Professional Documents
Culture Documents
Mesh Refine 内部调整网格
Mesh Refine 内部调整网格
CEA-DAM IDF
email: Philippe.Hoch@cea.fr
OUTLINE
Lagrangian step 1
Conservations Laws :
Dt
+ .u = 0,
Dt u + P = 0,
Dt E + .(Pu ) = 0.
(1)
internal energy.
CONS: The counter parts, the mesh quality can become very bad
very quickly (non convex quads, tangled mesh ... ). The mesh quality
may become very bad very quickly (non convex quads, tangled mesh ... )
Lagrangian Step 2
Staggered (Wilkins type: 1964, and new ones Shashkov and al.)
1
2
3
4
ALE Computation without and with special care to non convex cells
Consequences
1
Need
Rezoning process
Here, we want to use smoothing methods that keep the
same
(cost function).
M1
0.4454
0.4
YAxis
0.3
0.2
0.1
0.01336
1.2
1.3
1.4
1.5
1.56
everywhere!!
K3
i
Triangle ideal
(v)
= AW 1 .
xi = Sv
, with
|Sm ||m |
BUT
2m
m > 0
eme
simplex fm
. . . modication:
Escobar :
avec
Cell quality: q (S )
= S
2
K4
Triangle de reference
(u)
fm =
tr (t SS ),
= 1/q (S ),
|Sm ||m |
, is
m +
PNbSimplexe
fm .
m=1
2 +
m
W
cos ()
sin()
S
= AW1
S8
S9
S 10
S7
S2
S3
S1
S6
S4
S 11
S5
S 12
2
degree(i)
2
degree(i)(number_of _nodes(m) 2)
= 0),
Iter no 1
Iter no 2
iter no 4
iter no 5
= /3),
Escobar ortho
connectivity adaptation) .
iter no 3
iter no 6
Mesh
M1
2.0100
1.0100
7.1109
1.8109
1.0100
2.0100
Mesh M1
Mesh
2.0100
M1
1.67
1.0100
1
4.4109
2.1109
1.0100
2.0100
0.5078
0.4578
1.653
Mesh
M1
2.010
1.0100
0.0100
0.0100
1.0100
2.0100
What to remind
Escobar with adaptation W
Remarque
1
convergence.
2
Lagrangian grid.
never go until
K3
K2
v7
v10
v3
v4
K4
v12
v2
v1
v6
K1
v5
Denition
Qi Nodal Quality of i, 3 ex.:
1
2
3
Qi
min sin(v )
sin,rel =
Qi
max(max sin(v ),| min sin(v )|)
aire ,rel =
min Aire v
Qi
max(max Aire (v ),| min Aire (v )|)
1 Q i 1
' 1: locally optimal
Qi 0 : local degeneracy
Properties:
Qi
Denition
Q
M
.
.
: Quality of M,
2 choices:
M
M
Q = mini Qi
1 PNv M
M
Q1 =
Nv i Qi
M (Qref )
QM
i < Qref .
M >= Qmin,
If Q
arrival times).
et
< 0),
Denition
M1 is better than M2 i:
1
M
If Q1
.
.
.
M
and Q2 > 0
M2 and Nb M1 (Q
Q
M1 >
M1 Q M2 and
M1
M2 and
Q = Q
Q
Q
M
Q1
M
If Q1
.
.
>0
>0
M
and Q2
<0
<0
M
and Q2 < 0
M2 and Nb M1 (Q
Q
M1 >
M1 Q1M2 and
1
Q1
Q1
M
OR
ref ) Nb 2 (Qref ),
M1 (Q ) < Nb M2 (Q ),
OR
ref
ref
M1 (Q ) = Nb M2 (Q ) and Q M1 > Q M2
Nb
ref
ref
1
1
Nb
Nb
M1 (Q
M
ref ) Nb 2 (Qref )
M2 (Q )
)
<
Nb
ref
ref
1 and M 2 (close
i , i th vertex
M1 , Q M2 )
i
w (Qi
of
Mi
3 by :
1
2
= Mi1 + w (QiM , QiM )(Mi2 Mi1 )
Exemples
1
M 1 , Q M 2 ) = 1+Q1 iM 2
i
2+QiM +QiM
w (Qi
M1 , Q M2 ) =
i
w (Qi
0,
1,
M1
( Qi
M1 > Q M2 ,
i
sinon.
si Qi
M 2 6= 1)
and Qi
mesh.
"intersected "
Properties
Properties
1
3 is close to M 1 and M 2 .
M1 > Q M2
i
do not
Practically, we need to test the quality of the third mesh (see. Oxford
2005).
M Q ):
(Qi
Jun).
M Q )
(Qi
(Escobar type).
Remarque
1
sin
area
and Qi
and
N = KA
N
A
A
N
i
j V (KiA ) Ki Kj / j V (KiA ) Ki Kj
[Margolin,Shashkov]:Ki
KiN
Z
QdV
KiA
QdV
j V (K
A)
i
hR
i
R
Q def
= K N K A QdV K A K N QdV
i
j
i
j
Quantity Fij
Fij
(2)
Q = F Q ,
ji
Fij
(3)
iN = |KiA |Q
iA +
|KiN |Q
Positivity condition :
jA |KiA KjN |Q
iA
|KiN KjA |Q
j V (KiA )
P
|KiA | > j V (K A ) |KiA KjN |,
i
gives a local
(4)
regions Fik .:
Algebraic
eA
eN
mg
md
n 2 (e A)
A
e
e N
mg
md
n1 (eA)
n (e )
1
Fik
Positive scheme:
P
|KiA | > | V (Fik )<0 V ( Fik )|,
A e N = {S }
and S
6= {}
A
n 2
(e
)
A
e
e
N
mg
z
2
s
md
z
1
a
(e
Self-intersection ux
The rst order self intersection ux:
nblmt
X(e )
a e
Q=
sign(m, e , k )V (Zk )Qk
k =1
P
A
|Ki | > | V (Z A )<0 V (Zka )| gives a local maximum
Ge
Positivity:
principle
A N
and
(e )A (e )N
and e
N :=
(e )N
.
N
Linear representation (e ):
(e )1,h = (e )j + (e )j (x xj )
(5)
AND
e .
(e )1,h = j ej + ej j (x xj ) + j ej (x xj )
e
1,h
= ej +
and then:
j (x xj )
ej
j + j (x xj )
(6)
We consider 3 options:
I Volumes Fluxing computation (2 options) :
Initial Data
Initial conditions (density/temperature) and speed:
r
=0.001
=0
0.4
1./3.
=1
=1
4./15.
u=0
v=0
0.4
1./3.
=0.125
=0
u=3
v=4
4./15.
0.2
0.2
0.2
0.3
0.4
0.2
Densite ini
0.3
0.4
0.9
4.5
0.9
4
4
0.8
0.7
0.6
0.7
0.5
0.6
0.4
0.3
0.5
0.2
0.4
0.1
3
2
1.5
0.5
xi
xi
0.1
0
1
0.2
2.5
0.3
0.001
0
3.5
ZAxis
ZAxis
0.8
nx
50
50
Flux
0/1
0/1
Limit. maille
0
1
density min/max
0.001/0.953157874
0.001/0.953157874
Temperature min/max
0/0.999775443
0/1.023732158
0.7643
0.9
0.7
0.7
0.9
0.8
0.8
0.7
0.6
0.6
0.7
0.5
0.5
0.4
YAxis
YAxis
0.6
0.6
0.5
0.5
0.4
0.3
0.4
0.4
0.3
0.2
0.2
0.1
0.3
0.1
0.3
0.2214
0.2048
0.3
0.4
0.5
XAxis
0.6
0.7
0.2167
0.2167
0.3
0.4
0.5
XAxis
0.6
0.7
0.781
nx
50
50
50
50
Flux
1
1
0
0
Limit. cell
0
1
0
1
density min/max
0.001/0.999994937
0.001/0.999994937
0.000999984/0.999997086
0.000999984/0.999997086
Temperature min/max
0/0.999999724
0/1.000276494
-0.000008013/0.999999854
-0.000015653/1.000144011
1
0.9
0.8
0.8
0.7
0.7
0.6
0.6
0.5
0.4
YAxis
0.9
0.5
0.4
0.3
0.3
0.2
0.2
0.1
0.1
nx
50
50
50
50
50
50
Flux
1
1
1
0
0
0
Node Limit
-1
0
1
-1
0
1
e vit. finale, ap flux 1, limit maille 0, limit noeud 1 Norme vit. finale, ap flux 1, limit maille 0, limit noeud 0 Norme vit. finale, ap flux 1, limit maille 0, limit noe
4.5
4.5
4
4
3.5
2.5
2
2
1.5
ZAxis
0.5
YA
Z
xis
XAxis
4.8
3.5
4.2
4
2.5
2
3.6
2.4
1.5
1
00
5.4
5.705
ZAxis
4.977
1
00
0.5
YA
xis
XAxis
1.8
1
1.2
00
0.6
YA
xis
XAxis
Conclusion on projection
The projection with the following options gives the best results:
1
self-intersection
Linear limitation on
...)
3
Linear limitation on
t=2
t=1
P1
v=0
Evolution
purement
Lagrangienne
P2
Evolution avec
adaptation non conforme
apres phase Lagrangienne
Project (cont)
Lagrangian step,
Adaptation,
(rst
Welding Line
welding line
This straight line will take place of edges for which the aspect ratio is
under some tol.
A way to obtain a conservative step, is the computation of the
intersection between this line and some (very few) elements.
Strategy
(1)
Lagrange Step:
Convex
OR
non convex
Strategy (Cont)
(2)
Adaptation Step:
(a) See Previous Fig. Orthogonal projection of slaves nodes on the line
dened by master endpoints in the case of non convex cells around
any of slaves (or we can do it each times). Computation of new
conservative quantities over the new moved cells by self intersection
ux (=exact problem).
(b) Renement study with some criteria (geometrical/physic), see details
after. Computation of new conservative quantities is straigthforward
for rst order scheme. To obtain convex sub-cells there are 3 ways:
Figure: Isotropic
1D
Quasi 1D
(c) Unrenement study with some criteria (geometrical/physic), see
details ... Computation of new conservative quantities is done by
clipping (or self intersection...).
Strategy (Cont)
(3)
ALE Step:
(a) Slaves nodes are moved after the movement of Masters endpoints.
Smoothing
0.2528
0.2528
0.2
0.2
0.1
0.1
0
6.41017
1.0101
2.0101
0
6.41017
1.0101
2.0101
0.2742
0.2742
0.2
0.2
0.1
0.1
0
6.01017
1.0101
2.0101
2.7101
0
6.01017
1.0101
2.0101
2.7101
0.2822
0.2822
0.2
0.2
0.1
0.1
0
5.81017
1.0101
2.0101
2.8101
0
5.81017
1.0101
2.0101
2.8101
PLOT
YAxis
PLOT
1.708
YAxis
YAxis
0
0
0
1.101
1.2
1.3
XAxis
1.4
1.5
1.6
1.7
XAxis
PLOT
PLOT
1.595
1.6
1.5
1.4
YAxis
YAxis
1
1.3
1.2
1.1
0.9
1
1.1
1.2
1.3
1.4
XAxis
1.5
1.6
1.7
1.8
0
1.019
1.1
1.2
1.3
XAxis
1.4
1.5
1.6
Prospects
(0) Strategy more or less generic wrt Lagrangian hydro (at least
centered scheme of Despres/Mazeran and Ph Maire/Breil).
(1) For renement, the AMR-ALE is recover as a special case among the
3 generic cases that we propose here (genericity for any convex
polygonal cells).
(2) Test and implement a simplied computation of a generic building of
welding line for unrenement step.
(3) Improving tests for adaptation... Help us for good error estimators
(anisotropic) !!