Smo1264 Imortant
Smo1264 Imortant
Smo1264 Imortant
Abstract This paper discusses issues related to designing band-gaps in periodic plane grid structures. Finite element analysis is used to solve the dynamic behavior of a
representative unit cell and Bloch-Floquet theory is used
to extend the results to the infinite structure. Particular attention is given to the addition of non-structural
masses that are introduced as design variables. These
are used to create desirable features in the dispersion
diagram. Physical insight is presented into the optimal
choice of locations where masses should be added and
the results of several numerical examples are provided to
highlight this and other features of how band-gaps can
be created and located at desired frequency ranges. The
effect of the skew angle of the underlying grid structure
is also explored, as are mathematical refinements of the
modelling of the beam elements and the rotational inertia of the added masses. A scaling feature between the
size of the reducible and the irreducible reference cell
is exploited and the manner in which this can simplify
optimization approaches is discussed.
Key words Grids, phononic structures,band-gaps, periodic structures
1
Introduction
Structures exhibiting band-gaps prevent the propagation
of waves at certain frequencies. These structures may be
phononic (sonic) or photonic, depending on their bandgap frequency range. Sonic band-gap structures can be
used as frequency filters and wave guides, while photonic
Received: date / Revised version: date
A. R. Diaz, A. G. Haddow and L. Ma
Department of Mechanical Engineering, Michigan State University, East Lansing, MI 488241226, USA
e-mail: diaz@egr.msu.edu, haddow@egr.msu.edu
?
Corresponding author
band-gap structures have applications in optics and microwaves. Because of their potential use in emerging technologies, particularly those involving MEMS and optical
devices, the study of band-gap structures is receiving significant attention in the current literature.
Band-gaps appear in many kinds of periodic structures. In this paper we investigate the presence of bandgaps in plane grid structures, i.e., plane structures made
of intersecting beams which deflect in the plane perpendicular to the plane of the structure. The goal is to create
and maximize band-gaps in such structures by adding
non-structural masses at strategic locations of the grid,
while retaining the periodicity of the structure. A standard optimization problem is formulated whereby optimal mass distributions are identified that maximize gaps
above specific bands while limiting the amount of added
material.
The literature dealing with photonic band-gaps is
very extensive. The popular web site Photonic & Sonic
Band-Gap Bibliography (Dowling and Everitt 2004) lists
over 3000 articles on photonic band-gaps as of this writing. Work on sonic or phononic band-gaps is much less
abundant, although the pace of activity in this area is
increasing rapidly. In early work, a theoretical analysis of vibration in periodic plane grids was discussed by
Heckel (1964). Heckel found bands of high attenuation
and bands of no attenuation for bending-wave transmission through plane grids. In recent work related to
the present paper, Martinsson and Movchan (2003), and
Jensen (2003) studied in-plane wave propagation in twodimensional (2D) periodic lattice structures, including
2D truss structures, plane frames and 2D mass-spring
models. The authors showed that complete band-gaps
exist for certain distributions of stiffness and mass and
demonstrated how band-gaps can be created at low frequency ranges by introducing a local resonator into periodic structures. An optimization perspective has also
been used to introduce band-gaps in materials with a
periodic structure. In Cox and Dobson (1999, 2000), a
gradient based optimization method and an evolution algorithm were used to produce band-gaps in 2D photonic
crystals. In Sigmund (2001) and Sigmund and Jensen
(2002, 2003), topology optimization techniques were used
to optimize band-gaps in elastic materials and to create
slave dof
master dof
slave dof
u7
u1
u3
u5
u8
u2
u4
u6
m1
m2
reference cell
beam:
E, I, A, L,
non-structural
masses:
m1 m2
Fig. 1 Schematic of the introductory example
(1)
u1
u2
u0 =
u3 ,
u4
u5
u1
u
6
u2 ik
u1 =
.. = .. e ,
..
..
and
u1
..
..
.. .. ik
=
u7 = u3 e
u8
u4
u7=u3e-ik
u8=u4e-ik
1
u1
u3
u2
u4
2
u5=u1eik
u6=u2eik
8000
4
3
4000
(2)
Frequency (Hz)
2
600
2 ( )
400
Gap
200
1
0
1 ( )
/2
Mode 1
1 ( )=200 Hz
20
40
60
80
100
Mode 2
2 ()=395 Hz
m1
m1
m2
t(2)
t(1)
and the optimal mass distribution will be further discussed in subsequent sections.
We now move on to the 2D problem involving plane
grids, the subject matter of this paper.
3
Geometry of the periodic grid structure
A periodic plane grid p is a plane grid that can be
constructed by periodically repeating a sub-section
of the grid through translations along two non-collinear
vectors t(1) and t(2) , i.e.,
p = ( + n1 t(1) + n2 t(2) ),
2
nZ
n1 , n 1 Z
(b) Representative
cell P
L = {x R2 : x = Tn, n Z2 }
The direct lattice characterizes the periodicity of the
grid, which in turn introduces constraints on the shape
of the waves that can propagate through the structure.
Using Bloch-Floquet theory (Brillouin 1953), the vibration properties of p can be extracted from the analysis of an irreducible representative cell . In this paper,
(3)
(4)
Mmm Mms
M =
Msm Mss
(5)
4.2
Stiffness and mass matrices for a 2D grid
representative cell
A standard finite element model is used here to model
the grid. The element used is a two-node, three degree
of freedom per node element, with degrees of freedom as
follows: translation perpendicular to the grid plane, rotation to account for out of plane bending, and rotation
about the longitudinal axis of the element. Bending is
modelled using either Euler-Bernoulli or Timoshenko assumptions. Joints at points of intersection of the grid are
assumed to be rigid and one or more elements are used to
model each segment between two adjacent intersections.
Consistent mass matrices are used in the analysis.
The stiffness and mass matrices of a representative
cell, respectively Kc and Mc , can be constructed following a standard finite element assembly procedure. After partitioning of the cells degrees of freedom uc
CN M D+N SD as uc = {um , us }T , these matrices are expressed as
Kmm Kms
K =
Ksm Kss
and
T =
1 c c c
(u ) M u
2
(9)
V =
1 c c c
(u ) K u
2
(10)
(6)
where B C(NSD,NMD). B is called here the quasiperiodicity matrix. Clearly, B depends on the wave vecT
tor k. Since slave nodes in are members of (1,0) or
T
(0,1) , B has only entries that are either 0, 1, eik1 , or
eik2 .
(8)
(7)
Here () denotes Hermitian transpose. In view of the periodicity of the structure, um and us are not independent
degrees of freedom and therefore, the energy expressions
(9) and (10) involve a non-minimal set of generalized coordinates. We seek mass and stiffness matrices such that
kinetic and potential energies in one cell are expressed
in terms of only um , i.e., matrices M and K such that
T =
1 m
(u ) Mu m
2
(11)
V =
1 m
(u ) Kum
2
(12)
T =
1 m
(u ) (Mmm + B Msm + Mms B + B Mss B)u m
2
t(2)
(13)
60o
t(1)
(14)
^k2
k2
d(2)
A
M
um + Kum = f
(15)
B
120o
d(1)
(b) Cell in reciprocal lattice
(17)
where = 2 is an eigenvalue and a CN M D is the corresponding eigenvector. Note that because of the presence
of the quasi-periodicity matrix B in (13) and (14), M
and K are functions of the wave vector k = (k1 , k2 ) and
therefore both and a depend on k. As discussed in
the introductory example, a dispersion diagram displays
bands of eigenfrequencies as functions of the wave vector,
where each band is associated with a given mode (this
was illustrated in Fig. 3). In principle, in order to draw
this diagram, equation (17) should be solved for each
k R2 . However the periodicity of the structure results
in significant simplifications. In fact, all eigenfrequencies
on the periodic structure are produced by exploring only
a small subset of R2 . This leads to the concept of reciprocal lattice.
that yield waves with
The set of all wave vectors k
the same periodicity of the direct lattice L is known as
the reciprocal lattice LR . This lattice can be described
using two tiling vectors d(1) and d(2) , as the set
R2 : k
= m1 d(1) + m2 d(2) , (m1 , m2 ) Z2 }
LR = {k
where d(1) and d(2) are the solution to
k1
4.4
The dispersion diagram
um = aeit
^1
C k
(18)
belongs to
for i, j=1, 2. In view of (3), a wave vector k
ikx
ik(x+X)
LR provided that e
= e
for any x L and
LR if eikX
all X L. This implies that k
= 1 for
(19)
2
t(2)
p
p
5
Optimization problem
k2
t(1)
k1
4.5
Further Simplifications
It is possible to show that all frequencies in bj in (19) can
be obtained even when a smaller subset of the (k1 , k2 )
plane is explored. For instance, it is easy to show that the
mass and stiffness matrices M and K (see (13) and (14))
are invariant with respect to transformations (k1 , k2 ) 7
(k1 , k2 ) within the first Brillouin zone. Therefore, if
a(k) is a mode with eigenfrequency , so is a(-k). As a
result, the j-th band can be obtained simply by searching
in the upper-half of the Brillouin zone, i.e.,
bj = {j (k) : k [, ] [0, ]}
To account for this and perhaps other simplifications,
we define as a subset of [, ]2 such that all solutions = 2 to the eigenvalue problem (17) are obtained
whenever bj is defined as
bj = {j (k) : k }
(20)
(21)
subject to
0 xi xmax = mref
(22)
(23)
G(x)
=
1p,qm
(24)
maximizes G(x)
=
subject to
1p,qm
0 xi xmax = mref
(25)
maximize z
subject to Gpq (x) = j+1 (kp , x) j (kq , x) z
0 xi xmax = mref
(26)
(27)
where
2j j0 = (aj ) K0 aj j (aj ) M0 aj
(28)
j
j X
=
(ajm ) Fajm
xi
2
(29)
mDi
where entries ajm represents the component in the massnormalized eigenvector aj corresponding to the degrees
of freedom of node m and F is 3 3 is diagonal with
diagonal entries (1, fx , fy ) (recall that for each design
variable xi the mass matrix is modified by adding xi ,
fx xi , and fy xi to the affected degrees of freedom).
6
Examples
In the following examples we consider the design of grids
characterized by representative cells of different sizes,
including those shown in Fig. 8. The beam sections are
rectangular with cross section b h = 2 0.5 mm with
the wide side on the plane of the grid. Masses are placed
symmetrically on , at grid joints and also mid-side
in each grid segment (each grid segment between two
grid corners is discretized using two finite elements). The
bound on the maximum allowable additional added mass
is xmax = mref where mref = 8bhL0 and L0 = 10 mm
is the size of one grid segment. For simplicity, in what
follows x is normalized by x x/xmax so that in re-
9
1
4
4
4
2
3
2 1
(b) 2x2 cell
2
2
3
1 2 1
(a) 1x1 cell
1 2 3 2
4
6
8
4
3
7
10 9
6
8
11
1
2
3
10 11 12 11
10 7
11
10 9
3
2
2
1
6
2
4 5 4
(c) 4x4 cell
2
2
dG
=
dx3
1
2
P
mD3
1
2
6.1
Example 1
This simple example will serve to illustrate the nature
of solutions to the 2D problem. We seek to design a
band-gap into the grid by adding masses in a symmetric
arrangement characterized by a 22 square irreducible
cell of side L = 2L0 = 20mm. The gap will be placed
above the second band.
At the solution to this problem, the dispersion diagram is as shown in Fig. 9 (a). The gap appears above 34
kHz and extends to 68 kHz. The solution consists of simply adding the largest possible mass at design position 3
of the cell (mid-edge, Fig. 9(b)).
The optimal mass distribution can be explained by
examining the modes at critical locations in the dispersion diagram. For example, the mode associated with
the maximum eigenvalue of band 2, at point (0,0) of the
dispersion diagram, achieves the maximum magnitude
at design position 3 (Fig. 9(b)). Since adding masses at
points with larger displacements introduces more substantial drops in the corresponding eigenvalue, adding a
mass at position 3 is most effective at lowering band 2.
2
2
(2 a2m 3 a3m ) =
P
mD3
2
2 a2m > 0
where ajm , m D3 are amplitudes of modes j=2,3 measured at design position 3. For this particular design
a3m = 0 at all four locations associated with position
dG
3, insuring that dx
> 0, which explains why x3 should
3
be made as large as possible. A similar argument can be
applied to design position 1, to conclude that no mass
dG
should be placed there since dx
< 0. While in general it
1
is cumbersome to carry this analysis on to larger, more
complex problems, the analysis does generate insights
into the nature of solutions to this problem.
6.2
Example 2
In this example we seek to design a band-gap into the
grid by adding masses in a symmetric arrangement characterized by a 44 irreducible cell of side L = 4L0 =40
mm. We consider both a square and a skewed lattice
with angles =90 and =60 and translation vectors
t(1) = (L, 0)T and t(2) = L(cos , sin )T . Gaps are introduced above bands 1, 2 and 5.
There are some general comments that can be made
regarding the results of this example. In general, the way
in which the mass is distributed around the reference cell
is dictated by the mode shapes that are involved and
the conflicting requirements of lowering the critical frequency of the lower band, while trying to increase the
critical frequency of the upper band. If we wish to create a gap between bands 1 and 2, it is advantageous to
add as much mass as possible close to the center of the
reference cell, as doing this lowers the critical point of
band 1 without a corresponding decrease in the critical
point of band 2. The situation is similar when the gap is
to be introduced above higher bands. A more quantitative analysis could be presented, similar to the approach
dG
in Example 1, but for brevity it is omitted.
using dx
3
Note that the upper limit of additional mass xmax
assigned in this example is not large enough to create a
gap for the square reference cell yet it is enough to open
a small gap for the skewed cell(see Fig. 10(a) and Fig.
10
frequency(kHz)
200
150
no gap could be
introduced
here
100
gap
50
0
(,)
(0,0)
(,0)
k vector
(,)
x=0
x=1
x=0.
x=0.06
x=0.57
x=0.65
x=0.71
x=1.
6.3
Example 3
In design applications one may be interested in creating
a gap near a particular frequency. Here we discuss how
this can be attempted without changing material or section properties, or grid dimensions. One alternative is to
fix the scale of the problem by selecting the size of the
irreducible cell - e.g., 4 4 - and proceed to solve several problems, each problem maximizing the gap above
a different band. From the solutions obtained one then
selects the one that best fits the design purpose. While
there is no guarantee that a suitable band-gap in the
desired frequency range will be found this way, this procedure results in designs which may then be scaled, if
necessary, by scaling the material properties.
In order to understand the gap positioning problem
better, it is useful to look into the effect of the size of the
irreducible cell used in the analysis. If the tiling period
of the mass distribution is characterized by a 1 1 irre-
11
50
200
no gap could be
introduced
here
150
30
20
10
gap
(,)
frequency(kHz)
(0,0)
(,0)
k vector
frequency(kHz)
frequency(kHz)
40
(,)
50
40
40
30
30
20
20
gap
50
50
100
(,)
(0,0)
(,0)
k vector
(,)
gap
10
10
0
(,)
(0,0)
(,0)
k vector
(,)
(,)
(0,0)
(,0)
k vector
(,)
50
frequency(kHz)
150
0
100
gap
50
40
40
40
frequency(kHz)
30
20
gap
gap
0
20
20
(,)
(,)
(0,0)
(,0)
k vector
(,)
00
(,)
(,)
(0,0)
(,0)
(0,0)
(,0)
vector
kk vector
40
50
(,)
200
(,)
(,)
30
(0,0)
(,0)
k vector
10
10
10
30
30
60
70
80
90
150
frequency(kHz)
frequency(kHz)
gap
100
gap
50
Gap/Gap Max
1
0
0.8
0.6
(,)
(0,0)
(,0)
k vector
(,)
1x1 G1
0.4
Fig. 13 Dispersion diagrams for an optimized periodic material distribution analyzed using different reference cell sizes
4x4 G2
0.2
2x2 G2
0
30
40
50
60
70
Angle
80
90
ducible cell, optimal solutions of the problem have dispersion diagrams as shown in Fig. 13(a). The maximum
gap above band 1 starts at L = 50 kHz and extends for
95 kHz. Design variables (scaled by xmax ) are x=(1,0).
If the same design were to be repeated over a 2 2 or
a 4 4 patch of the grid (thus forming a reducible cell),
the dispersion diagram would show exactly the same gap
frequency(kHz)
12
200
x=1.0
150
x=0.71
x=0.0
100
gap
gap
50
(,)
(0,0)
(,0)
k vector
(,)
(,)
(0,0)
(,0)
k vector
(,)
frequency(kHz)
150
100
50
gap
gap
0
(,)
(0,0)
(,0)
k vector
(,)
(,)
(0,0)
(,0)
k vector
(,)
6.5
Example 5. Model refinements: effect of Timoshenko
model
Experts in this field may object to the use of EulerBernoulli beam theory in this analysis, finding instead
that a Timoshenko model is more suitable. By means of
a couple of examples, here we expose some of the differences that one may expect to find should a Timoshenko
beam model be used. Again, we use solutions obtained in
Example 3 for 22 and 44 cells as a reference and compare the dispersion diagrams under Euler-Bernoulli and
Timoshenko beam assumptions. The results are shown
in Fig. 17.
The relaxation of the kinematic constraints introduced by the Timoshenko beam has a complex effect on
200
200
200
150
150
150
frequency(kHz)
frequency(kHz)
13
100
gap
50
(,)
(0,0)
(,0)
100
100
gap
gap
50
50
(,)
(,)
(,)
(0,0)
(0,0)
frequency(kHz)
(,0)
(,0)
(,)
(,)
kk vector
vector
k vector
200
200
150
150
100
100
50
50
6.6
Example 6. Finite size solutions
gap
0
(,)
(0,0)
(,0)
gap
0
(,)
(,)
(0,0)
k vector
(,0)
(,)
k vector
frequency(kHz)
the solution at higher frequencies, which in this case results in a net reduction of the gap size. The effect is less
significant at lower frequencies but there again the Timoshenko beam model has smaller band-gaps. It should be
noted that as the mass distributions that result in the
dispersion diagrams shown in Fig. 17(b) and (d) were
not designed to be optimal under the Timoshenko beam
assumptions, one may expect a small improvement if the
mass distribution were optimized under the new modelling assumptions. However, in these examples one can
show that the improvements obtained by further optimization are not significant.
200
200
150
150
100
100
gap
50
(,)
(0,0)
(,0)
(,)
(,)
(0,0)
(,0)
(,)
k vector
k vector
frequency(kHz)
gap
50
50
50
40
40
30
30
20
20
gap
gap
10
10
0
(,)
(0,0)
(,0)
(,)
k vector
(,)
(0,0)
(,0)
(,)
k vector
Fig. 17 Dispersion diagrams indicating the difference between using Euler-Bernoulli and a Timoshenko beam models
14
0.1
A/A0
0.1
Gap
Gap
30
Frequency (kHz)
(a) 4 4 patch
30
Frequency (kHz)
(b) 12 12 patch
7
Conclusions and discussion
A variety of examples demonstrate different strategies to
design band-gaps into a plane grid structure. The sensitivity of the gap width and its position relative to different bands has been studied with respect to the addition of non-structural masses. Solutions are found to be
related to specific critical points in the dispersion diagram. Typically, these points are at values of the wave
vector associated with standing waves in the structure.
An understanding of the sensitivity of the gap width to
the mass position can be gained by studying the modes
at these critical points. The influence of the skew angle of the grid on the distribution of the bands has also
been reported. The relationship between this angle, the
width of the band-gap, and the inter-band positioning
has been numerically investigated and the complexity of
this relationship was demonstrated. Clearly, this angle
could be employed as another design parameter in the
optimization scheme, but this was not detailed in the
current study. However, some effects of the underlying
mathematical modelling of the structure were investigated. For instance, it was observed that the difference
between an Euler and a Timoshenko beam model was
more significant at the higher frequencies. The effect of
including rotational inertia to the non-structural masses
was also shown to influence the higher frequencies. This
latter effect could be employed as a design variable as
the construction of the actual structure might be chosen
to allow this quantity to be varied in an real design.
A novel way of viewing the scaling issue of the reference cell was discussed. The relationship between small,
irreducible cells and larger, reducible cells was presented.
Exploiting this relationship so that only a low number of
bands at the lower frequencies need be considered when
trying to position a gap can lead to a reduction in the
level of complexity of the system to be optimized. A sequence of 1 1 , 2 2 , and 4 4 reference cell sizes
where used to numerically demonstrate this observation.
References
Brillouin, L. 1953: Wave propagation in periodic structures.
2nd edition, New York: Dover Publications Inc.
Cox, S.J.; Dobson, D.C. 2000: Band structure optimization of
two-dimensional photonic crystals in H- polarization, Journal
of Computational Physics 158, 214224.
Cox, S.J.; Dobson, D.C. 1999: Maximizing band-gaps in twodimensional photonic crystals, SIAM J.Appl. Math. 59, 2108
2120.
Diaz, A; Benard, A. 2001: On The Discretization Of Problems Involving Periodic Planar Tilings, Communications in
Numerical Methods in Engineering, 17, 543549.
Dowling J. and Everitt H. 2004: Sonic Band-Gap Bibliography, http: // home.earthlink.net /jpdowling/ pbgbib.html
Heckel, M.A. 1964: Investigations on the vibrations of grillages and other simple beam structures, Journal of the Acoustical Society of America 36, 13351343.
Jensen, J.S. 2003: Phononic band-gaps and vibrations in
one-and two-dimensional mass-spring structures, Journal of
Sound and Vibration 266, 10531078.
15
Kafesaki, M., Sigalas, M. M., Economou, E. N. 1995: Elastic
wave band-gaps in 3-D periodic polymer matrix composites,
Solid State Communications 96 (5), 285289.
Langlet, P.; Hladky-Hennion, A.-C.; Decarpigny, J.N. 1995:
Analysis of the propagation of plane acoustic waves in passive
periodic materials using the finite element method, Journal
of the Acoustical Society of America 98, 27922800.
Ma, L.; Diaz, A.; Haddow, A. 2004: Modeling and design of
materials for controlled wave propagation in plane grid structures, Paper DETC2004-57183, Proc. of DETC04, ASME
2004 Design Engineering Technical Conf. , Salt Lake City,
UT.
Martinsson, P.G.; Movchan, A.B. 2003: Vibrations of lattice
structures and phononic band-gaps, Q.Jl Mech. Appl. Math.
56 (1), 4564.
Orris, R.M.; Petyt, M. 1974: A finite element study of harmonic wave propagation in periodic structures, Journal of
Sound and Vibration 33, 223236.
Sigalas, M.; Economou, E. 1994: Elastic waves in plates with
periodically placed inclusions, Journal of Applied Physics,
75(6), 28452850.
Parmley, S.; Zobrist, T.; Clough, T.; Perez Miller,A.; Makela,
M.; Yu, R. 1995: Phononic band structure in a mass chain,
Appl. Phys. Lett. 67 (6), 777779.
Sigmund, O. 2001: Microstructural design of elastic band-gap
structures, Proc. of the Second World Congress of Struct. and
Multidisc. Optim. . Dalian, China.
Sigmund, O.; Jensen, J.S. 2002: Topology optimization of
phononic band-gap materials and structures, Fifth World
Congress on Computational Mechanics. Vienna, Austria.
Sigmund, O.; Jensen, J.S. 2003: Systematic design of phononic band-gap materials and structures by topology optimization, Phil. Trans. R. Soc. Lon. A 361, 10011019.
Svanberg, K. 1987: The method of moving asymptotes.
Int.J.Num. Mech. Eng. 24,459373.
Vasseur, J.O.; Deymier, P.A.; Frantziskonis, G.; Hong, G.;
Djafari-rouhani, B.; Dobrzynski, L. 1998: Experimental evidence for the existence of absolute acoustic band-gaps in twodimensional periodic composite media, Journal of Physics:
Condensed Matter 10, 60516064.