0 Up votes0 Down votes

88 views12 pagesOct 05, 2010

© Attribution Non-Commercial (BY-NC)

PDF, TXT or read online from Scribd

Attribution Non-Commercial (BY-NC)

88 views

Attribution Non-Commercial (BY-NC)

- Solid 65 ANSYS
- 1Mechanics of Deformable Bodies I.pdf
- EJERCICIOS COMBINADOS
- 5 Asme Viii a-24 Clamp Connection
- Shallow Water Hydrodynamics
- Well Performance-3.pdf
- Hardware Mechanical Handbook
- Course Handbook Mech 1054 2011-2012
- IESL Part II Mechanical
- 39-0433
- Engenering Agriculture Perry
- User's Guide __ 2
- Stress State
- Dynamic of Bow Rotor
- LECTURE 2
- Simulation Instructor WB 2011 ENG
- JL-67-April Design of Anchorage Zone Reinforcement in Prestressed Concrete Beams
- SOM_TYS Exp_924.pdf
- Intro
- Cooling Tower 1

You are on page 1of 12

www.elsevier.com/locate/compstruc

Carla Antoci *, Mario Gallati, Stefano Sibilla

Dipartimento di Ingegneria Idraulica e Ambientale, Università degli Studi di Pavia, via Ferrata, 1, 27100 Pavia, Italy

Available online 15 February 2007

Abstract

A Lagrangian model for the numerical simulation of ﬂuid–structure interaction problems is proposed in the present paper. In the

method both ﬂuid and solid phases are described by smoothing particle hydrodynamics: ﬂuid dynamics is studied in the inviscid approx-

imation, while solid dynamics is simulated through an incremental hypoelastic relation. The interface condition between ﬂuid and solid is

enforced by a suitable term, obtained by an approximate SPH evaluation of a surface integral of ﬂuid pressure.

The method is validated by comparing numerical results with laboratory experiments where an elastic plate is deformed under the

eﬀect of a rapidly varying ﬂuid ﬂow.

2007 Elsevier Ltd. All rights reserved.

Lagrangian formulations, whereas ﬂuids are often

In many engineering applications, the forces exerted by described by Eulerian formulations. The coupling of the

a ﬂuid ﬂow on the conﬁning solid boundaries do not mod- two media is usually obtained by an Arbitrary-Lagrang-

ify signiﬁcantly the geometry of the boundaries. In this ian–Eulerian (ALE) formulation for the ﬂuid. Signiﬁcant

cases, the ﬂuid ﬂow can be studied as occurring within rigid contributions [1–3] have been proposed in the simulation

boundaries, and the forces applied on the solid boundaries of FSI problems in this context. Rugonyi and Bathe [1] per-

can be obtained after the characteristics of the ﬂuid motion form a simpliﬁed stability analysis of the interface equa-

have been determined. tions and study the long-term dynamic stability of FSI

On the other hand, whenever the characteristic times of systems by use of Lyapunov characteristic exponents. They

the motion of the ﬂuid ﬂow and of the solid boundaries are also show the solution of some FSI problems, as the

comparable, it is necessary to couple the dynamics of the dynamics of spring-loaded valves in fuel pumps, that indi-

two media. These ﬂuid–structure interaction (FSI) prob- cate the actual possibility to simulate complex coupled phe-

lems can be solved by employing either a simultaneous nomena. Recent developments in the simulation of viscous

(or direct) solution or a partitioned (or iterative) solution. incompressible and compressible ﬂuid ﬂows with structural

A description of the two procedures can be found in [1], interactions are discussed in [2]. Le Tallec and Mouro [3]

together with the explanation of their main advantages simulate the dynamics of an hydroelastic shock absorber

and drawbacks. The simultaneous technique is particularly adopting an ALE formulation for the ﬂuid equations.

convenient when the interaction between the structure and An alternative approach to the numerical simulation of

the ﬂuid is very strong (and the displacements of the struc- FSI problems consists in the description of both the ﬂuid

and the structure motion by a Lagrangian formulation.

This can be especially eﬀective when studying problems

*

Corresponding author. Tel.: +39 0382 985321; fax: +39 0382 985589. characterized by large displacements of the ﬂuid–structure

E-mail address: carla.antoci@unipv.it (C. Antoci). interface and by a rapidly moving ﬂuid free-surface. An

0045-7949/$ - see front matter 2007 Elsevier Ltd. All rights reserved.

doi:10.1016/j.compstruc.2007.01.002

880 C. Antoci et al. / Computers and Structures 85 (2007) 879–890

example of these problems is the FSI inside safety valves tensor and the notation implies summation over repeated

for pressure reduction, where an elastic plate deforms indices.

owing to water pressure, allowing part of the ﬂuid to ﬂow The stress tensor can be decomposed into its isotropic

out at atmospheric conditions, thus causing a pressure and deviatoric parts:

relief in the connected pipe. In this kind of problems, the rij ¼ pdij þ S ij ; ð3Þ

use of Lagrangian techniques for both the solid and the

ﬂuid part of the problem appears promising, as it permits where p ¼ rkk =3 is pressure, Sij is the deviatoric stress ten-

to easily follow in time the motion of the ﬂuid–solid inter- sor and dij is the Kronecker tensor.

face and to simulate the free-surface of the ﬂuid without Pressure can be formally deﬁned in the same way for

any speciﬁc treatment. In particular, encouraging results both ﬂuid and solid by the following linearized equation

have been recently obtained by the smoothed particle of state, which holds for small variations of density:

hydrodynamics (SPH) technique (see [4] for a recent review p ¼ c20 ðq q0 Þ; ð4Þ

of the method), which allows to obtain numerical solutions qﬃﬃﬃﬃ qﬃﬃﬃﬃ

of the continuum equations by deﬁning the variables at a where c0 ¼ qe0 for the ﬂuid and c0 ¼ qK0 for the solid,

set of suitable moving points, reconstructing the continu- being e the compressibility modulus of the ﬂuid and K

ous ﬁeld by means of interpolation functions centred on the bulk modulus of the solid. Eq. (1) is strictly valid only

each moving point. for compressible ﬂows, while for incompressible ﬂows it re-

The SPH technique was ﬁrst developed in astrophysics duces to the divergence-free condition for the velocity ﬁeld.

by Lucy [5] and by Gingold and Monaghan [6]. It was then However, in order to avoid the complexity of the implicit

successfully applied to the study of various ﬂuid dynamics computation of pressure in a meshless method, the incom-

problems, such as free-surface incompressible ﬂows [7], and pressible ﬂuid can be studied as weakly compressible, thus

viscous ﬂows [8,9]. Since the early 1990s, SPH was applied retaining the validity both of (1) and of the equation of

also to the simulation of elasticity and fragmentation in state (4). However, since the stability of an explicit numer-

solids: in particular, Libersky et al. [10] modelled the elastic ical integration of Eqs. (1) and (2) depends on the Courant

response of solid structures by an incremental formulation condition, the maximum time step is inversely proportional

of Hooke’s law. to the sound speed c0. It is therefore often necessary to as-

SPH has been also used to simulate the interaction sign to the compressibility modulus a value which is lower

between diﬀerent ﬂuids [11,12], diﬀerent solids [13] and than the real one, in order to limit the computational time.

between ﬂuids and structures [14] in presence of explosions. This leads to errors that can be reduced if a proper value is

In some commercial codes, an SPH description of the ﬂuid assigned to e. In particular Monaghan [7] suggests that, in

motion is coupled to a ﬁnite element formulation for the order to limit density ﬂuctuations to 1%, the Mach num-

solid dynamics, in order to simulate FSI problems. ber, i.e. the ratio between the local ﬂow velocity and c0,

The present paper discusses a FSI model where both the must be everywhere lower than 0.1. Many applications of

ﬂuid and the solid parts are modelled by SPH. Aim of the weakly compressible SPH models (see, for instance

model is the analysis of FSI problems where large elastic [9,11,15,16]) conﬁrm that incompressible ﬂows can be sim-

displacements of the solid occur, while rapidly moving ulated with good precision in this way.

free-surfaces characterize the ﬂuid motion. If the dynamics of the ﬂuid ﬂow is dominated by inertial

The reliability of the numerical results yielded by the forces, viscosity eﬀects can be safely neglected, and Sij = 0

proposed SPH FSI model is checked against laboratory can be assumed for ﬂuids. For solids, the linear elastic rela-

data obtained during a simple 2D interaction experiment. tion between stress and deformation tensors can be derived

in time in order to obtain an evolution equation for Sij. The

2. Numerical model use of the corotational, or Jaumann, time derivative guar-

antees that the formulation is independent from superposed

2.1. Equations of motion rigid rotations, resulting in the incremental formulation of

Hooke’s law corrected by the Jaumann rate:

The motion of a continuum subjected to the action of

DS ij 1

gravity, in isothermal conditions, is described by the conti- ¼ 2l Dij dij Dij þ S ik Xjk þ Xik S kj ; ð5Þ

nuity equation Dt 3

Dq ovi where

þq ¼ 0; ð1Þ

Dt oxi 1 ovi ovj

Dij ¼ þ ð6Þ

and by the momentum equation 2 oxj oxi

Dvi orij is the rate of deformation tensor,

q ¼ qgi þ ; ð2Þ

Dt oxj 1 ovi ovj

Xij ¼ ð7Þ

2 oxj oxi

where t is time, q is density, vi is the velocity vector, xi is the

position vector, gi is the gravity vector, rij is the stress is the spin tensor and l is the shear modulus.

C. Antoci et al. / Computers and Structures 85 (2007) 879–890 881

8 2 3

Eq. (5) belongs to the class of hypoelastic constitutive > 3 j~xj 3 j~

xj j~

xj

>

relations, which relate the objective measure of stress rate >1 2 h

>

>

þ4

h

06

h

6 1;

to the rate of deformation. For this kind of relations a r< 3

W ð~

x; hÞ ¼ 2 j~

xj j~

xj ð9Þ

stored energy function can not be deﬁned: therefore, in h >>

>

1

2 16 6 2;

the case of large deformations, energy is not conserved, >4

> h h

:

as happens instead when hyperelastic laws are considered 0 elsewhere;

[17]. On the other hand, the relation (5) is rate-indepen-

dent, incrementally linear and reversible: therefore, when where r is a normalization constant equal to 10/7p, is used

small increments about a state of ﬁnite deformation are in the present work.

considered, the increments in stress are proportional to Neglecting viscous stresses, the SPH approximation of

those in strain and are recovered after unloading [18]. This the momentum equation (2) for the generic ﬂuid particle

means that, if (5) is integrated in time with a conveniently a reduces to

small time step, it can be adopted as a constitutive model Dvia X p pb

oW ab

a

when small ﬁnite deformations are considered. It can be ¼ mb 2 þ 2 dij þ gi ; ð10Þ

Dt b

q a q b oa x j

observed that the integration in time of Eq. (5) from an ini-

tial equilibrium conﬁguration preserves the proportionality where the symmetric formulation of the pressure gradient

between stress and deformation, which characterizes the term guarantees that the action–reaction principle between

elastic response. particles a and b is satisﬁed.

The use of the constitutive equation (5) for the solid, The SPH approximation of the momentum equation (2)

which has been proposed in an SPH context by [10,19], for solids becomes

has the advantage to allow for a common description of

both the ﬂuid and the solid dynamics in terms of pressure Dvia X p p S ij S ij

¼ mb a2 þ b2 dij þ 2a þ 2b

and velocity. Dt b

q a q b q a qb

oW ab

þ Pab dij þ Rijab f q þ gi ; ð11Þ

oa x j

2.2. SPH formulation

where the last two terms between brackets have been intro-

In the SPH formulation, the continuum is divided duced in order to solve numerical problems connected with

into pseudo-particles of constant mass m. This mass is the meshless nature of the SPH method. The ﬁrst one is an

distributed around the centre of mass of each particle, artiﬁcial viscosity term which has been proposed by [21] in

according to a density distribution deﬁned by a suitable order to smooth out the velocity oscillations which can

kernel function [4], which has non-zero values only on arise owing to non-uniform particle distribution in space

a circle of radius 2h centred on the particle itself. The when particles get too close to each other:

length scale h, deﬁned as ‘‘smoothing length’’, characterizes 8

the dimension of the domain of deﬁnition of the kernel < acab lab ð~ va ~

vb Þ ð~

xa ~

xb Þ < 0;

function, and hence the discretization of the numerical Pab ¼ qab ð12Þ

:

scheme. 0 ð~

va ~

vb Þ ð~

xa ~

xb Þ > 0;

According to the Lagrangian approach, each particle is

followed along its trajectory. The density and the velocity where cab ¼ 12 ðca þ cb Þ, qab ¼ 12 ðqa þ qb Þ and lab ¼

hð~

va ~

vb Þð~

xa ~

xb Þ

of each particle are updated by explicit integration of the xb j2 þð0:1hÞ2

xa ~

j~

.

continuity (1) and the momentum equations (2). Particle This term introduces numerical dissipation in the model,

trajectories are then computed by integrating in time the which must be kept to a minimum to reduce unwanted

material derivative of velocity obtained from Eq. (2). oscillations in the velocities without aﬀecting the solution

According to the continuity equation (1), the material in a sensible way. The sensitivity to parameter a of the

derivative for the density of the generic particle a, ﬂuid numerical solution of a test elastic problem is discussed

or solid, can be obtained by SPH interpolation as in Section 2.4.

The second term plays the role of an artiﬁcial stress and

Dqa X was proposed by [19] to eliminate the eﬀects of the so-called

¼ mb ð~

va ~

vb Þ ra W ab ; ð8Þ

Dt b ‘‘tensile instability’’ [14,22,23]. This instability, which is

strictly related to the interpolation technique of the stan-

where the summation is extended to all the particles b dard SPH method [24], is especially noticeable when simu-

around a and the use of the velocity diﬀerence ~ va ~vb in lating tension states in solids: particles tend then to clump

the SPH approximation of the divergence guarantees that, together, causing non-physical fractures in the material.

for a constant velocity ﬁeld, the material derivative of den- The ‘‘artiﬁcial stress’’ term in Eq. (11) becomes eﬀective

sity is zero. In (8) ra W ab is the gradient of the kernel func- when particle i is in tension and acts as a repulsive force

tion W ð~ x; hÞ evaluated at ~ x ¼~xa ~xb . The cubic spline avoiding particle clumping. In the form proposed by [19]

kernel proposed by [20] it is evaluated as

882 C. Antoci et al. / Computers and Structures 85 (2007) 879–890

q

W ðj~

xa ~xb jÞ velocity oscillations, but introduces some energy dissipa-

Rijab f q ¼ Rija þ Rijb ; ð13Þ

W ðdÞ tion in the numerical solution. However, its use avoids

the inclusion of the artiﬁcial viscosity term (12) in (10).

where q is a parameter, d is the mean initial distance be- A sensitivity analysis of the numerical solution of an

tween particles and Rij is obtained as follows. For each par- unsteady hydraulic problem to the smoothing coeﬃcient

ticle the stress tensor rij is diagonalised. Then an artiﬁcial h is discussed in Section 2.4.

stress term is evaluated for any of the diagonal components An Euler explicit scheme of integration in time is

i which are positive:

r adopted for both the ﬂuid and the solid structure. The time

r

ia integration scheme is staggered, i.e. the solution of momen-

Ria ¼ e ; ð14Þ

q2a tum equation is shifted by half time step with respect to

other variables. The choice of the time step depends on

where e is a parameter. the Courant stability condition. Since there are two media

The artiﬁcial stress in the original coordinates system Rij in the domain, the smallest of the time steps required by

is then calculated by rotating the coordinates back. each medium is chosen.

Gray et al. derive optimal values from the dispersion The adoption of an implicit integration, although possi-

equations, suggesting to use e = 0.3 and q = 4. An analysis ble in principle, would require the inversion of a large

of the sensitivity of the numerical solution to the parameter sparse matrix, thus being computationally intensive.

e is also discussed in Section 2.4.

The deviatoric stress Sij in (11) is calculated by integrat-

2.4. Sensitivity to the corrective terms

ing in time, by an implicit scheme, the rate of deviatoric

stress from the incremental hypoelastic relation (5). The

The sensitivity to the artiﬁcial viscosity and artiﬁcial

SPH estimate of velocity gradient, needed to compute the

stress terms in Eq. (11) has been checked by simulating

spin and the rate of deformation, can be obtained by ﬁrst

the free oscillations of an elastic plate [19,26] having one

introducing a Taylor expansion of the velocity of particle

end clamped and the other one free. The plate is initially

a truncated to the ﬁrst order, which leads to the following

horizontal (Fig. 1) and the initial velocity distribution is

system in the unknown components of the velocity gradient

assigned according to the analytical expression of the free

tensor:

oscillations of a thin plate:

X mb oW ab X mb oW ab

vib ¼ via f ðxÞ

qb oa xj qb oa xj vy ðxÞ ¼ vL0 c0 ;

b b

X f ðLÞ

ovi mb oW ab

þ ðxkb xka Þ: ð15Þ where

oxk a b qb oa xj

f ðxÞ ¼ ðcosðkLÞ þ coshðkLÞÞðcoshðkxÞ cosðkxÞÞ

The solution of system (15), although needing to be solved

at each velocity gradient evaluation, is preferred to a direct þ ðsinðkLÞ sinhðkLÞÞðsinhðkxÞ sinðkxÞÞ;

SPH evaluation of the velocity gradient because it accounts

while vL0 ¼ 0:01 determines the initial velocity of the free

for a non-uniform particle distribution, especially close to

end. Since we consider the fundamental mode, kL ¼ 1:875.

the boundaries. The plate oscillates around the initial position, with ampli-

tude and period depending on elastic and geometric

2.3. Velocity correction and time integration scheme properties.

The results discussed in the following are obtained

Particle velocities ~

v obtained by time integration of the for:L = 0.2 m, H = 0.02 m, q = 1000 kg/m3, K ¼ 3:25

momentum equation are corrected in order to smooth 106 N=m2 , l = 715000 N/m2.

out unwanted numerical peaks. The correction is obtained The computed non-dimensional amplitude and period

by of the ﬁrst oscillation (A/L = 0.124 and Tc0 =L ¼ 81:5) are

P mb

bq ð~

vb ~ va ÞW ab

v~a ¼ ~

ab

~ va þ ð1 hÞ P mb ; ð16Þ

bq

ab

W ab

For solid particles, these corrected values ~ v~ are used,

according to the XSPH scheme [25], in (8) and (15) and

in the computation of the updated position of particles,

while the uncorrected velocities ~ v are used for time integra-

tion of Eq. (11) at the following step. For ﬂuid particles, ~ v~

are used also in the time integration of the momentum Eq.

(10), according to the scheme proposed by [15]: this scheme

proves to be more eﬀective than XSPH in smoothing out Fig. 1. Scheme of the plate.

C. Antoci et al. / Computers and Structures 85 (2007) 879–890 883

(AG =L ﬃ 0:125 and T G c0 =L ﬃ 82), although they both diﬀer

to some extent from the analytical solution (Aa =L ¼ 0:115

and T a c0 =L ¼ 72:39).

The main eﬀect that has been veriﬁed while checking the

sensitivity to parameters e and a consists in a damping of

the free oscillations of the plate. Fig. 2, where the ratio

between the amplitude of the oscillations after ﬁve periods

(A5) and the amplitude of the ﬁrst oscillation (A1) is plot-

ted, shows that this damping occurs when e < 0.15.

Also the introduction of the artiﬁcial viscosity term con-

tributes to elastic energy conservation (Fig. 3): actually, for

a = 0 the computation is unstable, while damping of the

Fig. 4. Dam-break: proﬁles for diﬀerent values of h at t = 0.36 s (in

free oscillations reduces only for a > 1. This is probably simulations A, B and C the following values have been assigned to h

due to the fact that the local oscillations of velocity cause respectively: h = 0.96, h = 0.92 and h = 0.84).

disorder in the spatial distribution of solid particles, thus

compromising interpolated estimates. When this disorder

occurs, a decrease in time of the amplitude of the oscilla-

tions is observable. Artiﬁcial viscosity avoids this problem,

thus leading to a satisfactory conservation of elastic energy.

The amplitudes of the ﬁrst oscillation, as well as the oscil-

lation periods, do not vary signiﬁcantly with a. On the

other hand the higher is e, the larger the amplitude A1 of

the ﬁrst oscillation (Fig. 2), since the artiﬁcial stress reduces

the eﬀective stiﬀness of the plate, reducing the elastic

Fig. 5. Comparison between a snapshot by [15] at t = 0.36 s and an image

from a simulation with h = 0.92.

choice of e, in order not to aﬀect the solution excessively.

The sensitivity of ﬂuid ﬂow simulations to the velocity

correction scheme (16) has been checked on the classical

dam-break problem: a mass of ﬂuid at rest (here having

an initial height of 0.1 m) moves after the instantaneous

removal of the conﬁning wall at t = 0. The dissipative eﬀect

of the velocity correction scheme can be observed in Fig. 4,

where free-surface proﬁles for diﬀerent values of h at

t = 0.36 s are compared with the experimental data

Fig. 2. Inﬂuence of parameter e on the amplitude of the free oscillations of obtained by [15]. In particular, high values of h lead to

the plate. wave fronts speeds and to free-surface proﬁles closer to

the experimental ones. The best value of h must therefore

be chosen in order to assure stability to the computation

without introducing excessive numerical dissipation.

In Fig. 5, a snapshot of the dam-break experiment at

t = 0.36 s is compared with the corresponding particle dis-

tribution from the SPH simulation with h = 0.92.

ﬂuid. All the particles located farther than 2h from the

interface interact, of course, only with particles of the same

species (particle a2 in Fig. 6). On the other hand, when SPH

interpolation is performed on particles closer to the inter-

Fig. 3. Inﬂuence of parameter a on the amplitude of the free oscillations face (like particle a1 in Fig. 6), particles of both media

of the plate. are involved. The simplest approach in this case consists

884 C. Antoci et al. / Computers and Structures 85 (2007) 879–890

particles.

Fig. 6. Fluid (black) and solid (white) particles near the interface.

rijs nis njs ¼ pf : ð20Þ

The application of (19) and (20) requires the precise com-

in extending the summations in (10) and (11) to all the par-

putation of the position of the interface surface and of its

ticles b regardless of their nature. Although interpenetra-

normal direction.

tion of ﬂuid and solid particles can occur in principle, it

is prevented by the XSPH correction of velocity, as

3.1. Deﬁnition of the interface and its normal

observed by [13] in the case of contact between solids.

This method is equivalent to the introduction of cou-

A way to deﬁne the interface and its normal can be

pling conditions on the interface between a solid and a vis-

found in SPH estimates of constant functions and of their

cous ﬂuid, i.e.:

gradient, as proposed by [14] to deﬁne complex solid

vif ¼ vis ð17Þ boundaries. However, when no fragmentation takes place,

as in the problems here considered, solid particles maintain

and

the same regular pattern in time, making the identiﬁcation

rijs njs ¼ rijf njf ; ð18Þ of the interface easier. The interface is therefore deﬁned

here as a line distant d/2 from the row of solid particles

where the subscripts s and f identify the solid and the ﬂuid, closest to the ﬂuid (Fig. 8).

respectively, while the sign convention is the same as in For every solid particle closer than 2h to the interface

Fig. 7. (‘‘boundary’’ particles) the tangent unit vector is deﬁned by

This simple approach, which is the correct one for vis-

cous ﬂuids as it imposes automatically a no-slip condition xaþ1 xa1 y aþ1 y a1

^ta ¼ ðtax ; tay Þ ¼ ; ; ð21Þ

on the contact surface, is not valid when the eﬀect of vis- xaþ1 ~

j~ xaþ1 ~

xa1 j j~ xa1 j

cosity is neglected and an inviscid ﬂow formulation is where the subscripts a + 1 and a 1 identify the particles

adopted. The following kinematic and dynamic interface which immediately precede and follow particle a along

conditions, which imply the continuity of the normal com- the row parallel to the boundary.

ponent of velocity and of normal stress, must be therefore The normal unit vector is computed accordingly

applied in this case:

^na ¼ ðtay ; tax Þ: ð22Þ

The position of the interface point closest to each particle a

is also deﬁned

1

xinta ¼ ~

~ xa þ r þ d^na ; ð23Þ

2

where r is the number of rows between particle a and the

boundary.

Fig. 7. Normal convention. the action–reaction principle, is guaranteed by imposing

C. Antoci et al. / Computers and Structures 85 (2007) 879–890 885

that the force exerted by the ﬂuid on the solid has the same the two media separate (and C ceases to be an interface

modulus as the force exerted by the solid on the ﬂuid, but surface between solid and ﬂuid), introducing a smooth

opposite direction. To obtain the SPH evaluation of the decrease in the interface pressure value.

force applied by one phase on the other, one can ﬁrst con- The term ~F f!s;a =qa is therefore added to (11), which, for

sider that the ‘‘smoothed’’ estimate of the gradient of pres- boundary particles, is modiﬁed as follows:

sure in a domain X bounded by the surface CX is Dvia X rija rijb q oW ab

Z ¼ mb þ 2 þ Pab dij þ Rijab f

Dt q2a qb oa x j

hrpð~xÞi ¼ pð~ 0

x ÞW ð~

x ~

x0; hÞjCX þ pð~ x0 Þr~x W ð~ x0 ; hÞdX0 ;

x ~ b2Xs

X F if!s;a

ð24Þ þ gi þ : ð28Þ

qa

which becomes, when discretized by SPH: Since, in general, the position of ﬂuid and solid particles

Z X mb across the interface is not symmetric, the action–reaction

hrpð~

xÞi ¼ x0 ÞW ðj~

pð~ x0 j; hÞdC0 þ

x ~ p ra W ab : principle must be imposed through a linear interpolation

CX b

qb b procedure: an interpolated value ~ F f!s;a is evaluated at the

ð25Þ position a, symmetric to a* across the interface (Fig. 9)

and the reaction term ~ F s!f;a is calculated as

The surface term on the right-hand side of (24) vanishes if

~

F s!f;a ¼ ~

F f!s;a : ð29Þ

X coincides with the domain of deﬁnition of the kernel

function, since W is zero for j~ x ~x0 j ¼ 2h. On the other Thus the continuity of normal stresses through the inter-

hand, when computing the pressure gradient term for the face (20) is assured.

solid ‘‘boundary’’ particle a (Fig. 9), the summation in The momentum equation (10) for ﬂuid ‘‘boundary’’ par-

(25) is extended only to the solid particles b in Xs and ticles is modiﬁed by adding the force per unit mass

CX ¼ CXs . In this case, the surface term does not vanish ~

F s!f;a =qa :

any more and it accounts for the force per unit volume

~ Dvia X p p oW a b Fi

F f!s;a exerted by the ﬂuid on particle a. If the pressure ¼ mb a2 þ b2 dij þ gi þ s!f;a : ð30Þ

on the interface is approximated by its SPH interpolation Dt b2X

qa qb oa xj qa

f

limited to the ﬂuid particles in Xf:

X mb

pinta ¼ 2 p W ðj~

xinta ~

xb j; hÞ; ð26Þ 3.3. Kinematic interface condition

b2X

qb b

f

the surface term in (25) can be expressed as obtained as

Z P mb

~

F f!s;a ¼ pinta W ðj~ x0 j; hÞdC0

xinta ~ ð27Þ b2Xs qb vib W ðj~

xinta ~xb j; hÞ

C viinta ¼ P mb : ð31Þ

b2Xs qb W ðj~xinta ~ xb j; hÞ

where C is the portion of interface surface included in the

domain of deﬁnition of the kernel (Fig. 9). The interface surface represents a moving wall boundary

It should be noted that (26) contains in fact an approx- for the ﬂuid ﬂow. This moving boundary condition is en-

imate normalization for the interface pressure pinta : actu- forced by a modiﬁed version of the ‘‘ghost particle tech-

ally, the summation nique’’ [16], where the solid particle positions are used as

P on mthe right-hand side should have

a geometric support to assign a ﬂuid velocity distribution

been divided by b

b2Xf qb W ðj~

xinta ~

xb j; hÞ, which is an

SPH approximation for 0.5 when the liquid and solid able to reproduce, by SPH interpolation, the required nor-

phases are in contact. However, it has been found that mal component of ~ vinta on the interface.

the approximate normalization (26) works better when

4. Deformation of an elastic plate subjected to

time-dependent water pressure

by comparison of the numerical results with data measured

during suitable laboratory experiments. In these experi-

ments an elastic gate, clamped at one end and free at the

other one, interacts with a mass of water initially conﬁned

in a free-surface tank behind the gate.

Fig. 9. Dynamic condition for solid boundary particle a. vertical walls, thus creating a tank, as represented in

886 C. Antoci et al. / Computers and Structures 85 (2007) 879–890

of the plate and the water levels in the tank have been mea-

sured by digital image processing.

of magnitude of the plate height L and of the initial water

level H, the analysis of the experiments shows that the

water transient ﬂow and the plate deformation can be stud-

ied as a two-dimensional phenomenon. In particular, it

must be noted that no bending of the rubber plate was

observed along the direction normal to the lateral walls.

A 2D simulation in the plane (x, y) shown in Fig. 11 has

therefore been realized.

Water is considered as a perfect ﬂuid with density

qf = 1000 kg/m3. As explained in Section 2.1, the incom-

Fig. 10. Scheme of the tank and of the gate: frontal view, lateral and plan. pressible ﬂuid is analysed as weakly compressible to retain

the explicit solution of the continuity equation according to

(1) and to the equation of state (4). However, the time step

Fig. 10. One of the walls consists in an upper rigid part and yielded by the Courant condition [7] when adopting the

in a lower deformable plate made of rubber. The rubber real value of the compressibility module (approximately

plate is free at its lower end, thus representing an elastic 2 · 109 N/m2) is too small to obtain numerical solutions in a

gate closing the tank. The geometric dimensions of the sys- reasonable computing time. Therefore, a lower value of the

tem and the physical characteristics of the elastic gate are compressibility modulus ðe ¼ 2 106 N=m2 Þ is adopted,

reported in Table 1. The Young modulus E has been while the appearance of unphysical compressibility eﬀects

obtained by performing tension tests in the range of defor-

mations which occur in the phenomenon. Since the Young

modulus for rubber depends on deformation, the reported

value is intended as an average value.

Being the rubber plate clamped only along its upper

side, it is free to deform when subjected to the pressure

of the ﬂuid behind it. In order to assure sealing and mini-

mize friction along the lateral plexiglas walls of the tank,

a thin layer of transparent fat has been inserted between

the lateral sides of the plate and the tank, paying care to

distribute it uniformly along the sides of the plate, in order

to minimize its eﬀect on the motion.

While the rubber plate is held ﬁxed by an external rigid Fig. 11. Initial conﬁguration.

support, the tank is ﬁlled with water up to the desired level.

When the water in the tank is in hydrostatic conditions, the

rigid support is suddenly removed, thus allowing the plate

to deform while water ﬂows under it. The experiments were

recorded by a digital video camera, at a frequency of

25 frames per second. The displacements of the free end

Table 1

Dimensions of the system and physical characteristics of the rubber plate

Dimensions

A (m) 0.1

H (m) 0.14

B (m) 0.1

B* (m) 0.098

L (m) 0.079

s (m) 0.005

Rubber

q (kg/m3) 1100

E (MPa) ﬃ10

Fig. 12. Simulation: initial conﬁguration.

C. Antoci et al. / Computers and Structures 85 (2007) 879–890 887

Fig. 13. Frames and images from simulation every 0.04 s from t = 0 s (a) until t = 0.4 s (k).

888 C. Antoci et al. / Computers and Structures 85 (2007) 879–890

is avoided by requiring that the local Mach number is placements, whereas, close to the free end, the plate seems

everywhere lower than 0.1. to move almost as a rigid body.

The rubber plate is discretized by solid particles having The displacements of the plate in the simulation are

density qs ¼ 1100 kg=m3 . Since some uncertainty occurs in slightly larger than those in the experiment. This is possibly

the estimate of an average Young modulus for rubber, sim- due to the fact that the stiﬀness of the calculated plate is

ulations with diﬀerent values of E were run. The results of lower than the real one. However, the diﬀerences here high-

the simulation with E ¼ 1:2 107 N=m2 , which better lighted are consistent with those found in the examples dis-

reproduces the experimental phenomenon, are discussed cussed in Section 2.4.

in the following. In addition, it must be noted that during the experiment

The Poisson coeﬃcient is set equal to 0.4. This value, some leakage of water occurs besides the gate (as it can be

although lower than the theoretical one for incompressible viewed in Fig. 13 at times larger than 0.12 s). Owing to the

rubber, allows us to use larger time steps while respecting leakage, the pressure behind the plate in the experiment

the Courant condition; it has been checked that the inﬂu- might be slightly lower than the one predicted in the simu-

ence of the modiﬁcation of the Poisson coeﬃcient on the lation, leading to a hydrodynamic force on the plate lower

numerical results is negligible. than the simulated one.

The adopted values of Young modulus and Poisson coef- In Fig. 14, the horizontal and vertical displacements

ﬁcient correspond to a bulk modulus K ¼ 2 107 N=m2 computed for the plate are compared with those measured

and to a shear modulus l ¼ 4:27 106 N=m2 . in the digitalized images acquired during the experiments.

According to the results of the sensitivity analysis dis-

cussed in Section 2.4, the artiﬁcial stress parameter e and

the velocity smoothing parameter h have been set respec-

tively equal to 0.3 and 0.92. Since the phenomenon simu-

lated is rapid, a value a = 1 has been preferred for the

artiﬁcial viscosity parameter to avoid excessive numerical

dissipation. Fig. 12 shows the initial particle distribution

in the simulation.

Slip boundary conditions are imposed to the ﬂuid ﬂow

on the rigid walls: these conditions are imposed through

the method of ghost particles [16] on the right tank wall

and on the bottom (continuous lines in Fig. 12) and by

layers of ﬁxed ghost particles in the upper rigid part of

the left tank wall.

Fig. 14. Horizontal and vertical displacements of the free end of the plate.

The clamp condition is imposed to the plate through a

layer of ﬁxed solid particles. This layer is orthogonal to

the end of the plate. Only densities and stresses are updated

in time for these particles, whereas their velocities are set

equal to zero and their positions are ﬁxed in time.

Two diﬀerent reﬁnement levels (d 1 ¼ 8:33 104 m

close to the plate and d2 = 0.002 m far from the plate) have

been adopted in the discretization of the ﬂuid mass,

whereas for the solid is d = d1. The resulting total number

of particles is np ¼ 6012. The time step in the simulation is

dt ¼ 8:34 106 s. At the initial time, the ﬂuid is assumed

to be in hydrostatic conditions, while in the plate stresses

and deformations are equal to zero.

and images of SPH particle positions at corresponding

times are shown in Fig. 13. The dynamics of the phenom-

enon, in terms of deformation of the plate and of evolution

of the ﬂuid free-surface, is well reproduced. It can be

observed that the calculated shape of the plate is similar

to the observed one: the deformation is maximum near

the clamp, with null second derivative of the horizontal dis- Fig. 15. Water level (m) just behind the gate (a) and 5 cm far from it (b).

C. Antoci et al. / Computers and Structures 85 (2007) 879–890 889

Fig. 16. rxx distribution at t = 0.15 s: colour scale ranging from rxx = Fig. 18. rxy distribution at t = 0.15 s: colour scale ranging from rxy =

150 000 N/m2 (white) to rxx = 150 000 N/m2 (black). 200 000 N/m2 (white) to rxy = 200 000 N/m2 (black).

tively correct, showing the internal part of the plate in ten-

sion (dark particles), whereas the external part is

compressed (light particles). A maximum stress value of

0.75 MPa is found for the ryy component near the clamp,

while the other components show lower maximum values.

All the stress components tend to zero when approaching

the free end of the plate.

5. Conclusions

method has been described. In the model, both the ﬂuid

and the solid phases are discretized by SPH particles. Once

the ﬂuid–solid interface has been deﬁned, coupling condi-

tions are imposed to particles close to it. In particular,

the action of the ﬂuid on the solid is computed through

Fig. 17. ryy distribution at t = 0.15 s: colour scale ranging from ryy = the evaluation of an approximated surface integral of ﬂuid

750 000 N/m2 (white) to ryy = 750 000 N/m2 (black). pressure. On the contrary, the action on the ﬂuid is com-

puted by linear spatial interpolation from the previous

one, in order to satisfy the action–reaction principle.

It can be seen that the time evolution of the phenomenon is The SPH model, thanks to its Lagrangian nature, has

well described by the SPH simulation, even if the diﬀer- the advantage of allowing an easy deﬁnition of the ﬂuid–

ences discussed above lead to a 10% overprediction of the solid interface, even in presence of large displacements of

maximum horizontal displacement of the plate. the structure, and does not need any speciﬁc treatment

The evolution of the free-surface is also well reproduced for the free-surface of the ﬂuid. The contact between diﬀer-

by the simulation. Fig. 15 shows the water level history ent media is also automatically established when particles

immediately behind the gate and in the middle of the tank. of diﬀerent media are closer than twice as the smoothing

Consistently with the larger vertical displacement of the length of the kernel function.

gate, and hence with the larger gate opening, the computed In addition no time shift is introduced in the calculation

ﬂow rate is slightly higher than the real one, leading to a of the dynamics of the two media, since the variables of

faster decrease of the water level in the ﬁrst part of the tran- both media are updated simultaneously. Anyway, in prob-

sient. Later than t = 0.075 s, the computed and measured lems where the characteristic times of the dynamics of the

values of the water level evolve in the same way. two media diﬀer too much, the model can be easily modi-

The stresses rxx, ryy, rxy in the elastic plate are plotted ﬁed in order to integrate the equations for the two media

at t = 0.15 s, when the displacements are maximum with diﬀerent time steps.

890 C. Antoci et al. / Computers and Structures 85 (2007) 879–890

The model has been tested by comparing numerical [8] Takeda H, Miyama SM, Sekiya M. Numerical simulation of viscous

results and data collected during a laboratory experiment ﬂow by smoothed particle hydrodynamics. Progr Theor Phys 1994;

92:939–60.

concerning the outﬂow under an elastic sluice gate. This [9] Morris JP, Fox PJ, Zhu Y. Modeling low Reynolds number

experiment has been chosen because it concerns the interac- incompressible ﬂows using SPH. J Comput Phys 1997;136:214–26.

tion between a long and thin elastic structure and a ﬂuid [10] Libersky LD, Petschek AG, Carney TC, Hipp JR, Allahdadi FA.

mass, in presence of large displacements of the structure High strain Lagrangian hydrodynamics. J Comput Phys 1993;109:

and of a free surface ﬂow. 67–75.

[11] Monaghan JJ, Cas RAF, Kos AM, Hallworth M. Gravity currents

The results have shown that realistic predictions, both of descending a ramp in a stratiﬁed tank. J Fluid Mech 1999;379:39–69.

the displacement of the elastic structure subjected to ﬂuid [12] Colagrossi A, Landrini M. Numerical simulation of interfacial ﬂows

pressure and of the resulting ﬂuid ﬂow, can be obtained by smoothed particle hydrodynamics. J Comput Phys 2003;191:

by the SPH model, although some improvement in the 448–75.

treatment of the elastic solid dynamics still needs to be [13] Vignjevic R, De Vuyst T, Campbell J. The use of an homogeneous

repulsive force for contact treatment in SPH, WCCM V, Fifth world

achieved. congress of computational mechanics, Vienna, Austria, July, 7–12,

2002.

Acknowledgements [14] Randles PW, Libersky LD. Smoothed particle hydrodynamics: some

recent improvements and applications. Comput Methods Appl Mech

We are also grateful to Roberto Allieri and Ivano Brivio Eng 1996;139:375–408.

[15] Gallati M, Braschi G. Simulazione Lagrangiana di ﬂussi con

for their valuable help in the execution of the experiments. superﬁcie libera in problemi di idraulica. L’acqua 2000;5:7–18.

We acknowledge Dresser Italia S.r.l. for the ﬁnancial [16] Gallati M, Braschi G. Numerical description of rapidly varied ﬂows

support to the present research. via SPH method. In: IASTED Int Conf, ASM, Creta 2002.

[17] Simo JC. Numerical analysis and simulation of plasticity, Handbook

of numerical analysis, vol. VI, Part. 3. Elsevier Science B.V.; 1998.

References [18] Belytschko T, Liu WK, Moran B. Nonlinear ﬁnite elements for

continua and structures. John Wiley & Sons; 2000.

[1] Rugonyi S, Bathe KJ. On ﬁnite element analysis of ﬂuid ﬂows fully [19] Gray JP, Monaghan JJ, Swift RP. SPH elastic dynamics. Comput

coupled with structural interactions. Comput Model Eng Sci 2001; Methods Appl Mech Eng 2001;190:6641–62.

2:195–212. [20] Monaghan JJ. Smoothed particle hydrodynamics. Annu Rev Astron

[2] Bathe KJ, Zhang H. Finite element developments for general ﬂuid Astrophys 1992;30:543–74.

ﬂows with structural interactions. Int J Numer Methods Eng 2004; [21] Monaghan JJ, Gingold RA. Shock simulation by the particle method

60:213–32. SPH. J Comput Phys 1983;52:374–89.

[3] Le Tallec P, Mouro J. Fluid structure interaction with large structural [22] Swegle JW, Hicks DL, Attaway SW. Smoothed particle hydro-

displacements. Comput Methods Appl Mech Eng 2000:1–29. dynamics stability analysis. J Comput Phys 1995;116:123–34.

[4] Monaghan JJ. Smoothed particle hydrodynamics. Rep Progr Phys [23] Morris JP. Analysis of smoothed particle hydrodynamics with

2005;68:1703–59. applications, Ph.D. thesis, 1996.

[5] Lucy LB. A numerical approach to the testing of the ﬁssion [24] Rabczuk T, Belytschko T, Xiao SP. Stable particle methods based on

hypothesis. Astron J 1977;82:1013–24. Lagrangian kernels. Comput Methods Appl Mech Eng 2004;193:

[6] Gingold RA, Monaghan JJ. Smoothed particle hydrodynamics: 1035–63.

theory and application to non-spherical stars. Mon Not R Astron Soc [25] Monaghan JJ. On the problem of penetration in particle methods.

1977;181:375–89. J Comput Phys 1989;82:1–15.

[7] Monaghan JJ. Simulating free surface ﬂows with SPH. J Comput [26] Landau LD and Lifšits EM. Teoria dell’elasticità. Editori riuniti

Phys 1994;110:399–406. Edizioni Mir, 1979.

- Solid 65 ANSYSUploaded byRory Cristian Cordero Rojo
- 1Mechanics of Deformable Bodies I.pdfUploaded byChristian Galope
- EJERCICIOS COMBINADOSUploaded byandy_flores
- 5 Asme Viii a-24 Clamp ConnectionUploaded bypastyl
- Shallow Water HydrodynamicsUploaded byArpach Pacheco
- Well Performance-3.pdfUploaded byWashoo Hasoo
- Hardware Mechanical HandbookUploaded byapi-3723333
- Course Handbook Mech 1054 2011-2012Uploaded byAhmar Raza
- IESL Part II MechanicalUploaded byHarsha Suranjith Amarasiri
- 39-0433Uploaded byJohn Rong
- Engenering Agriculture PerryUploaded byDaniel López
- User's Guide __ 2Uploaded byadehriya
- Stress StateUploaded byGabriel Caicedo Russy
- Dynamic of Bow RotorUploaded bymh1999
- LECTURE 2Uploaded byMuhammed Sulfeek
- Simulation Instructor WB 2011 ENGUploaded byjorgemariovega4910
- JL-67-April Design of Anchorage Zone Reinforcement in Prestressed Concrete BeamsUploaded byMurali Krishna Reddy Arikatla
- SOM_TYS Exp_924.pdfUploaded byRajanVishwakarma
- IntroUploaded byMelmar Jangerald Sabio
- Cooling Tower 1Uploaded byDebabrata Tantubai
- caso real 5.pdfUploaded byCarlosDanielClavijoGranados
- 03_Stress_03_Internal_Stress.pdfUploaded byArunBelwar
- 4.9Uploaded byAnmol Kumar
- 7Uploaded bySony Jsd
- 5 Third Year First SemesterUploaded byRojan Pradhan
- 8. PT SLABUploaded byPre Sheet
- Modelling Dust Liberation.pdfUploaded bysergio71
- PIP-SS-007-SLUGUploaded byYogesh Pangare
- Kupdf.com Design of Concrete Ring Type Foundation for Storage TankUploaded byFadli Punye
- 47-106-3-PBUploaded byOMUNG GARG 14BME0154

- 25969174 Price and Output Determination Under Monopolistic onUploaded byShoaib Ahmed Khoso
- 4PH0_1P_que_20160119Uploaded bynaamashir
- ggUploaded byAnshuman Pradhan
- Endocrine GlandUploaded byRS
- Assessment and EvaluationUploaded byJohnson O.V
- Busting a Cap without Die-ing* (*With brief diversions in system & hardware REUploaded bypanicopticon
- Admit Card- RmatUploaded bySakshi Chouhan
- 10.1.1.104Uploaded byNipun Cp
- Do Rling 2019Uploaded byErinson Custodio Plasencia
- Related LiteratureUploaded byLeobel Baruelo Ariate
- completed-transcript-5595766Uploaded byapi-250565308
- The Lessons of the Hole ArgumentUploaded byCarlosGerardoSanjuánCiepielewski
- Abiding and Confiding Music sheetUploaded byhenchudi
- LEHH8194.pdfUploaded byLore Ch
- Pro700Uploaded byRamtel
- sta1503_2013_s01_-_assignment02_memo_---tl_202_2013_1_e[1]Uploaded bysal27adam
- BECKHOFF-EtherCAT Beckhoff Course (en)Uploaded byJorge_Andril_5370
- ADNOC Schools Yearbook 2016 - 2017Uploaded byJames Gleeson
- HOW TO INSTALL KDE4 APPLICATIONS ON WINDOWS.pdfUploaded byKurgan Gl
- SyllabusSampleQuestions AshokaUploaded byakanksha450
- Brexit, Trump, racism Bhambra-2017Uploaded byleidiapache
- Compressed+Air+Filters.pdfUploaded byDiana Moise
- Judith Becker - Deep Listeners LectureUploaded byRoberto Cruz
- The Archaeology of the Hellenistic Far EUploaded byeunaosoudesleal7223
- job20.docxUploaded bymainul arefeen
- 1561346816134_EE8511-C&I LAB MANUAL final 24.6.19Uploaded byChe Thiru
- Spooling tubing |Spooling foam|B12-Re-Spooler|ReelpowerwcUploaded byReelpowerWC
- Phosphorous Cycle (Slide Presentation)Uploaded byUmmiey Syahirah
- 3000 Litre Transcube Fuel TankUploaded byyulianus_sr
- Yamaha FZ16 SUploaded byBapi Sarkar

## Much more than documents.

Discover everything Scribd has to offer, including books and audiobooks from major publishers.

Cancel anytime.