You are on page 1of 10

--

: caoer@simwe qq:1694240258

09 simwe --
http://forum.simwe.com/viewthread.php?tid=912004 2010 12 31
simwe 17

--
1 --
2 --
3 --
4 --
5 --
6 --
--

--

05
simwe

simwer

--

1.1 PDE
PDE()
PDE
PDE

PDE

qq:1694240258
qq:1694240258

PDE
plasma

PDE ODE ODE


PDE PDE ODE

1.2 PDE
PDE PDE
PDE

FFT

periodic boundary condition


PDE

PDE
element
(node) 4
4

1.3 Galerkin approximation weak formation/


(weak formation) strong form PDE
weak form weak form
weak form
Galerkin
loop stiffness matrix
gauss quaradure
approximation Galerkin gauss
newton
residual
newton stiffness matrix

stiffness matrix 10x10x10


8 11x11x11 node T
stiffness matrix (11x11x11)x(11x11x11)=1771561

( gaussian elimination )

qq:1694240258
qq:1694240258

Bathe Finite Element Procedure



solver

1.4

stress strain
flux

--

2.1 ANSYS
ansys
ansys abaqus
ansys

ansys apdl ansys


Ansys Swanson

Swanson Swansea Z
Swansea Z 09 Finite
Element Method ansys

2.2 COMSOL
comsol femlab2.0

comsol PDE

comsol pde PDE


weak form PDE

qq:1694240258
qq:1694240258

ansys
ansys comsol
umfpack+mpich
umfpack 5.2 comsol umfpack
comsol blasc++ mpi comsol

2.3 FEAP

abaqus ansys abaqus


feap fortran
gui apdl
macro
tecplot. mpi
elastic plastichyperelastic
8.1 PETSc
FEAP FEAP
FEAP
FEAP feap linux
make

2.4 libmesh deal II


deal II
libmeshc++ robust
adaptive meshing
singularity libmesh
20
libmesh libmesh
c++
inux
Libmesh cluster

2.5 ALGOR_pipeline

pipeline


feap libmesh PETSc/blas
comsol ansys/abacus/adina

qq:1694240258
qq:1694240258

3 --

3
4

3.1 Jacobian

8
8
global local jacobian
global reference jacobian
global local
jacobian
jacobian Jacobian
0

3.3
8 gauss
stiffness matrix
4 stiffness matrix 21x21 k

damping damping matrix


3 M*u_tt+D*u_t+K*u=F source
pde
blas MKL
(gotoblas PETSc)

3.4 Ax=b

M*u_tt+D*u_t+K*u=F 2x2
u_tt u_t

newmarks HHTenergy conserved crank nicolson Euler

K_tilde*u=F u

3.5

pde
M1*u1_tt+D1*u1_t+K1*u1=F1 M2*u2_t+K2*u2=F2

qq:1694240258
qq:1694240258

M1*u3_tt+(D1+M2)*u3_t+(K1+K2)*u3=F1+F2
K_tilde*u3=F3

K_tilde Unsymmetric matrix

3.6

Dirichlet boundary condition


>1e10 u u0
natrual boundary condition q_n q_n
q_n source natural
boundary condition
q_n {F}

4 --

ansys 100

3d

()
pde
Pde

4.1 u1u2u3
small deformation finite
deformation ansys
small finite

displacementmixedenhanced strain displacement

pde 6 4

locking buckling convergence

qq:1694240258
qq:1694240258

cpu
cache data layout
small deformation
/

4.2 T.
pde T
T T 1 heat flux = gradient of T
Fourier heat conduction .

4.3 u1 u2 u3 1 2 3

5 u
shell
MITC Bathe 84
( degenerated solid method MITC
100


von karman nonlinear
plate/shell theory 50
nonlinear nonlinear shell/plate

4.4 :u1u2u3123
pde
u 4
Moment3 shear stress
2

simwe
pde
mesh

qq:1694240258
qq:1694240258

5 --

Ax=b


solver

5.1
gaussian elimination pivoting

Factorization
Factorization LU Factorization
ACholesky Factorization
simwe Cholesky QR
Factorization
gaussian elimination Factorization backsubstitution

5.2
relaxation, incomplete
factorization, multigrid
Jacobi
Gauss-Seidel relaxation
Jacobi Gauss-Seidel
Preconditioner
P^{-1}Ax = P^{-1}b,
P preconditioner, P 0
P Preconditioned Conjugate
Gradients(PCGs)
Krylov Krylov
General minimal residual (GMRES) multigrid Krylov
GMRES FGMRES .GMRES
GMRES 2

5.3
PDE quasi-static method Midpoint
static Method PDE thermal or diffusion equation Backward Euler
implicit methodbackward difference formular methed (BDF) Generalized midpoint method
BDF PDE

qq:1694240258
qq:1694240258

NEWMark method HHT alpha method

5.4 newton
-
x_new = x + f(x)/f'(x) ||x_new-x||
f(x) residual
residual x residual
x residual x residual
newton

6 --

6.1
cpu fem
fem
cpu
latency blocks ghost node/halos communication
time Nvidia
Fermi GPU
speedups
2-3

6.2

diffusiondislocation

Z
6 Monte
Carlo

qq:1694240258
qq:1694240258

--

simwe xx

simwe

simwe

caoer@simwe
--
--
--
--
--

qq:1694240258