Sie sind auf Seite 1von 36

Application of the unified

semi-analytical wall boundary


conditions to incompressible
SPH

A. Leroy1 , D. Violeau1 , C. Kassiotis1 , M. Ferrand2

1
Laboratoire dHydraulique Saint-Venant
EDF R&D, Cetmef, cole des Ponts ParisTech, France
2
MFEE department, EDF R&D, France

8th Spheric International Workshop, Trondheim (Norway), 3-6 June 2013


Introduction

Context
ISPH (incompressible SPH) performs well in predicting
pressure fields
However, von Neumann pressure wall BCs are difficult to
prescribe with dummy particles
In 2010, Ferrand et al. proposed new semi-analytical wall
BCs for SPH

2 / 22
Introduction

Context
ISPH (incompressible SPH) performs well in predicting
pressure fields
However, von Neumann pressure wall BCs are difficult to
prescribe with dummy particles
In 2010, Ferrand et al. proposed new semi-analytical wall
BCs for SPH

Goal of this work


adapting semi-analytical wall BCs to the resolution of the
pressure Poisson equation

2 / 22
SPH operators in the framework of
semi-analytical wall boundary conditions
a, b F : free particles
P =F V
v V : vertex particles
ns
s S : segments

s v see Ferrand et al., 2012

3 / 22
SPH operators in the framework of
semi-analytical wall boundary conditions
a, b F : free particles
P =F V
v V : vertex particles
ns
s S : segments

s v see Ferrand et al., 2012

SPH divergence of a field A at particle a :


1 X 1 X
Da {Ab } = mb Aab wab + s Aas as
a a a a
bP sS

where Aab = AaAb , wab = wh (ra rb ),


Z Z rv 2
a = wh (ra r0 )d 2 r0 and as = wh (ra rs )ns dS
a rv 1

3 / 22
SPH operators in the frame of unified
semi-analytical boundary conditions

SPH antisymmetric gradient of a field A at particle a:


   
,+ a X Aa Ab a X Aa As
Ga {Ab } = mb + wab s + as
a 2a 2b a 2a 2s
bP sS

The SPH symmetric gradient written in the same way is noted G,


a .

4 / 22
SPH operators in the frame of unified
semi-analytical boundary conditions

SPH antisymmetric gradient of a field A at particle a:


   
,+ a X Aa Ab a X Aa As
Ga {Ab } = mb + wab s + as
a 2a 2b a 2a 2s
bP sS

The SPH symmetric gradient written in the same way is noted G,


a .

SPH Laplacian of a field A at particle a with a diffusion coefficient B:

1 X Aab 2 X
La {Bb , Ab } = Vb (Ba + Bb ) 2 rab wab Bs (A)s as
a rab a
bP sS

4 / 22
Analytical computation of a in 2D

a can be written as the integral of a flux through the boundary


(see Feldman and Bonet, 2007):

kr a r 0 k
Z   X
(r a r 0 )n r 0 dS r 0 = 1
 
a = q= as
h
sS
Z
(q) (r a r 0 ) dS r 0

with: as + ns
s
q 5
 
1  5q
(for the 5th order Wendland kernel, (q) = 2 2
1 1+ + 2q 2 for q 6 2)
2h q 2 2

The as are calculated analytically, so that we obtain an exact


value for a (see the paper).

5 / 22
Navier-Stokes equations for an incompressible
flow

The Navier-Stokes equations in the case of an incompressible


flow read:
u=0


du = 1 p + (E u) + Fe

dt

2
where p = p + k and E = + T
3
(k : turbulent kinetic energy, T : eddy viscosity)

6 / 22
Navier-Stokes equations for an incompressible
flow

The Navier-Stokes equations in the case of an incompressible


flow read:
u=0


du = 1 p + (E u) + Fe

dt

2
where p = p + k and E = + T
3
(k : turbulent kinetic energy, T : eddy viscosity)

We use a projection method to solve this system.

6 / 22
The ISPH algorithm

ua = una + t La {E,b , unb } + Fea


 
1 Prediction of the velocity

7 / 22
The ISPH algorithm

ua = una + t La {E,b , unb } + Fea


 
1 Prediction of the velocity

a
2 La {1, pbn+1 } = D {u } Pressure Poisson equation (PPE)
t a b

7 / 22
The ISPH algorithm

ua = una + t La {E,b , unb } + Fea


 
1 Prediction of the velocity

a
2 La {1, pbn+1 } = D {u } Pressure Poisson equation (PPE)
t a b

t ,+ n+1
3 un+1 = ua G {pb } Correction of the velocity
a
a a

7 / 22
The ISPH algorithm

ua = una + t La {E,b , unb } + Fea


 
1 Prediction of the velocity

a
2 La {1, pbn+1 } = D {u } Pressure Poisson equation (PPE)
t a b

t ,+ n+1
3 un+1 = ua G {pb } Correction of the velocity
a
a a

4 rn+1 = rn + tun+1 Movement

We solve this linear system using Bi-CGSTAB or GMRES.

7 / 22
The ISPH algorithm

ua = una + t La {E,b , unb } + Fea


 
1 Prediction of the velocity

a
2 La {1, pbn+1 } = D {u } Pressure Poisson equation (PPE)
t a b

t ,+ n+1
3 un+1 = ua G {pb } Correction of the velocity
a
a a

4 rn+1 = rn + tun+1 Movement

We solve this linear system using Bi-CGSTAB or GMRES.

7 / 22
Solving the new PPE : Laplacian of the
pressure

n+1
2 X pab 1 X
La {1, pbn+1 }= Vb 2 rab wab (p)n+1
s as
a rab a
bP sS

imposition of a boundary condition on the pressure

8 / 22
Imposition of the boundary condition on the
pressure
Let s be a segment. We have:

un+1
s us 1
= psn+1
t s

9 / 22
Imposition of the boundary condition on the
pressure
Let s be a segment. We have:

un+1
s us 1
= psn+1
t s

We impose that s S, un+1


s = uwall
s :

s
psn+1 ns = (u uwall
s ) ns
t s

9 / 22
Imposition of the boundary condition on the
pressure
Let s be a segment. We have:

un+1
s us 1
= psn+1
t s

We impose that s S, un+1


s = uwall
s :

s
psn+1 ns = (u uwall
s ) ns
t s
so we have:
s
(p)n+1 as = (u uwall
s ) as
s
t s

9 / 22
Treatment of the boundary correction terms

Finally, we put the boundary term in the right-hand side and


solve a modified PPE:
n+1
2 X pab +a gzab a
Vb rab wab = D {u }
a 2
rab t a b
bP !

2 X us uwall
s
+ s +g as
a t
sS

La {pbn+1 +b gzb } = Ba
The unknowns are the pbn+1 . Note that 2 (gz) = 0.

10 / 22
Treatment of the boundary correction terms

Finally, we put the boundary term in the right-hand side and


solve a modified PPE:
n+1
2 X pab + a gzab a
Vb rab wab = D {u }
a 2
rab t a b
bP !

2 X us uwall
s
+ s +g as
a t
sS

La {pbn+1 +b gzb } = Ba
The unknowns are the pbn+1 . Note that 2 (gz) = 0.

10 / 22
Instability and free-surface treatments

Instability: application of the algorithm proposed by Lind et al.


(2012):
shift the particles positions according to:
rn+1
a rn+1
a + ra with ra = Cshift h2 Ca

1 X 1 X
where: Ca Ga {1} = Vb wab as
a a
bF sS

correct the velocity, turbulent kinetic energy and dissipation rate, e.g.:

un+1
a un+1
a + G, n+1
a {ub } ra

11 / 22
Instability and free-surface treatments

Instability: application of the algorithm proposed by Lind et al.


(2012):
shift the particles positions according to:
rn+1
a rn+1
a + ra with ra = Cshift h2 Ca

1 X 1 X
where: Ca Ga {1} = Vb wab as
a a
bF sS

correct the velocity, turbulent kinetic energy and dissipation rate, e.g.:

un+1
a un+1
a + G, n+1
a {ub } ra

Free-surface: a particle is identified as belonging to the free-surface


if Da {rb } 1.5 (Lee et al., 2007). Then its pressure is set to 0.

11 / 22
SPH k  turbulence model

The k  turbulence model can be integrated in this framework


(Ferrand et al., 2012) with:

ka2
T ,a = C
a
dka 1 T ,b
= Pa a + La {b + , kb }
dt a k
da a 1 T ,b
= (C1 Pa C2 a ) + La {b + , b }
dt ka a 

where C , k , C1 , C2 and  are model constants.

12 / 22
Lid driven cavity (Re = 1000): velocities

u+
-0.4 -0.2 0 0.2 0.4 0.6 0.8 1
0.4 1

0.2 0.8

0 0.6

v+

z+
-0.2 Ghia et al., v+ 0.4
Ghia et al., u+
-0.4 ISPH 0.2
WCSPH
FV
-0.6 0
0 0.2 0.4 0.6 0.8 1
x+

13 / 22
Lid driven cavity (Re = 1000): velocities

1%
u+
L2 error

-0.4 -0.2 0 0.2 0.4 0.6 0.8 1


0.4 1

0.1 % 0.2 0.8


1st order
2nd order
0 0.6
ISPH

v+

z+
WCSPH
0.01 % -0.2 Ghia et al., v+ 0.4
1e-02 1e-01 Ghia et al., u+
h/L ISPH
-0.4 0.2
WCSPH
FV
-0.6 0
0 0.2 0.4 0.6 0.8 1
x+

13 / 22
Lid driven cavity (Re = 1000): pressure

0.2 2
ISPH ISPH
0.15 WCSPH WCSPH
0.1 FV 1.5 FV
0.05
0 1
+

+
p

p
-0.05
-0.1 0.5
-0.15
0
-0.2
-0.25
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
x+ x+

Computational time for 60s of physical time (on 1 CPU)

Model Number of cells/part. Time

FV 512 512 38 h
ISPH 200 200 31 h
WCSPH 200 200 32 h

14 / 22
Water collapse on a wedge
2000 VoF

Pressure force (N per m)


WCSPH
1500 ISPH

1000

500

0
0 1 2 3 4 5 6
t+
ISPH simulation Pressure force on the left side of the
wedge vs. time
Computational time for 2s of physical time (on 1 CPU)

Model Number of cells/part. Time

3
VoF 6.e > 1h
ISPH 6.e3 20 min
WCSPH 6.e3 30 min

15 / 22
PERIODICITY
Water wheel

PERIODICITY

16 / 22
Water wheel: comparison with VoF

1.2
ISPH

Pressure force (kN per m)


1 WCSPH
VoF
0.8
0.6
0.4
0.2
0
0 10 20 30 40 50 60 70 80 90

VoF ISPH t+
Pressure force on one bucket vs. time

Computational time for 30s of physical time (on 1 CPU)

Model Number of cells/part. Time

4
VoF 8.e 5 days
ISPH 3.e4 15 h
WCSPH 3.e4 18.5 h

17 / 22
Turbulent channel flow (Re = 640)
Velocity profile
20

15

u+
10

5 DNS Kawamura et al.


ISPH
FV
0
10 100
z+
k profile  profile
5 120
DNS Kawamura et al. DNS Kawamura et al.
ISPH 100 ISPH
4 FV FV
80
3
k+

+
60
2
40
1 20
0 0
0 100 200 300 400 500 600 0 100 200 300 400 500 600
+
z z+

18 / 22
Fish-pass

1.5
0.132
z (m)

0.612
0.5
0.3

0
0 0.5 1 1.5 2 2.5 3
x (m)

Computational time for 20s of physical time

Model Number of cells/part. Time

4
FV 2.5e 2.8 h (4 CPUs)
ISPH 6.e4 49 h (1 CPU)
WCSPH 6.e4 38 h (1 CPU)

19 / 22
Profiles after 20s
Velocity Pressure
2 2
ISPH ISPH
WCSPH WCSPH
1.5 FV 1.5 FV
z (m)

z (m)
1 1

0.5
0.5

0
0 0.5 1 1.5 2 2.5 3 3.5 0
5000 5500 6000 6500
|u| (m s-1)
k p (Pa)
2

1.5
z (m)

0.5 ISPH
WCSPH
FV
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
k (m2s-2)

20 / 22
Conclusions and future work

The proposed modified ISPH model allows to :


model complex free-surface and confined, turbulent or
laminar flows presenting arbitrary wall boundary
geometries
predict correctly the kinematic fields and pressure

21 / 22
Conclusions and future work

The proposed modified ISPH model allows to :


model complex free-surface and confined, turbulent or
laminar flows presenting arbitrary wall boundary
geometries
predict correctly the kinematic fields and pressure

Future work :
Extend this work to 3D on a GPU code in order to perform
simulations of realistic cases
Model active scalars effects in case of turbulent or laminar
flows

21 / 22
Thank you for your attention.

22 / 22

Das könnte Ihnen auch gefallen