the fibers. Changing the angle from 0 to 45degrees
results in under 25 percent increase in frequency, but an
additional 45 degrees of rotation result inan increase in frequency
of over a factor of 6. Hence the action of the fibers is strongly
delayed until thefinal 15-20 degrees of rotation until they align
with the axial motion.
10
-
Paul R. Heyliger / Thin-Walled Structures 00 (2014) 1?? 11
The torsional frequencies are given in Table 5. The uncoupled
mode for = 90 is given by Song and Li-brescu for two conditions
associated with neglecting or enforcing warping restraint within
their theory. Thelatter case results in higher frequencies, which
are in good agreement with the present model. These resultsalso
indicate that the torsional stiffness for this material peaks near
60 degrees for the limited increments ofthis analysis.
Of most interest are the flexural modes since they are usually
the easiest to excite and possess the lowestfrequency. These modes
are classified according to their dominant displacement motion
using the phrasestransverse bending modes and lateral bending
modes. These incorporate flexural motion for which thedisplacements
in the plane of the cross-section are generally in the z
(transverse) or x (lateral) directions.Once again, for fiber angles
of 0 and 90 degrees the flexural motion uncouples from the
torsional and axialmodes, but for other angles there is significant
rotation of the cross-section coupled with the flexural motionalong
the lengths of the beam. The first three transverse modes are given
in Table 6 and the first two lateralmodes are given in Table 7. The
results of Song and Librescu are compared with the present model
for 15degree increments, and values for the Rayleigh theory are
also given for purposes of comparison for thelimiting cases of 0
and 90 degrees.
As expected, the resulting frequencies from the full equations
of elasticity are lower than those of theone-dimensional beam model
of Song and Librescu. The agreement for the lowest transverse mode
is quitegood, with the beam theory results being within six percent
of the elasticity results for all fiber orientations.The second
transverse mode has good agreement for the lower fiber angles but
overestimate the elasticityresults by over twenty percent as the
fibers approach the axis of the beam.
The lateral frequencies are in good agreement for fiber
orientations near 0 or 90 degrees, but for otherorientations are as
much as 50 percent higher than the elasticity results. The second
lateral modes are moreproblematic. Both the Rayleigh theories and
elasticity theory give a frequency of over 5000 rad/sec forthis
mode when the fibers are oriented at 0 degrees. The results of Song
and Librescu have a frequency of4331 rad/sec. This pattern
continues for all fiber orientations in that the results of Song
and Librescu areroughly 10-15 percent under the elasticity results.
This surprising result may be explained if the secondlateral mode
identified by Song and Librescu was actually the third transverse
bending mode, in which casethe comparison is far more in line with
the elasticity results given in Tables 6 and 7.
Typical modal shapes for two representative fiber orientations
are shown in Figures 10-13. The lowesttwo torsional modes are shown
in Figure 10, and demonstrate the coupled nature of the deformation
asthe fiber angle goes from 0 degrees to 60 degrees. In the former
case, the torsional motion is dominantand virtually exclusive. As
the fiber angle changes, the torsional motion still dominates the
nature of thecross-sectional deformation but there is far more
inclusion of deformation that is not strictly torsional
thatcontributes to the stored energy and explains in part the
increase in frequency for orientation angles between0 and 90
degrees. Similar behavior is demonstrated in Figures 11 and 12,
which show the lowest twotransverse bending modes. For = 0, the
motion is purely flexural. For other fiber angles, there is
combinedvisible twist of the cross-section along with the dominant
bending action. In the case of bending, thefibers are being
reoriented to change what amounts to the longitudinal modulus of
the beam, which alsoincreases the frequency. Only in the case of
the lowest two lateral modes shown in Figure 13 do the modesstay
visually uncoupled. This reflects the fact that the fibers are in
the plane of the primary cross-sectiondeformation rather than
perpendicular to the motion as is the case of transverse
bending.
The original material properties of Song and Librescu were
somewhat unusual for a fiber-reinforcedmaterial in the the Poisson
ratios were identical and the shear moduli in the plane of the
fibers were unequal.To present additional results for another
material, the analysis with the Song and Librescu geometry
wasrepeated for the graphite-polymer material properties tabulated
by Hyer [50]: E1 = 155.0 (all GPa), E2 =
11
tiganiHighlight
-
Paul R. Heyliger / Thin-Walled Structures 00 (2014) 1?? 12
E3 = 12.10, 12 = 13 = 0.248, 23 = 0.458, G13 = G12 = 4.40, G23 =
3.20. A density of = 1600 kg/m3
was used, which is typical for this class of material. The
results of this analysis are given in Table 8, withthe values given
for the first six frequencies for fiber orientation angles of
pi/12. These are mainly given forpurposes of comparison and to
serve as a benchmark for other computational models.
One crucial difference between the results using the material
properties of Hyer and the general behaviorof Song and Librescu is
the nature of the change in frequency of the primary torsion mode
as the fiberorientation angle changes. For the material properties
of Song and Librescu, this frequency peaked at =45 degrees with an
increase in frequency of about 25 percent. For the properties of
Hyer, this mode peaksat about 66 degrees with a frequency of 8780
rad/sec. This is a 73 percent increase over the value at =0. When
the fibers align with the global x or z coordinates ( = 0 or = 90,
respectively), the frequencydepends almost exclusively on the shear
modulii in the 23 plane (a lower value) and then the 13 plane
(ahigher value).
The Hollow Ellipse
The vibration of hollow shells of elliptic cross section, often
referred to as elliptic-cylindrical (EC)shells when the wall
thickness is thin, have a number of practical applications even
though the numberof comparative studies is relatively small. For
shells with constant thickness, Sewall and co-workers [51]provided
both experimental and analytical results and Yamada and co-workers
[52] gave numerical resultsusing classical shell theory. More
recently Hayek and Boisvert [53] studied similar EC shells with a
higher-order shell theory.
The shell geometry and constitutive properties considered in
this section matches that of of an identicalshell studied by both
Yamada and co-workers [52] and by Hayek and Boisvert [53]. The
ratio of the semi-major axis length a to that of the semi-minor
axis length of b is given as a/b=2.5, and the ratio L/a=2.198,where
L is the length of the beam. The ratio of the shell thickness h to
a is given by h/a=0.007325. Both endsof the cylinder are under
simple support. Two cases are considered: the isotropic shell and
the orthotropicshell.
A general spatial solution to this class of motion can be
expressed as
u(x, y, z) =m=1
Um(x, z) sinmpiyL
v(x, y, z) =m=0
Vm(x, z) cosmpiyL
w(x, y, z) =m=1
Wm(x, z) sinmpiyL
These displacements satisfy the boundary conditions at both ends
of the shell and, unlike the cantileversection, split the axial and
cross-sectional dependencies of the problem. This allows
frequencies to be de-termined for specified values of m. For this
specific problem, the dimensionless frequencies are sought forthe
isotropic case when the Poisson ratio = 0.3 and m=1. The are
computed as dimensionless frequencies,and are given by [53]
2 =2a2(1 2)
E(17)
12
tiganiHighlight
-
Paul R. Heyliger / Thin-Walled Structures 00 (2014) 1?? 13
where E is the modulus of elasticity.As was the case for the
hollow rectangle, the numerical integration over the thin walled
section can be
accomplished in closed form by evaluating the integrals of the
coefficient matrices for the outer semi-axislengths and then
subtracting similar integrals over the inner semi-axis lengths.
Since this section is isotropic, the vibrational modes can be
separated into various modal groups thatisolate the nature of the
motion. In the studies of others, the displacement functions over
the beam cross-section were explicitly separated in terms of
trigonometric functions that describe the circumferential
vari-ation around the circumferential coordinate . This effectively
makes the problem one-dimensional in thecircumferential direction
of the cross-section mid-plane. Both Yamada [52] and Hayak [53]
reported theirresults as a function of either symmetric or
antisymmetric modes along with the circumferential wave num-ber n
in the above expansion. In the present analysis, the frequencies
are grouped according to the symme-tries of the three displacement
components (u,v,w) in the cross-section (x,z) directions. A summary
of thesegroups is given in Table 9, which enforces antisymmetry
about an axis by the symbol O, indicating onlyodd polynomials for
that displacement component were given in that coordinate
direction. The numericalresults are presented in terms of these
four groups.
As was the case for the rectangular section, the number of terms
used to represent the displacementcomponents was adjusted until an
increase in number of terms did not yield a significant drop in
frequency.Since there is now a single axial term containing either
the sine or cosine axial dependence, only the cross-section
variables need adjustment. Approximations up to and including x9z7
for u, x9z8 for w, and x8z9 forv were used in the results that
follow.
The dimensionless frequencies for all four groups are tabulated
in Table 10 and compared with theresults of Yamada [52] and Hayak
[53] along with the two-dimensional finite element model
developedas part of the present study. These latter results were
obtained by meshing one quarter of the ellipticalcross-section with
4 divisions in the thickness direction and 541 elements along one
quadrant of the ellipse.These numbers were determined after both
halving and doubling the number of divisions with little changein
frequency. Hence there were a total of 2160 4-node elements and
2705 nodes used to represent thequadrant. Four different analyses
were ran, each of which had boundary conditions at the 0 and 90
degreeborders of the quadrant that matched those of the grouped
polynomials.
The results are all in very good agreement with the results from
the elasticity model giving, with onlyone exception, the lower
frequency. The average differences between the lowest 3 frequencies
from eachsymmetry group is 2.5, 2.8, and 2.1 percent between the
elasticity results and the higher-order shell theoryof Hayak [53]
with some of the frequencies from shell theory for the third modes
being closer to 4 percentdifference. This good agreement is to be
expected since the shell in this case is quite thin and hence
itskinematics should be captured by such a theory. Hence when the
beam is isotropic, the simplified shelltheory overpredicts the
elasticity frequencies by approximately three percent.
For orthotropic materials, the analysis was repeated using the
material properties of Hyer [50] listed forthe rectangular section.
The fibers are oriented along the beam axis ( = 90). In this case
the same symmetrygroups hold as for the isotropic solid. The
dimensionless frequencies in this case are given by
2 =2a2E1
(18)
These values are shown in Table 10 are are again compared with
the results from the semi-analytic finiteelement model. These
frequencies are significantly below those of the isotropic section.
This is partiallyinfluenced by the ratio of the longitudinal
modulus to that of the shear modulus. For the isotropic solid,
13
-
Paul R. Heyliger / Thin-Walled Structures 00 (2014) 1?? 14
this ratio is 2.6. For the orthotropic material, the ratio is
over 35. Hence the level of shear deformation isexpected to be
significantly higher than that of the isotropic solid, and any
simplified model would need totake this behavior into account.
Conclusions
An elasticity-based approximation scheme was used to develop
estimates for the frequencies of a widevariety of thin-walled
beams. In nearly every case, this method provided lower estimates
of all frequenciesstudied. The primary conclusions of the
application of this approach as applied to open and closed
sectionsare given below.
For open thin-walled sections:
1. The frequency results of the present model for the doubly
unsymmetric section of Yaman [39] are invery good agreement with
two and three-dimensional finite element results and the lowest
modes ofother researchers but are significantly lower for the
remaining modes.
2. For the cantilevered thin-walled semi-circle with an assumed
Poisson ratio of zero, the bending fre-quency results computed
using the elasticity-based model are similar to but slightly lower
than resultsfrom Vlasov and Timoshenko beam theories, the former of
which are exactly the same for the beamstudied as the results from
the Rayleigh theory. A similar pattern is true for torsional
modes.
3. For the cantilevered thin-walled semi-circle with a non-zero
Poisson ratio, the bending frequencyresults from the elasticity
model are slightly above (roughly five percent) those computed by
Vlasovand Timoshenko beam theories. There is little difference
between torsional frequencies for the presentmodel in terms of
influence of Poisson ratio.
4. For the simply-supported thin-walled semi-circle, the results
of the elasticity model are slightly belowthose of Vlasov and
Timoshenko model results regardless of the value of Poisson ratio
used. Theassumption of zero Poisson ratio in the elasticity model
developed here has a very small (well underone percent) influence
on the simply-supported frequencies.
5. The agreement between the present elasticity model used here
and a similar power series approachpresented by Petrolo and
co-workers [48] is mixed, and may have been influenced by either
truncatedapproximations or the influence of local axial polynomials
rather than the global polynomials used inthe present study.
For closed thin-walled sections:
1. For a hollow anisotropic rectangular section whose thickness
to cross-section length direction was1:6 for one wall and 1:21 for
another, good agreement was found for axial and torsional modes
whenfibers had 0 or 90 degree orientation angles between elasticity
theory and a refined beam theory.
2. The lowest (transverse) bending frequencies of the
rectangular section were also in good agreementfor all fiber
orientations, with a maximum difference of about 7 percent. The
second transversefrequencies had more significant differences that
increased with fiber orientation, starting at 4 percent( = 0) and
increasing to over 25 percent ( = 90).
3. The lateral bending frequencies of the rectangular section
were in good (under 4 percent difference)agreement for the cases
where the fibers were along or perpendicular to the beam axis, but
for otherangles of orientation the present method gave frequencies
that were significantly (over 50 percent)lower than those of the
one-dimensional beam model. It is possible that the second lateral
mode ofSong and Librescu [17] is actually the third transverse
mode.
14
-
Paul R. Heyliger / Thin-Walled Structures 00 (2014) 1?? 15
4. For the isotropic elliptical shell whose thickness to
cross-section dimension ratio was 1:136, excellentagremeent was
obtained with thin-shell theory results with most differences in
frequency being lessthan 3 percent.
5. The dimensionless frequencies of an orthotropic elliptical
beam are roughly a third of those for theisotropic beam, indicating
a dramatic loss in general stiffness even when the wall thickness
is small.
In addition to these conclusions, representative elasticity
results were presented for anisotropic (for therectangle) and
orthotropic (for the ellipse) thin-walled beams that can be used
for comparison with othersimplified models.
Acknowledgment
This work was sponsored by the Mountains-Plains Consortium. The
support is gratefully acknowledged.
15
-
Paul R. Heyliger / Thin-Walled Structures 00 (2014) 1?? 16
References
[1] Wagner H, Pretschner W. Verdrehung und Knickung von offenen
Profilen. Luftfahrtforschung 1934; 11: 174-180.[2] Bleich F and
Bleich H. Bending, torsion, and buckling of bars composed of thin
walls. Prelim. Pub. 2nd Congress of Interna-
tional Association for Bridge and Structural Engineers, English
edition 1936; 871.[3] Kappus R. Twisting failure of centrally
loaded open-section columns in the elastic range.
Luftfahrtforschung 1937; 14: 444-
457. Translated by J. Vanier, National Advisory Committee for
Aeronautics, No. 851 (1938).[4] Gjelsvik A. The theory of
thin-walled bars. John Wiley and Sons, New York, 1981.[5] Vlasov V.
Thin-walled elastic beams. 2nd ed. Jerusalem: Israel Program for
Scientific Translations, 1963.[6] Goodier NJ. The buckling of
compressed bars by torsion and flexure. Cornell University
Engineering Experiment Station,
Bulletin 27, 1941.[7] Timoshenko SP. Theory of bending, torsion
and buckling of thin-walled members of open cross section. Journal
of the
Franklin Institute 1945; 239: 201-219.[8] Gere, JM and Lin, Y.
Coupled vibrations of thin-walled beams of open cross-section.
Journal of Applied Mechanics, ASME
1958; 25: 373-378.[9] Banerjee JR, Guo S., and Howson WP. Exact
dynamic stiffness matrix of a bending-torsion coupled beam
including warping.
Computers and Structures 1996; 59: 613-621.[10] Mei C. Coupled
vibrations of thin-walled beams of open section using finite
element method. International Journal of Me-
chanical Sciences 2001; 12: 883-891.[11] Kollar, LP.
Flexural-torsional vibration of open section composite beams with
shear deformaion. International Journal of
Solids and Structures 2001; 38: 42-43.[12] Cortinez VH and
Piovan MT. Vibration and buckling of composite thin-walled beams
with shear deformability. Journal of
Sound and Vibration 2002; 258: 701-723.[13] Piovan MT and
Cortinez VH. Mechanics of shear-deformable thin-walled beams made
of composite materials. Thin-Walled
Structures 2007; 45: 37-62.[14] Machado SP and Cortinez VH. Free
vibration of thin-walled composite beams with static initial
stresses and deformations.
Engineering Structures 2007; 29: 372-382.[15] Benscoter SU. A
theory of torsion bending for multicell beams. ASME Journal of
Applied Mechanics 1954; 21: 25-34.[16] Hodges DH, Atilgan AR,
Fulton MV, and Rehfield LW. Free-vibration analysis of composite
beams. Journal of the American
Helicopter Society 1991; 36: 36-47.[17] Song O and Librescu L.
Free vibration of anisotropic composite thin-walled beams of closed
cross-section contour. Journal
of Sound and Vibration 1993; 167: 129-147.[18] Murray NW.
Introduction to the theory of thin-walled structures. Clarendon
Press, Oxford, 1984.[19] Librescu L and Song O. Thin-walled
composite beams: Theory and application. Springer, The Netherlands,
2006.[20] Hodges DH. Nonlinear Composite Beam Theory, AIAA, United
States, 2006.[21] Hajianmaleki M and Qatu MS. Vibrations of
straight and curved composite beams: A review. Composite Structures
2013;
100: 218-232.[22] Visscher WM, Migliori A, Bell TM, and Reinert
RA. On normal modes of free vibration of inhomogeneous and
anisotropic
elastic objects. Journal of the Acoustical Society of America
1991; 90: 2154-2162.[23] Reddy JN, Energy Principles and
Variational Methods in Mechanics. Wiley, New York, 2002.[24] Reddy
JN, Mechanics of Laminated Composite Plates and Shells: Theory and
Analysis. CRC Press, Boca Raton, 2004.[25] Demarest HH. Cube
resonance method to determine elastic constants of solids. Journal
of the Acoustical Society of America
1971; 49: 768-775.[26] Ohno I. Free vibration of a rectangular
parallelepiped crystal and its application to determination of
elastic constants of
orthorhombic crystals. Journal of the Physics of the Earth 1976;
24: 355-379.[27] Heyliger PR. Axisymmetric free vibrations of
finite anisotropic cylinders. Journal of Sound and Vibration 1991;
148: 507-
520.[28] Heyliger PR and Jilani A. The free vibrations of
inhomogeneous cylinders and spheres. International Journal of
Solids and
Structures 1992; 29: 2689-2708.[29] Migliori, A and Sarrao, JL.
Resonant ultrasound spectroscopy. Wiley, New York, 1997.[30]
Heyliger P, Ugander P, and Ledbetter H. Anisotropic elastic
constants: measurement by impact resonance. ASCE Journal of
Materials in Civil Engineering 2001; 13: 356-363.[31] Heyliger
PR. Ritz finite elements for curvilinear particles. Communications
in Numerical Methods in Engineering 2005; 22:
335-345.[32] Carrera E, Petrolo M, and Nali P. Unified
formulation applied to free vibrations finite element analysis of
beams with arbitrary
cross-section. Shock and Vibrations 2010; 18: 485-502.
16
-
Paul R. Heyliger / Thin-Walled Structures 00 (2014) 1?? 17
[33] Carrera E, Giunta G, and Petrolo M. Beam Structures:
Classical and Advanced Theories. John Wiley and Sons, New
York,2011.
[34] Mochizuki E. Application of group theory to free
oscillations of an anisotropic rectangular parallelepiped. Journal
of thePhysics of the Earth 1987; 35: 159-170.
[35] Han SM, Benaroya H, and Wei T. Dynamics of transversely
vibrating beams using four engineering beam theories. Journalof
Sound and Vibration 1999; 225: 935-988.
[36] Timoshenko SP. On the transverse vibrations of bars of
uniform cross-section. Philos Mag 1922;125:125-131.[37] Timoshenko
SP. On the correction for shear of the differential equation for
transverse vibrations of bars of uniform cross-
section. Philos Mag 1921;41:744-746.[38] Cowper, GR. The shear
coefficient in Timoshenkos beam theory. J Appl Mech 1966;
33:335-340.[39] Yaman Y. Vibrations of open-section channels: a
coupled flexural and torsional wave analysis. Journal of Sound and
Vibration
1997; 204: 131-158.[40] Yaman, Y., personal communication
(2014). The original dimension of the thickness was 0.05 inches and
the centerline leg
lengths of the channel were 0.5, 1.5, and 1 inches. These values
are extremely close to the full dimensions of the cross-sectionused
for purposes of comparison in this study.
[41] Ambrosini D. On free vibration of nonsymmetrical
thin-walled beams. Thin-Walled Structures 2009; 47: 629-636.[42]
Arpaci A and Bozdag E. On free vibration analysis of thin-walled
beams with nonsymmetrical open cross-sections. Comput-
ers and Structures 2002; 80: 691-695.[43] Tanaka M and Bercin A.
Free vibration solution for uniform beams of nonsymmetrical
cross-section using Mathematica.
Computers and Structures 1999; 71: 1-8.[44] Friberg PO. Beam
element matrices derived from Vlasovs theory of open thin-walled
elastic beams. International Journal
for Numerical Methods in Engineering 1985; 21: 1205-1228.[45]
Jun L, Wanyou L, Rongying S, and Hongxing H. Coupled bending and
torsional vibration of nonsymmetrical axially loaded
thin-walled Bernoulli-Euler beams. Mechanics Research
Communications 2004; 31: 697-711.[46] Jun L, Rongying S, and Jin X.
Coupled bending and torsional vibration of nonsymmetrical axially
loaded thin-walled Timo-
shenko beams. International Journal of Mechanical Sciences 2004;
46: 299-320.[47] de Borbon, F and Ambrosini D. On free vibration
analysis of thin-walled beams axially loaded. Thin-Walled
Structures 2010;
48: 915-920.[48] Petrolo M, Zappino E, and Carrera E. Refined
free vibration analysis of one-dimensional structures with compact
and bridge-
like cross-sections. Thin-Walled Structures 2012; 56: 49-61.[49]
Rao SS. Vibration of Continuous Systems. Wiley, New Jersey,
2007.[50] Hyer MW. Stress Analysis of Fiber-Reinforced Composite
Materials. McGraw-Hill, Boston, 1998.[51] Sewall JL, Thompson Jr
WM, and Pusey CG. An experimental and analytical vibration study of
elliptical cylindrical shells.
NASA TN D-6089, 1971.[52] Yamada G, Irie T, and Notoya S.
Natural frequencies of elliptical cylindrical shells. Journal of
Sound and Vibration 1985;
101: 133-139.[53] Hayek SI and Boisvert JE. Vibration of
elliptic cylindrical shells: higher order shell theory. Journal of
the Acoustical Society
of America 2010; 128: 1063-1072.
17
-
Paul R. Heyliger / Thin-Walled Structures 00 (2014) 1?? 18
Table 1. Frequencies of the doubly unsymmetric channelMode
Frequency
3D FEM 2D FEM [41] [39] 3DRE1 50.59 47.38 (k=pi) 51.39 51.87
48.72 (k=pi)2 66.98 64.60 (k=pi) 96.48 114.79 68.39 (k=pi)3 144.5
(k=2pi) 188.84 207.41 151.5 (k=2pi)4 159.5 158.5 (k=pi) 204.81
263.88 161.0 (k=pi)
Table 2. Frequencies of the cantilevered thin-walled
semi-circular beamMode Frequency
[44] Rayleigh [45] 3DRE (=0) 3DRE [48]1 31.80 31.80 31.8 31.77
33.47 31.002 63.76 63.79 63.66 64.04 67.333 137.5 137.7 137.1 137.8
357.84 199.0 199.0 199.3 197.5 207.8 193.35 278.2 278.4 276.6 278.0
593.86 483.9 480.7 483.87 556.3 556.3 558.1 545.9 573.8 533.08
657.3 637.9 642.09 767.5 759.7 765.710 1075. 953.2 953.211 1087
1051.5 1103.1
Table 3. Frequencies of the simply-supported thin-walled
semi-circular beamMode Frequency
[45] Rayleigh [46] [47] [48] 3DRE ( = 0) 3DRE 2DFEM1 2L 89.27
89.24 89.23 86.28 89.03 89.06 89.262 2L 150.4 149.66 149.74 181.0
149.9 150.0 149.03 2L 320.3 317.25 317.78 278.6 315.7 315.9 314.84
L 357.1 356.5 364.02 356.44 784.3 352.8 353.1 353.95 L 365.8 364.31
353.1 364.0 364.6 361.16 2L/3 604.1 599.4 601.3 596.17 2L/3 803.5
800.5 761.8 781.1 782.9 784.48 L/2 885.0 874.0 878.2 871.99 L 1107
1047. 1048. 1049.10 2L/5 1218. 1195. 1202. 1195.
18
-
Paul R. Heyliger / Thin-Walled Structures 00 (2014) 1?? 19
Table 4. Axial frequencies of hollow rectangular section.Ply
Angle Mode Type
First axial Second axial1D [17] 3DRE 1D [17] 3DRE
0 11376 11749 11384 34127 35247 3414915 11537 3459930 12191
3649745 13956 4159860 18442 5378575 35692 10095090 71947 72007
71944 215841 216021 216134
19
-
Paul R. Heyliger / Thin-Walled Structures 00 (2014) 1?? 20
Table 5. Torsional frequencies of hollow rectangular section.Ply
Angle Mode Type
First mode Second mode[17] 3DRE [17] 3DRE
0 4397.6 1314115 4620.1 1379530 5266.0 1566545 6302.1 1862960
7538.7 1959875 7397.9 1805590 4239/4640 4686.0 12717/14595
14371
20
-
Paul R. Heyliger / Thin-Walled Structures 00 (2014) 1?? 21
Table 6. Transverse bending frequencies of hollow rectangular
section.Ply Angle Mode Type
First Transverse Second Transverse Third TransverseRayleigh [17]
3DRE Rayleigh [17] 3DRE Rayleigh 3DRE
0 242.4 249 242.2 1517.2 1558 1497.5 4239.7 4106.715 253 245.8
1577 1520.3 4171.330 269 262.5 1676 1622.0 4446.645 318 306.8 1985
1888.6 5149.860 443 415.6 2786 2525.4 6730.275 765 730.2 4934
4123.1 1042390 1534.7 1502 1420.6 9595.9 8502 6706.7 26814
14810
21
-
Paul R. Heyliger / Thin-Walled Structures 00 (2014) 1?? 22
Table 7. Lateral bending frequencies of hollow rectangular
section.Ply Angle Mode Type
First lateral Second LateralRayleigh [17] 3DRE Rayleigh [17]
3DRE
0 848.4 871 844.5 5234.5 4331 5059.715 881 859.1 4383 5170.330
949 917.8 4657 5560.145 1236 1069.7 5490 6462.760 2182 1438.3 7554
8473.375 3916 2484.6 12333 1294490 5365.6 4583 4476.2 33106 18062
17269
22
-
Paul R. Heyliger / Thin-Walled Structures 00 (2014) 1?? 23
Table 8. The first six modes of hollow rectangle with modified
elastic constants.Angle 1 2 3 4 5 6
0 (Rayleigh) 362.8 1268 2268 6338 78250 363.1 1264 2223 5086
6013 743115 356.9 1244 2191 5459 5951 738830 354.8 1236 2187 5980
6471 747445 385.8 1337 2379 6519 7861 811060 489.0 1686 2992 8093
8982 1000175 779.4 2658 4517 7980 11720 1420290 1252 4094 5460 6649
15702 16697
90 (Rayleigh) 1298 4539 8118 22685 28008
23
-
Paul R. Heyliger / Thin-Walled Structures 00 (2014) 1?? 24
Table 9. Modal symmetry groups for beam with elliptical
cross-section.Group Displacement x z
u O1 v O
wu
2 v O Ow Ou O O
3 vw Ou O
4 v Ow O O
24
-
Paul R. Heyliger / Thin-Walled Structures 00 (2014) 1?? 25
Table 10. Frequencies of hollow isotropic beam with elliptical
cross-section.Ply Angle Method
Group 3DRE [53] [52] 2d FEM1 0.0706 0.0722 0.0721 0.07191 0.1285
0.1327 0.1327 0.13191 0.1631 0.1644 0.1644 0.16412 0.0762 0.0778
0.0782 0.07762 0.1098 0.1111 0.1111 0.11182 0.1810 0.1877 0.1874
0.18723 0.0763 0.0783 0.0778 0.07773 0.1087 0.1119 0.1110 0.11073
0.1793 0.1872 0.1879 0.18564 0.0702 0.0722 0.0721 0.07164 0.1291
0.1340 0.1339 0.13324 0.2026 0.2012 0.2005 0.2023
25
-
Paul R. Heyliger / Thin-Walled Structures 00 (2014) 1?? 26
Table 11. Frequencies of hollow orthotropic beam with elliptical
cross-section.Method
Group 3DRE 2D FEM1 0.0288 0.02921 0.0419 0.04281 0.0619 0.06312
0.0286 0.02902 0.0446 0.04552 0.0595 0.06043 0.0285 0.02893 0.0438
0.04453 0.0564 0.05744 0.0289 0.02934 0.0419 0.04314 0.0661
0.0665
APPENDIX
The elements of the matrix equations for both the Ritz model and
the three-dimensional finite elementmodel are give as:
[K11]i j =V
C11 uix
uj
x+C16
uix
uj
y+C55
uiz
uj
z+C16
uiy
uj
x+C66
uiy
uj
y
dV (19)
[K12]i j =V
C12 uix
vj
y+C16
uix
vj
x+C45
uiz
vj
z+C26
uiy
vj
y+C66
uiy
vj
x
dV (20)
[K13]i j =V
C13 uix
wj
z+C45
uiz
wj
y+C55
uiz
wj
x+C36
uiy
wj
z
dV (21)
[K22]i j =V
C22 viy
vj
y+C26
viy
vj
x+C44
viz
vj
z+C26
vix
vj
y+C66
vix
vj
x
dV (22)
[K23]i j =V
C23 viy
wj
z+C44
viz
wj
y+C45
viz
wj
x+C36
vix
wj
z
dV (23)
26
-
Paul R. Heyliger / Thin-Walled Structures 00 (2014) 1?? 27
[K33]i j =V
C33 wiz
wj
z+C44
wiy
wj
y+C45
wiy
wj
x+C45
wix
wj
y+C55
wix
wj
x
dV (24)M11i j =
Vui
ujdV (25)
M22i j =Vvi
vjdV (26)
M33i j =Vwi
wj dV (27)
The elements of the coefficient matrices for the two-dimensional
semi-analytic finite element model aregiven as:
K11i j =V
C11 uix
uj
xcos2ky +C66ui
uj sin
2ky +C55uiz
uj
zcos2ky
dV (28)
K12i j =V
C12 uix
vjkcos2ky C66ui
vj
xksin2ky
dV = K21ji (29)
K13i j =V
C13 uix
wj
z+C55
uiz
wj
x
cos2kydV = K31ji (30)
K22i j =V
C22vi vjk2cos2ky +C44 viz vj
zsin2 ky +C66
vix
vj
xsin2ky
dV. (31)
K23i j =V
C23 viz
wj k cos2 ky C44vi
wj
zksin2ky
dV = K32ji (32)
K33i j =V
C33 wiz
wj
zcos2ky +C44wi
wj k
2sin2ky +C55wix
wj
xcos2ky+
dV (33)
27
-
Paul R. Heyliger / Thin-Walled Structures 00 (2014) 1?? 28
d1
d2
d3
t2
Figure 1. The beam cross-section of the doubly unsymmetric
channel. For the results presented here, the origin of the (x,y)
systemwas located at the lower-left interior corner of the beam
cross-section. This required integration of the coefficient
matrices overfour rectangular subdomains.
28
-
Paul R. Heyliger / Thin-Walled Structures 00 (2014) 1?? 29
t
di
Figure 2. The beam cross-section of the cantilever semi-circle.
The origin of the (x,y) system is located at the center of
thefull-circle extension of this geometry.
29
-
Paul R. Heyliger / Thin-Walled Structures 00 (2014) 1?? 30
(a) Mode 1 (symmetric)
30
-
Paul R. Heyliger / Thin-Walled Structures 00 (2014) 1?? 31
b) Mode 2 (symmetric)
31
-
Paul R. Heyliger / Thin-Walled Structures 00 (2014) 1?? 32
c) Mode 3 (anti-symmetric)
32
-
Paul R. Heyliger / Thin-Walled Structures 00 (2014) 1?? 33
d) Mode 4 (symmetric)
Figure 3. The first four modes of the doubly unsymmetric
channel
33
-
Paul R. Heyliger / Thin-Walled Structures 00 (2014) 1?? 34
0 0.2 0.4 0.6 0.8 1Normalized distance
-1
-0.8
-0.6
-0.4
-0.2
0
No
rmal
ized
di
spla
cem
ent
Figure 4. The first symmetric mode of the semi-circle: u (short
dash), v (long dash), and w (solid).
34
-
Paul R. Heyliger / Thin-Walled Structures 00 (2014) 1?? 35
0 0.2 0.4 0.6 0.8 1Normalized distance
-1
-0.75
-0.5
-0.25
0
0.25
0.5
0.75
No
rmal
ized
di
spla
cem
ent
Figure 5. The second symmetric mode of the semi-circle: u (short
dash), v (long dash), and w (solid).
35
-
Paul R. Heyliger / Thin-Walled Structures 00 (2014) 1?? 36
0 0.2 0.4 0.6 0.8 1-1
-0.75
-0.5
-0.25
0
0.25
0.5
0.75
Figure 6. The third symmetric mode of the semi-circle: u (short
dash), v (long dash), and w (solid).
36
-
Paul R. Heyliger / Thin-Walled Structures 00 (2014) 1?? 37
0 0.2 0.4 0.6 0.8 1Normalized distance
-1
-0.75
-0.5
-0.25
0
0.25
No
rmal
ized
di
spla
cem
ent
Figure 7. The first unsymmetric mode of the semi-circle: u
(short dash), v (long dash), and w (solid).
37
-
Paul R. Heyliger / Thin-Walled Structures 00 (2014) 1?? 38
0 0.2 0.4 0.6 0.8 1Normalized distance
-1
-0.8
-0.6
-0.4
-0.2
0
No
rmal
ized
di
spla
cem
ent
Figure 8. The second unsymmetric mode of the semi-circle: u
(short dash), v (long dash), and w (solid).
38
-
Paul R. Heyliger / Thin-Walled Structures 00 (2014) 1?? 39
0 0.2 0.4 0.6 0.8 1Normalized distance
-1
-0.75
-0.5
-0.25
0
0.25
0.5
0.75
1
No
rmal
ized
di
spla
cem
ent
Figure 9. The third unsymmetric mode of the semi-circle: u
(short dash), v (long dash), and w (solid).
39
-
Paul R. Heyliger / Thin-Walled Structures 00 (2014) 1?? 40
a) = 0 b) = 60
Figure 10. The first (top) and second (bottom) torsional modes
for two fiber orientations.
40
-
Paul R. Heyliger / Thin-Walled Structures 00 (2014) 1?? 41
a) = 0 b) = 60
Figure 11. The first transverse bending mode from the side (top)
and end (bottom) for two fiber orientations.
41
-
Paul R. Heyliger / Thin-Walled Structures 00 (2014) 1?? 42
a) = 0 b) = 60
Figure 12. The second transverse bending mode from the side
(top) and end (bottom) for two fiber orientations.
42
-
Paul R. Heyliger / Thin-Walled Structures 00 (2014) 1?? 43
a) First mode b) Second mode
Figure 13. The lowest lateral bending modes from the side (top)
and end (bottom). These modal plots are visually identical forboth
= 0 and = 60.
43