You are on page 1of 21

Applied Mathematical Modelling 35 (2011) 3833–3853

Contents lists available at ScienceDirect

Applied Mathematical Modelling


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

A corrected smoothed particle hydrodynamics method for solving transient


viscoelastic fluid flows
Tao Jiang, Jie Ouyang ⇑, Qiang Li, Jinlian Ren, Binxin Yang
Department of Applied Mathematics, Northwestern Polytechnical University, Xi’an 710129, China

a r t i c l e i n f o a b s t r a c t

Article history: In this work, a corrected smoothed particle hydrodynamics (CSPH) method is proposed and
Received 26 May 2010 extended to the numerical simulation of transient viscoelastic fluid flows due to that its
Received in revised form 16 January 2011 approximation accuracy in solving the Navier–Stokes equations is higher than that of the
Accepted 2 February 2011
smoothed particle hydrodynamics (SPH) method, especially near the boundary of the
Available online 5 March 2011
domain. The CSPH approach comes with the idea of combining the SPH approximation
for the interior particles with the modified smoothed particle hydrodynamics (MSPH)
Keywords:
method for the exterior particles, this is because that the later method has higher accuracy
CSPH
Viscoelastic flows
than the SPH method although it also needs more computational cost. In order to show the
Oldroyd-B fluid validity of CSPH method to simulate unsteady viscoelastic flows problems, the planar shear
Pom–Pom model flow problems, including transient Poiseuille, Couette flow and transient combined Poiseu-
Free surface ille and Couette flow for the Oldroyd-B fluid are solved and compared with the analytical
and SPH results. Subsequently, the general viscoelastic fluid based on the eXtended Pom–
Pom (XPP) model is numerically investigated and the viscoelastic free surface phenomena
of impacting drop are simulated by the CSPH for its extended application and the purpose
of illustrating the ability of the proposed method. The numerical results are presented and
compared with available solutions, which shows a very good agreement. All the numerical
results show the higher accuracy and better stability of the CSPH than the SPH, especially
for larger Weissenberg numbers.
Ó 2011 Elsevier Inc. All rights reserved.

1. Introduction

In the non-Newtonian fluid mechanics community, viscoelastic free surface flows have been studied for more than
30 years. Originally many grid-based numerical methods were presented to simulate viscoelastic flows such as finite differ-
ence methods (FDM), finite volume methods (FVM) and finite element methods (FEM). However, it is usually difficult for the
simulation of moving free surface and large deformation using the grid-based methods.
Recently, in the area of computational mechanics there are developing mesh-free methods as alternatives to traditional
grid-based methods. The smoothed particle hydrodynamics (SPH) method is one of the earliest mesh-free methods employ-
ing Lagrangian description of motion. The SPH method was originally developed for astrophysical applications [1,2]. Since its
invention, it has been extensively studied in many areas such as the elastic material [3,4], the free surface flows [5], viscous
flows [6], incompressible fluids [7,8], heat transfer [9], multi-phase flows [10,11], geophysical flows [12,13], turbulence mod-
eling [14], viscoelastic flows [15,16], and free surface viscoelastic flows [15,16].

⇑ Corresponding author. Tel.: +86 29 88495234; fax: +86 29 88491000.


E-mail address: jieouyang@nwpu.edu.cn (J. Ouyang).

0307-904X/$ - see front matter Ó 2011 Elsevier Inc. All rights reserved.
doi:10.1016/j.apm.2011.02.014
3834 T. Jiang et al. / Applied Mathematical Modelling 35 (2011) 3833–3853

The SPH has several advantages over grid-based methods: it handles convection dominated flows and large deformation
problems without any numerical diffusion; it handles easily complex geometries in two and three dimensions; complex free
surfaces are modeled naturally without the need of any form of explicit surface tracking technique; programming for com-
plex problems is easy to implement compared with grid-based methods.
On the other hand, the SPH suffers from several drawbacks including low accuracy [17,18], tensile instability, and diffi-
culty in enforcing essential boundary conditions. The SPH kernel approximation does not have first order accuracy near the
boundary. Various methods have been developed to improve the accuracy of the conventional SPH. Liu et al. [19] improved
consistency of the SPH by introducing a corrective kernel function and named their method as the reproducing kernel par-
ticle method (RKPM), and the similar modified methods are applied to simulate elastic material problems [20,21] in 2007. A
normalized smoothing function algorithm was proposed in [22]. Dilts [23,24] introduced the moving least square approxi-
mation into SPH computations. In 2000, Chen and Beraun [25] developed a corrected smoothed particle hydrodynamics
method (CSPM) for non-linear dynamic problems using Taylor series expansion. Liu et al. [26] applied a similar idea to vis-
cous flows, which was called as finite particle method (FPM). Following, an improved SPH method coupling SPH and FPM
was presented by Jiannong Fang et al. [27] and applied to free surface flows of viscous fluids. By the Taylor series expansion’s
concept Zhang and Batra [28] proposed the so-called modified smoothed particle hydrodynamics (MSPH) and applied it to
transient problems in elastic dynamics and heat conduction. However, the MSPH needs to spend much computation because
a local matrix equation for each particle at each step will be solved.
Due to that the SPH method and the MSPH method have complementary virtues for each other, a new corrected SPH
method (CSPH) is proposed in this paper, which is inspired by using the SPH approximation for the interior particles and
the MSPH for the exterior particles for the purpose of reducing the computational cost of MSPH while still maintaining a
comparable accuracy with MSPH. In order to verify the validity and ability of the proposed method, the viscoelastic fluid
flows based on the Oldroyd-B model and the eXtended Pom–Pom model (XPP) are simulated by the proposed CSPH. The
paper is organized as follows: In Section 2, the governing equations of Oldroyd-B model fluid are outlined; In Section 3,
we give the brief introduction on three particle approximations, namely, SPH, MSPH, and CSPH, and at the same time some
special issues are also considered, such as the accuracy and consistency analysis, momentum conservation, the boundary
conditions and time integration scheme and so on; In Section 4, the analysis of convergence of CSPH is demonstrated first,
and then some numerical results for the transient planar shear viscoelatic flows based on the Oldroyd-B and XPP models are
presented for demonstrating the validity and ability of CSPH for simulating viscoelastic fluid flows; The CSPH is extended to
simulate the viscoelastic free surface problem of impacting drop in Section 5. The concluding remarks are reported in
Section 6.

2. Governing equations

In a Lagrangian frame, the basic governing equations of describing the isothermal fluid dynamics are the continuity equa-
tion (mass conservation) and the momentum equation (linear momentum conservation), which can be written as

dq @v b
¼ q b ; ð1Þ
dt @x
dv a 1 @ rab
¼ þ Fa; ð2Þ
dt q @xb
where q denotes the fluid density, vb the bth component of the fluid velocity, rab the (a, b)th component of the total stress
tensor and Fa is the ath component of the body force. The spatial coordinates xb and time t are the independent variables. As
usual, d/dt is the material derivative defined in a fixed Eulerian frame, which is

d @ @
¼ þ vb b : ð3Þ
dt @t @x

The total stress tensor in Eq. (2) is made up of the ordinary isotropic pressure p and the extra stress tensor s:

rab ¼ pdab þ sab ; ð4Þ


ab ab
where d = 1 if a = b and d = 0 if a – b. In order to study a viscoelastic material, the relating constitutive equation must be
provided.
In this paper, the governing equations are non-dimensionalised via length scale L, velocity scale U, time scale L/U, the
pressure and extra stress scale gU/L where g is the total viscosity. Then the non-dimensional form of the momentum equa-
tion can be obtained

dv a @p @ sab
Re ¼  a þ b þ F ar ; ð5Þ
dt x @x
where Re ¼ qgUL ; F ar is the non-dimensional body force.
T. Jiang et al. / Applied Mathematical Modelling 35 (2011) 3833–3853 3835

2.1. Oldroyd-B model

In the paper, we will discuss a type model of viscoelastic fluid, which is the Oldroyd-B model. Therefore, here we first
review the relating knowledge of the models.
Introducing the symmetric strain rate tensor D
 
@v a @v b
D ¼ ½Dab  ¼ þ : ð6Þ
@xb @xa
The constitutive equation of the Oldroyd-B fluid is given by
!
r r
sab þ k1 sab ¼ g Dab þ k2 Dab ; ð7Þ
r
where the upper-convected derivative sab is
r
dsab @ v a cb @ v b ac
sab ¼  cs  cs : ð8Þ
dt @x @x
In Eq. (7) k1 and k2 are the relaxation and retardation time, respectively, and g is the total viscosity. The extra stress tensor s
can be split into

sab ¼ ssab þ spab ; ð9Þ

where sab
s and sab p refer to the viscous stress and elastic stress component, respectively.
The sas b is usually computed according to
ssab ¼ gs Dab ; ð10Þ
k2 k2
where gs ¼ k1
g . Introducing the relaxation ratio bo ¼ k1
and according to Eqs. (7), (9) and (10), the elastic stress tensor is
given as
r
spab þ k1 sapb ¼ gð1  bo ÞDab : ð11Þ

The non-dimensional form for Eqs. (10) and (11) is

ssab ¼ bo Dab ; ð12Þ


r
spab þ We spab ¼ ð1  bo ÞDab ; ð13Þ

where We ¼ k1LU .

2.2. Equation of state

If incompressible fluid is to be simulated better, a further kinematic constraint on the velocity field should be taken into
account, which ensures that the divergence of velocity is free. Many previous works using the SPH method to simulate
incompressible flow have treated incompressible fluid as a slightly compressible by adopting a suitable equation of state.
In this paper, the following equation of state [27] is used

pðqÞ ¼ c2 ðq  q0 Þ; ð14Þ
where c is the speed of sound and q0 a reference density. It can be shown that the density variation is proportional to the
Mach number M(M  V/c, where V is a typical reference velocity) [7]. A suitable choice of c can produce the desired variation.
In this paper, the density variation is controlled within 1%.

3. Corrected SPH scheme

3.1. SPH discretization

In the SPH method, the fluid domain X is discretised into a finite number of particles, where all the relevant physical
quantities are approximated in terms of the integral representation over neighboring particles. Any function f defined at
the position x = (x, y) can be expressed by the following integral:
Z
hf ðxÞi ¼ f ðx0 ÞWðjx  x0 j; hÞ dx0 ; ð15Þ
X

where W is the so-called smoothing function or kernel function and h is the smoothing length defining the influence area of
W. As usual, W needs to satisfy the following properties:
3836 T. Jiang et al. / Applied Mathematical Modelling 35 (2011) 3833–3853

Z
Wðjx  x0 j; hÞdx0 ¼ 1; ð16Þ
X
0 0
lim Wðjx  x j; hÞ ¼ dðjx  x jÞ; ð17Þ
h!0

Wðjx  x0 j; hÞ > 0 over X ð18Þ


and
Wðjx  x0 j; hÞ ¼ 0 when jx  x0 j > kh; ð19Þ
where k is a constant which is usually chosen by the given kernel function. If the smoothing function W is an even function
over X, by using the Taylor series expansion of f(x0 ) around x, it can be shown that the integral representation of f(x) is of
second order accuracy. However, this is true only for interior regions.
We can get the following particle approximation formula from Eq. (15)
X mj
hf ðxÞi ’ fj Wðjx  xj j; hÞ; ð20Þ
j
qj
where mj and qj are the mass and density of the jth particle, and fj  f(xj). mj/qj represents the occupied volume by the jth
particle. In order to have an accurate interpolation, the smoothing length h should be chosen bigger than the mean inter-
particle distance.
The particle approximation for a function and its first derivative at particle i can be written in condensed form as
X mj
fi ¼ fj W ij ; ð21Þ
j
qj
  X mj
@f @W ij
¼ ðfj  fi Þ ; ð22Þ
@xi j
qj @xi

where Wij = W(jxi  xjj, h) and @ Wij/@xi = @W(jxi  xjj, h)/@xi.


The smoothing function is one of the most important ingredients of the SPH method. Its choice is related not only with the
accuracy but also with the efficiency and stability of the resulting algorithm.
In this work, the quintic spline function is chosen as the smoothing function which is the function about r = jxi  xj j and h.
Let k = 3 in Eq. (19), then it reads
8
>
> ð3  qÞ5  6ð2  qÞ5 þ 15ð1  qÞ5 ; 0 6 q <;
>
>
<
ð3  qÞ5  6ð2  qÞ5 ; 1 6 q < 2;
W ij ¼ Wðr; hÞ ¼ w0 ð23Þ
>
> ð3  qÞ5 ; 2 6 q < 3;
>
>
:
0; q P 3;
where q = r/h, the normalization factor w0 is chosen as 7/(478ph2) in two dimensional problems.
In this paper, the velocity and stress gradient are calculated using the Eq. (22). Then the discretization schemes of the
non-dimensional governing equations are obtained at the particle i as
  X mj  b  @W
dq ij
¼ qi v i  v bj ; ð24Þ
dt i j
q j @x b
i
 a
dv X mj @W ij X mj  ab  @W
ij
Re ¼ ðpi þ pj Þ a þ si þ sjab þ F ari : ð25Þ
dt i j
q j @x i j
qj @x b
i

Introducing the velocity gradient


  X mj
@v a @W ij
jiab ¼ ¼ ðv aj  v ai Þ ; ð26Þ
@xb i j
qj @xbi

the discretization schemes of constitutive equation for Oldroyd-B model can be defined as
  
siab ¼ bo jiab þ jbi a þ spab ; ð27Þ
i
!
dspab     1  ab  1  bo  ab
ac
¼ ji scpb þ jbi c sca
p  sp þ ji þ jbi a : ð28Þ
dt i i We i We
i

3.2. Corrected SPH method

Because of the low accuracy of the conventional SPH method, Zhang and Batra [28] proposed the so-called MSPH by the
Taylor series expansion’s concept and applied it to transient problems in elastic dynamics and heat conduction. Here, the
MSPH method is briefly described.
T. Jiang et al. / Applied Mathematical Modelling 35 (2011) 3833–3853 3837

By correcting the kernel and its first order derivative with the concept of Taylor series expansion (see [28]), the corrected
particle approximations of Eqs. (21) and (22) can be obtained as
X mj
fi ¼ c ij ;
fj W ð29Þ
j
qj
  X mj @ Wc ij
@f
¼ fj ; ð30Þ
@xi j
qj @xi

c ij and its first order derivative @ W


where the modified kernel function W c ij =@xi is obtained by

^ 6 ÞT ¼ A1 w6 T
ðw ð31Þ
i
P mj l P mj l P mj l
and the Ai ¼ ðA66 Þi ; ðAl1 Þi ¼ j qj w ; ðAl2 Þi ¼ j qj xji w ; ðAl3 Þi ¼ j qj yji w ;

1 X mj 1 X mj X mj
ðAl4 Þi ¼ ðxji Þ2 wl ; ðAl5 Þi ¼ ðy Þ2 wl ; ðAl6 Þi ¼ xji yji wl ;
2 j qj 2 j qj ji j
qj
where l = 1, . . ., 6,
8  
<w^6 ¼ W c ij ; @ W
c ij =@xi ; @ W c ij =@x2 ; @ 2 W
c ij =@y ; @ 2 W c ij =@y2 ; @ 2 W
c ij =ð@xi @y Þ ;
i i i i
ð32Þ
: w6 ¼ W ; @W =@x ; @W =@y ; @ 2 W =@x2 ; @ 2 W =@y2 ; @ 2 W =ð@x @y Þ ;
ij ij i ij i ij i ij i ij i i

Like in SPH, the velocity and total stress gradient are calculated in MSPH as
8
>
> @ v a P mj a b
a @ W ij
qj ðv j  v i Þ @xb ;
>
> @xb i ¼
>
>
>
> j i
>
<  ab   
@r
P m ab @ b
W b ji
ab @ W
@xb i
¼ q
j
ri b  rj
ij
b ; ð33Þ
>
> j
j @xi @xj
>
>
>
>  b ij
> jab ¼ @vba ¼ P mj ðv a  v a Þ @ W
>
> ;
: i @x i qj j i @xb
j i

b
@W @Wb ji
according to [23] we know that –  @x ij . @xj
i
Finally, the particle positions are updated by the following equation:

dxai
¼ v ai : ð34Þ
dt
The MSPH schemes (33) have higher accuracy and better consistency than the conventional SPH schemes mentioned in
Section 3.1. According to [18], the particle approximations of MSPH will have third-order accuracy and second-order consis-
tency. Moreover, the MSPH possesses at least second-order accuracy on the boundary regions, which has been illustrated in
[18,28].
However, the MSPH method involves solving a local matrix for each particle at each step as can be seen from Eq. (31),
which needs to spend more calculated amount than that of SPH. Compared with MSPH, in order to reduce the computational
cost but still maintain a comparable accuracy, a corrected SPH is proposed, in which the SPH method is used for the interior
particles and the MSPH approximation for the exterior particles. In this paper, a particle is regarded as an exterior particle
when it is a boundary/surface particle within its support domain; otherwise, the particle is regarded as an interior particle.
Thus, the CSPH method for simulating the Oldroyd-B fluid can be obtained using Eqs. (22), (23), (26) and (33) in the inte-
rior and exterior domain, respectively.

3.3. Momentum preservation

As is well known, the angular momentum and linear momentum are usually not strictly preserved by the SPH particle
approximations. In order to conserve the linear and angular momentum, many improved schemes and preserving properties
of SPH formulations are discussed in [24,29,30]. The variational formulation and an energy-based approach are adopted to
study the conservation of momentum by Bonet and Lok [29] and Fang et al. [27], respectively. According to [27], the discrete
SPH Eq. (33) preserves the linear momentum in the absence of body force, but does not conserve the total angular momen-
tum. The particle approximation based on FPM preserves the total linear and angular momentum in the absence of external
forces according to [27], which is because that the particle approximation of FPM restores the first-order consistency. When
the first-order derivatives retained in the Taylor series expansion process, the FPM may be obtained in [26]. However, the
MSPH was obtained by retaining the second-order derivatives in the Taylor series expansion. According to [18,28], we
can know that the MSPH kernel and particle approximations have second-order consistency if the second-order derivatives
are retained.
3838 T. Jiang et al. / Applied Mathematical Modelling 35 (2011) 3833–3853

Therefore, the MSPH particle approximation Eq. (33) possesses the conservation of the total linear and total angular
momentum in the absence of external forces by combining the detail analysis with an energy-based approach in [27],
and which possesses the better stability than SPH. In fact, the CSPH also preserves the linear and angular momentum by
the remarks in [23].

3.4. Time integration scheme

In this paper, the leapfrog method [31] with second-order accuracy is used for its computational efficiency. In the leapfrog
scheme, the particle velocities, viscoelastic stress and positions are offset by a half time step. The basic procedure of leapfrog
method is: At the end of the first time step t0, the density, velocity and viscoelastic stress are advanced at half a time step,
while the particle positions are advanced in a full time step. Namely
8
> t ¼ t 0 þ Dt;
>
>
>
> Dt
< qi ðt 0 þ Dt=2Þ ¼ qi ðt 0 Þ þ 2 dqi ðt0 Þ;
>
v i ðt0 þ Dt=2Þ ¼ v i ðt0 Þ þ D2t dv ai ðt0 Þ;
a a
ð35Þ
>
>
> v sab ðt0 þ Dt=2Þ ¼ v sab ðt0 Þ þ Dt dv sab ðt 0 Þ;
>
>
> i i 2 i
: a
xi ðt 0 þ DtÞ ¼ xai ðt 0 Þ þ Dtv ai ðt0 þ Dt=2Þ;
where vs represents the viscoelastic stress which is sp for Oldryod-B model. In order to keep the calculations consistent at
each subsequent time step, at the start of each subsequent time step, the density, velocity and viscoelastic stress of each par-
ticle need to be predicted at half a time step to coincide the position. That is
8 Dt
< qi ðtÞ ¼ qi ðt  Dt=2Þ þ 2 dqi ðt  DtÞ;
>

>
v i ðtÞ ¼ v i ðt  Dt=2Þ þ D2t dv ai ðt  DtÞ;
a a ð36Þ
: a
v si ðtÞ ¼ v sai ðt  Dt=2Þ þ D2t dv sai ðt  DtÞ;
At the end of the subsequent time step, these values in Eq. (35) and position are advanced in the following standard leapfrog
schemes.
8
>
> t ¼ t þ Dt;
>
>
>
< qi ðt þ Dt=2Þ ¼ qi ðt  Dt=2Þ þ Dtdqi ðtÞ;
>
v ai ðt þ Dt=2Þ ¼ v ai ðt  Dt=2Þ þ Dtdv ai ðtÞ; ð37Þ
>
>
> v sa ðt þ Dt=2Þ ¼ v sa ðt  Dt=2Þ þ Dtdv sa ðtÞ;
>
>
> i i i
: a
xi ðt þ DtÞ ¼ xai ðtÞ þ Dtv ai ðt þ Dt=2Þ;
To ensure the numerical stability, the time step and space step must satisfy the well-known Courant–Friedrichs–Lewy
(CFL) condition. According to [31], we may choose the following stability condition:
"   #
1=2 2
h h h
Dt 6 min ; ; 0:15 ; ð38Þ
c Fa v0
where Fa is the hydrodynamical force acting on the particle, and v0 = l/q0 is the kinematic viscosity.
3.5. Boundary conditions

In this work, two types of boundary conditions are taken into account for start-up flow, which are periodic boundary con-
ditions and solid wall boundary conditions, respectively (see Fig. 1).
Firstly, the periodic boundary condition is implemented [16]. Fig. 1 shows the imposition of this boundary condition,
namely before current time step, the physical quantities of the particles ina zone are sent to the corresponding ones in b
zone, and the same thing also happens for the particles in c zone and d zone.
Then in order to avoid numerical instability near the boundary [32], two types of virtual particles are used:

r The first type virtual particles are just located on the solid boundary, namely ‘‘boundary particles’’. In contrast to work of
Cummins and Rudman [7], boundary particles in this paper do not employ an artificial repulsive force on approaching real
particles to prevent them from penetrating solid walls. Additionally, unlike Morris [6], the density of boundary particles is
not evolved. Specifically, in our paper the non-slip condition is enforced on the solid boundary and the boundary particle
positions remain fixed as time goes. The pressure and components of the viscoelastic stress on the boundary particles B
are calculated according to the following normalized interpolation formula:
,
X mj X mj
fi ¼ fj W ij W ij ; ð39Þ
j
qj j
qj
T. Jiang et al. / Applied Mathematical Modelling 35 (2011) 3833–3853 3839

Fig. 1. The placement of fluid particles (FP), boundary particles (BP), periodic boundary particles (PP) and ghost particles (GP).

where i represents the index of a boundary particle and j represents the indices of its neighboring fluid particles only. The
normal elastic stress in the direction perpendicular to the solid wall surface is set to zero.
s The second type virtual particles are placed just outside the solid boundary and fill a certain domain with at least a range
of depth comparable with the compact support of the kernel used in the computations, which are called ‘‘ghost particles’’
and have fixed density and positions. The velocity and the viscoelastic stress tensor on the ghost particles are obtained by
interpolation in the following way: for each ghost particle G, a corresponding point B and point R are just on the solid wall
and inside the fluid domain, respectively. In order to calculate conveniently we can make the normal distances of the
points G and R to the solid boundary equal. So the velocity v aG , extra stress saGb and pressure pG on the ghost particles
are obtained through the following linear extrapolations:
8 a
< v G ¼ 2v B  v R ;
a a
>

>
sG ¼ 2sB  sRab ;
ab ab
ð40Þ
: a
pG ¼ 2paB  paR ;
where v aB ; saBb ; pB denote the physical quantity on the boundary particle and v aR ; saRb ; pR denote the variation in the fluid do-
main. To specify the values for v aR ; saRb ; pR , the normalized interpolation formula (39) is applied again. We should note that
the above two types of virtual particles are adopted in SPH, and only the first type virtual particles are adopted in CSPH.

4. Numerical examples

4.1. Comparisons of accuracy and convergence for the CSPH and other methods

In order to exemplify the higher accuracy of the CSPH than the SPH, the following L1 normal error is introduced in the
paper, which is
Z
kf  f h kL1 ¼ jf  f h jdX ð41Þ
X

and the corresponding relative L1 normal error for any function f is


Eðf Þ ¼ kf  f h krel h
L1 ¼ kf  f kL1 =kf kL1 ; ð42Þ
h
where f and f represent the analytical and numerical solution.
Consider the function of one variable

f ðxÞ ¼ ðx  0:5Þ6 ; x 2 ½0; 1:


The interval is uniformly distributed particles with one particle at each end point. In order to obtain better convergence, the
smoothing length h should be appropriately choice, Otherwise, the numerical results using SPH will be divergent (see [17]).
The smoothing length h is chosen as twice space step Dx in this subsection.
3840 T. Jiang et al. / Applied Mathematical Modelling 35 (2011) 3833–3853

Fig. 2 shows the comparison of the function f(x) and its first and the second derivatives among the exact solution, the
CSPH methods and other improved SPH methods. It is clear that the CSPH can obtain the better results than the SPH method,
especially on the boundary regions. For the purpose of improving precision of the numerical methods, the particle numbers
from 21 to 51 is exhibited in Fig. 3. As expected, the accuracy of the numerical method is enhanced, even to the second-order
derivatives near the boundaries, CSPH can still obtain rather high precise than the SPH. Fig. 4 presents the convergence (N
denotes the total particle numbers) of the methods mentioned above respectively, and we can find that the convergence rate
of CSPH is the fastest than the SPH not only for the function f(x) but also its first derivative.

4.2. Application of the CSPH to transient planar shear flows

In this subsection, the CSPH technique is applied to simulate the start-up flow problems including transient Poiseuille
flow, Couette flow, and the combined Poiseuille Couette flow. To show that CSPH has higher accuracy than SPH and the lower
computational cost than the MSPH, the comparison between the numerical results obtained using CSPH, SPH for Newtonian
fluid are demonstrated in Section 4.2.1, further, we also compare the above results with the analytical solution. In subsection
4.2.2, the CSPH method is extended and tested for simulating the combined Poiseuille Couette flow of Oldroyd-B and com-
pared with SPH. Additionally, their numerical results and the theory solution are also compared.
The planar flow involves flow between two parallel, stationary and infinite plates located at y = 0 and y = L0. The Poiseuille
flow is described that the initially stationary fluid is driven by a body force F parallel to the x-axis and then removes. The
Couette flow is generated after that the upper plate suddenly moves at a certain constant velocity U0 horizontally, see
Fig. 5. In order to effectively analyze the evolution of the physical quantities with time in numerical simulations, three dif-
ferent spatial positions at y = 0.236L0,0.5L0, 0.76L0 are chosen in Section 4.2.2.
The combined Poiseuille and Couette flow is that the upper plate starts to move at constant velocity U0 along the x-axis,
and an additional body force F is employed parallel to the x-axis too. In order to describe the resultant velocity profile of the
combined Poiseuille and Couette flow, we define a non-dimensional quantity according to [33], namely
L0 F
B¼ Re: ð43Þ
4U 20

0.02 0.2

Exact
SPH
0.015 CSPH
Exact 0.1
SPH
CSPH
f ( x)

f ′( x)

0.01 0

0.005
-0.1

0
-0.2
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x x

1.5 Exact
SPH
CSPH
1
f ′′( x)

0.5

-0.5
0 0.2 0.4 0.6 0.8 1
x
Fig. 2. The numerical results of a) the function f(x), b) the first-order derived function f0 (x) and (c) the second-order f0 0 (x), with 21 equally spaced particles.
T. Jiang et al. / Applied Mathematical Modelling 35 (2011) 3833–3853 3841

0.02 0.2
Exact
SPH
CSPH Exact
0.015 0.1 SPH
CSPH

f ′( x)
f ( x)
0.01
0

0.005
-0.1

0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x x
2

1.5
Exact
SPH
CSPH
1
f ′′( x)

0.5

-0.5

0 0.2 0.4 0.6 0.8 1


x
Fig. 3. The numerical results of a) the function f(x), b) the first-order derived function f0 (x) and (c) the second-order f00 (x), with 51 equally spaced particles.

0.7 0.8

SPH 0.7 SPH


0.6 CSPH
CSPH
0.6
0.5
0.5
0.4
E ( f ′)
E(f)

0.4
0.3
0.3
0.2
0.2
0.1 0.1

0 0
40 80 120 160 200 40 80 120 160 200
N N
0
Fig. 4. Rate of convergence for (a) the function f(x) and (b) the first-order function f (x), with the increasing equally spaced particles.

4.2.1. Newtonian fluid


In this subsection, the dimensional parameters are adopted for the better comparison with the numerical results in [26].
We make the same parameter values as those in [26]: L0 = 103 m, q0 = 103 kg/m3, the kinetic viscosity v0 = 106 m2/s, the
body force F = 2  104 m/s2 and the certain constant velocity U0 = 2.5  105m/s2 for the Poiseuille and Couette flow respec-
tively. This means that Reynolds number is Re = 0.025, and the sound speed is taken as c = 2.5  104 m/s.
For the simulation presented here, a rectangular problem domain with Lx  Ly = 0.0005 m  0.001 m is modeled with
20  40 real particles, 2  20 boundary particles and 5  20 ghost particles. The smoothing length has been chosen as 1.3
times the initial particle spacing. According to the previous discussion, the time step is set to Dt = 104 s. In this subsection
the two types of flow above all reach a steady state after about 5000 steps or 0.5 s. Fig. 6 shows the comparison between the
3842 T. Jiang et al. / Applied Mathematical Modelling 35 (2011) 3833–3853

t ≤0 vx = 0 t >0 vx = 0

F
L0 y vx = 0 L0 y vx ( y, ∞)

x x
vx = 0 vx = 0

t >0 U0

3
y 2
vx ( y, ∞)
L0
1
x
vx = 0

Fig. 5. Geometry for the Poiseuille flow and Couette flow at initially stationary and steady state.

0 0.001
Analytical
SPH 0.01s
CSPH
0.0002 0.0008

0.1s
0.0004 0.0006 0.5s
0.01s 0.1s 0.5s
y
y

0.0006 0.0004

0.0008 0.0002
Analytical
SPH
CSPH
0.001 0
0 5E-06 1E-05 1.5E-05 2E-05 2.5E-05 3E-05 0 5E-06 1E-05 1.5E-05 2E-05 2.5E-05
U U
Fig. 6. Velocity profiles at different instants for the Poiseuille (left) and Couette (right) flow with Re = 0.025.

velocity profiles obtained using CSPH, SPH and the analytical solution [6] at t = 0.01 s,0.1 s, 0.5 s at low Reynolds number
Re = 0.025. As shown in Fig. 6, the numerical results of CSPH are much closer to the analytical solution than the SPH.
Fig. 7 shows the numerical precision difference of CSPH, SPH based on the relative L1 normal error in Eq. (42). Further,
the Table 1 shows the comparisons of computational efficiency between the CSPH, SPH and MSPH. From Table 1 we can
know that the consumed CPU time for MSPH simulation is around 4–5 times more than that for SPH simulation in our com-
puter. With the increased particle numbers or for the three dimensional case, the consumed CPU time for the MSPH grows
the most quickly during these three methods. As far as the consumed CPU time for the CSPH and SPH concerned, the differ-
ence between them is small. Finally, it is noteworthy that two type ghost particles are adopted for the SPH and only the
boundary particles are used for the CSPH.

4.2.2. Oldroyd-B fluid


Due to that the phenomenon of the combined Poiseuille and Couette flow is different from the Poiseuille flow or Poiseuille
flow, and the combined Poiseuille and Couette flow has been rarely simulated, especially for the non-Newtonian fluid case,
and then we mainly consider this combined start-up flow problem in this subsection.
The CSPH and SPH code are used to simulate the combined Poiseuille and Couette flow, for the Oldryod-B fluid, and adopt
the following non-dimensional parameters:L0 = 1, q0 = 1, the body force Fr = 1 and the certain constant velocity U0 = 0.01,
which correspond to Reynolds number of Re = 0.05, and the sound speed c = 0.1 m/s. For the simulations presented here,
a rectangular problem domain with Lx  Ly = 0.2  1 is modeled with 10  50 real particles, 2  10 boundary particles and
5  10 ghost particles. The smoothing length has been chosen as 1.1 times the initial particle spacing. According to the pre-
vious discussion, the time step is set to Dt = 105.
For the simulation of Oldroy-B fluid, we need to solve the non-dimensional system of ordinary differential Eqs. (24)–(28),
(33)–(37), and take bo = 3/9. Here, the Weissenberg numbers of We = 0.01, 0.1 and 1 are used. Two different non-dimensional
quantities of B are considered, B = 1.25 and B = 1.25.
T. Jiang et al. / Applied Mathematical Modelling 35 (2011) 3833–3853 3843

0.03 0.03

0.025 0.025

0.02 0.02

E (U )
E (U )

0.015 0.015

0.01 0.01

0.005 SPH 0.005 SPH


CSPH CSPH

0 0
0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5
t t
Fig. 7. Evolution of relative L1 normal error about the numerical and analytical solution of velocity along thex-axis for the Poiseuille (left) and Couette
(right) flow with Re = 0.025.

Table 1
Comparisons of consumed CPU time using the proposed CSPH and the SPH, MSPH for the simulation in Fig. 6.

Step numbers The SPH CPU time (s) Proposed CSPH CPU time (s) The MSPH CPU time (s)
1000 118.234 136.706 536.813
2000 230.312 275.37 1098.81
3000 344.968 411.357 1659.16
4000 455.265 550.851 2230.06
5000 565.859 785.65 2802.13

Fig. 8 shows quantitative comparisons between the velocity profiles obtained using CSPH, SPH and the analytical solution
(Eqs. (A.1)–(A.4)) at five different time, which indicates that the maximum allowable time step is Dt = 5  105. The CSPH
results are much closer to the analytical solution than the SPH results. The parameter We is 0.01 here and the final steady
is achieved after about 104 steps (or no-dimensional time 0.1) for two situations B = 1.25 and B = 1.25. After about 0.018
and 0.009 the velocity reaches peak value with B = 1.25 and B = 1.25, respectively. Meanwhile, the overshooting phenom-
enon appears for Oldroyd-B fluid.
Further simulations have been performed in Fig. 9, for the purpose of testing the ability to obtain stable and accurate re-
sults for larger Weissenberg numbers between CSPH and SPH. Fig. 10 shows the evolution of velocity at spatial positions 1–3
(see Fig. 5) with the Weissenberg number from 0.01 to 1 and the other parameters keep the same. The velocity overshooting
behavior becomes more evident with the increasing Weissenberg numbers. The accuracy of CSPH is much higher than that of
SPH, especially for larger Weissenberg numbers (see Fig. 9(c)). Meanwhile, we can know that the larger the Weissenberg
number is, the more consumed time of reaching the final steady needs. In a word, the CSPH method is efficient for the sim-
ulation of the combined Poiseuille and Couette flow.
Fig. 10 shows quantitative comparisons between the elastic shear stress profiles obtained using CSPH, SPH and the ana-
lytical solution (Eqs. (A.5)–(A.9)) at four different time. The CSPH results agree very well with the analytical solution, and are

Fig. 8. Velocity profiles at different instants for the combined Poiseuille and Couette flow. Re = 0.05, We = 0.01 and two different non-dimensional
quantities B = 1.25 (left), B = 1.25 (right).
3844 T. Jiang et al. / Applied Mathematical Modelling 35 (2011) 3833–3853

Fig. 9. Evolution of the velocity for the combined Poiseuille and Couette flow at spatial positions 1–3 (see Fig. 5). Comparison between the CSPH, SPH results
and the analytical solution with different B = 1.25 (left column), B = 1.25 (right column) and various Weissenberg numbers: (a) We = 0.01; (b) We = 0.1; (c)
We = 1.

much closer to the analytical solution than the SPH results, in which the parameter We is 0.01. From the Figs. 9 and 10, the
ability of the proposed CSPH can be tested for simulating an Oldroyd-B fluid including the accurate velocity and elastic stress
at the higher Weissenberg number and the overshooting phenomenon for the non-Newtonian fluid planar shear flow.
In order to further demonstrate the merit of CSPH for simulating viscoelastic fluid flows, the convergence rate of numer-
ical solutions obtained using the CSPH for solving the combined Poiseuille and Couette flow based on Oldroyd-B model is
illustrated at different times and compared with SPH results in Fig. 11. In this case, the physical parameters Re = 0.05,
We = 1,B = 1.25, the total real particles are Nx  Ny, and the ratio of Nx: Ny is fixed to be 1:5, where the Nx denotes the real
particles along the x-axis and the Ny denotes the real particles along the y-axis. The situation of relative L1 normal errors
for the velocity and elastic shear stress sxy
p varying with the increased real particles Nx is shown in Fig. 11.
T. Jiang et al. / Applied Mathematical Modelling 35 (2011) 3833–3853 3845

Fig. 10. Elastic shear stress sxy


p profiles at different instants for the combined Poiseuille and Couette flow with Re = 0.05, We = 0.01 and two different non-
dimensional quantities B = 1.25 (left), B = 1.25 (right).

0.3
CSPH
SPH
0.25

0.2
1
E (U )

0.15
0.1 1

0.1 0.1
0.05
0.5
0.5
0.05
0.05

0 20 40 60 80 100
Nx

0.5
0.45 CSPH
SPH
0.4
0.35
1
0.3
E (τ pxy )

0.25 1
0.1
0.2 0.1

0.15 0.05 0.05


0.5
0.1 0.5
0.05
0
0 20 40 60 80 100
Nx

Fig. 11. Rate of convergence for the combined Poiseuille and Couette flow of Oldroyd-B fluid with We = 1, B = 1.25, at different times: (a) the velocity and (b)
the elastic shear stress sxy
p .

From Fig. 11, we can get that the convergence rate of CSPH results for planar flow based on Oldroyd-B model is obviously
faster than that of SPH at different times. Further, the merit of that possesses better accuracy for the CSPH method than the
conventional SPH method is exemplified by solving the viscoelastic fluid flows.

4.2.3. General viscoelastic fluid flows based on XPP model


Due to the fact that the XPP [34,35] model is more general than the Oldroyd-B model in the description of polymer melts,
the former has more complex rheological behavior than the latter, and the former model can degrade into the latter model
on special occasions. Therefore, the single equation version of XPP in multi-mode form [34,36] is considered in this
3846 T. Jiang et al. / Applied Mathematical Modelling 35 (2011) 3833–3853

subsection, for the purpose of further illustrating the ability of proposed CSPH for solving viscoelastic fluid flows. The con-
stitutive equation of XPP model and its relative rheological parameters are introduced in Appendix B.
The governing equations for the viscoelastic fluid of Oldroyd-B model possess an analytical transient solution for planar
flow [37,38]. However, the governing equations for the XPP model do not possess an equivalent analytical transient solution
[37]. Through the analysis and numerical results in Section 4.2.2, we can know that it is credible to use the proposed CSPH for
simulating planar shear viscoelastic fluid flows based on XPP model. Here, in order to expediently determine the true ten-
dency of velocity and viscoelastic stress at a steady state for the planar flow problems, we mainly focus on the numerical
simulations of CSPH for Couette flow of XPP model and exemplify the influences of the physical parameters on the velocity
variation, stretch and the viscoelastic-stress tensor at spatial positions 1–3 (see Fig. 5) with time. The parameters of fluid
domain keeps the same as those in subsection 4.2.2, and we adopt the rheological parameters q = 2, a0 = 0.15, e = 1/3,
Re = 0.05.
The influences of the Weissenberg number on the velocity variation, stretch, and polymeric contributions to the viscoelas-
tic shear stress sxy
p are shown in Fig. 12. In this case, the different values of We from 0.01 to 1 are considered, and the b0 is set

Fig. 12. Physical quantities variation for the Couette flow of XPP model at spatial positions 1–3 (see Fig. 5) with different We using CSPH: (a) velocity; (b)
viscoelastic-stress; (c) the difference between the stretch k and its equilibrium value 1.
T. Jiang et al. / Applied Mathematical Modelling 35 (2011) 3833–3853 3847

to be 1/3. The velocity overshooting phenomenon disappears as Weissenberg number increases. The value of shear stress sxy p
decreases as the Weissenberg number increases, but the time of reaching the final steady increases. The larger the
Weissenberg number is, the larger value of stretch is. The stretch is nearer to its value of equilibrium (k = 1) for small value
of Weissenberg number than high Weissenberg number case.
The influences of the viscosity ratio b0 on the velocity variation, stretch, and polymeric contributions to the viscoelastic
stress Txy are shown in Fig. 13. The different values of b0 from 0.15 to 0.5 are considered, the other model parameters are
fixed and the We is set to be 0.1. High value of b0 corresponds to dilute or less-entangled polymeric solution. The phenom-
enon of velocity overshooting becomes more evident as the value of b0 decreases. Larger value of viscoelastic stress sxy p is
generated for small value of b0. The magnitude of the polymeric stress decreases when the Newtonian solvent contribution
becomes more important just as the viscosity ratio increases (see [36]). But the stretch grows with increasing b0 at the final
state of steady for Couette flow which can be seen in Fig. 13.

Fig. 13. Physical quantities variation for the Couette flow of XPP model at spatial positions 1–3 (see Fig. 5) with different viscosity ratio b0 using CSPH: (a)
velocity; (b) viscoelastic-stress; (c) the difference between the stretch k and its equilibrium value 1.
3848 T. Jiang et al. / Applied Mathematical Modelling 35 (2011) 3833–3853

5. Further application for viscoelastic free surface flows

The problems of free surface hydrodynamic flows for polymers are important in today’s industry, such as extrusion and
pharmaceutical industries of polymers. In these industries, all the flows involved almost exhibit viscoelastic behavior. In or-
der to further demonstrate the merit of the proposed method and the non-linear viscoelastic behaviors of polymer melt, the
CSPH is extended and applied to simulate the problem of impacting drop based on Odroyd-B and XPP models in this section.
According to [32,39], an artificial viscosity term [32,39] and a simple artificial stress term [40] need to be employed in the
following simulations for the purpose of improving the numerical stability and eliminating the phenomenon of unphysical
clustering of particles arises (namely, ‘‘tensile instability’’). The adopted artificial viscosity and stress terms all have been suc-
cessfully applied to simulate the viscoelastic free surface problems, which can be seen in detail in [32,39,40]. Moreover, the
solid wall boundary condition (see subsection 3.5) is considered, and the total stress-free condition (rn = 0, where n denotes
a unit normal vector to the surface) is satisfied naturally by the SPH and CSPH method (see [32,39]).
The dimensional parameters are chosen as the same as those in [32,39], namely, the ratio between Newtonian viscosity
and total viscosity b0 = 0.1, the relaxation time k1 = 0.02s, and the dimensionless parameters Re = 5, We = 1.
The problems of impacting drop for Newtonian and Oldroyd-B fluid have been simulated by conventional SPH and finite
difference methods in [32,39,38]. Here, we mainly focus on the comparisons between the CSPH results and the results in
[32,38] for the purpose of illustrating the merit of CSPH for simulating viscoelastic free surface flows.
The deformation and the pressure fields distribution (p/(q0 U2)) of impacting drop obtained using the CSPH and SPH for
the Newtonian and Oldroyd-B fluid are shown in Figs. 14-1 and 14-2, respectively. The phenomenon of pressure oscillations
occurs for the SPH method in a short time after droplet impact. The pressure oscillations grow more violent near the solid
wall varying with time, and later the whole pressure field is destroyed, resulting in making its physical interpretation and

Fig. 14-1. The distributions of the particle positions and pressure fields (p/(q0U2)) for an impacting drop of Newtonian fluid at different dimensionless
times: CSPH (left column) and SPH (right column).
T. Jiang et al. / Applied Mathematical Modelling 35 (2011) 3833–3853 3849

Fig. 14-2. The distributions of the particle positions and pressure fields (p/(q0U2)) for an impacting drop of Oldroyd-B fluid at different dimensionless times:
CSPH (left column) and SPH (right column).

possible practical use difficult. However, the pressure filed maintains a much smoother character obtained using the CSPH
method than the SPH, especially on the boundary regions.
Fig. 15 shows the comparisons of the widths for a Newtonian and an Oldroyd-B fluid drop obtained respectively by CSPH
as well as the other methods in [32,38]. The results of using the CSPH are much closer to the results in [38] than those of
using SPH for both of Newtonian and Oldroyd-B fluid cases. From Figs. 14-1, 14-2 and 15, we can observe that the Newtonian
drop hits the wall and spreads out evenly while retaining its convex shape, but the flow process of the viscoelastic drop

0.045 0.05
Width of Oldroyd-B droplet
Width of Newtonian droplet

0.04
0.045

0.04
0.035
0.035
0.03
0.03
SPH
SPH
0.025 CSPH
CSPH 0.025 FDM[39]
FDM[39]

0.02 0.02
1 1.5 2 2.5 3 3.5 1 1.5 2 2.5 3 3.5
Dimensionless Time Dimensionless Time
Fig. 15. Comparison of the numerical results obtained using SPH/CSPH method for the width of a Newtonian (a) and Oldroyd-B (b) droplet impact on
horizontal rigid plate varying with dimensionless times.
3850 T. Jiang et al. / Applied Mathematical Modelling 35 (2011) 3833–3853

Fig. 16. The deformation of a falling droplet based on the XPP fluid using CSPH method with Re = 5 and We = 1 at different dimensionless times.

suffers three stages. In the first stage, the Oldroyd-B drop displays a greater tendency to spread horizontally than the case of
Newtonian fluid. The viscoelastic drop contracts because of the elasticity of the fluid and achieves a positive vertical velocity
in the second period (about 2.3 < t < 3.75). Finally, the elastic effects become weak and the drop spreads out slowly like its
Newtonian counterpart. So, we can get that the CSPH method is very effective for simulating the viscoelastic free surface
problem of polymer melts and possesses higher accuracy than the SPH method.
In this section, the numerical simulation of impacting drop of a XPP fluid is also considered. We set a0 = 0.15, q = 2, e = 1/3
and the other parameters keep the same as those in Fig. 15 except for the parameter g and k0b. Fig. 16 illustrates the shape of
a falling droplet based on the XPP fluid using CSPH method with Re = 5 and We = 1 at different dimensionless times. It can be
seen that the main difference is that of the smaller contracting trend of the XPP fluid drop than that of Oldroyd-B fluid drop
in the second stage by comparing Fig. 16with Fig. 14-2 due to that the XPP type model has more complex rheological behav-
ior than the Oldroyd-B type model (see [34,35]).

6. Conclusions

In this paper, a corrected SPH (CSPH) method is proposed for simulating transient viscoelastic fluid flows by coupling the
SPH and MSPH. The proposed CSPH is inspired by combining the complementary virtues of the SPH and the MSPH, in which
we use the SPH approximation for the interior particles and the MSPH for the exterior particles. The start-up flow based on
the Oldroyd-B model is first considered for testing the accuracy and convergence of CSPH at low Reynolds number. The CSPH
results are much closer to the analytical solution than the SPH results for simulating the planar flow problems, especially for
larger Weissenberg numbers. Subsequently, the Couette flow based on the XPP model is numerically investigated by CSPH
for the purpose of illustrating the ability of the proposed method. In order to further demonstrate the advantage and the
extended application of the proposed method, the viscoelastic free surface phenomenon of impacting drop is simulated in
this work. All the numerical results show that some advantages of the CSPH method over the conventional SPH: (1) it has
higher accuracy and numerical stability for simulating viscoelastic planar flows, especially for larger Weissenberg numbers;
(2) The particle positions are more uniformly and the pressure fields are far smoother for simulating visoelastic free surface
flows; (3) it is more credible for capturing the non-linear behavior of viscoelastic fluid flows. The proposed method is ex-
pected to be further improved and extended to more complex fluid dynamic problems.
T. Jiang et al. / Applied Mathematical Modelling 35 (2011) 3833–3853 3851

Acknowledgements

The support from the National Natural Science Foundation of China (NSFC) (No. 10871159) and National Basic Research
Program of China (No. 2005CB321704) are fully acknowledged.

Appendix A

The following section provides the non-dimensional analytical solution of Oldroyd-B fluid for the combined Poiseuille and
Couette flow in a 2D channel, which may be obtained by Waters et al. [37] and Van et al. [38]. The exact solution for the
velocity field is

Re X1
4Re
Uðy; tÞ ¼  yðy  1ÞF þ yU 0  3
F sinðN1 yÞHn ðtÞ
2 n¼0 N 1
X1
2ð1Þn
þ U0 sinðN2 yÞHm ðtÞ; ðA:1Þ
m¼1
N2

where

We We We am
N1 ¼ ð2n þ 1Þp; N2 ¼ mp; an ¼ 1 þ N21 bo ; am ¼ 1 þ N22 bo ; cn ¼ 1  N22 ð2  bo Þ ; cm ¼
Re Re Re We
and
8 b c b pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 a < cosh 2We n
t þ bnn sinh 2Wen
t ; an P 2N 1 We=Re
n
Hn ðtÞ ¼ exp  t     pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ðA:2Þ
2We : cos bn t þ cn sin bn t ; an < 2N1 We=Re
2We bn 2We

where
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
We We
bn ¼ a2n  4N21 ; bn ¼ 4N21  a2n ;
Re Re
8  h       i
> 2bo N 22 We
< exp  c2m t cos 2We bm
t þ abmm  Re m
b
bm
sin 2We t ; Cm < 0
Hm ðtÞ ¼ h    i ðA:3Þ
> 2
: Tm;2 þ bo N2 expðTm;1 tÞ  Tm;1 þ bo N2 expðTm;2 tÞ =Tm ; Cm P 0
2

Re Re

where
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
We b c b c b We
Cm ¼ a  2
m 4N22 ; Tm;1 ¼ m  m ; Tm;2 ¼ m  m; Tm ¼ m ; bm ¼ a2m  4N22 ;
Re 2We 2 2We 2 We Re
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
We
bm ¼ 4N22
  a2m : ðA:4Þ
Re
For the analytical solution of the elastic stress tensor sp may be obtained by Eq. (A.1) and the way [38] of solving the vis-
coelastic stress tensor under shear flow. The sxy p is given below:
(    )
xy U0 t 4 X 1
2U 0 X1
sp ðy; tÞ ¼ ð1  bo Þ ð0:5  yÞF þ 1  exp  þ cosðN1 yÞK n ðtÞ þ cosðN2 yÞK m ðtÞ ;
Re We WeN 21 n¼0 WeRe m¼1
ðA:5Þ

where
 an
exp  2We t
K n ðtÞ ¼  fxn sinðxn tÞ  En cosðxn tÞ  Rn ½En sinðxn tÞ þ xn cosðxn tÞg;
ðEn Þ2 þ ðxn Þ2
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
if an < 2N1 We=Re; ðA:6Þ

or
" ! ! #,
We bo N 21 We bo N21
K n ðtÞ ¼ Tn;2 þ expðTn;1 tÞ  Tn;1 þ expðTn;2 tÞ Tn
1 þ WeTn;1 Re 1 þ WeTn;2 Re
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
if an P 2N1 We=Re; ðA:7Þ
3852 T. Jiang et al. / Applied Mathematical Modelling 35 (2011) 3833–3853

and
 am
exp  2We t
K m ðtÞ ¼ fxm sinðxm tÞ  Em cosðxm tÞ  Rm ½Em sinðxm tÞ þ xm cosðxm tÞg
ðEm Þ2 þ ðxm Þ2
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
if am < 2N2 We=Re; ðA:8Þ
or
" ! ! #,
We b N2 We b N2
K m ðtÞ ¼ Tm;2 þ o 2 expðTm;1 tÞ  Tm;1 þ o 2 expðTm;2 tÞ Tm
1 þ WeTm;1 Re 1 þ WeTm;2 Re
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
if am P 2N2 We=Re; ðA:9Þ
where
n
b m
b an1 am 1
xn ¼ ; xm ¼ ; En ¼ ; Em ¼
  ;
2We 2We We
2We 2We We
an N2 b am 2
N b b an
Rn ¼  1 o; Rm ¼  2 o ; Tn;1 ¼ n  ;
2Wexn Re 2Wexm Re 2We 2We
bn an b
Tn;2 ¼   ; Tn ¼ n :
2We 2We We

Appendix B

The extra stress tensor sp of Eq. (9) for the XPP model [34,35] can be defined as
r a0
f ðk; sp Þsp þ k0b sp þG0 ½f ðk; sp Þ  1I þ sp  sp ¼ 2k0b G0 D; ðB:1Þ
G0
where the function f(k,sp) is given by
  !
k0b mðk1Þ 1 1 a0 Isp sp
f ðk; sp Þ ¼ 2 e 1 þ 2 1 ; ðB:2Þ
k0s k k 3G20

here, the k0b and k0s are the orientation and backbone stretch relaxation time respectively, G0 is the linear relaxation mod-
ulus, I refers to the trace of a tensor. In the XPP model, the backbone stretch k is related to the viscoelastic stress tensor
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
jIs j
k¼ 1þ ; ðB:3Þ
3G0

where the symbol ‘‘jj ’’ represents the absolute value. The parameter v in the exponential term in Eq. (B.2) is incorporated
into the stretch relaxation time to remove the discontinuity from the gradient of the extensional viscosity. The following
parameters are introduced gp = G0k0b, We = k0bU/L, e = k0s/k0b, where e is the ratio of the stretch to orientation relaxation time
which shows that small values of e correspond to highly entangled backbones [34].
The above equations can be written in the following non-dimensional form
r
h 4
i
f ðk; sp Þsp þ We sp þ 1b
We
0
f ðk; sp Þ  1þka20 k I
; ðB:4Þ
We
þ ð1b s  sp ¼ ð1  b0 ÞD
Þk2 p
0

  (  2 )
2 1 1 We Isp sp
f ðk; sp Þ ¼ emðk1Þ 1  þ þ a0  ; ðB:5Þ
e k k4 ð1  b0 Þ 3k4
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
We jIsp j
where k ¼ 1 þ ð1b 0Þ 3
.
Specially, the viscoelastic fluid model can degenerate the Oldroyd-B fluid when a0 = 0 and f(k,s) = 1 in Eq. (B.1).

References

[1] R.A. Gingold, J.J. Monaghan, Smoothed particle hydrodynamics theory and application to non-spherical stars, Mon. Not. R. Astron. Soc. 181 (1977)
375–389.
[2] L.B. Lucy, A numerical approach to the testing of the fission hypothesis, Astron. J. 83 (1977) 1013–1024.
[3] R. Das, P.W. Cleary, Effect of rock shapes on brittle fracture using smoothed particle hydrodynamics, Theor. Appl. Fract. Mech. 53 (2010) 214–226.
[4] G.L. Chin, K.Y. Lam, Advances in Mesh-less and X-FEM methods, in: First Asian Workshop on Mesh-free Methods, Singapore, 2002.
[5] J.J. Monaghan, Simulating free surface flows with SPH, J. Comput. Phys. 110 (1994) 399–406.
[6] J.P. Morris, P.J. Fox, Y. Zhu, Modeling low Reynolds number incompressible flows using SPH, J. Comput. Phys. 136 (1997) 214–226.
[7] S.J. Cummins, M. Rudman, An SPH projection method, J. Comput. Phys. 152 (1999) 584–607.
T. Jiang et al. / Applied Mathematical Modelling 35 (2011) 3833–3853 3853

[8] S. Shao, E.Y.M. Lo, Incompressible SPH method for simulating Newtonian and non-Newtonian flows with a free surface, Adv. Water Resources 26 (7)
(2003) 787–800.
[9] P.W. Cleary, J.J. Monaghan, Conduction modeling using smoothed particle hydrodynamics, J. Comput. Phys. 148 (1999) 227–264.
[10] J.J. Monaghan, A. Kocharyan, SPH simulation of multi-phase?ow, Comput. Phys. Commun. 87 (1995) 225–235.
[11] J. Ha, P.W. Cleary, M. Prakash, M. Sinnott, Multi-phase and mutli-material flow modeling using smoothed particle hydrodynamics, in: Proceedings of
the II Spheric Workshop, Spain, 2007.
[12] R. Ata, A. Soulaı¨mani, A stabilized SPH method for inviscid shallow water flows, Int. J. Numer. Methods Fluids 47 (2005) 139–159.
[13] L. Oger, S.B. Savage, Smoothed particle hydrodynamics for cohesive grains, Comput. Methods Appl. Mech. Eng. 180 (1999) 169–183.
[14] J.J. Monaghan, SPH compressible turbulence, Mon. Not. R. Astron. Soc. 335 (2002) 843–852.
[15] M. Ellero, M. Kröger, S. Hess, Viscoelastic flows studied by smoothed particle dynamics, J. Non-Newtonian Fluid Mech. 105 (2002) 35–51.
[16] M. Ellero, R.I. Tanner, SPH simulations of transient viscoelastic flows at low Reynolds number, J. Non-Newtonian Fluid Mech. 132 (2005) 61–72.
[17] N.J. Quinlan, M. Basa, M. Lastiwka, Truncation error in mesh-free particle methods, Int. J. Numer. Methods Eng. 66 (2006) 2064–2085.
[18] M.B. Liu, G.R. Liu, Restoring particle consistency in smoothed particle hydrodynamics, Appl. Numer. Math. 56 (2006) 19–36.
[19] W.K. Liu, S. Jun, Y.F. Zhang, Reproducing kernel particle methods, Int. J. Numer. Methods Fluids 20 (1995) 1081–1106.
[20] C.S. Li, X.H. Liu, G.D. Wang, Ring upsetting simulation by the mesh-less method of corrected smooth particle hydrodynamics, J. Mater. Proc. Tech. 183
(2007) 425–431.
[21] Y. Vidal, J. Bonet, A. Huerta, Stabilized updated Lagrangian corrected SPH for explicit dynamic problems, Int. J. Numer. Methods Eng. 69 (2007) 2687–
2710.
[22] G.R. Johnson, S.R. Beissel, Normalised smoothing functions for SPH impact computations, Int. J. Numer. Methods Eng. 39 (1996) 2725–2741.
[23] G.A. Dilts, Moving least square particle hydrodynamics I, consistency and stability, Int. J. Numer. Methods Eng. 44 (1999) 1115–1155.
[24] G.A. Dilts, Moving least-squares particle hydrodynamics II: conservation and boundaries, Int. J. Numer. Methods Eng. 48 (2000) 1503–1524.
[25] J.K. Chen, J.E. Beraun, A generalized smoothed particle hydrodynamics method for nonlinear dynamic problems, Comput. Methods Appl. Mech. Eng.
190 (2000) 225–239.
[26] M.B. Liu, W.P. Xie, G.R. Liu, Modeling incompressible flows using a finite particle method, Appl. Math. Model. 29 (2005) 1252–1270.
[27] J. Fang, A. Parriaux, M. Rentschler, C. Ancey, Improved SPH methods for simulating free surface flows of viscous fluids, Appl. Numer. Math. 59 (2) (2009)
251–271.
[28] G.M. Zhang, R.C. Batra, Modified smoothed particle hydrodynamics method and its application to transient problems, Comput. Mech. 34 (2004) 137–146.
[29] J. Bonet, T.-S.L. Lok, Variational and momentum preservation aspects of smooth particle hydrodynamic formulations, Comput. Methods Appl. Mech.
Eng. 180 (1999) 97–115.
[30] G.L. Vaughan1, T.R. Healy, K.R. Bryan, A.D. Sneyd, R.M. Gorman, Completeness, conservation and error in SPH for fluids, Int. J. Numer. Meth. Fluids 56
(2008) 37–62.
[31] L. Hernquist, N. Katz, TreeSPH-A unification of SPH with the hierarchical tree method, Astrophys. J. Suppl. Ser. 70 (1989) 419–446.
[32] J. Fang, R.G. Owens, L. Tacher, A. Parriaux, A numerical study of the SPH method for simulating transient viscoelastic free surface flows, J. Non-
Newtonian Fluid Mech. 139 (2006) 68–84.
[33] F. Jiang, M.S.A. Oliveira, A.C.M. Sousa, SPH simulation of low Reynolds number planar shear flow and heat convection, Mat.-Wiss. u. Werkstofftech 36
(10) (2005) 613–619.
[34] M. Aboubacar, J.P. Aguayo, P.M. Phillips, T.N. Phillips, H.R. Tamaddon-Jahromi, B.A. Snigerev, M.F. Webster, Modelling pom–pom type models with
high-order finite volume schemes, J. Non-Newtonian Fluid Mech. 126 (2005) 207–220.
[35] Q. Li, J. Ouyang, B.X. Yang, T. Jiang, Modelling and simulation of moving interfaces in gas-assisted injection moulding process, Appl. Math. Model. 35
(2011) 257–275.
[36] G.B. Bishko, O.G. Harlen, T.C.B. McLeish, T.M. Nicholson, Numerical simulation of the transient flow of branched polymer melts through a planar
contraction using the ‘pom–pom’ model, J. Non-Newtonian Fluid Mech. 82 (1999) 255–273.
[37] N.D. Waters, M.J. King, Unsteady flow of an elastico-viscous liquid, Rheol. Acta. 9 (1970) 345–355.
[38] M.F. Tome, N. Mangiavacchi, J.A. Cuminato, A. Castelo, S. Mckee, A finite difference technique for simulating unsteady viscoelastic free surface flows, J.
Non-Newtonian Fluid Mech. 106 (2002) 61–106.
[39] A. Rafiee, M.T. Manzari, M. Hosseini, An incompressible SPH method for simulation of unsteady viscoelastic free-surface flows, Int. J. Non-Linear Mech.
42 (2007) 1210–1223.
[40] T. Jiang, J. Ouyang, B.X. Yang, The SPH method for simulating a viscoelastic drop impact and spreading on an inclined plate, Comput. Mech. 45 (2010)
573–583.

You might also like