Professional Documents
Culture Documents
Random Acoustic Compmech
Random Acoustic Compmech
https://doi.org/10.1007/s00466-018-1548-y
ORIGINAL PAPER
Received: 23 August 2017 / Accepted: 14 January 2018 / Published online: 22 January 2018
© Springer-Verlag GmbH Germany, part of Springer Nature 2018
Abstract
Acoustic metamaterials (AMs) are man-made composite materials. However, the random uncertainties are unavoidable in the
application of AMs due to manufacturing and material errors which lead to the variance of the physical responses of AMs. In
this paper, an uncertainty model based on the change of variable perturbation stochastic finite element method (CVPS-FEM)
is formulated to predict the probability density functions of physical responses of AMs with random parameters. Three types
of physical responses including the band structure, mode shapes and frequency response function of AMs are studied in the
uncertainty model, which is of great interest in the design of AMs. In this computation, the physical responses of stochastic
AMs are expressed as linear functions of the pre-defined random parameters by using the first-order Taylor series expansion
and perturbation technique. Then, based on the linear function relationships of parameters and responses, the probability
density functions of the responses can be calculated by the change-of-variable technique. Three numerical examples are
employed to demonstrate the effectiveness of the CVPS-FEM for stochastic AMs, and the results are validated by Monte
Carlo method successfully.
Keywords Acoustic metamaterials · Random parameters · Perturbation stochastic finite element method · Change-of-variable
technique · Probability density function
123
1024 Computational Mechanics (2018) 62:1023–1036
resonant mechanism, Lu et al. [1] proposed a topology opti- FEM) has been proposed by Xia et al. [39] to calculate
mization method for the optimal design of LRAMs to obtain the probability density functions of responses of stochas-
negative bulk modulus at certain frequency ranges. tic structures. However, the formulations of CVPS-FEM
An efficient numerical algorithm for the simulation of to solve the probability distributions of natural frequen-
AMs is of great importance in practical application of AMs. cies and mode shapes of stochastic structures are still not
Recently, some computational techniques have been pre- available.
sented to simulate the wave propagation in AMs, such as To obtain the reliable performance of the AMs with
the plane-wave expansion (PWE) method [19] and the finite- random parameters, the physical responses of stochastic
difference time-domain (FDTD) method [20]. PWE method AMs including band structure, mode shapes and frequency
is one of the most widely used techniques for the calcula- response function must be predicted accurately. This is
tion of AMs, however, it has an inherent drawback of poor extremely important in the design and application of AMs
convergence [21]. Furthermore, it is difficult for FDTD to in engineering practices. In order to predict the probability
simulate AMs with complex geometries because of the stair- density functions of band structure, mode shapes and fre-
case effect [22]. Recently, finite element method (FEM) has quency response function of AMs with random parameters,
become a popular method to simulate AMs. Based on the the CVPS-FEM is applied to formulate the random uncer-
frame of FEM, lumped mass (LM) method was proposed to tainty model of stochastic AMs in this work. In this random
calculate the wave propagation in phononic crystals by Wang uncertainty model, the random physical responses of AMs
et al. [23]. In addition, the finite element time-domain method are approximated as linear functions of the random param-
was developed to analyze the optical black hole in metamate- eters by using the first-order Taylor series expansion and
rials by Yang et al. [24]. A mixed finite element methods was perturbation technique, then, based on the linear function
proposed to analyze the wave propagation in double negative relationships of parameters and responses, the probability
metamaterials by Li [25]. Consequently, smoothed finite ele- density functions of the responses can be calculated by the
ment method (S-FEM) has been applied to compute the band change-of-variable technique.
gap of AMs [26–29]. The remainder of this paper is organized as follows. In
It is well known that the manufacturing tolerances, unpre- Sect. 2, the calculation of the probability density functions
dictable environment factors, and the material properties for linear functions by using change-of-variable technique
errors always exist, leading to the variance of the physical is presented, and the elastic wave in AMs is discussed. The
response of AMs. Therefore, in the design and application of formulation of the uncertainty model by using the CVPS-
AMs, it is of great significance to propose an efficient uncer- FEM to predict the random band structure, mode shapes and
tainty model to predict the physical responses of AMs with frequency response function of stochastic AMs is illustrated
consideration of uncertainty effects. Generally, the uncer- in Sect. 3. Three numerical examples are used to demonstrate
tainties of engineering structural parameters are defined as the accuracy of the uncertain model of AMs in Sect, 4. Several
random variables whose probability density functions can conclusions are given in Sect. 5.
be obtained on the basis of sample data statistical analysis.
Recently, the sources of uncertainty with functions of space
or time combined with machine learning algorithm have been
studied [30,31]. 2 Mathematical backgrounds
In general, the uncertain problems with random param-
eters can be treated as stochastic problems, and a lot of 2.1 Change-of-variable technique for linear
probabilistic approaches have been widely used to solve functions
the these problems, such as the Monte Carlo method
(MCM) [32,33], the spectral stochastic finite element method Assume that g = h (x1 , . . ., xm ) is a linear function with m
(SSFEM) [34,35] and the perturbation stochastic finite ele- independent random parameters, which is defined as
ment method (PSFEM) [36,37]. Among these probabilistic
approaches, the PSFEM has become more and more popu-
lar in the analysis of stochastic problems [36]. However, the
m
g = h (x1 , . . ., xm ) = A0 + Ai xi , i = 1, . . ., m (1)
random results obtained by PSFEM are given in the form
i=1
of mean value and standard deviations, and the probability
density functions of results cannot be obtained unless the
results are treated as Gaussian random fields [38]. In order where the probability density function of xi can be given as
to obtain the probability density functions of the random pxi (xi ).
results when non-Gaussian variables exist, the change of vari- Based on change-of-variable technique [39], the probabil-
able perturbation stochastic finite element method (CVPS- ity density function of g can be given as
123
Computational Mechanics (2018) 62:1023–1036 1025
m - 1
where ym = Am xm , Ym -2 = i=1 Ai x i .
2.2 Elastic wave in AMs Fig. 1 a The first Brillouin zone (square) and the irreducible Brillouin
zone (triangle) for a square lattice; b the first Brillouin zone (hexagon)
and the irreducible Brillouin zone (triangle) for a hexagon lattice
Assume that the media where the elastic wave propagates is
isotropic, and the governing equation of the elastic wave in
the frequency domain can be given as Combined with Eq. (7), Eq. (6) can be calculated for any
wave vector k limited in the first Brillouin zone. Furthermore,
(λ + μ) ∇ (∇ · u (r)) + μ∇ 2 u (r) = −ρω2 u (r) (3) this area is limited in the irreducible Brillouin zone (IBZ) for
the square and hexagon lattice as plotted in Fig. 1. The band
where r = (x, y, z) is the position vector, u (r) is the dis- structure can be obtained when the wave vector k is adopted
placement vector, ∇ = (∂/∂ x, ∂/∂ y, ∂/∂z) is the vector along different directions of M → → X → M on the
differential operator and ω is the angular frequency of the boundary of the IBZ. The band gap exists along the ω axis
propagating wave. λ and μ are Lamé’s first parameter and when two adjacent dispersion curves do not overlap.
shear modulus, respectively, which can be given as
vE E
λ= , μ= (4) 3 CVPS-FEM for AMs with random
(1 + v) (1 − 2v) 2 (1 + v)
parameters
where E is the Young’s Modulus, and v is the Poisson’s ratio.
Based on the Bloch’s theorem of wave propagation in the In this work, all uncertain parameters existing in AMs are
periodic structure, the displacement vector u (r) for the two- described as random parameters, which belong to a ran-
dimensional periodic AMs can be expressed as dom vector a = (a1 , a2 , . . ., am ), and the probability density
functions
of these random parameters are given as pa (a) =
u (r) = ei(k−r) uk (r) (5) pa1 (a1 ) , pa2 (a2 ) , . . ., pam (am ) .
As mentioned in Sect. 2, based on FEM and periodic
where uk (r) is the periodic function which has the same Bloch conditions, the band structure and mode shapes of
AMs can be calculated. Furthermore, the amplitude of fre-
periodic
characteristic with the AMs, the wave vector k = quency response of AMs is sharply reduced at the frequency
k x , k y is limited in the first Brillouin zone, and r = r (x, y)
is the position vector. range where the band gaps exist, hence the band gaps can
By using the Bloch’s theorem, the solution of the periodic be verified by calculating the frequency response function.
AMs structure system can be reduced one unit cell. Then, However, when the uncertain random parameters are intro-
based on the finite element method, Eq. (3) of the unit cell duced, these physical responses of AMs mentioned above are
can be expressed as changed. In this section, the CVPS-FEM is formulated, and
the probability distributions of the random physical responses
including the band structure, mode shapes and frequency
K − ω2 M U = 0 (6)
response function of AMs are calculated.
123
1026 Computational Mechanics (2018) 62:1023–1036
UTj MU j = 1 (9)
Therefore, the random natural angular frequency ω j (a) can
It has been proven [41] that the mass matrix M is always sym- be expressed as
metric positive definite because the shape functions satisfy
the partitions of unity property, and Eq. (9) will be satisfied ω j (a) = ωej + ω j
based on the so-called Rayleigh Quotient [42].
m e
1 eT ∂Ke e2 ∂M
Considering the existence of the random parameters, Eq. = ωej + U j −ω Uej ai −aie
e
2ω j ∂ai j
∂ai
(8) can be expressed as i=1
(15)
K (a) U j (a) = ω2j (a) M (a) U j (a) (10)
Then, Eq. (15) can be rewritten as
where K (a), M (a), ω j (a) and U j (a) are the stiffness
matrix, mass matrix, jth natural angular frequency and jth
m
ω j (a) = B0 + Bi ai (16)
mode shape of the AMs with considering random parameters,
i=1
respectively.
The first-order Taylor series expansion for the linear func- where
tion is exact, and the accuracy of it for the non-linear function
m
1 eT ∂Ke e
is also acceptable if the deviation of uncertain parameter is e2 ∂M
B0 = ωej − U − ω Uej aie
small [43]. On the other hand, by using the higher order 2ωej j ∂ai j
∂ai
i=1
Taylor expansion with expensive computational cost, the e
1 eT ∂Ke e2 ∂M
improvement of accuracy is rather small [39]. As our main Bi = U − ω Uej (17)
focus in this work is to study the small deviation of random 2ωej j ∂ai j
∂ai
parameters of AMs, only the first-order Taylor series expan-
By using the change-of-variable technique mentioned in
sion is used.
Sect. 2, the probability density function of the random natural
Based on the first-order Taylor series expansion, the stiff-
angular frequency ω j (a) can be given as
ness matrix K (a) and mass matrix M (a) can be expressed
as
pω j (a) ω j (a)
⎧
m
∂K (ae ) ⎪ 1 ω j (a)−B0
(m = 1)
K (a) = K ae + ai − aie = Ke + K (11) ⎪
⎪ |B1 | pa1
⎨ B1
∂ai ∞ ω (a)−B −B a
i=1 = |B1 | a =−∞ pa1 (a1 ) pa2 j 0 1 1
da1 (m = 2)
⎪
⎪ 2 1 B2
m ⎪
∂M (ae ) ⎩ ∞
M (a) = M ae + ai −aie = Me +M (12) am =−∞ pam (am ) p Fm−2 ω j (a) − B0 − Bm am dam (m > 2)
∂ai (18)
i=1
m−1
where ae is the mean value of a = (a1 , a2 , . . ., am ), Ke and where Fm−2 = i=1 Bi ai .
K are the mean value and deviation value of K (a), Me and The natural frequency f j can be given as
M are the mean value and deviation value of M (a).
The first-order perturbation of Eq. (10) can be given as ωj
fj = (19)
2π
e
K + K Uej + U j Similarly, based on the change-of-variable technique, the
2 probability density function of the random natural frequency
= ωej + ω j Me + M Uej + U j (13) f j (a) can be given as
where Uej and U j are the mean value and deviation value of p f j (a) f j (a) = 2π pω j (a) 2π f j (a) (20)
U j (a), ωej and ω j are the mean value and deviation value
of ω j (a). 3.2 CVPS-FEM for mode shape of AMs with random
Due to the compact supports of FEM shape functions, it parameters
can be easily obtained that K and M are symmetric [44].
T
Pre-multiplying both sides of Eq. (13) by Uej , using the For convenience, Eq. (10) can be rewritten as
normalization condition and ignoring the higher-order terms,
ω j can be given as K (a) U j (a) = λ j (a) M (a) U j (a) (21)
123
Computational Mechanics (2018) 62:1023–1036 1027
where λ j (a) = ω2j (a) is the jth eigenvalue of AMs unit cell. Substituting Eqs. (26) and (30) into Eq. (24), gives
The first-order perturbation of Eq. (21) can be given as T
n Ue K − λej M Uej
U j =
q 1 T
Uqe − Uej MUej Uej
Ke + K Uej + U j λej − λqe 2
q=1
q= j
= λej + λ j Me + M Uej + U j (22)
(31)
where λej and λ j are the mean value and deviation value of
Therefore, the random mode shape U j (a) can be expressed
λ j (a), respectively.
as
Ignoring the higher-order terms, Eq. (22) can be rewritten
as U j (a) = Uej + U j
⎡⎛
∂Ke
− λej ∂M
T e
Ke U j + KUej = λej MU j + λej MUej + λ j Me Uej m ⎢⎜
n Ue Uej
⎢⎜ q ∂ai ∂ai
= Uej + ⎢⎜ Uqe
(23) ⎣⎝ λej − λqe
i=1 q=1
q= j
The deviation value of the jth mode shape U j can be given ⎞ ⎤
as the superposition of these deterministic mode shapes [45]
1 T ∂Me e e ⎟ ⎟ ⎥
⎥
− Uej U j U j ⎟ ai − aie ⎥ (32)
n
2 ∂ai ⎠ ⎦
U j = Cq Uqe (24)
q=1
T T
Uej Me U j + Uej MUej + UTj Me Uej = 0 (28) The random mode shape U j (a) can be given in another
more apparent vector forms as
T ⎧ ⎫ ⎧ ⎫ ⎧ ⎫
Pre-multiplying each side of Eq. (24) by Uej Me , gives ⎪
⎪ U j,1 (a) ⎪
⎪ ⎪
⎪ D0,1 ⎪
⎪ ⎪
⎪ Di,1 ⎪
⎪
⎪
⎪ .. ⎪
⎪ ⎪ .. ⎪
⎪ ⎪ ⎪
⎪ .. ⎪
⎪
⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
T ⎨ . ⎬ ⎨ . ⎬ ⎨ . ⎪ ⎬
Cq = Uej Me U j (q = j) (29) U j (a) = U j,k (a) ; D0 = D0,k ; Di = Di,k
⎪
⎪ .. ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪
⎪
⎪
⎪ .
⎪
⎪
⎪ ⎪ ... ⎪
⎪
⎪ ⎪
⎪ ⎪ ... ⎪
⎪
⎪ ⎪
⎪
Substituting Eq. (29) and its transpose into Eq. (28), Cq can ⎪
⎩ ⎪
⎭ ⎪
⎩ ⎪
⎭ ⎪
⎩ ⎪
⎭
be given as U j,n (a) D0,n Di,n
m
1 T U j (a) = D0 + Di ai (35)
Cq = − Uej MUej (q = j) (30)
2 i=1
123
1028 Computational Mechanics (2018) 62:1023–1036
m
where Ze and Z are the mean value and deviation value of
U jk (a) = D0,k + Di,k ai (36)
Z (a), Fe and F are the mean value and deviation value of
i=1
F (a).
By using the change-of-variable technique mentioned in The first-order perturbation of Eq. (41) can be given as
Sect. 2, the probability density function of the random dis- e
placement U jk (a) can be given as Z + Z Re + R = Fe + F (44)
The structural dynamic equilibrium equation of AMs model Equation (46) can be rewritten as
can be given as
m
R (a) = H0 + Hi ai (47)
K − ω2 M R = F (38) i=1
where
where F and ω are the external unit load and the cor-
responding angular frequency, respectively. R donates the
m !
e −1 ∂Fe ∂Ze e e
displacement response of the AMs. H = Re − Z − R ai
∂ai ∂ai
To simplify the calculation process, Eq. (38) can be rewrit- i=1
!
ten as −1 ∂Fe ∂Ze e
Hi = Ze − R (48)
∂ai ∂ai
ZR = F (39)
The random displacement response R (a) can be given in
where Z is the structural dynamic stiffness matrix of the AMs another more apparent vector forms as
and can be given as ⎧ ⎫ ⎧ ⎫ ⎧ ⎫
⎪
⎪ R1 (a) ⎪
⎪ ⎪
⎪ H0,1 ⎪
⎪ ⎪
⎪ Hi,1 ⎪
⎪
⎪ . ⎪
⎪ ⎪ ⎪
⎪ .. ⎪⎪ ⎪
⎪ .. ⎪⎪
Z = K − ω2 M ⎪
⎪ . ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪
(40) ⎨ . ⎬ ⎨ . ⎬ ⎨ . ⎬
R (a) = Rk (a) ; H0 = H0,k ; Hi = Hi,k
⎪
⎪ ⎪ ⎪ .. ⎪ ⎪ .. ⎪
When the random parameters are introduced, the structural ⎪ .. ⎪
⎪ ⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪ . ⎪
⎪ ⎪
⎪ . ⎪
⎪ ⎪
⎪ . ⎪
⎪
⎪
dynamic equilibrium equation of AMs with the random ⎩ ⎭ ⎩ ⎭ ⎩ ⎭
Rn (a) H0,n Hi,n
parameters can be given as
m
R (a) = H0 + Hi ai (49)
Z (a) R (a) = F (a) (41)
i=1
Similarly, based on the first-order Taylor expansion, Z (a) The random displacement response Rk (a) of kth degree
and F (a) can be developed as can be expressed as
m
∂Z (ae )
m
Z (a) = Z ae + ai − aie = Ze + Z (42) Rk (a) = H0,k + Hi,k ai (50)
∂ai
i=1 i=1
123
Computational Mechanics (2018) 62:1023–1036 1029
12000 12000
Frequency (Hz)
Frequency (Hz)
10000 10000
a
8000 8000
r
6000 6000
Q1
4000 4000
y
x 2000 2000
0 0
-6 -5 -4 -3 -2 -1 0 1 2 3
Fig. 2 Sketch of a cross-section of AMs with a hexagon lattice of M X M
inclusions with a circular cross-section. a The unit cell of AMs and b Wave vector k log 10 ( Amplitude of u y )
the frequency response calculation model of AMs
Fig. 3 The band structure and frequency response function of AMs with
mean properties. a The band structure and b the frequency response
Similarly, by using the change-of-variable technique men- function
tioned in Sect. 2, the probability density function of the
random displacement response Rk (a) can be given as
4.1 Binary AMs with hexagon lattice
pRk (a) (Rk (a))
⎧
⎪ 1 p Rk (a)−H0
(m = 1)
Consider uncertain prediction of the physical response for a
⎪
⎪ |H | a1
⎨ 1
∞
H 1 Bragg-type AMs with hexagon lattice as shown in Fig. 2a,
1 Rk (a)−H0 −H1 a1
= |H | a =−∞ pa1 (a1 ) pa2 da1 (m = 2)
⎪
⎪ 2 1 H2 and the microstructure is made of epoxy with periodical
⎪
⎩ ∞
am =−∞ pam (am ) p Sm−2 (Rk (a) − H0 − Hm am )dam (m > 2) circular-shaped lead inclusions. In this numerical example,
(51) the structure parameters are given as follows: lattice con-
m−1 stant a = 0.086603m, radius of lead columns r = 0.03m,
where Sm−2 = i=1 Hi ai . length side of the hexagon unit cell l = 0.05m, and the
material properties of epoxy and lead are defined as random
parameters as presented in Table 1. To obtain the frequency
4 Numerical examples response function of AMs, a simplified AMs model com-
posed of several unit cells is proposed as shown in Fig. 2b.
In this section, three numerical examples are employed to In this frequency response model, it can be seen that the unit
illustrate the effectiveness of the formulated CVPS-FEM forced displacement is applied to the lower side boundary in
model. The probability distributions of band structures, mode y direction and the response is obtained at the center of the
shapes, and frequency responses for three different types upper side boundary.
stochastic AMs are predicted in these numerical examples. In Fig. 3, the band structure and frequency response func-
Furthermore, the accuracy of the results obtained by CVPS- tion of AMs with mean properties are firstly calculated. As
FEM is verified by MCM. The material properties for the expected, the amplitude of frequency response is sharply
random structure in this paper are given as follows: dropped at the frequency range where the band gap exists.
123
1030 Computational Mechanics (2018) 62:1023–1036
-3
In this example, the Young’s modulus E, Poisson’s ratio x 10
5.5
ν and density ρ of epoxy and lead are assumed to be ran- MCM
5
dom variables, and the parameters of these random variables CVPS-FEM
4.5
are given in Table 1. In the following analysis, the uncer-
123
Computational Mechanics (2018) 62:1023–1036 1031
Displacement 180
0.9 MCM
160
CVPS-FEM
N1 0.8
40
20
0
1.068 1.073 1.078 1.083 1.088
Amplitude of uy (200Hz)
Fig. 6 The 3rd mode shape along direction of AMs unit cell with
mean properties Fig. 8 The probability density function of the frequency response at
200 Hz of AMs with random parameters
(a) 240
MCM
(a)
a
(b) . Response (uy)
210 CVPS-FEM
Probability density function
180
Rin
150
Rex
120
90
60 y
x
30
Fig. 9 Sketch of a cross-section of LRAMs with a square lattice of
0 inclusions with a circular cross-section. a The unit cell of LRAMs and
0.29 0.293 0.296 0.299 0.302 0.305 0.308 b the frequency response calculation model of LRAMs
Displacement in x direction of node N1
(b) 160
MCM also be predicted by CVPS-FEM for other specific locations
140 CVPS-FEM of AMs with random parameters.
Probability density function
120
Figure 8 illustrates the probability density function of the
frequency response using CVPS-FEM and MCM models at
100 200 Hz. The numerical solutions of the probability density
80 function obtained from CVPS-FEM and MCM models are
almost identical. It is obviously noticed that the 95% confi-
60
dence interval for the response amplitudes is about 1.073 to
40 1.083.
20
4.2 Ternary LRAMs with square lattice
0
0.419 0.422 0.425 0.428 0.431 0.434 0.437 0.44 0.443
Displacement in y direction of node N1 A ternary LRAMs with lattice constant a = 0.02 m shown
in Fig. 9a is investigated in this example, and the structure is
Fig. 7 The probability density function of the normalized nodal dis- made up of epoxy with lead cylinders of radius Rin = 0.006m
placements in x and y direction of node N1 of the 3rd mode shape coated with rubber, the external radius of rubber cylinders is
along direction of AMs unit cell with random parameters. a The nor-
malized nodal displacement in x direction and b the normalized nodal Rex = 0.008m. The material properties for lead, rubber and
displacement in y direction epoxy are defined as random parameters, which are given in
Table 1. As shown in Fig. 9b, 5 × 5 unit cells of LRAMs
are constructed to calculate the frequency response function.
123
1032 Computational Mechanics (2018) 62:1023–1036
Frequency (Hz)
Frequency (Hz)
0.6
1000 1000
0.5
800 800 0.4
600 0.3
600
0.2
400 400 0.1
Q2 200
200
0
0 -1.2 -0.8 -0.4 0 0.4 0.8 1
M X M
Wave vector k log 10 ( Amplitude of u y ) N2
Fig. 10 The band structure and frequency response function of LRAMs
with mean properties. a The band structure and b the frequency response Fig. 12 The 3rd mode shape along X direction of LRAMs unit cell with
function mean properties
0.035
MCM
(a) 60
0.03 CVPS-FEM MCM
Probability density function
CVPS-FEM
50
0.025
Probability density function
0.02 40
0.015 30
0.01
20
0.005
10
0
320 330 340 350 360 370 380 390 400 410 420 430
0
Frequency (Hz) 0.296 0.304 0.312 0.32 0.328 0.336 0.344 0.352 0.36
Displacement in x direction of node N2
Fig. 11 The probability density function of the 3rd natural frequency
along X direction (Q 2 ) of LRAMs with random parameters (b) 40
MCM
35 CVPS-FEM
Probability density function
123
Computational Mechanics (2018) 62:1023–1036 1033
100 With the development of AMs, a new type AMs with consid-
ering defects has attracted great attention recently as it has
50
great application prospect such as the waveguide field [46].
0
Unfortunately, due to the manufacturing error, the discrep-
1.024 1.025 1.026 1.027 1.028 1.029 1.03 1.031 1.032 1.033 1.034 ancy between the design and actual manufacturing of AMs
Amplitude of uy (200Hz) is unavoidable, which leads to the uncertainty of waveg-
uide. Hence, it is of great significance to analyze this type
Fig. 14 The probability density function of the frequency response at
200 Hz of LRAMs with random parameters
AMs with considering the manufacturing error. As shown in
Fig. 15a, a 5×5 super-cell AMs with point defect is proposed
and investigated in this example. The square AMs structure is
and MCM. And the probability density function of random made up of epoxy with periodical square-shaped inclusions
natural frequency computed by CVPS-FEM is in perfect of lead columns. The lattice constant is a = 0.5 m, and the
agreement with the result obtained by the MCM, which side length of columns lead is b = 0.06 m.
clearly validates that CVPS-FEM is an effective method to Similarly, in order to calculate the frequency response, the
predict the random band structure of LRAMs with random unit forced displacement is applied to the lower side boundary
parameters. of the 5 × 5 super-cells in y direction and the frequency
Figure 12 shows the 3rd mode shape along X direction response is obtained at the center of its upper side boundary,
of LRAMs unit cell with mean properties. The probability which can be seen clearly in Fig. 15b.
density functions of the normalized x-displacement and y- The band structure and frequency response function of
displacement of node N2 in the 3rd mode shape computed by the AMs with mean properties are calculated as shown in
CVPS-FEM and MCM with considering the random material Fig. 16. Similarly, it can be seen clearly from Fig. 16 that
properties are plotted in Fig. 13. It can be easily observed that the vibration is reduced effectively at the frequencies where
an excellent agreement is given between the results obtained the band gaps exist. In addition, because of the existence of
by CVPS-FEM and MCM. The robustness and accuracy of the point defect, the defective modes are unavoidable and the
the proposed CVPS-FEM is proved clearly in dealing with frequency responses increase at these frequencies where the
ternary LRAMs with random variables. defective modes exist. As shown in Fig. 16b, the frequency
The probability density functions of frequency response response increases obviously at the frequency range around
of the ternary LRAMs with random parameters at 200 Hz 8000 Hz.
from CVPS-FEM and MCM are shown in Fig. 14. Again, the Similarly, the material parameters of epoxy and lead are
numerical results for the probability density function of the assumed to be random variables as listed in Table 1, and
y
x
123
1034 Computational Mechanics (2018) 62:1023–1036
Frequency (Hz)
8000 8000 0.6
0.5
6000 6000 0.4
0.3
4000 4000 0.2
0.1
2000 2000
0
0 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4
M X M
Wave vector k log 10 ( Amplitude of u y )
-3
x 10
4
(a) 100
MCM
MCM
3.5 CVPS-FEM
CVPS-FEM
Probability density function
Probability density function
80
3
2.5
60
2
40
1.5
1
20
0.5
0
0 0.192 0.196 0.2 0.204 0.208 0.212 0.216 0.22 0.224 0.228 0.232
7900 8040 8180 8320 8460 8600 8740 8880
Frequency (Hz) Displacement in x direction of node N3
(b) 70
Fig. 17 The probability density function of the 75th natural frequency MCM
along M direction (Q 3 ) of the defective AMs with random parameters 60 CVPS-FEM
Probability density function
50
the uncertainty method based on CVPS-FEM is employed to
40
predict the probability density functions of band structure,
mode shape and frequency response function of stochastic 30
defective AMs.
Without loss of generality, the uncertain natural frequency 20
123
Computational Mechanics (2018) 62:1023–1036 1035
100
paper. In this uncertainty model, based on the first-order Tay-
MCM
CVPS-FEM
lor series expansion and perturbation technique, the physical
responses of random AMs can be approximated as linear
Probability density function
80
functions of the random parameters, and then the probability
60 density functions of the responses can be calculated by the
change-of-variable technique. The results obtained by MCM
are used as the reference results to demonstrate the accu-
40
racy of the proposed method. It can be clearly seen from the
numerical examples that the CVPS-FEM can predict the band
20 structure, mode shapes and frequency response function of
AMs very well compared with MCM. Thus, the uncertainty
0 model based on CVPS-FEM formulated in this paper is very
1.096 1.1 1.104 1.108 1.112 1.116 1.12 1.124 1.128 1.132 1.136
effective to predict the uncertain physical responses of AMs
Amplitude of uy (200Hz)
with random parameters, which is of great significance for
the design and manufacturing of AMs in practical applica-
Fig. 20 The probability density function of the frequency response at
200 Hz of the 5 × 5 super-cells with random parameters tion.
123
1036 Computational Mechanics (2018) 62:1023–1036
12. Zhang S (2010) Acoustic metamaterial design and applications. 30. Bessa MA, Bostanabad R, Liu Z, Hu A, Apley DW, Brinson C et al
Dissertations & Theses—Gradworks (2017) A framework for data-driven analysis of materials under
13. Zhu R, Huang GL, Huang HH, Sun CT (2011) Experimental and uncertainty: countering the curse of dimensionality. Comput Meth-
numerical study of guided wave propagation in a thin metamaterial ods Appl Mech Eng 320:633–667
plate. Phys Lett A 375:2863–2867 31. Bostanabad R, Bui AT, Xie W, Apley DW, Chen W (2016)
14. Peng H, Pai PF, Deng H (2015) Acoustic multi-stopband meta- Stochastic microstructure characterization and reconstruction via
material plates design for broadband elastic wave absorption and supervised learning. Acta Mater 103:89–102
vibration suppression. Int J Mech Sci 103:104–114 32. Feng YT, Li CF, Owen DRJ (2010) A directed Monte Carlo solution
15. Liu Z, Zhang X, Mao Y, Zhu YY, Yang Z, Chan CT et al (2000) of linear stochastic algebraic system of equations. Finite Elem Anal
Locally resonant sonic materials. Science 289:1734–1736 Des 46:462–473
16. Sheng P, Zhang XX, Liu Z, Chan CT (2003) Locally resonant sonic 33. Tartakovsky DM, Xiu D (2006) Stochastic analysis of transport in
materials. Science 338:201–205 tubes with rough walls. J Comput Phys 217:248–259
17. Jensen JS (2003) Phononic band gaps and vibrations in one- and 34. Ghanem RG, Spanos PD (1991) Stochastic finite elements: a spec-
two-dimensional mass-spring structures. J Sound Vib 266:1053– tral approach. Springer, Berlin, p 224
1078 35. Wang M, Huang Q (2016) A new hybrid uncertain analysis method
18. Yao S, Zhou X, Hu G (2008) Experimental study on negative effec- for structural-acoustic systems with random and interval parame-
tive mass in a 1D mass spring system. New J Phys 10:043020 ters. Comput Struct 175:15–28
19. Shi S, Chen C, Prather DW (2004) Plane-wave expansion method 36. Kang Z, Cheng GD (2006) Structural robust design based on per-
for calculating band structure of photonic crystal slabs with per- turbation stochastic finite element method. Chin J Comput Mech
fectly matched layers. J Opt Soc Am A 21:1769–1775 23:129–135
20. Hao Y, Mittra R (2008) FDTD modeling of metamaterials. Theory 37. Papadimitriou C, Katafygiotis LS, Beck JL (1995) Approximate
and applications. Artech House Publishers, Norwood, MA analysis of response variability of uncertain linear systems. Probab
21. Shen L, He S, Xiao S (2002) A finite-difference eigenvalue algo- Eng Mech 10:251–264
rithm for calculating the band structure of a photonic crystal. 38. Culla A, Carcaterra A (2007) Statistical moments predictions for
Comput Phys Commun 143:213–221 a moored floating body oscillating in random waves. J Sound Vib
22. Argyropoulos C, Kallos E, Hao Y (2010) FDTD analysis of the 308:44–66
optical black hole. J Opt Soc Am B 27:2020–2025 39. Xia B, Yu D, Liu J (2014) Transformed perturbation stochastic
23. Wang G, Wen J, Liu Y, Wen X (2004) Lumped-mass method for finite element method for static response analysis of stochastic
the study of band structure in two-dimensional phononic crystals. structures. Finite Elem Anal Des 79:9–21
Phys Rev B 69:1324–1332 40. Li J, Wang Y, Zhang C (2012) Dispersion relations of a periodic
24. Yang W, Li J, Huang Y (2016) Modeling and analysis of the opti- array of fluid-filled holes embedded in an elastic solid. J Comput
cal black hole in metamaterials by the finite element time-domain Acoust 20:237–247
method. Comput Methods Appl Mech Eng 304:501–520 41. Liu GR (2016) On partitions of unity property of nodal shape func-
25. Li J (2007) Error analysis of mixed finite element methods for tions: rigid-body-movement reproduction and mass conservation.
wave propagation in double negative metamaterials. J Comput Appl Int J Comput Methods 13:1640003
Math 209:81–96 42. Liu G, Xi Z, Horie Y (2001) Elastic waves in anisotropic laminates.
26. He ZC, Li E, Wang G, Li GY, Xia Z (2016) Development of an Appl Mech Rev 56:B23
efficient algorithm to analyze the elastic wave in acoustic metama- 43. Xia B, Yu D (2012) Interval analysis of acoustic field with
terials. Acta Mech 227:1–16 uncertain-but-bounded parameters. Comput Struct 112–113:235–
27. Li E, He ZC, Wang G (2016) An exact solution to compute the 44
band gap in phononic crystals. Comput Mater Sci 122:72–85 44. Liu GR (2009) Meshfree methods: moving beyond the finite ele-
28. Li E, He ZC, Hu JY, Long XY (2017) Volumetric locking issue ment method, Second edn. CRC Press, Boca Raton
with uncertainty in the design of locally resonant acoustic meta- 45. Wang C, Gao W, Song C, Zhang N (2014) Stochastic interval
materials. Comput Methods Appl Mech Eng 324:128–148 analysis of natural frequency and mode shape of structures with
29. Li E, He ZC, Wang G, Liu GR (2017) An ultra-accurate numer- uncertainties. J Sound Vib 333:2483–2503
ical method in the design of liquid phononic crystals with hard 46. Dossou KB, Botten LC, Wilcox S, Mcphedran RC, Sterke CMD,
inclusion. Comput Mech 60:1–14 Nicorovici NA et al (2007) Exact modelling of generalised defect
modes in photonic crystal structures. Physica B 394:330–334
123