Introduction to Finite Element...
Transcript of Introduction to Finite Element...
Vincent Chiaruttini, Georges CailletaudVincent Chiaruttini, Georges [email protected]@onera.fr
Non Linear Computational Mechanics Athens MP06/2012–Non Linear Computational Mechanics Athens MP06/2012–
Introduction to Finite Element computationsIntroduction to Finite Element computations
2 FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–
OutlineOutline
Continuous to discrete problemStrong to weak formulation
Galerkin method for approximate solution computation
Isoparametric finite elementsFinite element mesh
Geometrical element
Interpolation of displacements
FE method for linear elastic problemsVariational formulation with isoparametric elements
Elemental computations
Global problem
Linear solution process
FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–
Continuous mechanical problemContinuous mechanical problem
Solving PDE on a given time-space domain
Studied domain
Local equations
Boundary conditions
u = ud on @u
¾ ¢ n = F on @f
@u
@f
Unable to get exact solutionUnable to get exact solutionHow to obtain an approximated solution?How to obtain an approximated solution?
r¾ + f = 0
"(u) =1
2
¡ru+rT u
¢
¾ = K : "
FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–
Continuous mechanical problemContinuous mechanical problem
Local equations
Admissible spaces for displacements
Variational formulation of the displacement problemFinding that verifies, for all
r¾ + f = 0¾ = K : "
" =1
2
¡ru+rT u
¢
Z
¾ : "¤ ¡ f:u¤d =Z
@F
F:u¤dS
u = ud on @u
¾ ¢ n = F on @f
Uad = fuju continuous and regular on ; u = ud on @ug
U0ad = fuju continuous and regular on ; u = 0 on @ug
u¤ 2 U0adu 2 Uad
, a(u; u¤) = L(u¤)
FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–
Continuous mechanical problemContinuous mechanical problem
Potential energy definition
Minimization problemLax-Milgram theorm insures
Variational formulation = weak formulation on a stationary functional
Galerkin based methodsGalerkin based methodsApproximate minimization problemApproximate minimization problem
P (v) = 1
2
Z
"(v) : K : "(v)d¡Z
f:vd¡Z
@F
F:vdS
P (v) = 1
2a(v; v)¡L(v)
u 2 Uad veri¯es 8u¤ 2 U0ad; a(u; u¤) = L(u¤), u = argminv2UadP(v)
FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–
From Galerkin method to FE methodFrom Galerkin method to FE method
Galerkin methodFinding the solution of a variational formulation in an admissible subspace ofUsing polynomial base functions defined on the domain, for instance
The variational formulation produces a NxN linear system
The approximate solution is optimal in the sense that it constitutes the projection of the solution on the given subspace (ie the error is orthogonal to the given subspace)
Advantages/drawbacksEasy to define a regular base (polynomials)Optimal approximationNon-sparse operators, not so easy for complex geometries
Mesh and compact supportMesh and compact supportbase functions => FE methodbase functions => FE method
uD(x) 2 Uad Ák(x) 2 U0adv(x) = uD(x) +NX
k=1
®kÁk(x)
Uad
8h 2 f1; : : : ;Ng;NX
k=1
a(Ák; Áh)®k = L(Ák)
FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–
Some popular discrete methods for PDESome popular discrete methods for PDE
Finite differences
Strong formulation of PDE
Differential approximations
Regular grid
Finite volume
Integral approximations
Very popular for conservation laws
Structured or unstructured mesh
Finite elements
Variational formulation
Optimal solution for a given approximation space (Galerkin approach)
Free meshes, efficient a priori and a posteriori error estimation
Ti-1 Ti Ti+1
h
dTi/dx=(Ti+1-Ti-1)/(2h)
Flux
8 FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–
OutlineOutline
Continuous to discrete problemStrong to weak formulation
Galerkin method for approximate solution computation
Isoparametric finite elementsFinite element mesh
Geometrical element
Interpolation of displacements
FE method for linear elastic problemsVariational formulation with isoparametric elements
Elemental computations
Global problem
Linear solution process
FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–
Isoparametric finite elementIsoparametric finite element
MotivationIndustrial need for a robust and efficient numerical tool to obtain accurate results in the simulation of mechanical problems with complex behaviour and geometries
Meshing processGiven a geometrical model (usually from CSG built using a CAD software)
235 parts45 parts
How to obtain a suitable discrete representationHow to obtain a suitable discrete representationfor FE computations?for FE computations?
FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–
Finite element meshFinite element mesh
A 2D example triangular elements–A continuous domain is approximated using a conform partition based onrectilinear triangular regions=> a triangular Mesh
To insure convergence of the solution process the elements size can be reducedIf the studied domain boundary is polygonal, it is possible to get
Interpolation functionsRepresent a scalar field on the domain using nodal values
= h
Ti
h
Ti
T1
T2
12
x 2 in element T
Can functions TCan functions Tii be less dependant of be less dependant of
the mesh?the mesh?x
T (x)
T3
3
Th(x) = N1(x) T1 +N2(x)T2 +N3(x)T3
Th(x)
FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–
Deformed and reference configurationsDeformed and reference configurations
Parametrization of an element positioned anywherein the studied space (linear triangular example)
Physical element Reference element
are base or shape functions defined on the reference domain for the geometrical representation of the element, usually linear:
»1
»2
x »
Ni(»)
X3
X1
X2
x = N1(»)X1 +N2(»)X2 +N3(»)X3
8<:N1(»1; »2) = 1¡ »1 ¡ »2N2(»1; »2) = »1N3(»1; »2) = »2
1(0; 0) 2(1; 0)
3(0; 1)
FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–
Deformed and reference configurationsDeformed and reference configurations
Parametrization of an element positioned anywherein the studied space (quadratic triangular example)
Physical element Reference element
Reference to physical transformation regularity ( nodes element)
The transformation between the reference and the physical configurations must be a bijection, the Jacobian must not reach the zero value.Misappropriate physical element geometry can produce such failureElement integration , if N is polynomial then J is also
»1
»2
x »
1 2 3X1
X2
X3
X4X5
X64
5
6
x =
nX
k=1
Nk(»)Xk J(») = detJ(»)
dV (x) = J(»)dV (»)
n
J(») =
·@xi@»j
¸=
"nX
k=1
@Nk@»j
(»)Xkj
#
FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–
Interpolation of displacementsInterpolation of displacements
Isoparametric elementThe same base functions are used for geometry and displacement interpolation
where are the vectorial nodal displacementsUkuh(x) =
nX
k=1
Nk(»)Uk
FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–
Operational computation of displacementsOperational computation of displacements
Local vector of displacement unknowns for element with nodes A vector of elementary degrees of freedoms is usually built, for a 2D problem by
where is the nodal displacement in direction i at node j Thus for any point inside the specified element of coordinate in the reference system, the displacement is obtained by
where is a matrix built using the base function, in 2D:
To insure correct convergence properties (ie representation of constant gradient and solid body displacement field), the following “partition of unity” property must be verified anywhere in the element:
U(j)i
x »
e
Ue =hU(1)1 ; U
(1)2 ; U
(2)1 ; U
(2)2 ; : : : ; U
(j)i ; : : :
iT
Ne(»)
N(») =
·N1(») 0 N2(») 0 : : : Nj(») 0 : : :0 N1(») 0 N2(») : : : 0 Nj(») : : :
¸
nX
k=1
Nk(») = 1
uh(x) = Ne(») ¢ Ue
n
FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–
Operational computation of displacement gradientsOperational computation of displacement gradients
Vectorial Voigt writing of second order tensors Symmetric gradients
Linear behaviour relationship as a matrix-vector product
Gradient computationPhysical system of coordinate gradient definitionTo link with the reference coordinate
with
using we getSymmetric gradient computation
That can be expressed as using the symmetric gradient operator
¾v = [A] "v
duh(x) = ruh(x) ¢ dx
dx = J(») ¢ d»H(») = [Hij ] =
"nX
k=0
@Nk@»j
(»)U(k)i
#duh(x) = H(») ¢ dx
ruh(x) = H(») ¢ J¡1(»)
"v =h"11 "22 "33
p2"12
p2"13
p2"23
iT
¾v =h¾11 ¾22 ¾33
p2¾12
p2¾13
p2¾23
iT
²vh(») = B ¢ Ue
²h[u(»)] =
1
2
³H(») ¢ J¡1(») +
¡H(») ¢ J¡1(»)
¢T´
16 FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–
OutlineOutline
Continuous to discrete problemStrong to weak formulation
Galerkin method for approximate solution computation
Isoparametric finite elementsFinite element mesh
Geometrical element
Interpolation of displacements
FE method for linear elastic problemsVariational formulation with isoparametric elements
Elemental computations
Global problem
Linear solution process
FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–
Finite element method for linear elasticityFinite element method for linear elasticity
Global algorithm from the variational formulation in displacementContinuous problem
Finite element discrete problem
@u
@f
Uad = fuju continuous and regular on ; u = ud on @ug
8u¤ 2 U0ad;
Sf
Suh
ek
uh(x) = Nke(») ¢ UeOn each element ek
Uks = Ud(xks)
Verifying null value on Su nodes
Virtual displacement field u¤h(x) = Nke(») ¢ U¤e
Prescribed displacements on Su nodes
Uk¤s = 0
8U j¤;, 8W; [W ]T [K][U ] = [W ]T [F ]
Z
h
"vh ¢ [A] : "v¤h dh =Z
h
f:u¤hdh +Z
@Fh
F:u¤hdS
Z
¾ : "¤d =Z
f:u¤d +Z
@F
F:u¤dS
FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–
Computing elementary contributionsComputing elementary contributions
Domain formulation based on element by element contributions
Usually the two parts are separated in
Internal force contributions
External force contributions
[W ]T [F ] =X
fekg
Z
ek
f:u¤hdV +Z
¡eF
F:u¤hdS
[Fext] =X
fekg
Z
ek
f:u¤hdV +Z
¡eF
F:u¤hdS
[Fint] + [Fext] = [0]
[W ]T [K][U ] =X
fekg
Z
ek
"vh ¢ [A] ¢ "v¤h dV
[Fint] = ¡X
fekg
Z
ek
"vh ¢ [A] ¢ "v¤h dV
FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–
Computing elementary contributionsComputing elementary contributions
Elementary rigidity matricesUsing the FE gradient operators and the local vectors of unknowns we need to compute:
Even for polynomial base/shape function, the inverse of Jacobian used to compute the [B] symmetric gradient operator, make impossible to evaluate such elementary integral exactly for non-trivial meshes. Thus a Gauss integration process is used:
Such process requires the knowledge of some predefined integration points(usually called Gauss points) and the associated weight for the reference configuration of the element.The parametric representation of both the geometry and the displacement field of the element constitutes one of most important aspect to insure the generic aspect of the FE method: only a few patterns of integration scheme are required corresponding to the associated finite element in reference configuration.
Z
ek
"vh ¢ [A] ¢ "v¤h dV = [We]T
Z
ek
[B]T [A][B]dV [Ue] = [We][Ke][Ue]
»g
wg
Z
ek
f(»)dV (») ¼GX
g=1
wgf(»g)
FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–
Computing elementary contributionsComputing elementary contributions
Gauss integration1D example
FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–
Computing elementary contributionsComputing elementary contributions
Gauss integration1D example
FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–
Computing elementary contributionsComputing elementary contributions
Gauss integration1D example
FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–
Computing elementary contributionsComputing elementary contributions
Gauss integration1D example
FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–
Computing elementary contributionsComputing elementary contributions
Gauss integrationIn 3D
FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–
Computing elementary contributionsComputing elementary contributions
External forces
For the volume force a Gauss integration process is requiredFor the nodal force nodal value must be correctly calculated
[Fext] =X
fekg
Z
ek
[W ]T [N ][f ]dV +
Z
¡eF
[W ][N ]T [F ]dS
FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–
Computing elementary contributionsComputing elementary contributions
External forces
FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–
Computing elementary contributionsComputing elementary contributions
External forces
For the volume force a Gauss integration process is requiredFor the nodal force nodal value must be correctly calculated
[Fext] =X
fekg
Z
ek
f:u¤hdV +Z
¡eF
F :u¤hdS
FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–
Computing elementary contributionsComputing elementary contributions
External forces
For the volume force a Gauss integration process is requiredFor the nodal force nodal value must be correctly calculated
[Fext] =X
fekg
Z
ek
f:u¤hdV +Z
¡eF
F :u¤hdS
FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–
Computing elementary contributionsComputing elementary contributions
External forces
For the volume force a Gauss integration process is requiredFor the nodal force nodal value must be correctly calculated
[Fext] =X
fekg
Z
ek
f:u¤hdV +Z
¡eF
F :u¤hdS
FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–
Assembling the global problemAssembling the global problem
Local to global DOF indices (thermal problem for shake of simplicity)
1
2
322
11
33
66 88
1010
99
55
9944
77
E1E2
E3 E4
E5
E6E7
E8E9
E10
E3
Global DOF numbering Elementary DOF numbering
1 2 3 4 5 6 7 8 9 10
3
4
5
6
7
8
1
2
9
10
Binary trace operators
Assembling global matrix
[K] =X
fekg[¦e]
T [Ke][¦e]
Global matrix
[U ] = [¦e]T [Ue][Ue] = [¦e][U ]
Prescribed displacementsWhen assembling is processed, prescribed displacement unknowns can be replaced and eliminated for the linear system
band
wid
th
FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–
Solving efficiently a sparse linear systemSolving efficiently a sparse linear system
[K][U ] = [F ]Large linear system to be solved
sparse with n equations (n from 104 to 109)K is a symmetric, definite positive for a linear-elastic problemReduce bandwidth using an accurate numbering process (reduce computational cost) Optimization in memory requirement: sparse storage only keep non zero terms
Direct solversFactorization process with L a lower-triangular matrix
Solving (2 successive triangular systems solving)Interest: matrix storage, less memory consuming than required by inverse computation, triangular system solving is of the same complexity order than matrix vector productOptimized multifrontal, dissection, multithreated, solvers
Iterative solversKrylov type solvers (conjugate gradient, GMRes) need good preconditionnersDomain decomposition solvers
Split the structure in many subdomains (for supercomputers parallel computation)Efficient iterative strategy to achieve interface equilibrium condition(very good preconditionners)
[K ] = [L][L]T
[L][Y ] = [F ] then [L]T [U ] = [Y ]
FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–
Convergence proprietiesConvergence proprieties
Displacement convergencePolynomial approximations are used, on any element Ee, Taylor expansion gives:
Committed error on displacement using p order shape function and h characteristic size
Strain and stress convergenceFE error
Energy normDefined by
For order the FE method is convergent in energy:
and if the problem is regular enough
(x; x0) 2 E2e u(x) = u(x0) +ru(x0):(x¡ x0) + : : :+O(kx¡ x0kp+1)
k"[u](x)¡ "h[u](x)k = O(hp); x 2 Ee
k¾[u](x)¡ ¾h[u](x)k = O(hp); x 2 Ee
kx¡ x0k < h gives ku(x)¡ uh(x)k = O(hp+1); x 2 Ee
kvk2E =Z
"[v] : A : "[v]d
p ¸ 1 ku¡ uhkE ! 0 if h! 0
ku¡ uhkE · C hp kukE
FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–
Global FE solution process for a linear problemGlobal FE solution process for a linear problem
Approximate geometry build a mesh using a priori analysis→
Loop on elements
Compute the local variational formulation contribution
Loop on elements Gauss points
Compute Jacobian, gradient matrices, get behaviour, multiply matrices to get local rigidity contribution
Integrate the rigidity matrix and the local vector for external forces
Assemble local rigidity matrix and local external forces
Apply Dirichlet BC, MPC (linear relationship between unknowns)
Compute global external forces
Solve the linear system
FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–FE analysis for non linear mechanics Athens MP06 - V.Chiaruttini–
ReferencesReferences
Bonnet M., Frangi A. (2006)Analyse des solides déformables par la méthode des éléments finis. Editions Ecole Polytechnique.Belytschko, T., Liu, W., and Moran, B. (2000).Nonlinear Finite Elements for Continua and Structures.Besson, J., Cailletaud, G., Chaboche, J.-L., and Forest, S. (2001).Mecanique non linéaire des matériaux. Hermes.–Ciarlet, P. and Lions, J. (1995).Handbook of Numerical Analysis : Finite Element Methods (P.1), Numerical Methods for Solids (P.2). North Holland.Dhatt, G. and Touzot, G. (1981).Une présentation de la méthode des élements finis. Maloine.Hughes, T. (1987).The finite element method: Linear static and dynamic finite element analysis. Prentice Hall –Inc.Simo, J. and Hughes, T. (1997).Computational Inelasticity. Springer Verlag.Zienkiewicz, O. and Taylor, R. (2000).The finite element method, Vol. I-III (Vol.1: The Basis, Vol.2: Solid Mechanics, Vol. 3: Fluid dynamics). Butterworth Heinemann.–