HAL Id: pastel-01068284 https://pastel.archives-ouvertes.fr/pastel-01068284 Submitted on 25 Sep 2014 HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers. L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés. Nonlinear Vibrations of Thin Rectangular Plates: A Numerical Investigation with Application to Wave Turbulence and Sound Synthesis Michele Ducceschi To cite this version: Michele Ducceschi. Nonlinear Vibrations of Thin Rectangular Plates: A Numerical Investigation with Application to Wave Turbulence and Sound Synthesis. Vibrations [physics.class-ph]. ENSTA ParisTech, 2014. English. pastel-01068284
166
Embed
Nonlinear Vibrations of Thin Rectangular Plates: A ...
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
HAL Id: pastel-01068284https://pastel.archives-ouvertes.fr/pastel-01068284
Submitted on 25 Sep 2014
HAL is a multi-disciplinary open accessarchive for the deposit and dissemination of sci-entific research documents, whether they are pub-lished or not. The documents may come fromteaching and research institutions in France orabroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, estdestinée au dépôt et à la diffusion de documentsscientifiques de niveau recherche, publiés ou non,émanant des établissements d’enseignement et derecherche français ou étrangers, des laboratoirespublics ou privés.
Nonlinear Vibrations of Thin Rectangular Plates: ANumerical Investigation with Application to Wave
Turbulence and Sound SynthesisMichele Ducceschi
To cite this version:Michele Ducceschi. Nonlinear Vibrations of Thin Rectangular Plates: A Numerical Investigationwith Application to Wave Turbulence and Sound Synthesis. Vibrations [physics.class-ph]. ENSTAParisTech, 2014. English. pastel-01068284
Abstract Nonlinear vibrations of thin rectangular plates are considered, using the von Kármán equations inorder to take into account the effect of geometric nonlinearities. Solutions are derived through an expansion overthe linear eigenmodes of the system for both the transverse displacement and the Airy stress function, resultingin a series of coupled oscillators with cubic nonlinearities, where the coupling coefficients are functions of thelinear eigenmodes. A general strategy for the calculation of these coefficients is outlined, and the particularcase of a simply supported plate with movable edges is thoroughly investigated. To this extent, a numericalmethod based on a new series expansion is derived to compute the Airy stress function modes, for which ananalytical solution is not available. It is shown that this strategy allows the calculation of the nonlinear couplingcoefficients with arbitrary precision, and several numerical examples are provided. Symmetry properties arederived to speed up the calculation process and to allow a substantial reduction in memory requirements.Finally, analysis by continuation allows an investigation of the nonlinear dynamics of the first two modesboth in the free and forced regimes. Modal interactions through internal resonances are highlighted, and theiractivation in the forced case is discussed, allowing to compare the nonlinear normal modes (NNMs) of theundamped system with the observable periodic orbits of the forced and damped structure.
1 Introduction
Plates elements are commonly found in a variety of contexts in structural mechanics. An understandingof their vibrational properties is crucial in many contexts, e.g. fluid-structure interaction problems, plateand panel flutter in aeronautics [13], energy harvesting of fluttering flexible plates [18], piezoelectric andlaminated plates [15,21], as well as their coupling with electro-magnetic and thermal fields [22]. When theplates are thin, vibration amplitudes can easily attain the same order of magnitude as the thickness. In thiscase, the nonlinear geometric effects cannot be neglected, resulting in a rich variety of dynamics [2,38].Examples can be given ranging from weakly to strongly nonlinear cases: nonlinear vibrations of plates withmoderate nonlinearity [2,45], fluid–structure interaction problems [25] and the transition from periodic tochaotic vibrations [4,37,50]. Aside from typical engineering problems, the chaotic dynamics exhibited by thinplates excited at large amplitudes finds application in the field of musical acoustics, as it accounts for the brightand shimmering sound of gongs and cymbals [6,7,12,29]. It was pointed out recently, from the theoretical,numerical and experimental viewpoints that the complex dynamics of thin plates vibrating at large amplitudes
M. Ducceschi · C. Touzé (B )Unité de Mecanique, ENSTA-ParisTech, 828 Boulevard des Maréchaux, Palaiseau, FranceE-mail: [email protected]
S. Bilbao · C. J. WebbJames Clerk Maxwell Building, University of Edinburgh, Edinburgh, Scotland
214 M. Ducceschi et al.
displays the characteristics of wave turbulence systems, and thus, it should be studied within this framework[9,20,34,35,49].
A widely used model in nonlinear plate modelling is due to von Kármán [54]. This model takes into accounta quadratic correction to the longitudinal strain, as compared to the classical linear plate equation by Kirchhoff[16,33,38,46]. The type of nonlinearity introduced is thus purely geometrical. The von Kármán equations areparticularly appealing because they describe a large range of phenomena while retaining a relatively compactform, introducing a single bilinear operator in the classic linear equations by Kirchhoff.
Pioneering analytical work in the analysis of rectangular thin plate vibrations with geometrical nonlineari-ties was carried out in the 1950s by Chu and Herrmann [17], demonstrating for the first time the hardening-typenonlinearity that has been confirmed by numerous experiments; see, e.g. [1,28]. Restricting the attention tothe case of rectangular plates, the work by Yamaki [55] confirms analytically the hardening-type nonlinearityfor forced plates. The case of 1:1 internal resonance for rectangular plates (where two eigenmodes have nearlyequal eigenfrequencies) has been studied by Chang et al. [14] and by Anlas and Elbeyli [3]. Parametricallyexcited nearly square plates, also displaying 1:1 internal resonance, have also been considered by Yang andSethna [56]. All these works focus on the moderately nonlinear dynamics of rectangular plates where only afew modes (typically one or two) interact together. In these cases, the von Kármán plate equations are projectedonto the linear modes, and the coupling coefficients are computed with ad-hoc assumptions that appear difficultto generalise. Finite element methods have also been employed—see, e.g. the work by Ribeiro et al. [42–44],and Boumediene et al. [10] to investigate the nonlinear forced response in the vicinity of a eigenfrequency. Re-cently, numerical simulations of more complex dynamical solutions, involving a very large number of modes inthe permanent regime, have been conducted, in order to simulate the wave turbulence regime and to reproducethe typical sounds of cymbals and gongs. For that, Bilbao developed an energy-conserving scheme for finitedifference approximation of the von Kármán system [5], which allows the study of the transition to turbulence[49] and the simulation of realistic sounds of percussive plates and shells [6,7]. Spectral methods with a verylarge number of degrees of freedom have also been employed in [20] to compare theoretical and numericalwave turbulence spectra.
This work aims at extending the possibilities of the modal approach to simulate numerically the non-linear regime of rectangular plates. Instead of introducing ad-hoc assumptions, a general model is herepresented; this model retains a vast number of interacting modes, making possible the investigation of theglobal dynamics of the plate while making it very precise. Within this framework, the advantages of themodal approach are retained (accuracy of linear and nonlinear coefficients, flexibility in setting modal damp-ing terms in order to calibrate simulation with experiment, . . .), and its limitations are overcome: there isno restriction with respect to the amount of modes that one wants to keep. In this work, the possibility ofsimulating dynamical solutions with a large number (say a few hundred) of modes is detailed. The case un-der study is that of a simply supported plate with in-plane movable edges. For this particular choice, thetransverse modes are readily obtained from a double sine series [26]; the in-plane modes, however, are notavailable in closed form. Interestingly, it was shown in [46] that the problem of finding the in-plane modesfor the chosen boundary conditions corresponds mathematically to the problem of finding the modes of afully clamped Kirchhoff plate. To this extent, a general strategy proposed in [31] is here adapted to find theclamped plate modes. To validate the results, the resonant response of the plate in the vicinity of the firsttwo modes is numerically investigated, for vibration amplitudes up to three to four times the thickness. Sec-ondly, a thorough comparison of the modal approach with the finite difference method developed in [5,6] isalso given. Calculation of the free response allows the study of the first two nonlinear normal modes of theplate and to highlight the complicated dynamics displayed at large amplitudes. Modal couplings, resonantand nonresonant, are investigated. Finally, the forced response is also computed, and the link between thebackbone curve and the forced response is investigated, showing the role of internal resonance and damp-ing.
2 Model description
Plates whose flexural vibrations are comparable to the thickness are most efficiently described by the vonKármán equations [17,33,39,46]. In the course of this paper, a rectangular plate of dimensions Lx , L y andthickness h (with h ≪ Lx , L y) is considered. The plate material is homogeneous, of volume density ρ,
Young’s modulus E and Poisson’s ratio ν. Its flexural rigidity is then defined as D = Eh3/12(1 − ν2). Thevon Kármán system then reads
where ∆ is the Laplacian operator, w = w(x, y, t) is the transverse displacement, and F = F(x, y, t) isthe Airy stress function. The equations present a viscous damping term cw and a sinusoidal forcing termδ(x − x0) f cos(Ωt) applied at the point x0 on the plate. The damping will take the form of modal viscousdamping once the equations are discretised along the normal modes. The bilinear operator L(·, ·) is known asvon Kármán operator [46] and, in Cartesian coordinates, it has the form of
L(α, β) = α,xxβ,yy + α,yyβ,xx − 2α,xyβ,xy, (2)
where ,s denotes differentiation with respect to the variable s. This operator, although itself bilinear, is thesource of the nonlinear terms in the equations. All the quantities are taken in their natural units, so that Eq.(1.1) and Eq. (1.2) have the dimensions, respectively, of kg m−1 s−2 and kg m−2 s−2. The term L(w, w) in Eq.(1.2) is quadratic in w and its derivatives, so once the solution for F is injected into (1.1), a cubic nonlinearitywill appear, leading to a Duffing-type set of coupled ordinary differential equations (ODEs).
2.1 Linear modes
The strategy adopted here to solve the von Kármán system makes use of the linear modes for the displacementw and Airy stress function F . This strategy is particularly useful for investigating the free and forced vibrationsof the system, in the sense that it allows for the reduction of the dynamics of the problem from an infinitenumber of degrees of freedom to a finite one. The eigenmodes for the displacement w will be denoted by thesymbol Φk(x, y), and thus w(x, y, t) is written as
w(x, y, t) = Sw
Nw∑
k=1
Φk(x, y)
‖Φk‖qk(t), (3.1)
where Φk is such that
∆∆Φk(x, y) =ρh
Dω2
kΦk(x, y). (3.2)
Note that the sum in Eq. (3.1) is terminated at Nw in practice. The linear modes can be defined up to a constantof normalisation that can be chosen arbitrarily. For the sake of generality, Sw here denotes the constant of
normalisation of the function Φ = SwΦk(x,y)‖Φk‖
. The norm is obtained from a scalar product (α, β) between two
functions α(x, y) and β(x, y), defined as
< α, β >=
∫
S
α β d S −→ ‖Φk‖2 =< Φk, Φk >. (4)
Eq. (3.2) is the eigenvalue problem definition, and it is a Kirchhoff-like equation for linear plates.The Airy stress function is expanded along an analogue series:
F(x, y, t) = SF
NF∑
k=1
Ψk(x, y)
‖Ψk‖ηk(t), (5.1)
∆∆Ψk(x, y) = ζ 4k Ψk(x, y). (5.2)
Boundary conditions for w and F will be specified in the next subsection. The linear modes so defined areorthogonal with respect to the scalar product and are therefore a suitable function basis [26]. Orthogonalitybetween two functions Λm(x, y),Λn(x, y) is expressed as
< Λm, Λn >= δm,n‖Λm‖2, (6)
where δm,n is the Kronecker delta.Once the linear modal shapes are known, system (1.1) may then be reduced to a set of ordinary differential
equations, each referring to the kth modal coordinate qk(t), k = 1, . . . , Nw. Nw represents the order of thesystem of ODEs.
216 M. Ducceschi et al.
2.2 Reduction to a set of ODEs
The introduction of the expansion series (3.1) and (5.1) allows for the decomposition of the original vonKármán problem onto a set of coupled, nonlinear ordinary differential equations (ODEs). As a starting point,Eq. (5.1) is substituted into Eq. (1.2) to obtain
ηk = −Eh
2ζ 4k
S2w
SF
∑
p,q
qpqq
∫
SΨk L(Φp, Φq)dS
‖Ψk‖‖Φp‖‖Φq‖. (7)
Integration is performed over the area of the plate, and the orthogonality relation is used. Injecting Eqs. (3)and (7) into Eq. (1.1) gives
ρhSw
∑
k
ω2kΦk
‖Φk‖qk + ρhSw
∑
k
Φk
‖Φk‖qk + cSw
∑
k
Φk
‖Φk‖qk
= −EhS3
w
2
∑
n,p,q,r
1
ζ 4n
L(Φp, Ψn)
‖Ψp‖‖Φn‖
∫
SΨn L(Φq , Φr )dS
‖Φq‖‖Φr‖‖Ψn‖qpqqqr + δ(x − x0) f cos(Ωt). (8)
Then, the equation is multiplied on both sides by Φs and integrated over the surface of the plate. The result is
qs + ω2s qs + 2χsωs qs = −
E S2w
2ρ
n∑
p,q,r
Hnq,r E s
p,n
ζ 4n
qpqqqr +Φs(x0)
‖Φs‖ρhSw
f cos(Ωt), (9)
where a modal viscous damping is introduced in the equation, scaled by χs = c/(2ρhωs) (a dimensionlessparameter). A practical advantage of the modal description is that χs can be estimated experimentally for alarge number of modes [11], and so the modal approach allows the simulation of complex frequency dependentdamping mechanisms with practically no extra effort.
Two third order tensors, Hnq,r and E s
p,n , appear in Eq. (9). These are defined as
Hnp,q =
∫
SΨn L(Φp, Φq)dS
‖Ψn‖‖Φp‖‖Φq‖, E s
r,n =
∫
SΦs L(Φr , Ψn)dS
‖Φr‖‖Φs‖‖Ψn‖. (10)
It is seen that the ODEs are cubic with respect to the variables qs , so a fourth-order tensor Γ can convenientlybe introduced in the equations, as
Γ sp,q,r =
NF∑
n=1
Hnp,q E s
r,n
2ζ 4n
. (11)
Once the tensor Γ is known, one is left with a set of coupled ODEs that can be integrated in the time variableusing standard integration schemes. Alternatively, continuation methods can be employed to derive a completebifurcation analysis of the nonlinear dynamics.
2.3 Boundary conditions
To recover the von Kármán equations, one may define the potential and kinetic energies of a bent plate, in thefollowing way:
V =
3∑
i,k=1
h
2
∫
S
σikuikdS, (12.1)
T =ρh
2
∫
S
w2dS, (12.2)
U =
2∑
i,k=1
h
2
∫
S
σik uikdS (12.3)
Nonlinear dynamics of rectangular plates 217
where V, T are the potential and kinetic energies for pure bending, and U is the potential energy for thestretching in the in-plane direction. Note that two strain tensors (uik and uik) and two stress tensors (σik andσik) are introduced, in order to account for the pure bending and in-plane energies; note also that the indices ofthe in-plane tensors can take only two values. Suppose that the displacement vector is u = (ux , u y, w) definedin a Cartesian set of coordinates x = (x, y, z). The symmetric strain tensor uik is linear and can be given interms of the vertical displacement w as follows [23]:
Note that the only nonlinear term that appears in the definitions of the energies is the quadratic factor in uik .It is possible to make use of Hamilton’s principle, stated in the form
t1∫
t0
δ(T − V − U ) dt = 0, (18)
to recover the equations of motion (1.1) plus the boundary conditions. These can be categorised as follows[46] (here ,n, ,t denote differentiation along the normal and tangent directions, respectively):
– In-plane direction– free edge: F,nt = F,t t = 0– immovable edge (w = 0 along the boundary): F,nn − νF,t t = F,nnn + (2 + ν)F,nnt = 0
This constraint has to be imposed as an extra condition only when the edge is transversely free. It is evident thatthe boundary conditions must be fulfilled by all the linear modes Φk, Ψk that appear in the expansions (2.1),(5.1). For the transverse function, simply supported boundary conditions are considered for the remainder of
218 M. Ducceschi et al.
the paper. These describe a fixed, rotationally free edge and permit a simplified analysis because a solution isreadily available:
Φk = sin
(
k1πx
Lx
)
sin
(
k2πy
L y
)
; ω2k =
D
ρh
[
(
k1π
Lx
)2
+
(
k2π
L y
)2]2
. (20)
For the in-plane direction, a free edge is considered. However, a different form of the boundary conditionswill be used, i.e. F = F,n = 0. It is evident that the assumed conditions are sufficient to satisfy the properconditions F,nt = F,t t = 0. Note that, mathematically speaking, the assumed conditions on F turn the stressfunction problem into a transversely clamped plate problem.
The selected boundary conditions are also known as simply supported with movable edges [1].
3 A solution for the clamped plate
As shown in the previous section, the eigenvalue problem for F with the chosen boundary conditions isequivalent to that of a clamped Kirchhoff plate. To this extent, the Galerkin method is employed, as ananalytical solution for the problem is not available.
The starting point of the Galerkin method is to express the eigenfunction Ψk of Eqs. (5) as a series of thisform:
Ψk(x, y) =
Nc∑
n=0
aknΛn(x, y) (21)
where Λn(x, y) are the expansion functions depending on some index n, and akn are the expansion coefficients:
these depend on the index n and of course on the index k. The total number of expansion functions is Nc,and obviously, the accuracy of the solution improves as this parameter is increased. The Λ’s must be carefullyselected from the set of all comparison functions [48]; this is to say that they need to satisfy the boundaryconditions associated with the problem that they are at least p times differentiable (where p is the order of thePDE), and they form a complete set over the domain of the problem. Completeness is quite a rather involvedproperty to prove; however, one generally resorts to variations of sine or cosine Fourier series, for whichcompleteness follows directly from the Fourier theorem.
For this work, the expansion functions were selected according to a general method proposed in [31], whereit is shown how a Kirchhoff plate problem can be solved by means of a double modified Fourier cosine series,i.e.
Λn(x, y) = Xn1(x)Yn2(y) =
(
cos
(
n1πx
Lx
)
+ pn1(x)
)(
cos
(
n2πy
L y
)
+ pn2(y)
)
, (22)
where pn1(x), pn2(y) are fourth-order polynomials in the variables x and y, and depending as well on theintegers n1, n2. Note that the order of the polynomials corresponds to the order of the PDE. The role of thepolynomial is to account for possible discontinuities at the edges due to the boundary conditions. Li [31]is mainly concerned with a general solution strategy, where the plate is equipped with linear and rotationalsprings at the edges to simulate the effect of different boundary conditions. In [31], the polynomials of Eq.(22) do not appear explicitly, as they are obtained through matrix inversion in order to comply with the generalform of the boundary conditions. In turn, these matrices present the values of all the springs, and the generalexpression of the Λ’s is rather involved. However, given that the focus here is on the clamped plate only, theanalytical limit of all the springs having infinite stiffness is taken, so that an explicit form for (22) can indeedbe recovered, and this is:
Xn1(x) = cos
(
n1πx
Lx
)
+15(1 + (−1)n1)
L4x
x4 −4(8 + 7(−1)n1)
L3x
x3 +6(3 + 2(−1)n1)
L2x
x2 − 1, (23)
and similarly for Yn2(y). Note that for the clamped plate satisfaction of the boundary conditions is essential fora fast converging solution. This is because the conditions at the edges for the clamped plate are geometrical,as they prescribe the vanishing of the displacement and of the slope. Thus, an expansion function that does notsatisfy these conditions could lead to slow converging solutions, if not to wrong results.
Nonlinear dynamics of rectangular plates 219
Table 1 Convergence of clamped plate frequencies, ζ 2k Lx L y, ξ = 2/3
It is seen that this expansion satisfies the clamped plate conditions, but not the differential equation. It ispossible to show however that one particular choice for the expansion coefficients ak
n will render the function Ψk
an eigenfunction for the problem. The Galerkin method describes how to build up stiffness and mass matricesin order to calculate the coefficient vector ak
n and the corresponding eigenfrequency ζ 4k . For the problem (5.2),
where L(·, ·) is the von Kármán operator. Note that the integrals can be calculated analytically, because of thesimple form of the expansion function. Explicit forms of the integrals are presented in “Appendix A”. Then,
K a = ζ 4 Ma, (25)
which is the required eigenvalue problem that leads to the expansion coefficients and the eigenvalues.
3.1 Numerical results for the clamped plate
In this section, the results obtained by Galerkin’s method are compared to the classical results found in Leissa’stables [30]. A finite difference scheme (FD) developed by Bilbao [5] is as well used as a benchmark. A usefulparameter in plate problems is the aspect ratio, here defined as Lx/Ly and denoted by the symbol ξ . Assumethat two plates present the same aspect ratio: then, it is straightforward to show that the quantity ζ 2Lx L y is
constant for the two plates, where ζ is defined in Eq. (5.2) (thus making ζ 2Lx L y a nondimensional parameter).As a first step, the rate of convergence of the eigenfrequencies is proposed in Table 1. The plate has an aspectratio of 2/3. Nc denotes the number of modes kept in the expansion (21). Note that convergence for thefirst 100 eigenfrequencies is obtained up to the fifth significant digit when Nc = 400. This corresponds to acalculation time of less than 10 s in MATLAB on a standard machine equipped with an Intel Core i5 CPU 650@ 3.20 GHz, and a memory of 4 GB. In Table 2, the results obtained by Galerkin’s method are compared tothose found in Leissa as well as to the outcome of the FD scheme. For this, the plate parameters have been setas: Lx = 0.4 m, L y = 0.6 m, ρ = 7, 860 kg/m3, ν = 0.3, h = 0.001 m, E = 2 × 1011 Pa. The FD scheme
employs 241 × 161 discretisation points, so that∆x∆y
S= 2.6 × 10−5. Even though Leissa’s book represents
220 M. Ducceschi et al.
Table 3 Convergence of clamped plate frequencies, FD scheme, ζ 2k Lx L y, ξ = 2/3
Fig. 1 First four modes for the clamped plate, ξ = 2/3
one of the main references in the area of plate eigenmodes and frequencies, its results are somehow outdated,being about 40 years old. Thus, discrepancies between the presented Galerkin’s method and the numbers fromLeissa’s book are not at all concerning. On the other hand, it is known that FD schemes converge at a slowerrate than a pure modal approach. This is a consequence of the fact that FD schemes rely on discrete gridmeshes. Convergence for the first eigenfrequencies for the plate using the FD scheme is presented in Table3. Note that the eigenfrequencies tend to converge to the same values as Galerkin’s method. However, thecalculation time in MATLAB for a mesh grid of 280 × 419 points is much slower (about 20 min). Table 4presents the eigenfrequencies for the square plate, using Galerkin’s method. It is possible to appreciate thesame rate of convergence as for the previous case. Again, the results are compared with Leissa and to the FDscheme outcome (161 × 161 grid points) in Table 5. Plots of some clamped plate eigenmodes are presented inFig. 1. These results show that the Galerkin method, with the carefully chosen expansion (23), is indeed a fastconverging strategy for the calculation of the eigenfrequencies, as it allows for precisely computing hundredsof modes within seconds.
Nonlinear dynamics of rectangular plates 221
4 The nonlinear coupling coefficients
4.1 Symmetry properties
In this section, symmetry properties for the coupling coefficients Γ that appear in Eq. (11) are presented. First,it is obvious that
H ip,q = H i
q,p, (26)
because of the symmetry of the operator L(·, ·). Secondly, integrating by parts, the integral in the definition ofE in Eq. (10) gives
‖Ψq‖‖Φn‖‖Φp‖Enp,q =
∮ [
ΦnΨq,yΦp,xx − 2ΦnΨq,xΦp,xy − Ψq
∂
∂y
(
ΦnΦp,xx
)
]
y · n dΩ +
+
∮ [
ΦnΨq,xΦp,yy + 2Ψq
∂
∂y
(
ΦnΦp,xy
)
− Ψq
∂
∂x
(
ΦnΦp,yy
)
]
x · n dΩ
+
∫
Ψq L(Φp, Φn)dS. (27)
It is easy to see that the selected boundary conditions make the surface integrals vanish, so that the followingproperty holds:
Enp,q = H
qp,n . (28)
In this way, the tensor Γ may then be conveniently written as
Γ sp,q,r =
NF∑
n=1
Hnp,q Hn
r,s
2ζ 4n
. (29)
Note that the tensor H as defined in Eq. (10) is divided by the norms of the modes, so the value of Γ isindependent of the particular choice for the constants Sw, SF in Eqs. (3.2), (5.2). Basically, the symmetryproperties for Γ mean the following sets of indices will produce the same numerical value:
(s, p, q, r), (r, p, q, s), (s, q, p, r), (r, q, p, s), (q, r, s, p), (p, r, s, q), (q, s, r, p), (p, s, r, q). (30)
These symmetry properties can lead to large memory savings when the number of transverse and in-planemodes is a few hundred.
4.2 Null coupling coefficients
For the sake of numerical computation, it would be interesting to know a priori which coupling coefficients arenull. In actual fact, empirical observations of the Γ tensor suggest that only a smaller fraction of coefficientsis not zero. As an example, consider Table 6 where the nonzero values for the coefficients Γ 1
5,q,r for a plate
with ξ = 2/3 were collected (with p, q = 1 . . . 10): the table presents only 24 nonzero coefficients out ofa total of 100. These coefficients measure the amount of interaction between the different transverse modes.As a matter of fact, the modes can be classified according to the symmetry with respect to the x and y axiswhere the origin is placed at the centre of the plate. Four families exist, and they are: doubly symmetric(SS), antisymmetric-symmetric (AS and SA) and doubly antisymmetric (AA). For instance, the first mode isa doubly symmetric mode because it presents one maximum at the centre of the plate, and is thus symmetricwith respect to the two orthogonal directions departing from the centre of the plate in the x and y directions,whereas mode 5 is AA. The first sixteen modes for the case under study may be classified in the followinggroups:
Value q r Modal shape groups Value q r Modal shape groups21.36 1 5 SS AA SS AA 27.55 6 2 SS AA AS SA−21.75 1 10 SS AA SS AA 150.98 6 7 SS AA AS SA48.46 2 3 SS AA SA AS 36.52 6 9 SS AA AS SA7.55 2 6 SS AA SA AS −72.47 7 3 SS AA SA AS122.11 3 2 SS AA AS SA 119.51 7 6 SS AA SA AS−169.47 3 7 SS AA AS SA 56.36 8 5 SS AA SS AA−69.44 3 9 SS AA AS SA −64.89 8 10 SS AA SS AA56.71 4 5 SS AA SS AA 10.19 9 3 SS AA SA AS9.8 4 10 SS AA SS AA 65.63 9 6 SS AA SA AS3.1 5 1 SS AA AA SS −51.96 10 1 SS AA AA SS144.68 5 4 SS AA AA SS 97.76 10 4 SS AA AA SS46.47 5 8 SS AA AA SS 30.75 10 8 SS AA AA SS
This list will become useful when interpreting the free vibration diagrams of the next section. Remarkably, thenumber of indices of the Γ coefficients (four) matches the number of modal shape sets. Table 6 presents themodal families to which the interacting modes belong; observation of alike tables permits to state the followingheuristic rule:
the indices (s, p, q, r) will give a nonzero value for Γ sp,q,r if and only if modes s,p,q,r come all from
distinct modal shape groups or if they come from the same group two by two.
For example, the combinations (SS, SS, AS, SA) and (SS, SS, SS, AS) will definitely give a zero value; on theother hand, the combinations (SS, SS, SS, SS), (SS, AA, SS, AA) and (SS, AS, SA, AA) will give a nonzerovalue. A rigorous mathematical proof is not carried out as it involves a rather lengthy development which isbeyond the scope of the present work. However, it has been numerically checked for a large number of Γ ’sinvolving a few hundred modes, providing an exhaustive verification of this rule.
This rule, in combination with the previous remarks on symmetry, can be used to speed up the calculationof the Γ tensor (for example by pre-allocating the zero entries when using a sparse matrix description). Insome way, this observation relates to the already noted property of von Kármán shells [47]. There, the couplingrules are actually more involved, but they can be somehow more directly proved mathematically.
4.3 A few words on the FD scheme
To validate the computational results for the Γ tensor, an FD scheme developed in [5] has been extensivelyused. In this sense, the role of the discretised L operator in Eq. (11) is central. For two discrete functions α, β
defined over the plate grid, the form for the discrete counterpart l(α, β) has been selected as
The δ’s are discrete derivative operators, and the µ’s are averaging operators, as follows from
δxx =1
h2x
(ex+ − 2 + ex−); δx+ =1
hx
(ex+ − 1); µx− =1
2(ex− + 1), (32)
where ex+ (ex−) is the positive (negative) shifting operator, and hx is the step size along the x direction. Notethat this particular choice for the l operator is due to the fact that it produces an energy-conserving scheme, asexplained exhaustively in [5]. The eigenmodes are obtained by solving discrete counterparts of Eqs. (3.2) and(5.2), and thus, a discrete double Laplacian is needed. At interior points, it can be approximated by
Enforcing of boundary conditions (simply supported and clamped) is described in [6]. Once the modes areknown, one makes use of (31) to get the values of the coupling coefficients in Eq. (11).
Nonlinear dynamics of rectangular plates 223
Table 7 Convergence of coupling coefficients, Γ kk,k,k(Lx L y)
In this subsection, some numerical results are presented. It is somehow useful to note that the Γ ’s depend onlyon the aspect ratio. In other words, the quantity
Γ sp,q,r (Lx L y)
3 (34)
is constant for all the plates sharing the same aspect ratio. Table 7 presents a convergence test for a plate ofaspect ratio ξ = 2/3. The convergence in this case depends on two factors: the first is the amount of stressfunction modes retained in the definition of Γ [NF in Eq. (11)]; the second is the accuracy on the Airy stressfunction modes and frequencies [quantified by the number Nc in Eq. (21)]. For clarity, in the following Tables,NF is always the same as Nc. It is seen that a four-digit convergence up to the Γ 100
100,100,100 coefficient isobtained when NF = 484, and thus, the convergence rate for these coefficients is slower than that of the stressfunctions eigenfrequencies alone. For the FD scheme, convergence depends on the number of modes retainedand also on the grid size. Thus, Tables 8 and 9 present some values for NF = 100 and NF = 200, respectively.Note that, contrary to what happens for the eigenfrequencies, convergence for the coupling coefficients is fromabove for FD and from below for the modal approach. It is also evident that a sufficiently large number ofstress modes has to be retained to calculate reasonable approximate values for the Γ ’s: failing to do so mayresult in completely erroneous estimates (see for instance the last row of Table 8 compared to the last row ofTable 9).
5 Analysis of the periodic solutions
The nonlinear dynamics of the plate is now analysed in terms of periodic solutions. The periodic orbits ofthe conservative system, also called the nonlinear normal modes (NNMs) [53], are first computed thanks to apseudo arc-length numerical continuation method implemented in the software AUTO [19]. The amplitude–frequency relationships (i.e. the backbone curves) are exhibited for the first two modes up to 3–4 times thethickness, displaying a complicated network of bifurcation branches generated by internal resonances andmodal couplings. Secondly, the forced responses of the damped plate are computed and their relationship withthe backbone curve illustrated.
Figure 2 is an illustration of the backbone convergence, for mode 1. The backbone is the curve obtained byplotting the maxima of the periodic solutions, in the case of undamped, unforced vibrations, which can bestable (continuous lines) or unstable (dashed lines). Note that only the principal branch is represented, andthus, the figure does not take into account the secondary branches departing from the bifurcation points. Thefigure presents the six backbones obtained when Nw = 6, 8, 10, 14, 16, 18. It is evident that the period ofthe vibration decreases as the amplitude increases, and thus, the curves bend to the right in the diagram;this behaviour is known in the literature as hardening-type nonlinearity. The backbone curves obtained forNw = 14, 16, 18 are almost exactly superimposed showing the convergence of the main solution branch forvibration amplitudes up to 4h. Note also that the cases Nw = 8, 10 are exactly superimposed because modes 9and 10 do not belong to SS (the family of mode 1); hence, the shape of the backbone does not change, althoughthe stability intervals do not coincide. No stable solutions are detected by AUTO for vibrations larger than 4h:this result is consistent with numerous experimental and numerical simulations of large amplitude vibrationsof plates; higher vibration amplitudes give way to unstable solutions, in quasiperiodic or turbulent regimes[49,50]. The range of convergence of the backbone decreases when less modes Nw are considered; particularlyfor the case of Nw = 6, the backbone displays significant differences from the converged solution. In addition,unstable solutions in this case set in much earlier, leading to the conclusion that when Nw = 6 the backbonecurve depicts an unrealistic scenario for amplitudes larger than 1.8h. The principal branch for the casesNw = 14, Nw = 16, Nw = 18 undergoes an internal resonance around ω/ω1 ≈ 1.27. This is a resonancebetween mode 1 and mode 11, and will be commented later. It is seen that the cases Nw = 16, Nw = 18 areperfectly superimposed, and thus, a total number of Nw = 16 modes is sufficient for full convergence; hence,this is the number of modes that will be considered in the remainder of the paper. Figure 3 shows the completeresonance scenario for mode 1, and in other words, it presents the backbone and the bifurcated branches.Figure 3 is basically a representation of the first NNM as a function of the frequency of vibration for the firstmode. For clarity, only the most significant modal coordinates are represented. Branches are denoted by thesymbol Bi
k where the index i refers to the branch number and k is the coordinate involved. Thus, B1 is the main
(backbone) branch, and B2, B3, . . . are secondary branches featuring a sudden loss of energy of q1 in favour ofother nonlinearly resonant modes. The appearance of internal resonance tongues due to the exchange of energybetween modes at nonlinear frequencies of vibration has been previously observed for systems involving afew degrees of freedom, or for continuous systems with local nonlinearities [8,24,27,41]; in turn, these works
Nonlinear dynamics of rectangular plates 225
(a)(b)
(c)
(d)
Fig. 3 a Free vibration diagram for mode 1, Nw = 16. b– d Bifurcated branches and internal resonances (colour figure online)
show that NNM branches may fold in the presence of internal resonances. In this paper, internal resonancefoldings in the NNM branches are reported for a continuous structure with distributed geometric nonlinearity.The bifurcated branches are composed mainly by unstable states along intricate paths and are difficult tocompute numerically when using continuation. Note, however, that the free NNM is a physical abstraction:when damping and forcing are introduced in the system, most of the complicated details disappear, as it willbe shown in the next subsection.
Observing B1 before the first bifurcation point, it is easily seen that modes 4 (B14, green), 8 (B1
8, light green),
11 (B111, magenta) and 12 (not shown) bear a relatively important contribution. Here, a typical nonresonant
coupling is at hand. As it can be deduced from Sect. 4.2, the only nonvanishing coefficients Γp
1,1,1 withp = 1, . . . , 16 are obtained for p = 1, 4, 8, 11, 12. These coefficients are of prime importance as they giverise to a term of the form Γ
p1,1,1q3
1 in the equation for qp. Thus, when q1 is large, modes 4,8,11 and 12 acquire
nonnegligible energy through the nonresonant coupling terms Γp
1,1,1, which act on the modal equations asforcing terms. These coefficients have been referred to as invariant-breaking terms because they have theproperty of breaking the invariance of the linear normal modes through modal coupling [51,52]. The couplingin these cases is nonresonant because no commensurability relationship exists between the frequencies ofvibration.
The first bifurcated branch is B2 and develops along a very narrow frequency interval between 1.2435 <
ω/ω1 < 1.248. It is a very small branch, and it is visible in Fig.3b (B21) and Fig.3d (B2
2). The modes involvedin this bifurcation are 1 and 2. It is evident that mode 2, so far quiescent, is activated by an internal resonancewith mode 1. The order of the internal resonance can be obtained from a temporal simulation of the systemcomprising Nw = 16 modes, fed at the input by the maximum displacements and velocities for all the modalcoordinates along B2. In this work, a fourth-order Runge–Kutta scheme is used for the time integration, givingat the output the oscillation in time for all the modes in the periodic regime. Figure 4a represents modes 1 and2 in the time domain on the point at ω/ω1 = 1.246 along the branch B2. The figure shows that the periodof vibration for mode 2 is exactly half the period of mode 1, resulting in a 1:2 internal resonance. Note thatstarting the simulation on any other point of the same branch will lead to the same resonance ratio.
In the next section, it will be seen that the bifurcation giving rise to B2 is key to the dynamics of the drivendamped oscillations: this branch tends to occupy larger portions of the phase space as the forcing and dampingterms increase, modifying the local structure of the invariant NNM manifold.
Following the principal branch in Fig. 3b, one encounters a second bifurcation giving rise to B3. This isan interesting branch where again quiescent modes are activated by internal resonances. Figure 3d revealsthat these are modes 2 (B3
2, red), 14 (B314, grey) and 16 (B3
16, black). Note that the branch B3 emerges at
226 M. Ducceschi et al.
(b) (c)(a)
Fig. 4 a Modes 1 (blue) and 2 (red) along B2 displaying 1:2 internal resonance. b Modes 1 (blue) and 2 (red) along B3 displaying
1:2 internal resonance. c Modes 1 (blue), 14 (grey) and 16 (black) along B3 displaying 1:10 internal resonance (colour figureonline)
ω/ω1 = 1.285 and first develops to the left towards decreasing frequencies. The branch is characterised atfirst by a strong coupling between modes 1 and 2 (visible in Fig. 3d) and then by a coupling amongst modes1,14 and 16. The order of the resonance can again be extrapolated from a Runge–Kutta time-domain schemefed with the AUTO output. This gives Fig. 4b, c where it is seen that modes 1 and 2 undergo a second 1:2internal resonance, whereas modes 1–14 and 1–16 display a 1:10 internal resonance. Thus, the dynamics ofthis branch is again dominated by even-order internal resonances. The last branch is B4. This is an improperlabelling because this branch is actually the principal branch undergoing an internal resonance with mode 11(B4
11, magenta). This branch is almost entirely unstable, and the Runge–Kutta time-domain simulation does notreturn stable periodic solutions. There is no doubt, however, that the branch is activated by internal resonancebetween modes 1 and 11, given the rapid growth of the latter in the bifurcation diagram at the expense of mode1.
The analysis of the first NNM revealed some important aspects of the nonlinear system: in particular, it wasshown that the bifurcated branches are generated by even-order internal resonances which, in turn, break thesymmetry of the cubic nonlinearity possessed by the system. This symmetry-breaking bifurcation has alreadybeen observed for the simple Duffing equation [32,40], as well as in systems with material nonlinearity [36].Physically speaking, the most important properties returned by the analysis of the free NNM are as follows:(i) the loss of stability of the periodic solutions for amplitudes above 3h; (ii) the pitchfork bifurcation givingrise to B2 presenting a strong coupling between modes 1 and 2. The next subsection will treat in some detaila few examples of forced-damped vibrations, and it will be seen how the shape of the NNM gets modified bythe damping and forcing terms.
5.1.2 Forced-damped vibrations
In this section, forced-damped vibrations are considered. The plate is forced with a sinusoid of maximumamplitude f and frequency Ω [see Eq. (9)] varied around the eigenfrequency of the first mode, ω1. In turn,damping and forcing terms modify the shape of the invariant manifold corresponding to the NNM of theprevious section. Internal resonances change too: some are basically unseen by the modified NNM, whereasothers play a major role.
The first case under study presents a forcing amplitude of f = 0.17 N, and a damping coefficient χi = 0.001(same for all modes). The result is pictured in Fig. 5. In the figure, the forced branches are represented withthe usual colouring scheme (blue for mode 1 and red for mode 2), whereas the black lines are the branchesfrom the Hamiltonian dynamics. The point labelled G in Fig. 5 corresponds to a pitchfork symmetry-breakingbifurcation, driven by the underlying Hamiltonian dynamics and by the existence of the 1:2 internal resonance.The main branch becomes unstable in favour of stable periodic orbits where both modes 1 and 2 are activated ina 1:2 internal resonance. Hence, branch B2 reveals its importance as it has a major effect in the damped-drivencase. One can also notice that, for this small amount of damping, the turning point J is located just before theresonant tongue along the original backbone curve.
In order to understand more deeply the role of the branch B2, two more cases of interest are portrayedin Figs. 6 and 7. Here, f = 1.36 N for both cases, and χi = 0.005 for Fig. 6 and 0.001 for Fig. 7. The firstimportant remark is the location of the pitchfork bifurcation along the main branch: q1/h = 1.899 for Fig. 6 andq1/h = 1.824 for Fig. 7. It is seen that the invariant manifold of the Hamiltonian dynamics is largely affectedby the damping and forcing terms: the bifurcation G is located at very different points in the phase spacewhen comparing free and forced-damped vibrations. The 1:2 internal resonance giving rise to B2 becomes
Nonlinear dynamics of rectangular plates 227
Fig. 5 Forced response for mode 1 with f = 0.17 N, χ = 0.001. G: pitchfork bifurcation point leading to the coupled solution;J turning point. Mode 1: blue, mode 2: red (colour figure online)
Fig. 6 Forced response for mode 1 with f = 1.36 N, χ = 0.005. G: pitchfork bifurcation point leading to the coupled solution;J turning point. Mode 1: blue, mode 2: red (colour figure online)
in the latter case a dominant part of the dynamics, taking up a large portion of the phase space composedmainly of stable solutions. As a consequence, stable solutions are found on B2 at amplitudes larger then 3h.In addition, there is no trace of the other bifurcations giving rise to B3, B4 in the Hamiltonian dynamics.This observation leads to the conclusion that the free and forced-damped analyses are complementary: onone hand, it is not straightforward to understand which bifurcations are key to the forced-damped vibrationswhen looking solely at the Hamiltonian dynamics; on the other hand, the forced-damped system is more easilyinterpreted by making use of the free vibrations diagrams. Hence, a complete scenario for the forced-dampedvibrations cannot be obtained if a preliminary analysis of free vibrations is disregarded.
228 M. Ducceschi et al.
Fig. 7 Forced response for mode 1 with f = 1.36 N, χ = 0.001. G: pitchfork bifurcation point leading to the coupled solution;J turning point. Mode 1: blue, mode 2: red (colour figure online)
Fig. 8 Backbone for mode 2 obtained when Nw = 16. Modes 7 (pink) and 9 (dark blue) are activated by the nonresonant couplingwithin the SA family; mode 5 (brown) from the AA family is activated by 1:2 internal resonance (see inset) (colour figure online)
5.2 Mode 2
5.2.1 Free vibrations
Figure 8 shows the second NNM for Nw = 16. Convergence in this case is not shown for the sake of brevity;note, however, that the convergence study gave results comparable to those of mode 1. Thus, the same modelincluding Nw=16 modes is kept for the remainder of the study. Once again, one can notice that no stablesolutions are found beyond a certain amplitude limit, which is numerically found at 1.5h for mode 2. Actually,the principal branch loses its stability at the appearance of the coupled branch. As for mode 1, some modes areactivated by nonresonant coupling, and these are the modes belonging to the same family as mode 2 (SA): the
Nonlinear dynamics of rectangular plates 229
(a) (b) (c)
Fig. 9 Examples of forced-damped vibrations around the NNM for mode 2. a f = 1.2 N, χ = 0.001; b f = 2.0 N, χ = 0.001;c f = 3.2 N, χ = 0.01. Mode 2: red, mode 5: brown (colour figure online)
Figure shows for clarity only modes 7 (B17, pink) and 9 (B1
9, dark blue). The most salient feature of the dynamics
is the internal resonance between modes 2 and 5: a time integration was performed on B2 at ω/ω1 = 2.0515,leading to the solution visible in the inset of Fig. 8 showing a 1:2 internal resonance. Interestingly, this branchis almost entirely unstable, except on the interval 2.051 ≤ ω/ω1 ≤ 2.052. As for mode 1, the Hamiltonianmanifold will be modified when damping and forcing are introduced in the system.
5.2.2 Forced-damped vibrations
Examples of forced-damped solution are presented in Fig. 9. The cases (a) and (b) present the same dampingcoefficient, χi = 0.001, and the forcing values are, respectively, f = 1.2 N, f = 2.0 N. Both forcing valuesare sufficient to reach amplitudes high enough to detect the internal resonance with mode 5. For case (a), thebifurcated branch remains almost completely unstable, as for the Hamiltonian dynamics. When the forcing ishigh enough, however, stable solutions appear along the interval 2.2 ≤ ω/ω1 ≤ 2.3. As a consequence, mode2 possesses a secondary branch of stable periodic orbits of amplitude larger than 1.5h, which was seen to bethe limit of stability for the Hamiltonian manifold. As for mode 1, it is seen that the introduction of forcing anddamping may lead to extended stable solutions on the coupled branches. Another case of interest is portrayedin Fig. 9c. Here, the maximum forcing is f = 3.2 N, and the damping coefficient is χi = 0.01. In this case,the damping effects are so evident that the turning point is located away from the backbone. Distortion is atypical effect of damping: the forced response does not fit tightly along the backbone, and the turning pointmoves away from it.
In turn, the analysis of the forced responses for mode 1 and 2 revealed some interesting aspects of the globaldynamics: (i) symmetry-breaking resonances are common and key to the dynamics of the dynamical response;(ii) stable solutions on the coupled branches may reach higher amplitudes than the Hamiltonian manifold, forparticular combinations of damping and forcing factors.
6 Conclusions
The nonlinear dynamics of rectangular plates has been investigated. A robust numerical method has beendeveloped to obtain accurate modal solutions for a very large number of modes. In this sense, a fast convergingsolution strategy has been derived for the calculation of the eigenmodes of a fully clamped plate (neededhere to solve for the Airy stress function of a plate in a nonlinear regime). Formal symmetry properties andcoupling rules have been illustrated to allow large computational and memory savings when calculating thecoupling coefficients Γ ’s. Reference values for some of these coefficients, previously unavailable in the caseof a rectangular geometry, have been presented.
Free and forced vibrations have then been taken under consideration for the first two modes. For thefirst time, the NNM branches of solution (conservative case) have been drawn out to very large amplitudes,showing the existence of internal resonance branches. An important feature, the nonexistence of periodicsolutions beyond some vibration amplitude (4h for mode 1, 1.8h for mode 2) has been found. A thoroughcomparison of the Hamiltonian dynamics with the forced-damped (observable) dynamics has been derived, inorder to highlight: (i) the necessity of a preliminary analysis of the free vibrations, (ii) the main differences one
230 M. Ducceschi et al.
can expect between the NNMs of the conservative systems and the observable periodic orbits of the forced-damped system. Simple features such as the shift of the turning point from the backbone for large values ofthe damping have been found. More interestingly, the importance of certain internal resonance tongues (thosewith the simpler ratio) has been underlined, whereas other are mostly undetected in the forced case. Finally, ithas been found that some coupled branches may override the amplitude limit of existence of periodic solutionspredicted by the backbone curve.
Even though the results presented here involve at most 16 modes, the numerical scheme developed isable to consider a few hundreds of them interacting together. The results shown here have been necessary tovalidate the model, which will be used to undertake further study of more involved dynamical problems (i.e.
wave turbulence or sound synthesis of damped impacted plates for the reproduction of gong-like sounds).
Appendix A: Matrices for the clamped plate problem
To set up the eigenvalue problem, Eq. (25), one may proceed as follows. First, it is necessary to define the sizeof the square matrices Ki j , Mi j . Suppose this size is A2 × A2 (where A is an integer). Then, the indices n1, n2
for the expansion function (22) range from 0 to A − 1. In this way, the total number of eigenvalues calculatedwill be A2. Note that all the quantities that appear in the definition of the matrices are quadratic, so one needsreally four indices to define the i j entry in each matrix. Suppose these indices are (m, n) and (p, q). Then,
and similarly for the integrals involving the functions Y .
References
1. Amabili, M.: Nonlinear vibrations of rectangular plates with different boundary conditions: theory and experiments. Comput.Struct. 82, 2587–2605 (2004)
2. Amabili, M.: Nonlinear Vibrations and Stability of Shells and Plates. Cambridge University Press, Cambridge (2008)3. Anlas, G., Elbeyli, O.: Nonlinear vibrations of a simply supported rectangular metallic plate subjected to transverse harmonic
excitation in the presence of a one-to-one internal resonance. Nonlinear Dyn. 30, 1–28 (2002)4. Awrejcewicz, J., Krysko, V.A., Krysko, A.V.: Spatio-temporal chaos and solitons exhibited by von Kármán model. Int. J.
Bifurc Chaos. 12, 1465–1513 (2002)5. Bilbao, S.: A family of conservative finite difference schemes for the dynamical von Kármán plate equations. Numer.
Methods Partial Differ. Equ. 24, 193–216 (2008)6. Bilbao, S.: Numerical Sound Synthesis Finite Difference Schemes and Simulation in Musical Acoustics. Wiley, New
York (2009)7. Bilbao, S.: Percussion synthesis based on models of nonlinear shell vibration. IEEE Trans. Audio Speech Lang.
Process. 18, 872–880 (2010)8. Blanc, F., Touzé, C., Mercier, J.-F., Ege, K., Bonnet Ben-Dhia, A.-S.: On the numerical computation of nonlinear normal
modes for reduced-order modelling of conservative vibratory systems. Mech. Syst. Signal Process. 36, 520–539 (2013)9. Boudaoud, A., Cadot, O., Odille, B., Touzé, C.: Observation of wave turbulence in vibrating plates. Phys. Rev.
Lett. 100, 234504 (2008)10. Boumediene, F., Duigou, L., Boutyour, E.H., Miloudi, A., Cadou, J.M.: Nonlinear forced vibration of damped plates by an
asymptotic numerical method. Comput. Struct. 87, 1508–1515 (2009)11. Chaigne, A., Lambourg, C.: Time-domain simulation of damped impacted plates. I. Theory and experiments. J. Acoust. Soc.
Am. 109, 1422–1432 (2001)12. Chaigne, A., Touzé, C., Thomas, O.: Nonlinear vibrations and chaos in gongs and cymbals. Acoust. Sci. Technol. 26, 403–
409 (2005)13. Chang, S.I., Bajaj, A.K., Krousgrill, C.M.: Nonlinear oscillations of a fluttering plate. AIAA J. 4, 1267–1275 (1966)14. Chang, S.I., Bajaj, A.K., Krousgrill, C.M.: Non-linear vibrations and chaos in harmonically excited rectangular plates with
one-to-one internal resonance. Nonlinear Dyn. 4, 433–460 (1993)15. Chen, W.Q., Ding, H.J.: On free vibration of a functionally graded piezoelectric rectangular plate. Acta Mechanica 153, 207–
216 (2002)16. Chia, C.Y.: Nonlinear Analysis of Plates. Mc Graw Hill, New York (1980)17. Chu, H.N., Herrmann, G.: Influence of large amplitudes on free flexural vibrations of rectangular elastic plates. J. Appl.
Mech. 23 (1956)18. Doaré, O., Michelin, S.: Piezoelectric coupling in energy-harvesting fluttering flexible plates: linear stability analysis and
Auto2000: Continuation and Bifurcation Software for Ordinary Differential Equations (with HomCont). Technical report,Concordia University, Canada (2002)
20. Düring, G., Josserand, C., Rica, S.: Weak turbulence for a vibrating plate: can one hear a Kolmogorov spectrum? Phys. Rev.Lett. 97, 025503 (2006)
21. Fu, Y.M., Chia, C.Y.: Nonlinear bending and vibration of symmetrically laminated orthotropic elliptical plate with simplysupported edge. Acta Mech. 74, 155–170 (1988)
22. Gao, Y., Xu, B., Huh, H.: Electromagneto-thermo-mechanical behaviors of conductive circular plate subject to time-dependent magnetic fields. Acta Mech. 210, 99–116 (2010)
23. Géradin, M., Rixen, D.: Mechanical Vibrations. Wiley, New York (1997)24. Golinval, J.C., Stephan, C., Lubrina, P., Peeters, M., Kerschen, G.: Nonlinear normal modes of a full-scale aircraft. In: 29th
International Modal Analysis Conference, Jacksonville, USA (2011)25. Gordnier, R.E., Visbal, M.R.: Development of a three-dimensional viscous aeroelastic solver for nonlinear panel flutter. J.
Fluids Struct. 16, 497–527 (2002)26. Hagedorn, P., DasGupta, A.: Vibrations and Waves in Continuous Mechanical Systems. Wiley, Chichester (2007)
232 M. Ducceschi et al.
27. Kerschen, G., Peeters, M., Golinval, J.C., Vakakis, A.F.: Nonlinear normal modes, part I: a useful framework for the structuraldynamicist. Mech. Syst. Signal Process. 23, 170–194 (2009)
28. Kung, G.C., Pao, Y.-H.: Nonlinear flexural vibrations of a clamped circular plate. J. Appl. Mech. 39, 1050–1054 (1972)29. Legge, K.A., Fletcher, N.H.: Nonlinearity, chaos, and the sound of shallow gongs. J. Acoust. Soc. Am. 86, 2439–2443 (1989)30. Leissa, A.: Vibration of Plates. Acoustical Society of America, New York (1993)31. Li, W.L.: Vibration analysis of rectangular plates with general elastic support. J. Sound Vib. 273, 619–635 (2003)32. Luo, A.C.J, Huang, J.: Analytical solutions for asymmetric periodic motions to chaos in a hardening Duffing oscillator. Non-
linear Dyn. 72, 417–438 (2013)33. Meenen, J., Altenbach, H.: A consistent deduction of von Kármán-type plate theories from three-dimensional nonlinear
continuum mechanics. Acta Mech. 147, 1–17 (2001)34. Mordant, N.: Are there waves in elastic wave turbulence? Phys. Rev. Lett. 100, 234505 (2008)35. Mordant, N.: Fourier analysis of wave turbulence in a thin elastic plate. Eur. Phys. J. B 76, 537–545 (2010)36. Moussa, M.O., Moumni, Z., Doaré, O., Touzé, C., Zaki, W.: Non-linear dynamic thermomechanical behaviour of shape
memory alloys. J. Intell. Mater. Syst. Struct. 23, 1593–1611 (2012)37. Murphy, K.D., Virgin, L.N., Rizzi, S.A.: Characterizing the dynamic response of a thermally loaded, acoustically excited
plate. J. Sound Vib. 196, 635–658 (1996)38. Nayfeh, A.H.: Nonlinear Oscillations. Wiley, New York (1995)39. Nayfeh, A.H., Pai, P.F.: Linear and Nonlinear Structural Mechanics. Wiley, New York (2004)40. Parlitz, U., Lauterborn, W.: Superstructure in the bifurcation set of the Duffing equation. Phys. Lett. A 107, 351–355 (1985)41. Peeters, M., Viguié, R., Sérandour, G., Kerschen, G., Golinval, J.-C.: Nonlinear normal modes, part II: toward a practical
computation using numerical continuation techniques. Mech. Syst. Signal Process. 23, 195–216 (2009)42. Ribeiro, P.: Nonlinear vibrations of simply-supported plates by the p-version finite element method. Finite Elem. Anal.
Des. 41, 911–924 (2005)43. Ribeiro, P., Petyt, M.: Geometrical non-linear, steady-state, forced, periodic vibration of plate, part I: model and convergence
study. J. Sound Vib. 226, 955–983 (1999)44. Ribeiro, P., Petyt, M.: Geometrical non-linear, steady-state, forced, periodic vibration of plate, part II: stability study and
analysis of multimodal response. J. Sound Vib. 226, 985–1010 (1999)45. Sathyamoorthy, M.: Nonlinear vibrations of plates: an update of recent research developments. Appl. Mech. Rev. 49, S55–
S62 (1996)46. Thomas, O., Bilbao, S.: Geometrically nonlinear flexural vibrations of plates: In-plane boundary conditions and some
symmetry properties. J. Sound Vib. 315, 569–590 (2008)47. Thomas, O., Touzé, C., Chaigne, A.: Non-linear vibrations of free-edge thin spherical shells: modal interaction rules and
1:1:2 internal resonance. Int. J. Solids Struct. 42, 3339–3373 (2005)48. Thomsen, J.J.: Vibrations and Stability. Springer, Berlin (2003)49. Touzé, C., Bilbao, S., Cadot, O.: Transition scenario to turbulence in thin vibrating plates. J. Sound Vib. 331, 412–433 (2011)50. Touzé, C., Thomas, O., Amabili, M.: Transition to chaotic vibrations for harmonically forced perfect and imperfect circular
plates. Int. J. Non-Linear Mech. 46, 234–246 (2011)51. Touzé, C., Thomas, O., Chaigne, A.: Hardening/softening behaviour in non-linear oscillations of structural systems using
non-linear normal modes. J. Sound Vib. 273, 77–101 (2004)52. Touzé, C., Thomas, O., Huberdeau, A.: Asymptotic non-linear normal modes for large-amplitude vibrations of continuous
structures. Comput. Struct. 82, 2671–2682 (2004)53. Vakakis, A.F.: Non-linear normal modes (nnms) and their applications in vibration theory: an overview. Mech. Syst. Signal
Process. 11, 3–22 (1997)54. von Kármán, T.: Festigkeitsprobleme im Maschinenbau. Encyklopädie der Mathematischen Wissenschaften 4, 311–
385 (1910)55. Yamaki, N.: Influence of large amplitudes on flexural vibrations of elastic plates. Zeitschrift für Angewandte Mathematik
und Mechanik 41, 501–510 (1961)56. Yang, X.L., Sethna, P.R.: Local and global bifurcations in parametrically excited vibrations of nearly square plates. Int. J.
Non-Linear Mech. 26, 199–220 (1991)
CHAPTER 5
Plates in a Strongly Nonlinear Regime: Wave Turbulence
This chapter corresponds to a paper published on Physica D [28] and dealing with the aspects of
Wave Turbulence in elastic plates. Section 2 gives the dynamical equations of von Karman plates and
the Finite Difference scheme to solve them, and thus it has been treated in great detail in Chapters
2 and 3. Section 4 is the core of the article, presenting the results and scaling properties of the
turbulent plate. Bibliographic references as well as appendices are included in this chapter, and not
at the end of the manuscript.
107
Physica D 280–281 (2014) 73–85
Contents lists available at ScienceDirect
Physica D
journal homepage: www.elsevier.com/locate/physd
Dynamics of the wave turbulence spectrum in vibrating plates: Anumerical investigation using a conservative finite difference scheme
Michele Ducceschi a, Olivier Cadot a, Cyril Touzé a,∗, Stefan Bilbao b
a Unité de Mécanique (UME), Ensta-ParisTech, 828, Boulevard des Maréchaux, 91762 Palaiseau Cedex, Franceb Room 7306B, James Clerk Maxwell Building, King’s Buildings, Mayfield Rd., Edinburgh EH9 3JZ, UK
h i g h l i g h t s
• Non-stationary wave turbulence in a vibrating plate is numerically studied.• Self-similar dynamics of the spectra are found with and without periodic external forcing.• The self-similar solutions are in agreement with the kinetic equation.• A realistic geometric imperfection is shown to have no effect on the global properties.
a r t i c l e i n f o
Article history:
Received 19 November 2013Received in revised form12 April 2014Accepted 25 April 2014Available online 10 May 2014Communicated by M. Vergassola
The dynamics of the local kinetic energy spectrum of an elastic plate vibrating in a wave turbulence(WT) regime is investigated with a finite difference, energy-conserving scheme. The numerical methodallows the simulation of pointwise forcing together with realistic boundary conditions, a set-up whichis close to experimental conditions. In the absence of damping, the framework of non-stationary waveturbulence is used. Numerical simulations show the presence of a front propagating to high frequencies,leaving a steady spectrum in its wake. Self-similar dynamics of the spectra are found with and withoutperiodic external forcing. For the periodic forcing, themean injected power is found to be constant, and thefrequency at the cascade front evolves linearlywith time resulting in a increase of the total energy. For thefree turbulence, the energy contained in the cascade remains constantwhile the frequency front increasesas t1/3. These self-similar solutions are found to be in accordance with the kinetic equation derivedfrom the von Kármán plate equations. The effect of the pointwise forcing is observable and introduces asteeper slope at low frequencies, as compared to the unforced case. The presence of a realistic geometricimperfection of the plate is found to have no effect on the global properties of the spectra dynamics. Thesteeper slope brought by the external forcing is shown to be still observable in amore realistic case wheredamping is added.
Wave Turbulence (WT) describes a system of waves interact-ing nonlinearly away from thermodynamical equilibrium [1,2]. Al-though the system under study is composed of waves only, theterm ‘‘turbulence’’ is used here in analogy with hydrodynamic tur-bulence, where the energy of the system is transferred throughscales (referred to as a cascade) resulting in a large bandwidth en-ergy spectrum. A particular property is that, for WT systems, theform of the spectrum can be derived analytically [3] and not just
in terms of dimensional analysis as for the Kolmogorov 41 theoryof hydrodynamics turbulence [4]. Using the assumption of weaknonlinearity, and an appropriate separation of timescales, a naturalclosure arises leading to an analytical expression for the equationfor the second ordermoment (e.g. the kinetic energy spectrum). So-lutions to this equation lead to two physically different scenarios:the first one represents the system at equilibrium, where the totalenergy of the system is equally spread among all the Fourier com-ponents of the system (knownas themodes), and thus correspond-ing to a Rayleigh–Jeans type of spectrum. The second scenario isout-of-equilibrium and leads to the Kolmogorov–Zakharov spec-trum that describes a flux of energy from the injection scale, whereenergy is input in the system, to the dissipation scale such as inhydrodynamics turbulence. In the latter scenario the modes re-
74 M. Ducceschi et al. / Physica D 280–281 (2014) 73–85
ceive and give energy to adjacent modes, thus creating a cas-cade of energy through scales. WT formalism has been appliedto many systems in a variety of contexts, ranging from quantum-mechanical to astrophysical systems, and includes many systemsencountered in the ordinary world. An exhaustive list may befound in [1]; here some examples are recalled: capillary [5,6]and surface gravity waves [7–9], Alfvén waves [10,11], and Kelvinwaves [12,13].
Flexural waves produced by large amplitude vibrations ofelastic plates have been studied within the framework of thewave turbulence theory [14] applied to the von Kármán equa-tion [15,16] for the transverse displacement w. The analytical Kol-mogorov–Zakharov spectrum is then given by
Pv(f ) =Ch
(1 − ν2)2/3ε1/3c log1/3
f ⋆c
f
; (1)
where εc is the constant flux of energy transferred through scales,Pv refers to the power spectrum of the transverse velocity v = w,h is the thickness of the plate, ν Poisson’s ratio of the material, andC a constant. Because the theory is fully inertial, f ⋆
c is the frequencyat which energy is removed from the system. In experiments, thisis ensured by the damping of the plate. At first order thespectrum is flat, but with a log-correction in the inertial range offrequencies. The WT theoretical result has been compared to ex-periments [17,18], showing discrepancies regarding the shape andscaling of the spectrumwith the energy flux. Thus, recentwork hasfocused on the investigation of thepossible causes for suchdiscrep-ancies. Experimentally, thewave-structure and dispersion relationwas checked in [18], leading to the conclusion that the nonlinearvibrations of a plate are indeed due to a set of waves following thetheoretical (linear) dispersion relation. The correct separation oftimescales, necessary assumption for the WT theory, was verifiedin [19]. A first discrepancy effectwas observed in [20], showing thatthe local forcing of the shaker is responsible for a steeper slope inthe supposed inertial range of the energy spectra. More recently,damping has also been shown to be the cause for a steeper slope ofthe spectrum, indicating that the inertial range might not exist forthin plates used in experiments, rendering then meaningless anycomparison with the WT theory [21]. From the numerical stand-point, it is worth mentioning that all the numerical methods usedso far are spectral schemes [14,22,21,23–25]. Hence the forcing isin the Fourier space, a feature that is different from a pointwise ex-citation used in experimental conditions. All available numericalresults recover the KZ spectrum of Eq. (1) when the damping is lo-calized at high frequency only. However, when realistic damping isadded, see e.g. [21,23], the same conclusions as for the experimentare met.
Other sources of discrepancies have not been addressed yet,such as the finite size effects or the possibility of three wave in-teractions (quadratic nonlinearities) in real plates. Because of thew → −w symmetry of the von Kármán equation, these non-linearities are not taken into account in [14]. Indeed, geometricalimperfections are unavoidable in real plates, and they are knownto break this symmetry and to produce quadratic nonlinearities[26,27]. In particular, it has been shown in [28,29] that imperfec-tions play an important role in the transition scenario to turbulenceand favor instabilities and the appearance of quasiperiodic vibra-tions.
The numericalmethod used in this work relies on a finite differ-ence, time domain, energy-conserving scheme [30,29]. The mainadvantages are that: (i) the time-stepping integrationmethod con-serves energy up to machine accuracy, so that essential propertiesof the underlying continuous Hamiltonian systems are preservedby the discretization [31]; (ii) the external forcing is pointwise inspace just as in the real experiments; (iii) realistic boundary con-ditions can be implemented instead of using periodic boundary
conditions as considered by previous numerical investigations us-ing spectral methods [14,22,25].
The aim of this article is to investigate numerically wave turbu-lence produced by the von Kármán plate equations. With a numer-ical scheme close to experimental conditions, unavoidable effectsin real experiments such as pointwise forcing and geometric im-perfections can be accounted for. In order to properly distinguishthe different effects, most of the presented results are obtained inthe absence of damping, where the framework of non-stationarywave turbulence should be used [32,33]. The theory predicts self-similar dynamics of the spectra with a front propagating to higherfrequencies. Such propagation has been observed for surface grav-ity waves in experiments [34]. On the contrary, capillary turbu-lence [35,36] exhibits a decay that begins from the high frequencyendof the spectral range. The discrepancywith the self-similar the-ory ofwave turbulence is ascribed to thepresence of finite dampingat all frequencies of the wave system [35,37].
The article is organized as follows: the governing equationstogether with the numerical approach are described in Section 2.Section 3 presents the data analysis tools used to study the spectraldynamics. The main results are given in Section 4. Periodicallyforced turbulence for a perfect plate is first considered. A self-similar propagation of a steep front towards the high frequencies,leaving in its wake a steady spectrum, is observed. The frequencyof the front is found to evolve linearly with time. The presenceof realistic geometric imperfections is then taken into accountand shown to have no influence on the spectral dynamics. InSection 4.2, the case of a free, undamped turbulence is exhibited.In that case, self-similar dynamics of the spectra are also observed,but now the front evolves with time as t1/3. Self-similar solutionsderived from the kinetic equation are found to display the samedependences, thus validating the numerical results that give inaddition the shape of the self-similar function. The pointwiseforcing is found to influence the shape of the universal spectrumleft in the wake of the front, with a steeper slope for the forcedcase. Finally, the effect of the pointwise forcing, underlined in theundamped cases, is confirmed in Section 4.3, where a decayingturbulence with a simple frequency-independent damping law isaddressed. Discussion and concluding remarks appear in Section 5.
2. Dynamical equations
2.1. Continuous time and space equations
The system under study is a rectangular elastic plate of thick-ness h, dimensions Lx, Ly, volume density ρ, Poisson’s ratio ν and
Young’s modulus E. Its flexural rigidity is defined as D = Eh3
12(1−ν2).
The dynamics of weakly nonlinear waves for the transverse dis-placement w(x, t) can be described by the von Kármán equa-tions [15,16]. The general case of an imperfect plate is hereconsidered. If w0(x) denotes the initial (static) imperfection, thenthe equations of motion read [26,38,27]
where ∆ is the Laplacian operator, 1a(x) = a,xx + a,yy, and L(·, ·)is the bilinear symmetric von Kármán operator, L(a(x), b(x)) =a,xx b,yy + a,yy b,xx − 2a,xy b,xy. F(x, t) is an auxiliary function calledthe Airy stress function which encapsulates the behavior of theplate in the in-plane direction, R(x, t) is a loss factor of some kindwhich will be specified shortly and F (x, t) is the external excita-tion load. In this work, the material parameters are chosen to cor-respond to a steel plate; thus E = 2 × 1011 Pa, ρ = 7860 kg/m3,ν = 0.3. The other geometrical and physical parameters will bereported case by case.
M. Ducceschi et al. / Physica D 280–281 (2014) 73–85 75
The dynamics of the plate is not complete until the boundaryconditions are not selected. Physical boundary conditions can bederived by conducting an energy analysis based on the Lagrangianof the system [16,39,40]. For this work, the particular case of atransversely simply supported plate with movable in-plane edgesis considered. In turn, the following conditions hold along theboundary ∂S
w = w,nn = 0 ∀x ∈ ∂S, (3a)
F = F,n = 0 ∀x ∈ ∂S, (3b)
where n is the direction normal to the boundary. This is an im-portant difference with respect to previously presented numericalsimulations, where periodic boundary conditions were employed.
The term R(x, t) represents losses. An artificial damping lawmay be used,
R(w, t) = 2σ0w, (4)
that dissipates energy at equal rates at all scales. In the contextof time-domain simulations of damped plates, the problem of anaccurate representation of the damping law with an ad-hoc timeoperator is complex and has led numerous authors to variouslaws, the implementation of which still remains a numericalchallenge; see e.g. [41]. Here the simplest time-domain operatorhas been chosen allowing us to explore numerically its effect onthe dynamics of the cascade. The reader should however keep inmind that it is ad-hoc and does not correspond to a real case.
The forcing is pointwise and of the form:
F (x, t) = δ(x − xF )A(t) sin(2π fpt). (5)
The injection point has been chosen at xF = (0.42Lx, 0.57Ly) forall the simulations. The forcing frequency fp is selected to be closeto the fourth eigenfrequency of the system, in order to activate thecascade more easily [29]. A(t) is chosen to be:
A(t) =
A0 t/t0 for 0 ≤ t ≤ t0;A0 for t0 ≤ t ≤ t1;0 for t ≥ t1.
(6)
In the above definition, t0 corresponds to the ramp time: theforcing ramps linearly from zero up to A0 in t0 seconds. Then,the forcing remains constant at A0 for t1 − t0 seconds, where t1corresponds to the total length of the simulation in the case ofperiodic forcing.
The injected power is defined in this work as
ε(t) = F (x, t) · w(xp, t)/ρS. (7)
After division by the factor ρS, where S = LxLy is the area of theplate, the injected power has the dimension of a velocity cubed.
2.2. Finite difference time domain scheme
In this section the numerical solution to system (2) togetherwith boundary conditions (3) is presented. Although numericalsimulations of von Kármán plates in the context of WT have beensuccessfully developed in previous studies [14,22,25,23,24], herea time domain simulation in physical space is presented. Timeand space are discretized so that the continuous variables (x, y, t)are approximated by their discrete counterparts (lδx,mδy, nδt),where (l,m, n) are integer indices and (δx, δy, δt) are the steps.Boundedness of the domain implies that (l,m) ∈ [0,Nx] × [0,Ny]so that the grid size is given by (Nx +1)× (Ny +1). The continuousvariablesw(x, t), F(x, t) are then approximated bywn
l,m, Fnl,m at the
discrete time n for the grid point (l,m). Time shifting operators areintroduced as
et+wnl,m = wn+1
l,m , et−wnl,m = wn−1
l,m . (8)
Time derivatives can then be approximated by
δt· =1
2ht
(et+ − et−), δt+ =1
ht
(et+ − 1),
δt− =1
ht
(1 − et+), δtt = δt+δt−. (9)
Time averaging operators are introduced as
µt+ =1
2(et+ + 1), µt− =
1
2(1 + et−),
µt· =1
2(et+ + et−), µtt = µt+µt−. (10)
Similar definitions hold for the space operators. Hence, theLaplacian ∆ and the double Laplacian ∆∆ are given by
δ∆ = δxx + δyy, δ∆∆ = δ∆δ∆. (11)
The von Kármán operator at interior points L(w, F) can then bediscretized as
l(w, F) = δxxwδyyF + δyywδxxF
− 2µx−µy−(δx+y+wδx+y+F). (12)
Thus the discrete counterpart of (2) is
Dδ∆∆w + ρhδttw = l(w + w0, µt·F) + Pnl,m − Rn
l,m; (13a)
µt−Dδ∆∆F = −Eh
2l(et−(w + 2w0), w). (13b)
The damping terms are
r0(l,m, n) = 2σ0δt·wnl,m; r1(l,m, n) = −2σ1δ∆wn
l,m. (14)
When σ0 = 0, the scheme is energy conserving, where thediscrete energy is positive definite and yields a stability condition,as proved in [30,42]. Implementation of boundary conditions isexplained thoroughly in [42].
3. Data analysis
The work is focused on the turbulent response at one point ofthe plate chosen as (0.3Lx, 0.2Ly). The kinetic energy spectrum isgiven by the velocity power spectrum which is calculated startingfromavelocity discrete-time series. For the remainder of the paper,the symbol vn will identify the discrete velocity at the output point,at the time t = nδt . Spectra analyses are performed on timewindows of duration τ . The discrete-time velocity power spectrumis then defined as:
Pv(f ) =(δt)2
τ
N
n=1
vne−i2π fn
2
(15)
where N = τ/δt is the total number of samples within the timewindow. For the typical case of a thickness h = 1 mm and surface0.4×0.6m2, the sampling frequency is chosen as 1/δt = 400 kHzand a time window of τ = 0.05 s is selected for the analysis ofthe spectra. In order to obtain a better convergence of the shape ofthe spectra, a mean is taken over M = 3 consecutive spectra; inother words, the symbol ⟨Pv(f )⟩ will identify the mean take over 3spectra covering a total timewindow T = Mτ . When the thicknessof the plate changes, timewindow and sampling frequency changeaccordingly. So, for instance, for a thickness h = 0.1 mm, the timewindow is multiplied by a factor of 10, τ = 0.5 s and the samplingfrequency is divided by a factor of 10, 1/δt = 40 kHz. The numberM remains instead fixed. In the following, the brackets ⟨· · ·⟩ willdenote an averaging on T which will generally depends on thetime.
76 M. Ducceschi et al. / Physica D 280–281 (2014) 73–85
Fig. 1. (a) Displacement field in the turbulent regime for an undamped, perfectly flat plate of thickness h = 0.1 mm and dimensions Lx × Ly = 0.4 × 0.6 m2 .(b) Corresponding velocity field.
The analysis for the injected power follows the same averagingrules. The injected power discrete time series is denoted by εn,from which the mean ⟨ε⟩ and the variance ⟨ε2⟩ are calculated. Thetemporal average ε is defined as the mean over the total data.
A characteristic frequency fc for the velocity power spectrum ishere introduced as
fc =
⟨Pv(f )⟩ f df
⟨Pv(f )⟩ df, (16)
with Pv(fc) also defining a characteristic spectral amplitude.Note that fc should not be confused with the theoretical cut-offfrequency f ⋆
c defined in Eq. (1). The characteristic frequency fc willbe used in the next section in order to quantify the self-similardynamics of the spectra in the non-stationary cases.
4. Numerical results
This section presents the results obtained for the followingcases:
Simulations are also conducted by varying the dimensions Sand the thickness h of the plate for different forcing amplitudes A0
and frequencies fp. The first part is devoted to periodically forcedturbulence and the second to free turbulence (or decaying, whendamping is added) after the forcing is stopped.
4.1. Periodically forced undamped turbulence
4.1.1. Perfect, undamped plates
Typical numerically obtained displacement and velocity fieldsare shown in Fig. 1 for illustration. The displacement field presentslow frequency patterns; taking the velocity filters out these lowfrequencies resulting in amuchmore homogeneous field, meaningthat velocity measurements at one point are relevant for theturbulent property of the whole plate as already mentionedin experiments having similar forcing schemes [17–19]. Theanisotropy effects due to the local forcing have been evidenced andcharacterized experimentally by Miquel and Mordant [19].
A case study is first examined to serve as a master example ofthe type of analysis that has been conducted on all the simulations.It corresponds to case 1 in Table 1 considering a plate of thicknessh = 1 mm, forced at fp = 75 Hz with a forcing of amplitude A0 =10 N and a ramp time t0 = 0.5 s (see Eq. (6), where t1 is the wholeduration of the simulation). The surface is Lx × Ly = 0.4 × 0.6 m2
and the grid size is 102 × 153 points, corresponding to a samplingrate of 400 kHz for the time integration.
Fig. 2(a) shows the spectrogram (evolution of the frequencyspectra with respect to time) of the velocity at the measurementpoint. It reveals the activated frequencies of the turbulent cascadeas a function of time. The energy keeps flowing into the system,creating a never ending cascade where modes of higher frequencyreceive energy from the adjacent lower frequency modes. Fig. 2(b)shows the velocity power spectra at different stages of thedynamics. It is evident that for these simulations no stationarystate exists: the spectra tend to occupy larger portions of theavailable frequency range as time goes by. It should be pointedout that the cascade front will develop up to half the samplingfrequency of the computation (200 kHz in this case): when thecascade hits this limit, an artificial boundary reflects the energyback into the box, towards smaller frequencies. This is a peculiar,
M. Ducceschi et al. / Physica D 280–281 (2014) 73–85 77
a
b
[m2/s
2/H
z]
Fig. 2. (a) Spectrogram of the velocity for the perfect undamped plate of thickness h = 1 mm, forcing from 0 to 10 N in 0.5 s (case 1 from Table 1), and then kept constant.(b) Corresponding velocity power spectra computed every 2.5 s from 5 to 25 s.
unwanted numerical phenomenon that is not taken into accountin the analysis. The simulation is stopped before the boundaryreflection happens; in this way, the cascade can be regarded asdeveloping within an infinite frequency domain. Fig. 3(a) showsthat the evolution of the characteristic frequency fc(t) is linear, fc =cf · t . The cascade front in Fig. 2 then develops to larger frequencieswith a constant cascade velocity cf . The spectral amplitude at thecharacteristic frequency ⟨Pv(fc)⟩ in Fig. 3(b) is seen to be fairlyconstant over time. The power velocity spectra, rescaled using boththe characteristic frequency fc and amplitude Pv(fc), are displayedin Fig. 4. They all satisfactorily superimpose, indicating that thedynamics of the energy spectrum is self-similar. This allows us towrite for the spectra
⟨Pv(f )⟩ = ⟨Pv(fc)⟩φP
f
fc
, (17)
where their shapes are given by the unique function φP (f /fc) (thesubscript P stands for periodically forced turbulence).
The injected power during the self-similar dynamics is shownin Fig. 5(a): the fluctuations increase with time while the averagestays constant. More precisely, Fig. 5(b) shows that ⟨ε2⟩ = Dt ,and ⟨ε⟩ = ε. Hence, the self-similar dynamics originate with the
a b
Fig. 3. (a) Time evolution of the characteristic frequency fc , (b) correspondingspectral amplitude of the spectra shown in Fig. 2(b) (case 1 in Table 1). Thecharacteristic frequency evolves as fc = cf t with cf = 412.05 s−2 and the meanamplitude is ⟨Pv(fc)⟩ = 1.11 · 10−4 m2/s2/Hz.
injection of a stationary energy flux characterized by ε. Meanwhile,the fluctuations of the injection flux grow following a diffusion-type behavior characterized by the coefficient D.
The analysis described above is now applied to 15 differentcases, summarized in Table 1. For all cases, the self-similardynamics display a constant injected power ε, a linear growth ofthe variance of injected power
ε2
, a linear increase of fc over timeand constant ⟨Pv(fc)⟩ has been observed. It is worth noting thatthe forcing values cover about four decades; this results in a largerange for the mean injected power ε. The thickness values coverone decade also. For each one of the cases, the cascade velocity cf ,the spectral amplitude at the characteristic frequency ⟨Pv(fc)⟩, thediffusion coefficient D are calculated. These quantities are plottedin Fig. 6 as functions of combinations of ε and h having the samedimensions. It can be seen that for all cases a linear relationshipis found, confirming the consistency of the dimensional argument.The constants of proportionalities are found from best linear fits:
⟨Pv(fc)⟩ = 2.51h(ε)1/3, (18a)
cf = 0.20(ε)2/3
h2, (18b)
D = 2.07 · 104 (ε)7/3
h. (18c)
In conclusion, the main result arising from the numericalsimulations of the periodically forced undamped plate is a self-similar evolution of the power spectra. It is characterized by theprogression towards higher frequencies of a steep cascade front,which leaves a steady self-similar spectrum in its wake. The self-similar progression is found to be linear with time and has beencharacterized by nondimensional numbers. The spectral amplitudeat fc is found to have a dependence on (ε)1/3 (see Fig. 6) and theself-similar spectrum can be expressed as
Pv(f ) = 0.42h(ε)13 ΦP
f
fc
, (19)
where ε is the mean injected power. In the absence of damping,the mean injected power can be confounded with the energy flux
78 M. Ducceschi et al. / Physica D 280–281 (2014) 73–85
Fig. 4. Spectra of Fig. 2(b) but normalized using the characteristic frequency fc and amplitude Pv(fc). Continuous red line shows a power law f − 14 . Continuous black lines
show the log correction log1/3
f ⋆cf
of the KZ spectrum, see Eq. (1), with f ⋆c = fc and f ⋆
c = 5fc . (For interpretation of the references to color in this figure legend, the reader
is referred to the web version of this article.)
Fig. 5. Time evolution of the injected power for the perfect undamped plate (case 1 in Table 1). (a) Time series, (b) ⟨ε⟩ and εrms =
⟨ε2⟩. Continuous lines are best fits thatgive ε = 9.65 · 10−5 m3/s3 , and D = 1.6 · 10−6 m6/s7 (see the text).
transfer εc through scales. The progression of the cascade front
towards higher frequencies is given by fc(t) = cf t ∝ ε23
h2t
(from Eq. (18b)). The function ΦP displayed in Fig. 4 increases asfrequencies decrease towards the forcing frequency fp. A best-fitapproximation of the slope of ΦP indicates that it follows a power-law for low frequencies with a small exponent close to −1/4; seeFig. 4.
The self-similar solutions for the kinetic equation derivedfrom the von Kármán plate equations are given in Appendix A.Considering a self-similar solution for the wave spectrum n(k, t)of the form:
n(k, t) = t−qf1(kt−p) = t−qf1(ξ), (20)
one finds for the power frequency spectrum Pv(ω, t):
Pv(ω, t) ∼ f1
ω
t
= g1
ω
t
. (21)
This relationship clearly evidenced that the frequency of the frontmust evolve linearlywith time,which is retrieved by the numericalsimulation. The function g1 can be identified with the function ΦP
found numerically.Let us now compare the self-similar spectrum with the KZ
solution. As the theoretical cut-off frequency f ⋆c cannot be related
to a given physical quantity in our numerical framework, the KZspectrum is built from Eq. (1) by selecting f ⋆
c = fc and f ⋆c = 5fc , and
reported in Fig. 4. As one is interested in the power-law behaviorin the low-frequency range, one can observe that selecting f ⋆
c = fcor f ⋆
c = 5fc has little influence on the slope comparison. It appearsthat even though the log-correction of the KZ spectrum cannot be
discarded, the slope of the self-similar numerical solution appearsto be a bit steeper.
The injected power fluctuation is found to increase as adiffusive law during the self-similar dynamics. A comprehensiveinterpretation of this behavior may be given by the model ofinjected power proposed in [43,44] for this system. In this work,the velocity w(xF , t) at the forcing point is assumed to result froma turbulent feedback v described by the velocity spectrum, and alinear response of the deterministic forcing F (x, t), say:
w(xF , t) = v + LF , (22)
with L a linear operator. The feedback turbulent velocity isassumed to be statistically independent of the forcing. Thus, usingEq. (22) and the periodic forcing in Eq. (5)withA(t) = A0, themean
M. Ducceschi et al. / Physica D 280–281 (2014) 73–85 79
Fig. 6. Results of simulations for the perfect, undamped plate with a periodicforcing, for all the 15 cases reported in Table 1. (a) Spectral amplitude ⟨Pv(fc)⟩ /h,(b) cascade velocity cf and (c) diffusion coefficient D.
a
b c
Fig. 7. Plate of dimensions 0.4 × 0.6 m2 with imperfection in the form of a raisedcosine. (a) 3D view, (b) and (c) x and y axes views.
of the squared injected power becomes:
⟨(F w)2⟩ =A20
2⟨v2⟩ + ⟨(LF )2F 2⟩. (23)
After a sufficiently long time, the stationary forcing term will benegligible comparedwith the quadratic term that keeps increasing
with time as the cascade propagates. Using Parseval’s identity:
⟨v2⟩ =
∞
0Pv(f )df (24)
and the expression of the self-similar time-dependent spectrum inEq. (19), Eq. (23) becomes:
⟨(F w)2⟩ ∼A20
2⟨v2⟩ ∝ A2
0
εt
h, (25)
then
⟨ε2⟩ ∝ A20
εt
h(ρS)2(26)
which gives the expected diffusive behavior. Hence, the injectedpower fluctuation is the consequence of a direct feedback of thepropagation of the kinetic energy spectrum during the self-similardynamics.
4.1.2. Imperfect, undamped plates
The effect of the presence of a plate imperfection on theturbulent dynamics is now investigated. Results are presentedfollowing the same procedure as for the perfect plate.
The static deformation w0(x) appearing in Eq. (2) is chosen inthe form of a raised cosine
w0(x) =Z
2
1 + cos
π
(x − x0)2 + (y − y0)2
L
, (27)
when (x − x0)2 + (y − y0)
2 ≤ L2, and zero otherwise. HereZ is the static (vertical) deflection, L is the width and x0 is thecenter of the deformation. The plate area is 0.4 × 0.6 m2 and thewidth is here selected to be 0.2 m, and x0 is the center of theplate; see Fig. 7. Z is then a free parameter that changes case bycase. This form of imperfection has been selected as it is closeto what can be observed in experiments, where large plates aregenerally affected by a pattern of largewavelength. Our goal is thusto quantify the effect of a selected realistic geometric imperfectionin order to assess its potential effect on the turbulence spectra.For the perfect plate with w0(x) = 0, the internal restoringforce is symmetric so that only cubic nonlinearities are presentin the von Kármán equations. However when an imperfection isconsidered, quadratic nonlinearity appears in themodel equationsand so three-wave processes are present in the dynamics.
A case study (case 11 in Table 2) is first examined. It correspondsto a platewith a thickness h = 1mm, and a deformation Z = 5mmas defined in Eq. (27). As the eigenfrequencies increase with theimperfection (see e.g. [29,27]), the excitation frequency is nowshifted so as to remain in the vicinity of the fourth eigenfrequency,so that now fp = 103 Hz, and the forcing amplitude is selected asA0 = 90 N.
During the dynamics, it is observed that the velocity powerspectra evolve almost identically to the case of the perfect plate,so that the spectrogram and power spectra of the imperfect plateare similar to those shown in Fig. 2. The characteristic frequencyincreases linearly with time while the characteristic amplituderemains fairly constant as shown in Fig. 8. The normalized spectrain Fig. 9 are superimposed according to a curve φP(f /fc) =⟨Pv(f )⟩/⟨Pv(fc)⟩ indicating self-similar dynamics. The self-similardynamics is also produced during a mean constant injection fluxwith diffusive-type fluctuations, as seen in Fig. 10.
A total of 12 simulations are considered for imperfect plates.The parameters are listed in Table 2. Note that the magnitude ofthe imperfection considered is large (Z ≥ h), and of the orderof what can be expected in real experiments. In particular, it hasbeen shown in [45,27,46] that an imperfection of the order of the
80 M. Ducceschi et al. / Physica D 280–281 (2014) 73–85
Fig. 8. Imperfect, undamped plate, case 11 of Table 2. (a) Time evolution of the characteristic frequency, (b) corresponding spectral amplitude. Continuous lines are best fitfc = cf t with cf = 226 s−2 (a), and the mean amplitude ⟨Pv(fc)⟩ = 2.66 · 10−4 m2/s2/Hz.
Fig. 9. Normalized velocity spectra using the characteristic frequency fc and amplitude Pv(fc) (case 11 in Table 2). Continuous black line shows the log correction log1/3
f ⋆cf
of the KZ spectrum, see Eq. (1), with f ⋆c = fc . Dashed red line shows a power law f − 1
4 . (For interpretation of the references to color in this figure legend, the reader is referredto the web version of this article.)
thickness h is able to change the type of nonlinearity of the lowfrequency modes. For each one of the cases, the cascade velocitycf , the spectral amplitude at the characteristic frequency Pv(fc) andthe coefficient D are plotted as functions of combinations of ε andh. It can be seen that for all cases a linear relationship is found(Fig. 11):
⟨Pv(fc)⟩ = 2.30hε1/3 (28a)
cf = 0.19ε2/3
h2(28b)
D = 1.86 · 104 ε7/3
h. (28c)
The scaling laws are identical to the perfect case, althoughthe data are a bit more scattered in Fig. 11 than in Fig. 6.
The obtained values for the proportional constants are also veryclose. The quadratic nonlinearity introduced by an imperfection isthen hardly discernable in the turbulent cascade dynamics whichindicates that the vibration amplitudes are sufficiently importantso that the cubic term dominates the quadratic one; hence onlythe cubic nonlinearity seems to drive the main characteristics.In conclusion, the geometric imperfection retained in this study,and which has been selected as it provides insight into realisticimperfections one may encounter in experimental situations, hasno effect on the main characteristics of the turbulent spectra.Hence it appears that plate imperfections should not be consideredas a potential cause for explaining the discrepancies observedbetween theory derived for perfect plates and real experimentswith unavoidable imperfections.
Fig. 10. Time evolution of the injected power for the imperfect undamped plate (case 11 in Table 2). (a) Time series, (b) ⟨ε⟩ and εrms =
⟨ε2⟩. Continuous lines are best fits:ε = 1.15 · 10−5 m3/s3 , and D = 0.0015 m6/s7 (see the text).
M. Ducceschi et al. / Physica D 280–281 (2014) 73–85 81
Fig. 11. Results of simulations for the imperfect, undamped plate with a periodicforcing, for the 12 cases reported in Table 2. (a) Spectral amplitude ⟨Pv(fc)⟩/h,(b) cascade velocity cf and (c) coefficient D.
In the remainder of the paper, the plate imperfections are nolonger considered. The next section is devoted to the study offree (unforced) turbulence in order to highlight the effect of thepointwise forcing.
4.2. Free undamped turbulence
We now consider the case where the perfect, undamped plate,given an initial turbulent spectrum energy, is left free to vibratein the absence of forcing and develops a cascade. The platedimensions are Lx×Ly = 0.4×0.6m2, and the thickness is selectedas h = 0.1 mm. The sampling rate is chosen as 40 kHz resulting ina grid size of 102 × 153 points. The excitation frequency is in thevicinity of the fourth eigenmode at 7.5 Hz. The forcing amplitudereaches A0 = 0.1 N linearly after a duration t0 = 0.1 s and isthen abruptly stopped. The response of the system is shown overa long time duration in the spectrogram of Fig. 12(a). Even afterstopping the external excitation, the number of excited modeskeeps increasing slowly. Because of the slowness of this dynamics,the data analysis has been exceptionally changed with respectto the standard procedure explained in Section 3. Here the timewindow is τ = 0.1 s and the number of spectra over which theaverage is taken is M = 100, resulting in a time T = Mτ = 10 s.
The velocity power spectra of the free decaying turbulence areshown at different stages of the dynamics in Fig. 12(b). The shapeof the spectra changes abruptly just after the forcing is stopped.There is an evidence of a flattening in the low-frequency part of
Fig. 12. (a) Spectrogram of the velocity of the perfect, undamped plate for whichthe forcing is stopped after 0.1 s. The plate is of thickness h = 0.1 mm and thesampling rate 40 kHz. (b) Corresponding velocity power spectra averaged over 10 s,displayed for time intervals of 30 s. The first one (red) is computed from0.1s (i.e. theend of the forcing) to 10.1 s. Straight red line corresponds to the power law f −1/4 .(For interpretation of the references to color in this figure legend, the reader isreferred to the web version of this article.)
the spectra, indicating once again the effect of the external forcinghas on the power-law slope. On the other hand, the cascade frontstill progresses towards high frequencies even without forcing.The corresponding characteristic frequency evolution is shown inFig. 13(a) and follows a clear 1/3 power law, significantly differentfrom the linear dependence found for the case with externalforcing. The energy conservation during the dynamics justifies the−1/3 power law best fit for the spectra amplitude in Fig. 13(b).More precisely, the characteristic frequency is found to behave as
fc = at1/3 with a = 331.5 s−43 , whereas the spectral amplitude
reads Pv(fc) = bt−1/3 with b = 1.1 · 10−7 m2 s−23 . In order
to express these dependences with nondimensional numbers, onecan introduce the conserved quantities of the system, i.e. the totalenergy ξ = h
2
Pv(f )df of the turbulent fluctuations – oncethe forcing stopped the system is conservative – and the platethickness h. The energy ξ may locally fluctuate since it is actuallythe energy of the whole plate that is conserved. However, forthe point considered, it is found to keep reasonably constant atξ ≈ 2 · 10−8 m3/s2 during the self-similar dynamics, as shown inFig. 14. Using the relationships derived from the best fits obtainedin Fig. 13 together with a dimensional analysis, one can reexpressthe dependences as
fc(t) = 0.45ξ
23
h2t13 Pv(fc) = 0.41hξ
13 t−
13 . (29)
The two constants now appearing in Eq. (29) should be universal,as are the nondimensional numbers derived from the analyses inprevious sections.
The normalized spectra using both the characteristic frequencyand corresponding spectra amplitude are shown in Fig. 15 for timeslarger than 10 s (i.e. after the low frequency spectra flattening).They all superimpose showing that the dynamics becomes self-similar with a spectrum universal shape ΦF such that Pv(f ) =
Pv(fc)ΦF
f
fc
.
The progression of the cascade front towards higher frequenciesmust be accomplished by the presence of an energy flux εc . It canbe estimated from the energy dξc of the activated modes betweenfc and fc + dfc as the cascade propagates during the time interval
82 M. Ducceschi et al. / Physica D 280–281 (2014) 73–85
Fig. 13. Undamped free turbulence. (a) Time evolution of the characteristic frequency fc and (b) corresponding spectral amplitude. Continuous lines are best fits fc = at1/3
with a = 331.5 s−43 (a), and Pv(fc) = bt−1/3 with b = 1.1 · 10−7 m2 s−
23 (b).
Fig. 14. Time evolution of the kinetic energy ξ = h2
Pv(f )df for the freeundamped turbulence.
dt , εc =dξcdt
= h2Pv(fc)
dfcdt. Using both evolutions in Eq. (29), the
estimation gives:
εc = 0.03ξ t−1. (30)
The spectrum in the self-similar dynamics of free turbulence canthus be expressed as:
Pv(f ) = 13.34hε13c ΦF
f
fc
. (31)
As for the first case with periodic forcing, the dependence ofthe frequency front in t1/3 can be derived from the theoreticalkinetic equation governing the dynamics of the wave spectrumfor vibrating plates. Following the calculations presented inAppendix A, and considering now that, for the self-similar solutionof the form given by Eq. (20), the total energy of the system isconserved, one finds that the power frequency spectrum should
fulfill the relationship:
Pv(ω, t) ∼ t−1/3f2
ω
t1/3
= t−1/3g2
ω
t1/3
. (32)
This theoretical results clearly exhibits the fact that the frequencyfront should evolve as t1/3 while the total energy as t−1/3. Inthis case of free turbulence, the function g2 can now be directlyidentified from the numerical solution ΦF .
Let us now compare the self-similar solution with the theoreti-cal KZ spectrum for vibrating plates. Because of the spectral flatten-ing highlighted in Fig. 12(b), one can observe that the function ΦF
is now very close to the log-correction of the theoretical KZ spec-trum for vibrating plates, as displayed in Fig. 15, and shown herefor f ⋆
c = fc . The similarity between the self-similar spectrum ofdecaying turbulence with the stationary KZ spectrum has alreadybeenmentioned for surface gravity waves [34] and capillary waves[35,37,36]. The comparison between the self-similar spectra of pe-riodically forced turbulence (Fig. 4) and free turbulence (Fig. 15)shows a steeper slope when forcing is present. This result shouldbe retrieved in a more realistic case where damping is also con-sidered and should corroborate the experimental results shownin [20]. The aim of the last section is thus to verify this numericallyin the case of a decaying turbulence.
4.3. Damped turbulence
The effect of the forcing is now studied in a damped case. Theplate dimensions are Lx × Ly = 0.4 × 0.6 m2, the grid size is102 × 153 and h = 1 mm. The damping introduced in Eq. (4) isselected as σ0 = 0.5 s−1. The forcing frequency is fp = 75 Hz, witha forcing amplitude of A0 = 140 N and a ramp time t0 = 0.5 s. Theforcing remains periodic from 0.5 to 3.5 s (t1 = 4 s in Eq. (6)) andthen abruptly stopped at t = 4 s.
Fig. 15. Free undamped turbulence. Normalized velocity power spectra of Fig. 12(b) for t > 10 s during the self-similar dynamics. Black line shows the log correction
log1/3
f ⋆cf
of the KZ spectrum, Eq. (1), with f ⋆c = fc .
M. Ducceschi et al. / Physica D 280–281 (2014) 73–85 83
Fig. 16. (a) Spectrogram of the velocity for the perfect damped plate. (b) Corre-sponding velocity power spectra averaged over T = 0.15 s computed every 5 sfrom 5 to 20 s.
Fig. 17. Time evolution of the injected power, ⟨ε⟩ and εrms =
⟨ε2⟩ for thedecaying turbulence experiment shown in Fig. 16.
The response of the damped system is shown over 20 s ofduration in the spectrogram in Fig. 16(a). The spectra reach anearly steady state just before t = 4 s that corresponds to thetime at which the forcing is stopped. Meanwhile, the injectedpower remains fairly constant in Fig. 17, ⟨ε⟩ (t) ≃ ε, and thecharacteristic frequency grows, just as for the undamped casestudied in Section 4.1. Themain difference is that the characteristicfrequency (Fig. 18) will saturate to a constant value once thestatistical steady state of turbulent energywill be reached. In otherwords, the cascade velocity front decreases towards zero whenapproaching the steady state.
As the cascade progresses to higher frequencies,more andmoremodes are activated, which results in an increase of the dissipationflux εd since each mode has a linear energy loss parameterizedby σ0 that should be compensated by the incoming flux. Hence,less and less energy flux εc is available to propagate the cascadefront velocity, since ⟨εc⟩ (t) = ε − ⟨εd⟩ (t). Once the forcing isstopped at t = 4 s, the characteristic frequency overshoots asshown in Fig. 18 and then sharply saturates. The drastic increaseof the characteristic frequency is provoked by the flattening of thespectral shape at low frequencies as observed in Fig. 16(b).
The effect of the pointwise forcing evidenced in previoussections is here retrieved for the damped dynamics. The numerical
experiment shown here shares similarities with the experimentalresult of [20], where the spectral flattening was also observed inthe decaying turbulence regime. Once the forcing stopped, thespectrum simply decreases exponentially as e−2σ0t as expectedby a pure damping linear dynamics; see Fig. 18(b). Actually onecan observe that the nonlinear dynamics are still present but veryweak since the nonlinear propagation depends on the vanishingturbulent energy ξ(t). Note that selecting other damping lawsshould lead to different behaviors in the decaying regime, resultingfrom the competition between the nonlinear propagation effectwith the energy losses, both of which having different frequency-dependences associated to different timescales. Here the dampinglaw is frequency independent so that the results lend themselvesto an easy physical interpretation.
5. Discussion and concluding remarks
The nonlinear dynamics of turbulent vibrating plates has beenstudied numerically with a finite-difference, energy-conservingscheme including a pointwise forcing together with realisticboundary conditions. The most important results have been ob-tained in the absence of damping, in the framework of non-stationary wave turbulence. Self-similar solutions for the energyspectrum have been exhibited for a wide range of parameter vari-ations. The simulations display the presence of a front propagat-ing to high frequencies. With pointwise forcing, this propagationis linear with time, whereas for free turbulence the dependence isin t1/3. These self-similar behaviors can be directly retrieved fromthe kinetic equation by analyzing the admissible self-similar solu-tions. From the numerical solutions, one is thus able to get a nu-merical value for the self-similar functions in non-stationary waveturbulence for plates, for the two cases studied in this paper, withand without external forcing. Comparing the shape of these nu-merically obtained functions, one observes that they share simi-larities with the theoretical KZ spectrum computed by [14], albeitexhibiting interesting differences. In the case of a pointwise forc-ing, a steeper slope is observed as compared to the free undampedturbulence. Note also that the power 1/3 dependence on the en-ergy flux is numerically retrieved.
This observation is robust to adding the damping in thesimulations and thus recovers experimental results shown in [20].It can thus be concluded that the local pointwise forcing has ameasurable effect on the slope in the low-frequency range. Thiseffect has been related in [20] to an anisotropy induced by thepresence of the shaker. A direct extension of the results presentedherein should thus to compute spatial spectra in order to verifynumerically that the same argument holds.
For the first time, our numerical set-up allows for an investiga-tion of the effect of a geometrical imperfection on the turbulentdynamics. The results, obtained in a non-stationary framework,clearly indicates that perfect and imperfect plates present identi-cal characteristics in the WT regime. This emphasizes the fact thatin this regime the cubic nonlinear terms dominate the quadraticones, which thus have no measurable effect on the spectral char-acteristics of the WT. Note however that this is not true for theregimes of transition to turbulence that involve weaker excitationamplitudes [28,29]. Note also that only a simple, low-frequencypattern has been introduced as a geometric imperfection, in orderto present numerical results close to what can be expected in real-life situations. The conclusions, based on numerical experiments,are only valid for those cases. Extensions of the present work couldconsider more complex geometric imperfections, with smallerwavelengths, in order to continue the quantification of the tran-sition between perfect and imperfect plates’ turbulent dynamics.
Finally, dimensional arguments have been used in order toproperly quantify the results in non-stationary cases. As no
84 M. Ducceschi et al. / Physica D 280–281 (2014) 73–85
Fig. 18. (a) Time evolution of the characteristic frequency fc and (b) corresponding spectral amplitude decreasing as ⟨Pv(fc)⟩ = 0.012e−2σ0 t of the spectra shown in Fig. 16,with σ0 = 0.5 s−1 (blue thick line).
theoretical prediction for the non-stationary evolution of systemswith log-corrected spectra exist at the present time, we believe theresults shown here could be used so as to ascertain a theoreticaldevelopment that may predict the observations reported in thiscontribution.
Acknowledgments
S. Bilbao was supported by the European Research Council,under Grant number StG-2011-279068-NESS. The authors want tothank Christophe Josserand for a fruitful discussion on the collisionintegral.
Appendix. Self-similar solutions for non-stationary wave tur-
bulence in plates
This appendix is devoted to the derivation of self-similarsolutions from the kinetic equation describing thewave turbulencein the von Kármán plate equations. Following the theoreticalcalculations reported by Düring et al. [14], the 4-waves kineticequation has the general expression given by Zakharov et al. [2],and reads
∂n(k, t)
∂t= I(k), (A.1)
with n(k, t) ≡ nk the wave spectrum and I(k) the collisionintegral, the expression of which can be found in [14]:
I(k) = 12π
|Jk123|2fk123δ(k + s1k1 + s2k2 + s3k3)
× δ(ωk + s1ω1 + s2ω2 + s3ω3)dk1dk2dk3, (A.2)
where Jk123 stands for the interaction term and fk123 is such that
fk123 =
s1,s2,s3
nknk1nk2nk3
1
nk
+s1
nk1
+s2
nk2
+s3
nk3
. (A.3)
Following [2], let us introduce a self-similar solution for the non-stationary evolution, depending only on the wavevector modulus,as
n(k, t) = t−qf (kt−p) = t−qf (η). (A.4)
Plugging this ansatz in the kinetic equation (A.1), and taking intoaccount the expression of |Jk123|2 found in [14], one gets
−t−q−1
qf (η) + pηf ′(η)
= I(η)t−3q+2p, (A.5)
so that a solution of the form (A.4) is possible only if the condition−q − 1 = −3q + 2p is satisfied. It can be rewritten as
2(q − p) = 1. (A.6)
Let us introduce the total energy of the distribution
ξ =
ω nk dk (A.7)
and consider the two cases numerically studied:
Case 1: The plate is forced by a sinusoidal pointwise forcing ofconstant amplitude and excitation frequency. In this casethe total energy increases linearly with time so that ξ ∼ t .
Case 2: The plate is left free to vibrate, given an amount of energyas the initial condition. In this case the total energy isconstant so that ξ ∼ t0.
Substituting (A.4) into (A.7) one obtains a second relationshipbetween p and q, which reads, depending on the case considered
4p − q =
1 for case 10 for case 2. (A.8)
Solving for (p, q) in both cases give
case 1 : p = 1/2, q = 1, (A.9)
case 2 : p = 1/6, q = 2/3. (A.10)
The last step consists in expressing the self-similar solution forPv(ω) the power spectrum of the transverse velocity v = wused in the analysis, which is related to the power spectrum ofthe displacement Pw(ω) by a proportionality relationship Pv(ω)∝ ω2Pw(ω). Using the space–frequency relationship Pw(ω)dω ∝Pw(k)kdk, together with the dispersion relation, one finds Pw(ω) ∝Pw(k), such that Pv(ω) ∝ k4Pw(k). Finally, using the relationshipPw(k) ∝
nkω
given in [14], one obtains finally Pv(ω, t) ∝ k2n(k, t),so that the self-similar solutions for Pv(ω, t) finally reads for thegeneral case with nk given by Eq. (A.4):
Pv(ω, t) ∼ t2p−qf (ω1/2t−p). (A.11)
Specifying now the solutions for (p, q) found for the two casesunder study, one obtains for case 1:
Pv(ω, t) ∼ f1
ω
t
= g1
ω
t
, (A.12)
and for case 2:
Pv(ω, t) ∼ t−1/3f2
ω
t1/3
= t−1/3g2
ω
t1/3
, (A.13)
where g1,2 (or f1,2) have been indexed with respect to case 1 andcase 2, and are functions to be defined.
M. Ducceschi et al. / Physica D 280–281 (2014) 73–85 85
References
[1] S. Nazarenko, Wave Turbulence, Springer-Verlag, Berlin, Heidelberg, 2011.[2] V. Zakharov, V. L’vov, G. Falkovich, Kolmogorov Spectra of Turbulence 1:Wave
Turbulence, in: Series in Nonlinear Dynamics, Springer, 1992.[3] A. Newell, S. Nazarenko, L. Biven, Wave turbulence and intermittency, Physica
D 152–153 (2001).[4] U. Frisch, Turbulence, Cambridge University Press, Cambridge, 1995.[5] V.E. Zakharov, N.N. Filonenko, Energy spectrum for stochastic oscillations of
surface of a liquid, J. Appl. Mech. Tech. Phys. 8 (1967) 37.[6] A.N. Pushkarev, V.E. Zakharov, Turbulence of capillary waves, Phys. Rev. Lett.
76 (1996) 3320–3323.[7] V.E. Zakharov, N.N. Filonenko, Energy spectrum for stochastic oscillations of
surface of a liquid, Sov. Phys. Dokl. 11 (1967) 881–883.[8] M. Onorato, A.R. Osborne, M. Serio, D. Resio, A. Pushkarev, V.E. Zakharov, C.
Brandini, Freely decayingweak turbulence for sea surface gravity waves, Phys.Rev. Lett. 89 (2002) 144501.
[9] E. Falcon, C. Laroche, S. Fauve, Observation of gravity—capillary waveturbulence, Phys. Rev. Lett. 98 (2007) 094503.
[10] S. Galtier, S. Nazarenko, A. Newell, A. Pouquet, A weak turbulence theory forincompressible magnetohydrodynamics, J. Plasma Phys. 63 (2000) 447–488.
[11] S. Galtier, S. Nazarenko, A. Newell, A. Pouquet, Anisotropic turbulence of shear-Alfvèn waves, Astrophys. J. Lett. 564 (2002) L49–L52.
[12] W. Vinen, Classical character of turbulence in a quantum liquid, Phys. Rev. B61 (2000) 1410–1420.
[13] W. Vinen, Decay of superfluid turbulence at a very low temperature: theradiation of sound from a Kelvin wave on a quantized vortex, Phys. Rev. B 64(2001) 134520.
[14] G. Düring, C. Josserand, S. Rica, Weak turbulence for a vibrating plate: can onehear a Kolmogorov spectrum? Phys. Rev. Lett. 97 (2006) 025503.
[15] T. von Kármán, Festigkeitsprobleme im Maschinenbau, Encyklopadie Math.Wiss. 4 (1910) 311–385.
[16] O. Thomas, S. Bilbao, Geometrically nonlinear flexural vibrations of plates: in-plane boundary conditions and some symmetry properties, J. Sound Vib. 315(2008) 569–590.
[17] A. Boudaoud, O. Cadot, B. Odille, C. Touzé, Observation of wave turbulence invibrating plates, Phys. Rev. Lett. 100 (2008) 234504.
[18] N. Mordant, Are there waves in elastic wave turbulence? Phys. Rev. Lett. 100(2008) 234505.
[19] B. Miquel, N. Mordant, Nonlinear dynamics of flexural wave turbulence, Phys.Rev. E 84 (2011) 066607.
[20] B.Miquel, N.Mordant, Nonstationarywave turbulence in an elastic plate, Phys.Rev. Lett. 107 (2011) 034501.
[21] T. Humbert, O. Cadot, G. Düring, C. Josserand, S. Rica, C. Touzé,Wave turbulencein vibrating plates: the effect of damping, Europhys. Lett. 102 (2013) 30002.
[22] N. Yokoyama,M. Takaoka,Weak and strongwave turbulence spectra for elasticthin plate, Phys. Rev. Lett. 110 (2013) 105501.
[23] B. Miquel, A. Alexakis, N. Mordant, Role of dissipation in flexural waveturbulence: from experimental spectrum to Kolmogorov–Zakharov spectrum,Phys. Rev. E (2014) submitted for publication.
[24] B. Miquel, Études expérimentales et numériques de la turbulence d’ondes deflexion (Ph.D. thesis), Université Pierre et Marie curie, Paris, 2013.
[25] B. Miquel, A. Alexakis, C. Josserand, N. Mordant, Transition from waveturbulence to dynamical crumpling in vibrated elastic plates, Phys. Rev. Lett.111 (2013) 054302.
[26] Z. Celep, Free flexural vibration of initially imperfect thin plates with largeelastic amplitudes, ZAMM Z. Angew. Math. Mech. 56 (1976) 423–428.
[27] C. Camier, C. Touzé, O. Thomas, Non-linear vibrations of imperfect free-edgecircular plates and shells, Eur. J. Mech. A Solids 28 (2009) 500–515.
[28] C. Touzé, O. Thomas, M. Amabili, Transition to chaotic vibrations forharmonically forced perfect and imperfect circular plates, Internat. J. Non-Linear Mech. 46 (2011) 234–246.
[29] C. Touzé, S. Bilbao, O. Cadot, Transition scenario to turbulence in thin vibratingplates, J. Sound Vib. 331 (2012) 412–433.
[30] S. Bilbao, A family of conservative finite difference schemes for the dynamicalvon Kármán plate equations, Numer. Methods Partial Differential Equations24 (2008) 193–216.
[31] E. Hairer, C. Lubich, G. Wanner, Geometric Numerical Integration, Structure-Preserving Schemes for Ordinary Differential Equations, Springer, 2006.
[32] G. Falkovich, A. Shafarenko, Nonstationary wave turbulence, J. Nonlinear Sci.1 (1991) 457–480.
[33] C. Connaughton, A. Newell, Y. Pomeau, Non-stationary spectra of local waveturbulence, Physica D 184 (2003) 64–85.
[34] R. Bedard, S. Lukaschuk, S. Nazarenko, Non-stationary regimes of surfacegravity wave turbulence, JETP Lett. 97 (2013) 459–465.
[35] G.V. Kolmakov, A.A. Levchenko, M.Y. Brazhnikov, L.P. Mezhov-Deglin,A.N. Silchenko, P.V.E. McClintock, Quasiadiabatic decay of capillary turbulenceon the charged surface of liquid hydrogen, Phys. Rev. Lett. 93 (2004) 074501.
[36] L. Deike, M. Berhanu, E. Falcon, Decay of capillary wave turbulence, Phys. Rev.E 85 (2012) 066311.
[37] G. Kolmakov, Decay of capillary turbulence on the surface of a viscous liquid,JETP Lett. 83 (2006) 58–63.
[38] G.L. Ostiguy, S. Sassi, Effects of initial geometric imperfections on dynamicbehaviour of rectangular plates, Nonlinear Dynam. 3 (1992) 165–181.
[39] L.D. Landau, L.P. Pitaevskii, E.M. Lifshitz, A.M. Kosevich, Theory of Elasticity,third ed., Butterworth–Heinemann, 1986.
[40] M. Ducceschi, C. Touzé, S. Bilbao, C. Webb, Nonlinear dynamics of rectangularplates: investigation of modal interaction in free and forced vibrations, ActaMech. 225 (2014) 213–232.
[41] A. Chaigne, C. Lambourg, Time-domain simulation of damped impacted plates.I: theory and experiments, J. Acoust. Soc. Am. 109 (2001) 1422–1432.
[42] S. Bilbao, Numerical Sound Synthesis: Finite Difference Schemes andSimulation in Musical Acoustics, Wiley, 2009.
[43] O. Cadot, A. Boudaoud, C. Touzé, Statistics of power injection in a plate set intochaotic vibration, Eur. Phys. J. B 66 (2008) 399–407.
[44] O. Cadot, C. Touzé, A. Boudaoud, Linear versus nonlinear response of a forcedwave turbulence system, Phys. Rev. E 82 (2010) 046211.
[45] C. Touzé, O. Thomas, Non-linear behaviour of free-edge shallow sphericalshells: effect of the geometry, Internat. J. Non-Linear Mech. 41 (2006)678–692.
[46] C. Touzé, C. Camier, G. Favraud, O. Thomas, Effect of imperfections anddamping on the type of non-linearity of circular plates and shallow sphericalshells, Math. Probl. Eng. 2008 (2008) Article ID 678307.
CHAPTER 6
Sound Synthesis of Gongs
In the previous chapters, the von Karman equations were shown to be an appealing system to study
because they give rise to a vast range of nonlinear phenomena despite the apparent simplicity of the
system under consideration (a vibrating thin plate). An attractive possibility of numerical modelling
of thin plates resides in the synthesis of gong-like sounds.
The aim of this chapter is to present some simulation results coming from running the modal code. For
the first time, a modal approach is used in reproducing the weakly and strongly nonlinear vibrations
of thin plates. The chapter presents at the beginning a discussion on how to implement an excitation
mechanism to set the plate into motion, and convenient ways of extracting the output.
A quick discussion on loss is then offered, to show that the modal approach is very flexible in terms
of the damping law that one wants to implement. Successively, a discussion on the number of modes
to be retained in the simulation is offered, and finally a comparison with a Finite Difference scheme
is given. See also [30].
6.1. Simulation Parameters
In this section the setup for the simulations is presented. The excitation, output and damping factor
choices are explained and examples given.
6.1.1. Plate excitation. In order to produce a gong-like sound, a plate must be excited in a
way that approximates how one may actually set a gong into motion. The most obvious way is to
strike a plate. For that, one may think of the excitation mechanism P (x, t) to be in the form of
(6.1) P (x, t) = f(x)p(t),
where f(x) is some kind of spacial distribution and p(t) is a temporal forcing. The spacial distribution
can be safely chosen to be a Dirac delta, f(x) = δ(x − xi), where xi is the input location. This is
largely sufficient, in many cases, for striking. When the ratio between mallet size and area of the
plate gets larger, however, one may choose a different form for f(x); for example a 2D raised cosine
distribution [8]. Such a function, here denoted by fr(x), is
(6.2) fr(x) =
H2
[
1 + cos(
π√
(x− xi)2 + (y − yi)2/r)]
if (x− xi)2 + (y − yi)
2 ≤ r2;
0 if (x− xi)2 + (y − yi)
2 > r2.
121
122 6. SOUND SYNTHESIS OF GONGS
0 0.1 0.2 0.3 0.40
0.2
0.4
0.6
0.8
1
x [m]
fr[m
]
0 0.2 0.4 0.60
0.2
0.4
0.6
0.8
1
y [m]
fr[m
]
(a) (b) (c)
Figure 1. 2D raised cosine distribution, Eq. (6.2), with amplitude H = 1m and
width r = 0.2m, centered around the point xi = [0.2m 0.3m]. (a): 3D view; (b):
x-section; (c): y-section
Such a function depends on 2 parameters, a radius r and an amplitude H; an example of such a
function is given in Fig. 1.
Turning the attention to the temporal excitation, one may wish to represent it as interaction between
the surface of the plate and a mallet. Let wH(t) be the position of the mallet at the time t. Then
the force acting on the plate due to the mallet interaction is given by Newton’s second law
p(t) = −mHwH,tt,
where mH is the mass of the mallet. This force is usually represented as a ”one-sided” nonlinearity,
i.e. a nonlinear function of which acts when its argument is positive and is zero otherwise. Hence
−mHwH,tt = kα+1H
(
[wH − 〈fr, w〉]+)α
,
where kH a stiffness parameter and
([wH − 〈fr, w〉]+)α =
0 if wH − 〈fr, w〉 ≤ 0;
(wH − 〈fr, w〉)α if wH − 〈fr, w〉 > 0.
Such a contact model has been used by Rhaouti et al. for membranes [75], as well as by Lambourg
et al. for anisotropic plates [55]. The most attractive feature of such a nonlinear model is that the
contact duration depends on the velocity of the striking mallet, leading to brighter sounds when the
velocity is increased [8].
Given the short contact duration of the mallet and the plate (usually of the order of a few millisec-
onds), one may wish to make use of a simplified model for p(t). This form is again in the form of a
raised cosine, this time in the form of
(6.3) p(t) =
pm2
[1 + cos (π(t− t0)/Twid)] if |t− t0| ≤ Twid;
0 if |t− t0| > Twid.
6.1. SIMULATION PARAMETERS 123
0.1 0.103 0.1050
20
40
60
t [s]
pm
[N]
Figure 2. Time-dependent raised cosine to simulate the impact of a timpani mallet
(blue) or a drumstick (black). Twid is selected as 2ms for the mallet and 1ms for the
stick; pm is 20N for the mallet and 70N for the stick.
In this case, the contact distribution is symmetrical about the time t0, and Twid represents half of the
contact duration. When activating the plate using a raised cosine distribution, one is free to select
the contact duration and the amplitude of the distribution. Typically, one may select Twid ≈ 1−3ms
and pm ≈ 10 − 50N for a timpani mallet (softer strike, longer contact); Twid ≈ 0.25 − 1ms and
pm ≈ 40− 100N for a drumstick (harder strike, shorter contact). Fig. 2 depicts two examples of such
distribution.
Note that the examples presented in the next sections will be derived using a raised cosine as input
conditions on one point of the plate; hence, for the remainder of this work
P (x, t) = δ(x− xi)p(t),
where p(t) is as in Eq. (6.3).
Striking is not the only way that one may set a plate into motion. Very interesting effects can be
obtained by bowing the plate at one edge. In this case, the function f(x) is a Dirac delta at an input
point xi of a free edge. The temporal function can be looked for as
(6.4) p(vrel) = −pbφ(vrel),
where vrel = w(xi, t),t− vbow is the relative velocity of the bow and the input point on the plate, and
φ(vrel) is, in general, an antisymmetric function of the velocity presenting a steep slope near zero (see
[8] for details and examples). The attractive feature of such an input is that the variety of sounds
that can be produced is very large, ranging from quasi-harmonic tones to noisy samples. However,
examples of that will not be considered in this work.
In a scheme such the Stormer-Verlet or the energy conserving scheme of sec. 2.7, initial conditions can
be inserted directly on q(0), q(1) (or, equivalently, in a displacement/velocity fashion, by considering
an approximation to the velocity at the initial time, v = δt+q(0) such that q(1) = kv + q(0)). An
interesting effect might be obtained by considering a random excitation, i.e. by inputting random
noise at the initial instant. This gives rise to a broadband excitation which resembles a strike, and
124 6. SOUND SYNTHESIS OF GONGS
that can be used as an alternative to the input strategies described before. Note, however, that not
only such an excitation cannot be justified on a physical basis, but also that it cannot be controlled
(it is a random sequence). Hence, implementing such initial conditions should really be regarded as a
helpful shortcut while prototyping a code; this choice should however be discarded in refined models.
6.1.2. Output. When listening to a vibrating plate, the ears are sensitive to changes in the air
pressure due to the waves produced on the surface of the plate and transmitted in the surrounding
medium. The problem may be further compounded by the presence of walls and refractive objects
in a closed space. Thus, a complete model of a sounding plate is achievable only when the sur-
rounding environment is somehow taken under consideration in the simulations. Needless to say, for
such simulations the computational requirements in terms of memory and time are enormous, even
for small-sized rooms. Recently, the use of graphic cards and parallel computing has allowed the
simulation of large scale environments and complex geometries; see, for instance, the work by the
NESS project [9, 93, 10, 85], and the work by Chabassier on the piano [13, 14].
As a first approximation, however, one may extract the output directly on the surface of the plate, by
recording the displacement of a particular point or by moving the output point around the surface of
the plate (creating an interesting ”phaser” effect). For the present work, the output is always taken at
a point of the plate. Once the displacement waveform is extracted, it is possible to obtain the velocity
and acceleration of the point by considering for instance the forward time derivative approximation
in Eq. (2.70). These are high-pass filters that render the sound brighter, should one want to get rid
of the low-frequency component of the displacement waveform. An example are presented in Fig. 3.
In the figure, the time series of the displacement, velocity and acceleration are shown (normalised to
have a maximum amplitude of 1, so to be played as audio samples). It is evident that the velocity and
acceleration tend to privilege the high-frequency range of the output, and thus they are preferable
over the displacement when trying to reproduce a bright, shimmering sound such that of a gong.
This said, the low-frequency part of the spectrum is also important because it gives an idea of the
dimensions of the gong that is being simulated. Hence, in many cases the velocity time series should
be preferred over the acceleration. Notice, however, that the choice cannot be made in terms of pure
physical arguments, and choosing amongst displacement, velocity and acceleration really becomes a
question of taste.
6.1.3. Damping. Another factor that influences the perception of sound is damping. A distinct
advantage of the modal approach is that one is able to tune the damping coefficients at will with
practically no extra effort. This is indeed the most attractive feature of the modal approach applied to
sound synthesis of gongs. A question remains to what damping law one should use for this problem.
The question of damping in plates is an open one, and it has to do with the fact that the perceived
”loss” is the contribution of a number of independent factors and underlying physical mechanisms.
Generally speaking, one may try to categorise loss in terms of
6.1. SIMULATION PARAMETERS 125
0 0.2 0.4 0.6 0.8 1−1
−0.5
0
0.5
1
t [s]
w(x
o,t)
0 0.2 0.4 0.6 0.8 1−1
−0.5
0
0.5
1
t [s]
˙ w(x
o,t)
0 0.2 0.4 0.6 0.8 1−1
−0.5
0
0.5
1
t [s]
¨ w(x
o,t)
(a) (b) (c)
0 0.2 0.4 0.6 0.8 10
500
1,000
1,500
t [s]
f[H
z]
0 0.2 0.4 0.6 0.8 10
500
1,000
1,500
t [s]
f[H
z]
0 0.2 0.4 0.6 0.8 10
500
1,000
1,500
t [s]
f[H
z]
(d) (e) (f)
Figure 3. Time domain simulation of a steel plate of thickness 0.001 and dimensions
0.4 × 0.6m2. The plate presents NΦ = 50 transverse modes and NΨ = 12 in-plane
modes. The plate is activated using raised cosine of amplitude pm = 100N and
Twid = 0.001s, applied at the point [0.3Lx 0.3Ly]. The output is recorded at
[0.7Lx 0.3Ly]. (a): time series of normalised displacement at output point; (b):
time series of normalised velocity at output point; (c): time series of normalised
acceleration at output point; (d): displacement spectrogram at output point; (e):
velocity spectrogram at output point; (f): acceleration spectrogram at output point.
• thermoelastic effects: the vibrational energy is in this case transformed into heat generated
by the friction between the molecules when the plate is deformed;
• visoelastic effects : the vibrational energy is lost to viscosity;
• radiation damping : the continuity equation between the air and the surface of the plate
allows to define a ”coincidence frequency” after which the flexural waves developing on the
plate are passed into the surrounding medium (air). This happens when the speed of the
flexural waves corresponds to the speed of the sound in air, denoted by c. The coincidence
126 6. SOUND SYNTHESIS OF GONGS
frequency can be given as
(6.5) fr =c2
2π
√
12ρ(1− ν2)
Eh2;
• edge damping : losses due to clamps and supports of the vibrating plate.
Lambourg [54] tried to quantify experimentally the modal damping factors. He obtained results
analogous to those presented in Fig. 4. In Fig. 4(a), one can appreciate that, after the coincidence
frequency, the damping ratios increase sensibly. This means that the radiation damping is prominent
in the frequency range past that frequency. On the other hand, at smaller frequencies the viscoelas-
tic and thermoelastic effects dominate, and the modal damping ratios present variations around a
supposedly mean value (Fig. 4(b)).
Arcas [5] points out that, for a thin plate with simply-supported boundary conditions, the contribution
of the thermoelastic effect can be given as
(6.6) γth(f) =4π2f2RC
2(4π2f2h2 + C/h2),
where R, C, are constants depending on the thermal properties of the material (temperature, con-
ductivity, ...). For steel, C ≈ 1.8 · 102 [rad/m2/s] and R ≈ 9.7 · 103 (nondimensional constant).
In view of an application to wave turbulence, Humbert experimented with a thin steel plate (h ≈
0.5mm) for which the coincidence frequency can be estimated to be fr ≈20kHz [47]. In the audi-
ble range for such a plate, then, the viscoelastic and thermoelastic effects dominate, and Humbert
measures a power-law dependence for the loss mechanism, as in
(6.7) γh(f) = 0.05f0.6 [s−1].
An estimate of the loss mechanism in plates is beyond the scope of the present Thesis. However,
the modal code was developed precisely because of the lack of a numerical models able to simulate a
general damping law. In a Finite Difference environment, even a (seemingly) simple law such as (6.7)
becomes a numerical challenge, but in the modal code, one just needs to set the damping coefficients
in Eq. (2.57) as
(6.8) χs(fs) =0.05
4π(fs)
−0.4.
In this way, the discrete set of points given by 2χsωs will belong to the continuous curve γh(f) of
Eq. (6.7), see Fig. 6. Hence, very complicated damping laws can be fitted in the modal code, which
is not possible in a Finite Difference scheme.
In his book [8], Bilbao proposes two damping laws for his finite difference scheme; these laws are
summarised in Eq. (2.58) and they are global laws proportional to, respectively, the velocity field
and the Laplacian of the velocity field. The coefficients σ0, σ1 can be chosen according to the plate
reverberation time, which is is the typical time after which the Sound Pressure Level drops by 60dB
(see [5]).
6.1. SIMULATION PARAMETERS 127
(a)
(b)
Figure 4. Damping ratios measured by Lambourg for a plate of dimensions Lx ×
Ly × h = 0.23m × 0.2205m × 2mm vibrating in a linear regime. Figures are taken