Journal of Sound and Vibration 332 (2013) 4324–4338

Contents lists available at SciVerse ScienceDirect

Journal of Sound and Vibration

journal homepage: www.elsevier.com/locate/jsvi

Free vibration of stepped thickness rectangular plates using

spectral finite element method
Gang Wang n, Ahmet Unal
Department of Mechanical and Aerospace Engineering, The University of Alabama in Huntsville, Huntsville, AL 35899, USA

a r t i c l e in f o abstract

Article history: Free vibration of stepped thickness rectangular plates is investigated in this paper using
Received 10 September 2012 the spectral finite element method (SFEM). It is impossible to obtain closed-form solutions
Received in revised form for a uniform rectangular plate with arbitrary boundary conditions except the Levy-type
25 February 2013
plates, i.e., at least simply-supported at two opposite edges. Moreover, the variation of
Accepted 4 March 2013
plate thickness adds complexity to the problem. The Kantorovich method is employed to
Handling Editor: S. Ilanko
Available online 17 April 2013 obtain an analytical approximation solution form. Then, a plate spectral finite element is
developed accordingly in the frequency domain. Similar to the conventional finite element
method (CFEM), after assembling elements and applying boundary conditions, each
modal frequency and associated mode shape function can be determined by iteratively
solving corresponding equations in plate x and y directions. Available literature results
along with NASTRAN simulations are used to validate our SFEM predictions of stepped
thickness plates with various boundary conditions. Compared to the CFEM, only a fraction
of the mesh is needed to achieve comparable accuracy in each mode. Substantial
computation cost can then be saved. Mode shape functions are extracted and presented
in a semi-analytical form. Physical insights of wave propagation characteristics in a
stepped thickness rectangular plate can be collected from these results. In summary, an
efficient and accurate SFEM is developed to conduct free vibration analysis of stepped
thickness rectangular plates with various boundary conditions.
& 2013 Elsevier Ltd. All rights reserved.

1. Introduction

Plate structures are widely used in almost every aspect of engineering applications, including aerospace, civil, and
electronics. Plates with thickness variations can be advantageous in modern structures due to their unique features, such as
weight savings, stiffness enhancement, optimized load carrying, and tailored dynamic design (e.g., shifting natural
frequency). Therefore, it is very important to understand free vibrations of such stepped thickness plates. Extensive
literature is available for the free vibration of a uniform rectangular plate [1–3]. Exact solutions exist only for Levy-type
plates, in which at least two opposite edges are simply-supported. Because plate mode shapes can be determined using the
separation of variables method so that two-dimensional plate problem is reduced to one-dimensional beam problem. Leissa
[1] summarized natural frequency results for uniform plates with a wide range of boundary conditions. Natural frequencies
were calculated by the assumed modes (Ritz) method, in which one-dimensional beam mode shape functions are used to
approximate the two-dimensional plate modes by assuming a product of separable variable in both x and y directions.
Chopra was the first to investigate free vibration of stepped thickness rectangular plates with simply-supported boundary

n
Corresponding author. Tel.: þ 1 256 824 6209; fax: þ 1 256 824 6758.
E-mail address: gang.wang@uah.edu (G. Wang).

0022-460X/$ - see front matter & 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.jsv.2013.03.008
 G. Wang, A. Unal / Journal of Sound and Vibration 332 (2013) 4324–4338 4325

Nomenclature Vxz, Vyz resultant shear force along plate edges

w(x,t) plate transverse displacement
a, c plate length, dimension along x direction Wmn(x, y, ωmn) mnth plate mode shape function
b plate width, dimension along y direction Xmi independent X component of the mnth plate
D, D1, D2 plate bending flexural stiffness mode shape, i ¼1, Nx
E Young's modulus of plate material Yn independent Y component of the mnth plate
h, h1, h2 plate thickness mode shape
m, n number of half-wavelengths of a mode along x ρ density of plate material
and y direction, respectively ν Poisson's ratio
Mxx, Myy resultant bending moment along plate edges δ variation operator
Mxy resultant twist moment along plate edges ωmn mnth plate modal frequency [rad/s]
Nx number of independent functions of X
component

conditions along all edges in which the Kirchhoff hypothesis was assumed [4]. First eight natural frequencies of a single
stepped thickness plate were calculated based on the Levy-type solution approach. The formulations contain error in shear
force and bending moment continuity conditions at junction as pointed out by Warburton [5]. However, the general solution
methodology and natural frequency trend were correct as commented by Warburton. Yuan and Dickinson corrected the
continuity conditions and provided exact natural frequency results of the same plate [6]. In addition, a Rayleigh–Ritz
method was developed to model the stepped thickness plates, in which artificial springs were introduced at the junctures
between each sub-plate and system boundaries using simple shape functions [6]. Lam and Amrutharaj developed a hybrid
method to investigate free vibration of single or multiple stepped thickness plates employing a set of beam polynomial with
subsectioning in Rayleigh–Ritz method [7]. Plates with fully clamped and cantilevered boundary conditions were examined
along with fully simply-supported case. Guo et al. corrected the continuity conditions and simplified solution expression by
dropping cosine and hyperbolic cosine terms in the mode shape functions in order to improve computational efficiency for
higher frequency modes [8]. A finite strip method (FSM) was proposed as well. Frequency predictions from the FSM were
compared with those from the finite element method (FEM) and classical method (CM) (i.e., the Levy-type exact solution for
both one and two stepped thickness plates with fully simply-supported conditions). These research works built a rich
database for free vibration analysis of stepped thickness rectangular plates.
It is no doubt that we can obtain natural frequency and mode shape functions of stepped thickness plates using
aforementioned numerical approaches, e.g., Rayleigh–Ritz, FSM, and FEM. With the advancement of computing capability,
the FEM has become a dominant computational tool for modern structural analysis since 1980s [9]. However, analytical
methods are equally important and have advantages in providing more physical insights and allowing results to be
presented in a compact form. Particularly, in free vibration of a plate, the fundamental natural frequency and nodal line
location are essential parameters in structural design and sensor placement. Unfortunately, closed-form solutions of
rectangular plate bending vibration exist only for Levy-type of plate. Moreover, when plate boundary conditions, step
location and number are changed, it is cumbersome to re-derive continuity conditions at junctions and re-assemble the
equations required in the Levy-type solution in order to accommodate them. Therefore, a simple but accurate analytical or
semi-analytical solution approach is needed for the stepped thickness plates with various boundary conditions.
The Kantorovich method is a variational method and can be an alternative approach to provide analytical approximation
solution in structural analysis [10]. Unlike the Rayleigh–Ritz or FEM, the governing equations (i.e., partial differential
equations (PDEs)) are reduced to a set of linear algebraic equations. The Kantorovich method reduces the PDEs to a set of
ordinary differential equations (ODEs) in terms of undetermined functions. It is expected to solve the ODEs to achieve better
accuracy compared to Rayleigh–Ritz or FEM method. A resurgence of interest of using the Kantorovich method was
observed in the literature during the past decade as a result of improvement in computational power [11–15]. All of these
successes prove that the Kantorovich method is a universal and powerful mathematical tool for PDEs. However, although an
analytical solution expression is obtained, one still needs to apply proper compatibility or continuity conditions in order to
solve the equations for stepped thickness plates if general ODE solution format is adopted. We face the same drawbacks as
discussed in Levy-type solution for plate with thickness variations. In order to remedy it, a standard matrix solution
representation and a simple assembly procedure as used in the conventional finite element method (CFEM) that can
automatically account for continuity conditions, will substantially enhance the capability of using the Kantorovich method
in structural analysis.
The spectral finite element method (SFEM) is an alternative structural analysis tool. It is also called dynamic stiffness
matrix method [16–18]. By analogy to CFEM, the displacement interpolation functions in the SFEM are derived from the
solution of governing equation compared to the polynomial functions used in the CFEM. The governing equations are first
transformed from the time domain to the frequency domain using a Discrete Fourier Transform (DFT). The spatial variation
is semi-explicitly obtained by solving the characteristic equation in wavenumber space. This results in a complex shape
function matrix representing the linear superposition of all the wave modes and the solution is accurate. Also, both the
4326 G. Wang, A. Unal / Journal of Sound and Vibration 332 (2013) 4324–4338

complex dynamic strain–displacement matrix and the dynamically consistent force vector can be derived. This methodology
has been successfully demonstrated in one-dimensional structures such as sandwich beam analysis [19], rotating beam [20],
and other structures as summarized by Lee [21]. For two-dimensional plate structures, typically there is no closed-form
solution available except Levy-type plate structures. The throw-off elements were introduced in order to reduce 2D problem
to 1D beam-like case as discussed by Doyle [16] and Gopalakrishnan et al. [18], in which the reflection waves are not
considered and wave is assumed to travel in a semi-infinite domain. SFEMs were developed for Levy-type plates to conduct
both dynamic response and free vibration analyses [22,23]. There is no attempt to develop two-dimensional spectral finite
element model for rectangular plate structures with non-Levy-type boundary conditions.
In this paper, a SFEM is developed to conduct free vibration analysis of stepped thickness rectangular plates. We leverage
the analytical solutions from the Kantorovich method and unique features used in CFEM such as standard matrix
representation and element assembly procedure. The goal is to provide an efficient and accurate solution to free vibration
of stepped thickness rectangular plates with various boundary conditions, i.e., not limited to Levy-type plates. Based on the
separable analytical solutions derived from the Kantorovich method, nodal line displacement vector is defined in each
direction and used to solve for the wave coefficients in the solution expressions. Then, a dynamic stiffness matrix is
developed to relate nodal line displacement vector with corresponding nodal force vector. The purpose is to formulate a
standard matrix representation so that we can assemble these spectral finite elements accordingly to account for the
thickness variations in a plate structure. However, unlike the 1D case, an iterative solution scheme is required to obtain
solutions in both x and y directions. Normally, it takes one or two steps to reach the convergence. The paper proceeds as
follows. In Section 2, we provide a detailed discussion on the derivation and formulation of the proposed spectral finite
element model. This is followed by the results section. Natural frequency and mode shape prediction results of stepped
thickness rectangular plates with different boundary conditions (e.g., clamped-free-clamped free (denoted as CFCF)),
simply-supported (denoted as SSSS), fully clamped (denoted as CCCC) are calculated and compared to available literature
results and NASTRAN predictions. A summary and some concluding remarks are given in Section 4.

2. Spectral finite element model

2.1. Equations

Fig. 1 shows a thin uniform rectangular plate structure element. Applying the Kirchhoff hypothesis, the potential energy
is given by
Z Z " 2  2 2  2 2 #
1 a b ∂2 w ∂ w ∂2 w ∂2 w ∂ w
U¼ D þ þ2υ þ 2ð1−υÞ dxdy (1)
2 0 0 ∂x2 ∂y2 ∂x2 ∂y2 ∂x∂y

The kinetic energy is

Z a Z b  2
1 ∂w
T¼ ρh dxdy (2)
2 0 0 ∂t
where the plate bending flexural stiffness, D, is defined as
3
Eh
D¼
12ð1−υ2 Þ
The free vibration governing equation and associated boundary conditions can be determined by applying Hamilton's
principle to Eqs. (1) and (2). As we discussed in the previous section, closed-form solution is only available for Levy-type
plates (i.e., at least two opposite edges with simply-supported boundary conditions). Instead of solving the governing
equation directly, we will determine the solution to satisfy the zero variation condition of a functional, which is equivalent
to solve the boundary value problem of the governing equation. First, we need to define a functional as shown in the

Fig. 1. Plate structure and resultant forces.

 G. Wang, A. Unal / Journal of Sound and Vibration 332 (2013) 4324–4338 4327

following equation:
I ¼ U þT (3)
In this case, it is the total energy of the plate structure. Note that this functional is carefully selected in order to yield
identical governing equation and associated boundary conditions when applying Hamilton's principle.
Secondly, the general plate displacement solution in the spectral form is
∞ ∞
wðx,y,tÞ ¼ ∑ ∑ amn W mn ðx,y,ωmn Þejωmn t (4)
m¼1n¼1
pffiffiffiffiffiffiffi
where j is equal to −1. W mn is the corresponding mnth plate mode shape function and amn is the amplitude of each mode.
For free vibration analysis, we need to determine both natural frequency ðωmn Þ and the corresponding mode shape function
ðW mn Þ. In the following derivation, subscripts m and n are omitted for simplification, i.e., W mn ¼ W.
Finally, we will take variation of the functional. By setting zero variation condition, we can equivalently obtain the
solution of plate free bending vibration.
Z aZ b  4  Z b  a Z b
∂ W ∂4 W ∂4 W ρhωmn 2 ∂W 
δI ¼ D þ 2 þ − W δWdxdy þ M xx δ  dy− V xz δðWÞja0 dy
0 0 ∂x4 ∂x2 ∂y2 ∂y4 D 0 ∂x  0
0
Z a  b Z a " a #b
∂W  ∂ 2
W
þ M yy δ  dx− V yz δðWÞjb0 dx þ 2Dð1−υÞ ¼0 (5)
0 ∂y  0 ∂x∂y 0
0 0

The internal resultant bending moment and shear force along four edges (i.e., x ¼0, a and y¼ 0, b) are
 2   2 
∂ W ∂2 W ∂ W ∂2 W
M xx ¼ D þ υ , M yy ¼ D þ υ
∂x2 ∂y2 ∂y2 ∂x2
 3   3 
∂ W ∂ W
3
∂ W ∂3 W
V xz ¼ −D þ ð2−υÞ , V yz ¼ −D þ ð2−υÞ
∂x3 ∂x∂y2 ∂y3 ∂x2 ∂y
Note that Eq. (5) is only valid to a uniform plate. We need to modify the integration limits to account for a stepped
thickness plate.

2.2. Solution

As shown in Fig. 2, a plate with single stepped thickness along x direction is used to demonstrate the development of
spectral finite element models along x and y directions and the iteration scheme. Mathematically, we would need to solve
two bi-harmonic partial differential equations (PDEs) with associated boundary conditions and continuity conditions to
determine the solutions, which yield two independent displacement functions. By assuming the separation of variables
solution for each mode shape function, two pairs of X and Y components are obtained. Because the plate exhibits stepped
thickness only along x direction, simplification can be conducted by assuming the same Y component. Boscolo and Banerjee
[23] applied the same assumption to develop SFEM models for stepped thickness Levy-type plates. Key steps are outlined as

 Assume the mnth plate mode shape function with a separation of variables form
8
Nx < X m1 Y n 0≤x≤c 0≤y≤b
W mn ðx,y,ωmn Þ ¼ ∑ X mi ðx,ωmn ÞY n ðy,ωmn Þ ¼ (6)
i¼1 : X m2 Y n c≤x≤a 0≤y≤b

Fig. 2. Spectral finite elements in a stepped thickness plate.

4328 G. Wang, A. Unal / Journal of Sound and Vibration 332 (2013) 4324–4338

Such treatment along x direction is analogous to a ‘stepped thickness beam’ and the treatment along y direction is
analogous to a ‘uniform beam’, as shown in Fig. 2. Here, Nx is the number of independent mode shape functions of X
component. In this case, Nx is equal to 2. We need two independent mode shape functions to describe wave propagation
characteristics along x direction. Typically, Nx coincides with the number of impedance changes along x direction. We need
one independent mode shape function to describe the wave propagation along y direction as aforementioned.

 Apply the Kantorovich method and obtain ODEs in terms of X m1 , X m2 , and Y n .

We need to assume either X or Y component as a priori to determine the corresponding ODEs, as discussed in Sections
2.2.1 and 2.2.2.
 Based on the analytical solution forms of X m1 , X m2 , and Y n , formulate spectral finite element models in both x and y
directions.

Similar to the CFEM, we need to develop a standard matrix formulation in structural analysis. In this case, the
interpolation functions are duplicated directly from the exact solutions of the ODEs instead of assumed polynomials
functions used in the CFEM. Detailed discussion is shown in Sections 2.2.1 and 2.2.2. Global dynamic stiffness matrix in each
direction can be obtained by assembling elements as in the CFEM. Typically, the total element number is equal to Nx in x
direction and 1 in y direction. However, we may have to include more elements to account for boundary conditions and
improve computational efficiency.

 Develop an iterative iteration scheme to determine the free vibration characteristics.

Unlike one-dimensional spectral finite element method [21], we need iteratively to solve for the mode shape functions.
An iteration scheme is required as discussed in Section 2.2.3.
Detailed derivations of each step will be presented in the subsequent sections. We must emphasize that the same
methodology can be applied to a plate with multiple stepped thickness in either x or y direction.

2.2.1. Analytical solution and spectral finite element model in x direction

2.2.1.1. Analytical solution.
If we assume that Y n is prescribed as a priori, then we have
8
< Y n δX m1 0≤x≤c 0≤y≤b
δW mn ¼ (7)
: Y n δX m2 c≤x≤a 0≤y≤b

In addition, the assumed function Y n satisfies all boundary conditions along y direction to vanish corresponding terms in
Eq. (5). (Note that appropriate integration limits need to be reset in Eq. (5) to account for the stepped thickness plate
configuration). Substituting Eqs. (6) and (7) into Eq. (5) and integrating along the y direction, two ordinary differential
equations (ODEs) can be obtained in terms of unknown functions, X m1 and X m2 . In the following section, we will
demonstrate the solution of X m1 . Similarly, we can determine the solution of X m2 . The ODE is
4 2
d X m1 d X m1
þ 2βx þ γ x X m1 ¼ 0 (8)
dx4 dx2
where
Z b
αx ¼ Y n 2 dy
Z
0 !
2
b
d Yn
βx ¼ 2
Y n dy =αx
0 dy
Z b 4 !
d Yn ρh1 ωmn 2
γx ¼ 4
Y n dy =αx −
0 dy D1

Here h1 is the thickness of the first segment of the plate and D1 is the corresponding bending flexural stiffness. We will
define the shear force and moment along x direction later. The solution of Eq. (8) is a superposition of two waves, i.e.,
propagating wave and evanescent wave.
X m1 ¼ c1 sinðpx xÞ þ c2 cosðpx xÞ þc3 sinhðqx xÞ þc4 coshðqx xÞ (9)
where px and qx are wave numbers in terms of circular frequency ωmn. They are given as
qffiffiffiffiffiffiffiffiffiffiffiffiffiffi
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 
   
px ¼  −βx − βx −γ x , qx ¼  −βx þ βx 2 −γ x 
2
   
 G. Wang, A. Unal / Journal of Sound and Vibration 332 (2013) 4324–4338 4329

Fig. 3. Plate SFEM in the x-direction.

2.2.1.2. Spectral finite element formulation.

Four unknown wave coefficients, c1 to c4 can be determined by the boundary conditions along x direction in order to obtain
non-trivial solutions. Now we introduce a nodal line displacement vector and the corresponding nodal line force vector as
shown in Fig. 3. The purpose is to solve for the four unknown wave coefficients using four nodal line displacement
components. A dynamic stiffness matrix will then be derived to relate the nodal line displacement vector with corresponding
nodal line force vector. In order to account for the thickness variations along plate x direction, different elements can be
applied and assembled accordingly as used in the CFEM. Detailed derivations are shown below. The nodal line displacement
vector (i.e., displacements and slopes) can be re-written in terms of four wave coefficients as shown in Eq. (9).
2 3 2 3 2 32 3
f1 c1 0 1 0 1 c1
6 df1 7 6 7 6 76 7
6 dx 7 6 c2 7 6 px 0 qx 0 76 c2 7
6 7 6 7 6 76 7
6 f2 7 ¼ P6 c3 7 ¼ 6 sinðpx cÞ cosðpx cÞ sinhðqx cÞ coshðqx cÞ 7 6 7 (10)
4 5 4 5 4 54 c3 5
df2
dx
c4 px cosðpx cÞ −px sinðpx cÞ qx coshðqx cÞ qx sinhðqx cÞ c4

The corresponding nodal line force components in terms of bending moments and shear forces can be determined by
conducting integration of associated boundary conditions along y direction. They are given by
! !
2 3
d X m1 d X m1 dX m1
Mx ¼ D αx þ e X
1 m1 , V x ¼ −D αx þ e2 (11)
dx2 dx3 dx

where
Z b 2 Z b 2
d Yn d Yn
e1 ¼ υ Y n dy, e2 ¼ ð2−υÞ Y n dy
0 dy2 0 dy2
Finally, a 4  4 dynamic stiffness matrix can be determined by relating the nodal line displacement vector with the
corresponding nodal line force vector.
2 3 2 3 2 3
V x1 −V x ð0Þ f1
6 7 6 7 6 df1 7
6 M x1 7 6 −M x ð0Þ 7 6 7
6 7 6 7 −1 6 dx 7
6 V x2 7 ¼ 6 V x ðcÞ 7 ¼ DRP 6 f2 7 (12)
4 5 4 5 4 5
M x2 M x ðcÞ df2
dx

The dynamic stiffness matrix is given by

Dx1 ðω,aÞ ¼ DRP −1 (13)
where
2 3
−αx p3x þe2 px 0 αx q3x þ e2 qx 0
6 7
6 0 −ð−αx p2x þ e1 Þ 0 −ðαx q2x þ e1 Þ 7
R¼6
6 −ð−αx p3 þ e2 p Þcosðp cÞ
7
4 x x x −ðαx p3x −e2 px Þsinðpx cÞ −ðαx q3x þ e2 qx Þcoshðqx cÞ −ðαx qx þ e2 px Þsinhðqx cÞ 7
3
5
ð−αx p2x þ e1 Þsinðpx cÞ ð−αx p2x þ e1 Þcosðpx cÞ ðαx q2x þ e1 Þsinhðqx cÞ ðαx q2x þe1 Þcoshðqx cÞ

Because the dynamic matrix, Dx1, is constructed based on the exact solution of the ODE in Eq. (8), one single element is
sufficient to capture the dynamics if there is no thickness variation along x direction. Minimum number of elements
coincides with the number of thickness variations along plate x direction. Typical boundary conditions are

Clamped x ¼ 0,c X m1 ¼ dX m1
¼0
dx
Simply-supported x ¼ 0,c
2
X m1 ¼ d dxX2m1 ¼ 0
Free x ¼ 0,c V x ¼ Mx ¼ 0

Note that above spectral finite element was derived by assuming that the element length is equal to c. If more elements
are required in order to calculate X m1 , additional dynamic stiffness matrix can be determined by simply substituting the
corresponding element length in Eq. (10). The general solution of X m1 does not change as shown in Eq. (9).
4330 G. Wang, A. Unal / Journal of Sound and Vibration 332 (2013) 4324–4338

Similarly, we can determine the solution of X m2 and formulate the dynamic stiffness matrix as well. Applying the same
assembly procedure as used in the CFEM, the continuity conditions are automatically fulfilled at junctions. Finally, a global
dynamic stiffness matrix can be obtained. After the application of appropriate boundary conditions along x direction, we can
determine the circular frequency, nodal line displacement vector, and X components of a plate mode by satisfying the zero
determinant condition of the global dynamic matrix in free vibration analysis.

2.2.2. Analytical solution and spectral finite element model in y direction

2.2.2.1. Analytical solution.
After obtaining all solutions in x direction, we will determine the solutions in y direction using the same procedure.
In this case, we assume that both X m1 andX m2 are prescribed as a priori, the variation of W is given by
8
< X m1 δY n 0≤x≤c 0≤y≤b
δW ¼ (14)
: X m2 δY n c≤x≤a 0≤y≤b

Then, an ODE can be derived in y direction as shown below

4 2
d Yn d Yn
þ 2βy þ γy Y n ¼ 0 (15)
dy4 dy2
where
Z c Z a
αy ¼ X 2m1 dx þ X 2m2 dx
0 c
Z Z !
c 2 a 2
d X m1 d X m2
βy ¼ X m1 dx þ X m2 dx =αy
0 dx2 c dx
2

Z Z !
c 4 a 4
d X m1 d X m2 ρhωmn 2
γy ¼ X m1 dx þ X m2 dx =αy −
0 dx4 c dx4 D

Here h is the baseline plate thickness (we set h ¼h1) and corresponding D (D ¼D1) can be calculated. The solution of
Eq. (15) has the similar form as shown in Eq. (9).
Y n ¼ d1 sinðpy yÞ þ d2 cosðpy yÞ þ d3 sinhðqy yÞ þd4 coshðqy yÞ (16)

where py, and qy are wave numbers in terms of circular frequency ω. They are given below.
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
   
py ¼  −βy − βy −γ y , qy ¼  −βy þ βy 2 −γ y 
2
   

2.2.2.2. Spectral finite element formulation.

Similarly, we introduce nodal line displacement vector (i.e., displacements and slopes), as shown in Fig. 4, to solve for the
four unknown wave coefficients in Eq. (16).
2 3 2 3 2 32 3
θ1 d1 0 1 0 1 d1
6 dθ1 7 6 7 6 76 7
6 dy 7 6 d2 7 6 py 0 qy 0 76 d 2 7
6 7 6 7 6 76 7
6 θ2 7 ¼ Q 6 d3 7 ¼ 6 sinðpy bÞ cosðpy bÞ sinhðqy bÞ coshðqy bÞ 7 6 7 (17)
4 5 4 5 4 54 d 3 5
dθ2
dy d4 py cosðpy bÞ −py sinðpy bÞ qy coshðqy bÞ qy sinhðqy bÞ d4

The corresponding nodal line force components along y direction are defined as
! !
2 3
d Yn d Yn dY n
M y ¼ D αy þg 1 Y n , V y ¼ −D αy þ g 2 (18)
dx2 dy3 dy

Fig. 4. Plate SFEM in the y-direction.

 G. Wang, A. Unal / Journal of Sound and Vibration 332 (2013) 4324–4338 4331

where
Z Z ! Z Z !
c 2 a 2 c 2 a 2
d X m1 d X m2 d X m1 d X m2
g1 ¼ υ Xdx þ Xdx , g 2 ¼ ð2−υÞ X m1 dx þ X m2 dx
0 dx2 c dx2 0 dx2 c dx2

Finally, a 4  4 dynamic stiffness matrix can be determined to relate the nodal line displacement and force vector in y
direction.
2 3 2 3 2 3
V y1 −V y ð0Þ θ1
6 M 7 6 −M ð0Þ 7 6 dθ1 7
6 y1 7 6 y 7 6
−1 6 dy 7
7
6 7 6 7
6 V y2 7 ¼ 6 V y ðbÞ 7 ¼ DSQ 6 θ2 7 (19)
4 5 4 5 4 5
dθ2
M y2 M y ðbÞ dy

The dynamic stiffness matrix is given by

Dy ðω,bÞ ¼ DSQ −1 (20)
where
2 3
−αy p3y þ g 2 py 0 αy q3y þg 2 qy 0
6 7
6 0 −ð−αx p2x þg 1 Þ 0 −ðαy q2y þ g 1 Þ7
6 7
S¼6 7
6 −ð−αy p3y þg 2 py Þcosðpy bÞ −ðαy p3y −g 2 py Þsinðpy bÞ −ðαy q3y þg 2 qy Þcoshðqy bÞ −ðαy q3y þ g 2 qy Þsinhðqy bÞ 7
4 5
ð−αy p2y þ g 1 Þsinðpy bÞ ð−αy p2y þ g 1 Þcosðpy bÞ ðαy q2y þ g 1 Þsinhðqy bÞ ðαy q2y þg 1 Þcoshðqy bÞ

Above formulation assumes that single element is used along y direction. If more elements are needed, we can determine
the dynamic stiffness matrix accordingly by substituting different element length as shown in Eq. (17). Note that the
solution will not change as shown in Eq. (16).

2.2.3 Iteration algorithm

An algorithm as discussed by Wang and Wereley [14] was developed to determine natural frequency of a uniform plate
under in-plane vibration, in which single mode shape function in either x or y direction is sufficient to predict each mode.
Modifications are required to account for the thickness variation in plate bending vibration case. Detailed iteration
algorithm is shown below.

Step 1. In the y direction, prescribe the mode shape Y 0n a priori, for iteration k ¼0.
A uniform beam mode shape function matching plate boundary conditions is applied as tabulated in the textbook [24].
For example, the mode shape function of a beam with clamped–clamped boundary conditions is given by
Y 0n ¼ coshðβn yÞ−cosðβn yÞ−sn ðsinhðβn yÞ−sinðβn yÞÞ
here βn is the wavenumber for nth mode and coefficient αn is defined as well in Ref. [24]. We must emphasize that we
just need this mode shape once to initiate the process.
Step 2a. Increment k, Y kn ¼ Y k−1
n .
Step 2b. Obtain the ODEs as shown in Eq. (9). Since two independent functions assumed along x direction, we will obtain
two ODEs in terms of X km1 and X km2 . Then, two corresponding dynamic stiffness matrices can be determined accordingly.
The element assembly procedure is similar to the CFEM [9]. In this case, elements 1 and 2 are connected by a nodal line
instead of a single point. Final global dynamic stiffness matrix has the size of 6  6. Typically, it is a banded matrix as
shown in the CFEM [9]. The application of boundary conditions along x direction is also similar to that of CFEM [9]. For a
clamped–clamped plate along x direction, we will eliminate the 1st, 2nd, 5th, and 6th column and row. Then, a 2  2
matrix is obtained. The use of nodal line displacement components instead of wave coefficients allows us to
automatically satisfy the compatibility conditions at each junction. Finally, a frequency ωkxm is solved numerically to
result in a zero determinant of the final dynamic stiffness matrix in order to satisfy boundary conditions in x direction.
Associated non-zero solutions of the nodal line displacement components can be determined as well. These nodal line
displacements will be used to determine the corresponding X component in each element using Eqs. (9) and (10).
Step 2c. Using the newly calculated mode shape functions along x direction to obtain the ODE(s) in terms of Y kn . Based on
our assumption, we determine the Y kn using both X km1 and X km2 by conducting the integration over the entire plate x
direction. In this case, an approximate mode shape function, Y kn , is determined compared to exact solution of PDEs.
Following the similar procedure as discussed in Step 2b and by applying appropriate boundary conditions along y
direction, a frequency, ωkyn , can be solved numerically to result in a zero determinant in order to satisfy boundary
conditions in y direction. Finally, corresponding mode shape function, Y kn , can be calculated.
Step 3. Check convergence between ωkxm and ωkyn . If a tolerance is reached for the specified value, then the iteration is
stopped. Otherwise, go to step 2a, k ¼ k þ 1:
4332 G. Wang, A. Unal / Journal of Sound and Vibration 332 (2013) 4324–4338

Table 1
Natural frequency results of a clamped-free-clamped-free (CFCF) plate.

Description Mode (mn) SFEM (Hz) Expt (Ref. [15]) Error (%)

11 88.3 83.7 5.5%

12 111.3 107.3 3.7%
13 210.7 207.2 1.7%
21 243.1 233.7 4.0%
22 275.4 266.0 3.5%
23 389.8 381.4 2.2%
14 420.8 420.7 0.03%

Fig. 5. Single stepped thickness plate configurations: (a) fully simply-supported (SSSS); and (b) fully clamped (CCCC).

In summary, in order to calculate each modal frequency and mode shape function for a stepped thickness rectangular
plate, we first need to determine parameter Nx based on the plate thickness variations. From the boundary conditions along
plate edges, we then apply appropriate number of spectral finite element in each direction to solve for these independent
mode shape functions. Increasing the element number along each direction does not affect the solution characteristics.
However, it could improve the convergence because all coefficients are evaluated numerically. Typically, one or two
iterations are enough to yield the convergent natural frequency and the associated mode shape functions. MATLAB code was
developed to implement above scheme. As aforementioned, Boscolo and Banerjee [23] developed a dynamic stiffness for
Levy-type of rectangular plates because the mode shape is pre-determined along the direction with simply-supported
boundary conditions. Subsequently, the original two-dimensional problem is reduced to one-dimensional problem.
However, we iteratively solve for mode shape function in x and y directions based on the newly developed spectral finite
element models to accommodate plates with non-Levy-type of boundary conditions. Accurate predictions of modal
frequency and mode shape function are expected and will be discussed in the results section. In addition, the mode shape
functions are presented in a semi-analytical form. Physical insights can be collected, which lead to a fundamental
understanding of the wave propagation characteristics in a stepped thickness rectangular plate.

3. Results

3.1. Clamped-free-clamped-free plate

Table 1 shows the first seven modal frequency results of a uniform clamped-free-clamped-free (CFCF) rectangular plate.
In this case, a 2  1 mesh (i.e., two equal length elements in x direction and one element in y direction) was used. Moreover,
the same mesh size was used to calculate each modal frequency. The maximum error is 5.5%, which is comparable to the
other approach as discussed in Ref. [15].

3.2. Fully simply-supported stepped thickness plate (SSSS)

A single stepped thickness square plate with simply-supported conditions along four edges (denoted as SSSS), was
studied, as shown in Fig. 5a. The side to length ratio (i.e., c/a) is 0.25 and the thickness ratio (i.e., h1/h2) varies from 1 to 0.5.
 G. Wang, A. Unal / Journal of Sound and Vibration 332 (2013) 4324–4338 4333

Table 2
Dimensionless frequency λ ¼ ðρh1 ωmn 2 =D1 Þ1=2 =π 2 for a fully simply-supported square plate with single stepped thickness as shown in Fig. 5a.

Mode c/a¼ 0.25 h1/h2 ¼1 h1/h2 ¼0.9 h1/h2 ¼0.8 h1/h2 ¼0.7 h1/h2 ¼ 0.6 h1/h2 ¼ 0.5

1 (1,1) Lam et al. [7] 1.998 2.193 2.424 2.703 2.924 3.449
Chopra[4] 2 2.199 2.437 2.724 3.072 3.488
Exact [6] 2 2.1636 2.3617 2.6018 2.8937 3.2581
Present 2 2.1636 2.3617 2.6018 2.8937 3.2581

2 (2,1) Lam et al. [7] 4.996 5.392 5.843 6.361 7.004 7.992
Chopra [4] 5 5.4 5.855 6.379 7.1096 7.803
Exact [6] 5 5.4007 5.8623 6.4112 7.643 8.0978
Present 5 5.4007 5.8623 6.4112 7.1096 8.0978

3(1,2) Lam et al. [7] 4.999 5.469 6.019 6.719 7.643 8.349
Chopra [4] 5 5.495 6.078 6.757 7.519 8.279
Exact [6] 5 5.4414 5.9792 6.6469 7.505 8.6828
Present 5 5.4414 5.9792 6.6469 7.505 8.6828

4(2,2) Lam et al. [7] 8.0006 8.633 9.341 10.226 11.276 12.862
Chopra [4] 8 8.637 9.358 10.199 11.259 12.775
Exact [6] 8 8.6557 9.4503 10.4528 11.7975 13.7385
Present 8 8.6557 9.4503 10.453 11.798 13.738

Table 3
Mode shape functions of a fully simply-supported square plate with single stepped thickness as shown in Fig. 5a, c/a¼ 0.25, h1/h2 ¼ 0.5.

8 nπy
< X m1 sin b 0≤x≤c 0≤y≤b
W mn ¼ 
: X m2 sin nπy
b
c≤x≤a 0≤y≤b

X m1 ¼ a1 sinðpx1 xÞ þ a2 cosðpx1 xÞ þ a3 sinhðqx1 xÞþ a4 coshðqx1 xÞ 0≤x≤0:25

X m2 ¼ b1 sinðpx2 xÞþ b2 cosðpx2 xÞþ b3 sinhðqx2 xÞ þ b4 coshðqx2 xÞ 0:25≤x≤1 x ¼ x−0:25

Mode (1,1) Mode (2,1) Mode (1,2) Mode (2,2)

a1 1.4581 1.1475 1.0287 1.2111

a2 0 0 0 0
a3 −0.14336 0.0009617 −0.0024699 0.016747
a4 0 0 0 0
px1 4.7208 8.3698 6.7984 9.8038
qx1 6.4827 9.4759 11.188 13.231
b1 0.2764 −0.47662 −0.17926 −0.65616
b2 0.9006 0.69936 0.91221 0.7462
b3 −0.10003 −0.30065 −0.087792 −0.2538
b4 0.09994 0.30064 0.087791 0.2538
px2 2.4917 5.4856 1.8357 5.3215
qx2 5.0939 7.0591 9.0734 10.357

In Ref. [7], 6  6 terms were used to construct the eigenvalue problem for frequency predictions using the Ritz method. We
only need two elements along x direction to account for the thickness variation and one element along y direction. Finally,
a 2  1 mesh was used in the SFEM to yield a 4  4 global dynamic stiffness matrix in x direction and a 2  2 global dynamic
stiffness matrix in y direction. A dimensionless frequency is assumed and defined as λ ¼ ðρh1 ωmn 2 =D1 Þ1=2 =π 2 . As listed in
Table 2, our SFEM frequency predictions are compared with literature results and exact solutions. It is evident that the
present results are identical to the exact solutions because our SFEM is formulated based on the same solutions of governing
equations for the Levy-type plates. Table 3 shows the results of semi-analytical mode shape functions in which the thickness
ratio was set to be h1/h2 ¼0.5. Similar to the classical Levy-type solutions [4–6], two different sets of mode shape functions
were obtained along x direction to account for the thickness variation. The mode shape function solution contains both
propagating waves (sine and cosine terms) and evanescent waves (hyperbolic sine and cosine). Unlike the beam bending
vibration case, these two different waves are characterized by two different wave numbers. Wave coefficients associated
with cosine and hyperbolic cosine vanish on the left portion of the plate in order to satisfy zero displacement boundary
conditions along that edge. Fig. 6 shows the first four mode shape plots. Clearly, due to the thickness variation, each mode
does not exhibit symmetric property as observed in a uniform plate. The peak values of each mode shift toward the left
edge, i.e., the thinner portion.
4334 G. Wang, A. Unal / Journal of Sound and Vibration 332 (2013) 4324–4338

3.3. Fully clamped plate

A single stepped thickness plate with clamped boundary conditions along four edges (denoted as CCCC) was investigated,
as shown in Fig. 5b. Definitions of side length and thickness ratio (i.e., h1/h2) are the same as the SSSS plate (b ¼1). The
thickness ratio varies from 1.5 to 0.6. Our SFEM predictions of natural frequencies and mode shape functions were validated
by the NASTRAN results for a square plate and a rectangular plate. The same dimensionless frequency parameter is assumed
as defined in the SSSS plate case.

Case 1. Square Plate, c/a ¼0.25

Table 4 shows the first four natural frequency results of the plate with different thickness variations. Our SFEM frequency
predictions were validated by the NASTRAN results, in which the QUAD4 plate elements with four different mesh sizes, i.e.,
4  4, 8  8, 16  16, and 32  32, were used. These different mesh sizes were assumed to demonstrate the convergence of
the NASTRAN simulations. In order to accurately predict up to fourth modal frequency, we must use as high as 32  32
meshes in the NASTRAN model. However, in this case, a 2  2 mesh size was used in our SFEM, which yields a 2  2 global
dynamic stiffness matrix in x and y directions, respectively. Two elements along x direction were used to account for the
thickness variation. Similar to a clamped-clamped beam, we must use two elements along y direction in order to account for
the boundary conditions. This same mesh size was used to predict the first four modes without increasing the mesh size as
observed in the NASTRAN model for higher modal frequency predictions. Substantial reduction of mesh size is achieved.
Our SFEM frequency results match well with the NASTRAN predictions using 32  32 meshes. Typically, it takes a few steps
to iteratively solve the equations in both x and y directions in our SFEM for each mode. Semi-analytical mode shape
functions are shown in Table 5. Physical insights can be collected from these mode shape solutions. Two different sets of
mode shape functions along x direction were obtained due to the thickness variation. Similar to the SSSS plate case, the
propagating wave and evanescent wave are associated with different wave numbers. Two dependent mode shape functions
along y direction were obtained as well with two same wave numbers as listed in Table 5. Each mode shape can be re-
constructed easily due to its semi-analytical form. Different spatial resolutions can be simply set by specifying desired data
points along each direction. Comparisons of each mode shape function are plotted in Fig. 7 for the case of h1/h2 ¼1.5. Our
SFEM mode shape predictions using 2  2 meshes correlate well with the NATRAN results, in which 32  32 meshes were
used. In order to demonstrate the thickness variation effect on the mode shapes, we plotted the first mode shape function
under four different thickness ratios as shown in Fig. 8. A symmetric mode is observed in a uniform plate (i.e., h1/h2 ¼1.0)
and the mode shape peak value is located at center. As the thickness ratio varies, the peak value shifts to the thinner
segment of the plate as expected for the case of h1/h2 ¼1.5, 1.25 and 0.6.

Fig. 6. Mode shape plots for a SSSS Square plate (Side length ratio: c/a ¼0.25 and h1/h2 ¼ 0.5): (a) Mode: 1  1; (b) Mode: 1  2; (c) Mode: 2  1; and (d)
Mode: 2  2.
 G. Wang, A. Unal / Journal of Sound and Vibration 332 (2013) 4324–4338 4335

Table 4
Dimensionless frequency results of a fully clamped square plate with single stepped thickness as shown in Fig. 5b, c/a ¼0.25 and a ¼b.

Mode h1/h2 ¼1.5 h1/h2 ¼1.25 h1/h2 ¼ 1.0 h1/h2 ¼ 0.6

(1,1)
NASTRAN, 4  4 mesh 4.5720 4.2879 4.0352 3.6966
8  8 mesh 2.7721 3.1326 3.6805 5.4677
16  16 mesh 2.7421 3.1013 3.6445 5.3998
32  32 mesh 2.7419 3.1012 3.6443 5.3962
Present, 2  2 mesh 2.7446 3.1041 3.6475 5.4031

(1,2)
NASTRAN,4  4 mesh 11.3299 10.9764 10.6297 9.3892
8  8 mesh 5.5776 6.5013 7.8539 11.1432
16  16 mesh 5.3182 6.2031 7.5002 10.6423
32  32 mesh 5.2829 6.1606 7.4472 10.5697
Present, 2  2 mesh 5.2565 6.1484 7.4375 10.5590

(2,1)
NASTRAN, 4  4 mesh 11.5478 11.1292 10.6297 9.8063
8  8 mesh 5.8053 6.6337 7.8539 12.1386
16  16 mesh 5.5511 6.3400 7.5002 11.5555
32  32 mesh 5.5111 6.2946 7.4472 11.4565
Present, 2  2 mesh 5.5047 6.2867 7.4375 11.4390

(2,2)
NASTRAN, 4  4 mesh 19.0384 17.9607 17.2436 16.3844
8  8 mesh 8.6778 9.8976 11.7401 17.4533
16  16 mesh 8.1478 9.3129 11.0453 16.3319
32  32 mesh 8.0818 9.2430 10.9650 16.1922
Present, 2  2 mesh 8.0838 9.2449 10.9670 16.1930

Table 5
Mode shape functions of a fully clamped square plate with single stepped thickness as shown in Fig. 5b, c/a¼ 0.25, h1/h2 ¼ 1.5.

8
< X m1 Y n1 0≤x≤c 0≤y≤b
W mn ¼
: X m2 Y n1 c≤x≤a 0≤y≤b

X m1 ¼ a1 sinðpx1 xÞþ a2 cosðpx1 xÞþ a3 sinhðqx1 xÞ þ a4 coshðqx1 xÞ 0≤x≤0:25

X m2 ¼ b1 sinðpx2 xÞ þ b2 cosðpx2 xÞ þ b3 sinhðqx2 xÞþ b4 coshðqx2 xÞ 0:25≤x≤1 x ¼ x−0:25

Mode (1,1) Mode (1,2) Mode (2,1) Mode (2,2)

a1 2.3272 1.1172 0.72281 1.0974

a2 −1.2306 −0.57056 −0.6292 −0.51156
a3 −1.1264 −0.49161 −0.57362 −0.49137
a4 1.2306 0.57056 0.6292 0.51156
px1 2.707 3.8454 6.247 4.7322
qx1 5.5925 8.7385 7.8719 10.569
b1 2.4205 2.6687 0.77096 0.9241
b2 0.21924 0.51263 0.84769 0.82133
b3 −0.76511 −0.48659 −0.1535 −0.17879
b4 0.78076 0.48737 0.15231 0.17867
px2 4.8928 4.4707 8.2171 8.1815
qx2 6.9201 10.542 9.5111 12.499
Y n1 ¼ c1 sinðpy yÞ þ c2 cosðpy yÞþ c3 sinhðqy yÞþ c4 coshðqy yÞ 0≤y≤0:25
Y n2 ¼ d1 sinðpy yÞ þ d2 cosðpy yÞ þ d3 sinhðqy yÞ þ d4 coshðqy yÞ 0:25≤y≤1 y ¼ y−0:25
c1 1.4385 0.76573 1.571 0.86889
c2 −0.92636 −0.63748 −0.57121 −0.52004
c3 −0.92402 −0.63761 −0.57118 −0.52004
c4 0.92636 0.63748 0.57121 0.52004
py 4.2858 7.6717 3.8391 7.3618
qy 6.6722 9.2132 10.559 12.3
d1 1.5021 0.33896 1.3691 0.26977
d2 0.81929 0.93693 0.95902 0.97603
d3 −0.1683 −0.064345 −0.040564 −0.02407
d4 0.18071 0.063073 0.040979 0.023968
4336 G. Wang, A. Unal / Journal of Sound and Vibration 332 (2013) 4324–4338

Fig. 7. Mode shape contour plots for a fully clamped square plate (c/a¼ 0.25, h1/h2 ¼ 1.5, and b/a¼ 1); SFEM ¼2  2 meshes and NASTRAN¼ 32  32 meshes:
(a) SFEM mode (1,1); (b) NASTRAN (c) SFEM mode (1,2); (d) NASTRAN; (e) SFEM mode (2,1); (f) NASTRAN; (g) SFEM mode (2,2); and (h) NASTRAN.

Fig. 8. First mode shape of a fully clamped square plate with different thickness variations, c/a¼ 0.25: (a) h1/h2 ¼ 1.5; (b) h1/h2 ¼ 1.25; (c) h1/h2 ¼1.0;
(d) h1/h2 ¼0.6.

Case 2. Rectangular plate, c/a¼0.5, b/a ¼2

Further validations are conducted for a rectangular plate. A 2  4 mesh size was used in our SFEM. As discussed before,
we need two elements to account for the plate thickness variation along x direction. Instead of using a minimum of two
elements along y direction, we used four equal length elements, which resulted in quick convergent frequency solutions
when using FSOLVE code in MATLAB to find zero determinant solutions. Further investigation will be conducted to identify
efficient solvers for these transcendental equations. Similar to the square plate case, we need as many as 32  32 meshes in
NASTRAN model to obtain first four modal frequencies and mode shapes. Table 6 shows the frequency results. Our SFEM
frequency predictions match well with the NASTRAN results. However, only a fraction of the mesh was used in the SFEM.
Fig. 9 shows the contour plots for the first four modes and very good correlations were achieved.
 G. Wang, A. Unal / Journal of Sound and Vibration 332 (2013) 4324–4338 4337

Table 6
Dimensionless frequency results of a fully clamped rectangular plate with single stepped thickness as shown in Fig. 5b, c/a ¼0.5 and b/a¼ 2.

Mode h1/h2 ¼1.5 h1/h2 ¼1.25 h1/h2 ¼ 1.0 h1/h2 ¼ 0.6

(1,1)
NASTRAN, 4  4 mesh 8.6000 9.5860 10.8249 13.4721
8  8 mesh 8.0391 8.9377 10.0812 12.6198
16  16 mesh 7.9581 8.8454 9.9760 12.4954
32  32 mesh 7.9476 8.8330 9.9618 12.4789
Present, 2  4 mesh 7.9502 8.8342 9.9623 12.484

(1,2)
NASTRAN,4  4 mesh 13.7605 14.9329 16.6590 22.0231
8  8 mesh 10.8531 11.8786 13.3042 17.2550
16  16 mesh 10.5209 11.5324 12.9257 16.7068
32  32 mesh 10.4887 11.5010 12.8924 16.6474
Present, 2  4 mesh 10.5 11.511 12.902 16.664

1,3)
NASTRAN, 4  4 mesh 28.2792 30.6614 34.0362 44.8492
8  8 mesh 16.7327 18.1801 20.2870 26.7459
16  16 mesh 15.1352 16.4884 18.4246 24.1421
32  32 mesh 14.9185 16.2634 18.1787 23.7732
Present, 2  4 mesh 14.899 16.238 18.148 23.737

(1,4)
NASTRAN, 4  4 mesh 36.6365 39.1415 43.1914 58.8779
8  8 mesh 22.8667 24.7590 27.4640 36.2925
16  16 mesh 21.8319 23.6783 26.2672 34.6697
32  32 mesh 21.0984 23.0934 25.8524 33.4784
Present, 2  4 mesh 20.971 22.939 25.671 32.557

Fig. 9. Mode shape contour plots for a fully clamped rectangular plate (c/a¼ 0.5, h1/h2 ¼ 1.5, and b/a ¼2); SFEM ¼2  4 meshes and NASTRAN¼ 32  32
meshes: (a) SFEM mode (1,1); (b) NASTRAN (c) SFEM mode (1,2); (d) NASTRAN; (e) SFEM mode (1,3); (f) NASTRAN; (g) SFEM mode (1,4); and (h) NASTRAN.

4. Conclusions

A spectral finite element method was successfully developed to conduct free vibration analysis of stepped thickness
rectangular plates. This is the first attempt to solve rectangular plates with Levy-type or non-Levy-type boundary
conditions. The spectral finite element models were formulated based on the analytical solutions using the Kantorovich
method. A dynamic stiffness matrix was employed to relate the nodal line displacement vector with nodal force vector to
yield a standard matrix representation in each direction. The continuity conditions are automatically satisfied after element
assembly as used in the CFEM. Iteratively, both mode shape functions in x and y directions can be determined. We must
emphasize that no special functions are required to initiate the iteration. Typically, uniform beam bending mode shape
functions with corresponding boundary conditions are used. Other polynomial functions can be used as well, as long as they
satisfy the geometry boundary conditions.
Literature results and NASTRAN simulations were used to validate the SFEM predictions for rectangular plates with CFCF,
SSSS, and CCCC boundary conditions and very good correlations are achieved. Our SFEM only requires a fraction of mesh
4338 G. Wang, A. Unal / Journal of Sound and Vibration 332 (2013) 4324–4338

compared to the NASTRAN model. The mode shape functions predicted by the SFEM are presented in a semi-analytical form.
Physical insights of wave propagation characteristics can be collected accordingly.
Currently, the FSOLVE subroutine in MATLAB was used to solve the frequency equation and performs well in most cases.
It is always challenging to solve a transcendental equation in terms of unknown frequency, which satisfies zero determinant
condition. Further investigation will be conducted to enhance the robustness of numerical scheme. We must emphasize that
the formulation and development of spectral finite element model can be extended to non-uniform plates. However, in that
case, the solutions would become complicated. In order to directly apply current SFEM, we can approximate a non-uniform
plate using a series of stepped thickness plates. Then we are able to determine each modal frequency and mode shape
function.
In summary, an efficient and accurate spectral finite element model was developed to conduct free vibration analysis of
stepped thickness rectangular plates with various boundary conditions.

Acknowledgments

Research partially supported by the junior faculty start-up funding by the University of Alabama in Huntsville.
Encouragement and helpful discussions with colleagues and comments from reviewers are highly appreciated.

References

[1] A. Leissa, Vibration of plates, NASA SP-160, U.S Government Printing Office, Washington DC, 1969.
[2] D.J. Gorman, Vibration analysis of plates by the superposition method, World Scientific, New Jersey, 1999.
[3] J.N. Reddy, Theory and analysis of elastic plates, Taylor & Francis, Philadelphia, PA, 1999.
[4] I. Chopra, Vibration of stepped thickness plates, International Journal of Mechanical Sciences 16 (6) (1974) 337–344.
[5] G.B. Warburton., Comment on: vibration of stepped thickness plates, International Journal of Mechanical Sciences 17 (3) (1975) 239.
[6] J. Yuan, S.M. Dickinson, The flexural vibration of rectangular plate systems approached using artificial springs in the Rayleigh–Ritz method, Journal of
Sound and Vibration 159 (22) (1992) 39–55.
[7] K.Y. Lam, G. Amrutharaj, Natural frequencies of rectangular stepped plates using polynomial beam functions with subsectioning, Applied Acoustics 44
(4) (1995) 325–340.
[8] S.J. Guo, A.J. Keane, M. Moshrefi-Torbati, Vibration analysis of stepped thickness plates, Journal of Sound and Vibration 204 (4) (1997) 645–657.
[9] O.C. Zienkiewicz, OC, R.L. Taylor, 5th ed. The finite element method, the basis, 1, Butterworth-Heinemann, Oxford, UK, 2000.
[10] L.V. Kantorovich, V.I. Krylov, Approximate methods of higher analysis, Groningen, the Netherlands, Noordhoff, 1964..
[11] V.H. Cortinez, P.A.A. Laura., Analysis of vibrating rectangular plates of discontinuously varying thickness by means of the Kantorovich extended
method, Journal of Sound and Vibration 137 (3) (1990) 457–461.
[12] R.B. Bhat, J. Singh, G. Mundkur, Plate characteristic functions and natural frequencies of vibration of plates by iterative reduction of partial deferential
equation, ASME Journal of Vibration and Acoustics 115 (2) (1993) 177–181.
[13] C. Rajalingham, R.B. Bhat, G.D. Xistris, Closed-form approximation of vibration modes of rectangular cantilevered plates by the variational reduced
method, Journal of Sound and Vibration 197 (3) (1996) 263–281.
[14] G. Wang, N.M. Wereley, Free in-plane vibration of rectangular plates, AIAA Journal 40 (5) (2002) 953–959.
[15] G. Wang, N.M. Wereley, D.C. Chang, Analysis of bending vibration of rectangular plates using two dimensional plate modes, Journal of Aircraft 42 (2)
(2005) 542–550. March-April.
[16] J.F. Doyle, Wave propagation in structures, Second Edition, Springer-Verlag, New York, 1997.
[17] A.Y.T. Leung, Dynamic stiffness and substructures, Springer-Verlag, London, 1993.
[18] S. Gopalakrishnan, A. Chakraborty, D.R. Mahapatra, Spectral finite element method: wave propagation, diagnostics and control in anisotropic and
inhomogeneous structures, Springer, Berlin, 2008.
[19] G. Wang, N.M. Wereley, N.M., Spectral finite element analysis of sandwich beams with passive constrained layer damping, ASME Journal of Vibration
and Acoustics 124 (3) (June, 2002) 376–386.
[20] G. Wang, N.M. Wereley, Free vibration analysis of rotating blades with uniform tapers, AIAA Journal 42 (12) (2004) 2429–2437.
[21] U. Lee, Spectral element method in structural dynamics, John Wiley & Sons (Asia) Ltd, Singapore, 2009.
[22] U. Lee, J. Lee., Spectral element method for Levy-type plates subject to dynamic loads,, Journal of Engineering Mechanics 125 (2) (1999) 243–247.
[23] M. Boscolo, J.R. Banerjee, Dynamic stiffness elements and their applications for plates using first order shear deformation theory, Computers and
Structures 89 (3–4) (2011) 395–410.
[24] D.J. Inman, Engineering vibration, Prentice Hall, Englewood Cliffs, NJ, 1994.