Upload
others
View
5
Download
0
Embed Size (px)
Citation preview
Partition of Unity Based Shell Finite Elements
1
Hyungmin Jun
Ocean Systems Engineering, KAISTNov. 4th, 2014
Ph. D. Thesis Presentation
2
Goal
: To develop the efficient and robust shell finite elements based on partition of unity concept.
• Linear and nonlinear static analyses• Shell elements with the effects shear deformation (Reissner-Mindlin theory)
Committee
Prof. PhillSeung Lee, Chair (Ocean Systems Engineering, KAIST) Prof. HyukChun Noh (Civil and Environmental Engineering, Sejong University) Prof. Hyun Chung (Ocean Systems Engineering, KAIST) Prof. YeunWoo Cho (Ocean Systems Engineering, KAIST) Prof. Seunghwa Ryu (Mechanical Engineering, KAIST)
Limitations :
Ph. D. Thesis Presentation
3
Contents : Partition of unity based shell finite elements
1. Part I (Introduction to Partition of Unity (PU) Based FEM) Brief History of the Finite Element Method Application of the Finite Element Method Finite Element Analysis Procedures Ultimate Goal of the Finite Element Method Partition of Unity Based Shell Finite Element
2. Part II (PU Based Shell Element with Improved Membrane Behaviors) Introduction to Nonlinear Finite Element Analysis Introduction and Scope of Research MITC3+ in the Nonlinear Analysis
Key Concepts / Nonlinear Formulation Benchmark Problems
The Method with Improved Membrane Behaviors Comparison with Other Methods Key Concepts / Nonlinear Formulation Benchmark Problems
3. Future Works
4. Conclusions
Enriched MITC3
MITC3+
Enriched MITC3+
4
Part I:
Introduction to Partition of Unity (PU) Based FEM
Brief History of the Finite Element Method Application of the Finite Element Method Finite Element Analysis Procedures Ultimate Goal of the Finite Element Method Partition of Unity Based Shell Finite Elements
5
Brief history – Events in the Finite Element Methods
Events in the Finite Element Methods
1960 1970 1980 1990 2000 2010
1956: First paper by Turner, Clough, Martin and Topp
1960: FEM coined by Clough
1970s: Commercial software(ABAQUS, ANSYS, ADINA)
1982: ANS by MacNeal
1976: Incompatible modeby Taylor
2000: GFEM (PU FEM)by Strouboulis
UC Berkely 1950
1958: UC program using IBM 704 by Wilson
1967: First book by Zienkiewicz
1971: Reduced integrationby Zienkiewicz
1971: First FEAPby Taylor
1990: EAS by Simo
1984: MITCby Bathe
2005: IGAby Hughes
1999: XFEM by Belytschko
1962: Isoparametric element by Taig
1966: Numerical integration by Irons
1970: Curved shell by Ahmad
~1960• Direct Stiffness Method• Force Method
6
FEA software list
Commercial (48)ABAQUSANSYSADINANASTRANALGORLS-DYNAPAM-CRASHMARCLUSASCOMSOLVISUALFEAFEMAP……..
Open source (19)FreeFem++OOFEMjFEMGetFEM++Code Aster…….. Commercial FEA software prices start from around $1,500 to $60,000.
Proprietary / Commercial Finite Element Analysis (FEA) Software
List of finite element software packages – Commercial FEA software
ADINAABAQUS
VISUALFEA
ANSYS
ALGOR NASTRAN
Figure source : google image
7
Automobile Aerospace Electronics Civil engineering
Healthcare Sports Defense Ship building
Applications of FEM #1 – Industry fields
Figure source : google image
8
Olivier et al., Science 2008(Developmental patterning)
Dario et al., Nature Communications 2011(Frequency stabilization)
Diederik et al. Nature Photonics 2013(Disordered photonics)
McHenry et al. PNAS 2007(Predatory behavior in Smilodon)
Giorgio et al., Nature Communications 2012(Microscopic magnetic simulation)
Sampathkumar et al. Cell Current Biology 2014(Plant development)
Anderson et al. Cell Theory 2006(Tumor morphology)
Masson et al. PNAS 2006(Ionic contrast terahertz)
Applications of FEM #2 – Research fields (High profile journal)
Figure source : journals
9
A Study on the Finite Element Method
Of course, more research is neededdespite the remarkable developments and improvements.
ANSWER :
Many researchers and their contributions over 50 years
More than 10,000,000 papers to improve performance of the FEMs
Most widely used numerical methods in engineering
Stabilized commercial software products (e.g. ABQUS, ANSYS, ADINA,…)
Is there nothing left to improve FEMs?CLUE
10
FE analysis procedures – Step 1: Observation of the system
Step 1: Observation of the system
This eccentric load is perpendicular to the column, but it doesn’t pass through the column’s centroid.
Column under an eccentric compressive load
Figure source : google image
11
FE analysis procedures – Step 2: Idealization of the system
Step 2: Idealization of the system
Major assumption : Linear elastic material
Imposing the boundary condition
Appling an eccentric load
Idealization
See the dissertation, p. 121
12
FE analysis procedures – Step 3: Discretization of the domain
Step 3: Discretization of the domain (using ADINA founded in 1974, ADINA 9.0.5 latest released version (2014.10))(setting the material properties, imposing the boundary condition, and applying loads)
FE discretization
13
FE analysis procedures – Step 4: Solving the problem and interpreting the result
Step 4: Solving the problem and interpreting the result
Result : deformed shape of the column
More accurate result
See the dissertation, p. 122
14
The cause of this problem – In-plane shear locking
In-plane shear locking : The element has an excess of shear strain which contributes to the poor ability of the element to reproduce bending modes.
2D pure bending problem 2 2 2 2
2 2( , ) 1 12 2Ma r Mb sv r s
EI EIa b
( , ) Mu r s rsEI
0xyu vy x
Correct beam deformation
( , )u r s urs
0v
xy ur
Parasitic in-plane shear strain
Finite element deformation
See the dissertation, p. 7
15
Remedies – Books and manual books
Amazon search450 results in science & math, keyword : “finite element method”
Commercial software : ADINAManual book : 12 books
16
The ultimate goal of the finite element method #1 – User conveniences & Accuracy
Accurate solution & User conveniences
One click Reliable results
Controllability
What does “user conveniences” mean?
17
The ultimate goal of the finite element method #2 – User conveniences & Accuracy
Step 2: Idealization of the system
Step 3: Discretization of the domain
Step 4: Solving the problem
Step 1: Observation of the systemG
ener
al p
roce
dure
Add
ition
al p
roce
dure
Estimating the FE errors
Locally enrichment without mesh refinements
Accurate solution
See the dissertation, p. 50
18
The ultimate goal of the finite element method #3 – Error estimates
“A Finite element solution contains enough information to estimate its own error”
Cook RD et al., Concepts and applications of finite element analysis, 4th ed., Wiley.
Two types of error estimates serve very different purpose
A priori error estimates To check the order of convergence of a given FE method
A posteriori error estimates To indicate where the error is particularly high
I. Residuals – based methodBabuska I and Rheinboldt WC, Int. J. Numer. Meth. Eng.;12:1597-615, 1978 (1371 cited)
II. Recovery – based methodZienkiewiz OC and Zhu JZ, Int. J. Numer. Meth. Eng.;24:337-57,1987 – ZZ method (2147 cited)
“Nowadays, a booming activity in error estimation is fostered by better understanding of mathematical foundations”
Aninsworth M, Oden JT, A posteriori error estimation in finite element analysis, Wiley
Without mesh refinements
Without using transition elements
p-refinement >> h-refinement
Shell elements alleviating locking
Combining with adaptive mesh refinement
Partition of unity based shell finite element(Enriched MITC3 shell element)
19
The ultimate goal of the finite element method #4 – Error estimates
Hyang et al., DDMSE;78:53-74,2011
Error estimate / Adaptive mesh refinement
Disadvantages
Mesh refinement algorithms Only h-refinement 2D / 3D solid problem Transition elements
If shell problem,how can treat the locking phenomenon?
See the dissertation, p. 24
5-nodeelement
4-nodeelement
4-nodeelement
20
The ultimate goal of the finite element method #4 – Error estimates
Hyang et al., DDMSE;78:53-74,2011
Error estimate / Adaptive mesh refinement
Disadvantages
Mesh refinement algorithms Only h-refinement 2D / 3D solid problem Transition elements
If shell problem,how can treat the locking phenomenon?
See the dissertation, p. 24
5-nodeelement
4-nodeelement
4-nodeelement
21
3
1( ) i i
iu h u
xFEM
3
1( ) ( )i i
iu h u
x xPU based FEM : local approximation function( )iu x
ih : shape function
2 3( ) ( ) ( )i i i i i iu u x x a y y a x
3
1
3
2 3
3
21
1
13
3
( )
( ) ( )
( ) ( )
( )
ˆ
i ii
i i i i i i
i ii i i ii i
i
ii
u h u
h u
h x x a h y y a
x x a y y a
h u
uu
x x
: nodal displacement variableiu
T( ) ( )iu x p a x
Local approximation ,
T ( ) [1, , , , ]x y xyp x
1 2 3 4( ) [ , , , , ]i i i ia a a aa x
1 2 3( , )i i i i i i i i iu x y a x a y a u
1 2 3i i i i i ia u x a y a
1 2 3( )i i i iu a xa ya x
( )iu x
PU based Finite Element Method – Displacement interpolation
See the dissertation, p. 26
Use of the high order interpolation Local use
22
, ,i , ,i1 1 ˆ ˆ2 2ij i j j i j je g u g u g u g u
, ,i12ij i j je g u g u
The linear terms of the enriched covariant strain components
MITC3
1
2
rt
st
e a cs
e a cr
ˆ( )i i
i ir r
x x xg g
PU based MITC3 shell element – Locking treatment
MITC6
1 1 1
2 2 2
rt
st
e a b r c s
e a b r c s
Convergence study
# of elements
Dec
reas
e of
the
erro
rs
PU based MITC3 shell element(also called enriched MITC3)
23
Jeon HM, Lee PS, Bathe KJ. The MITC3 shell finite element enriched by interpolation covers. Comput Struct 2014;134:128-42.
Fine mesh
Coarse mesh
PU based MITC3 shell element – “Highly sensitive” shell problem
24
Journal paper – The MITC3 shell finite element enriched by interpolation covers
25
The ultimate goal of the FEMs – PU based shell finite elements
Locally enriched by the PU based shell elements
Conventional FEM processes
Error estimation
Reliable results
Con
trolla
ble
Fini
te E
lem
ent M
etho
d
BiologistEngineerChemistry Analysis of nan
Biological phenEngineering stru
Black box
26
Part II:
PU based Shell Element with Improved Membrane Response)
Introduction to Nonlinear Finite Element Analysis Introduction and Scope of Research MITC3+ Shell Element in the Nonlinear Analysis
Key Concepts / Nonlinear Formulation Benchmark Problems
The Method with Improved Membrane Behaviors Comparison with Other Methods Key Concepts / Nonlinear Formulation Benchmark Problems
27
• Geometric nonlinear analysis• Material nonlinear analysis• Changing in boundary condition
Classification
Linear analysis Nonlinear analysisCantilever beam subjected to end shear force
Introduction to nonlinear FEA #1 – Linear analysis vs Nonlinear analysis
4
1, 10,1.2 10 , 0.34.5
b h lE vP
28
Introduction to nonlinear FEA #2 – Linear analysis vs Nonlinear analysis
LINEAR ANALYSIS NONLINEAR ANALYSIS
1. Strain Infinitesimal strain 12
jiij
j i
uue
x x
Green-Lagrange strain 12
k k
i j
jiij
j i
uue
x xxu u
x
2. Rotation (curved beam/shell)
Infinitesimal rotation t t i t t i t i
n t n V Q V
t t i t t it 3 t
Q I Θ
3 2
3 1
2 1
00
0
t t i t t it t
t t i t t i t t it t t
t t i t t it t
Θ
▶ 2 1i i
i i V V
Finite rotation t t i t t i t i
n t n V Q V
t t i t t i t t i 2t 3 t t
1 ( )2!
Q I Θ Θ
3 2
3 1
2 1
00
0
t t i t t it t
t t i t t i t t it t t
t t i t t it t
Θ
▶ 22 2
11 ( )2
t t ii i n
i t ii i V V V
3. Framework
-
◇ Total Lagrangian formulation ◇ Update Lagrangian formulation ◇ Corotational formulation
4. Iterative scheme
-
○ Full Newton-Raphson method ○ Modified Newton-Raphson method ○ BFGS matrix update method
29
Three types of approaches : to improve the nonlinear finite element solution
Introduction to nonlinear FEA #3 – Approaches to improve nonlinear finite element scheme
1. Using the high performance computing
2. Obtaining the Equilibrium path
3. Development and improvement of the finite elements (present research)
30
Read geometric / material information
Assemble loads
Calculation of stresses
Solving the linear equation( 1) ( ) ( 1)
0 0t i i t t t i K U R F
Update position vector( ) ( 1) ( )t t i t t i i U U U
Assemble internal force vector
0
T 00 0t t ij
ijVd V F B S
Assemble tangential stiffness matrix
0
T 00 0t ijkl
L ij klVC d V K B B
0
00t t ij
NL ij oVS d V K N
End program
Start program
New
ton
Rap
hson
pro
cedu
re
No
No Inc. loads
Convergence
Incr
emen
tal l
oad
step
pro
cedu
reGeometric nonlinear FE procedure
Three types of approaches #1 – High performance computing
Two quad core Intel Xeon X5560Core : 8 running at 3.2 GHzRAM : 48 GB of DDR31333MHzParallel library : OpenMP
NVIDIA GeForce GTX480Core : 480 CUDA, 1401 MHzRAM : 1.5 GB of video memoryParallel library : CUDA
Dick et al., Simul Model Pract Th 2011;19:801-16.Karatarakis et al., Comput Methods Appl Mech Engrg 2014;269:334-55.
High performance computing
31
Read geometric / material information
Assemble loads
Calculation of stresses
Solving the linear equation( 1) ( ) ( 1)
0 0t i i t t t i K U R F
Update position vector( ) ( 1) ( )t t i t t i i U U U
Assemble internal force vector
0
T 00 0t t ij
ijVd V F B S
Assemble tangential stiffness matrix
0
T 00 0t ijkl
L ij klVC d V K B B
0
00t t ij
NL ij oVS d V K N
End program
Start program
New
ton
Rap
hson
pro
cedu
re
No
No Inc. loads
Convergence
Incr
emen
tal l
oad
step
pro
cedu
reGeometric nonlinear FE procedure
Load control / Displacement control (Batoz and Dhatt, 1979) Limit point analysis – Arc length method
(Riks(1972), Crisfiled(1980)) Others – Automatic load step(Ramm, 1982), Line search(Box 1969)
Obtaining the equilibrium path
Three types of approaches #2 – Obtaining the equilibrium path
32
Read geometric / material information
Assemble loads
Calculation of stresses
Solving the linear equation( 1) ( ) ( 1)
0 0t i i t t t i K U R F
Update position vector( ) ( 1) ( )t t i t t i i U U U
Assemble internal force vector
0
T 00 0t t ij
ijVd V F B S
Assemble tangential stiffness matrix
0
T 00 0t ijkl
L ij klVC d V K B B
0
00t t ij
NL ij oVS d V K N
End program
Start program
New
ton
Rap
hson
pro
cedu
re
No
No Inc. loads
Convergence
Incr
emen
tal l
oad
step
pro
cedu
reGeometric nonlinear FE procedure
S3, S3R, S4, S4R, SAX1, SAX2, SAX2T, SAXA1n, SAXA2n, STRI3, S4R5, STRI65, S8R, S8RT, S8R5, S9R5
SHELL28, SHELL41, SHELL43, SHELL51, SHELL57, SHELL61SHELL63, SHELL91, SHELL93, SHELL99, SHELL143, SHELL150SHELL157, SHELL163, SHELL181, SHELL185
Axisymmetric shell, DISP10 Collapsed MITC3, MITC4, MITC4i, MITC6, MITC8, MITC9, MITC16
QUAD4, QUAD8, QUADR, TRIA3, TRIA6, TRIAR, TRIA, CONEAX, RTRPLT
Different types of shell elements in commercial software
Three types of approaches #3 – Development and improvement of the finite elements
33
Validation of the nonlinear FE code – CODE vs ADINA
Commercial software (ADINA) CODE (C / C++ / Fortran)
MITC4 shell element Full Newton-Raphson method 100 load step
Deformed shapes• CODE (C++)• TECPLOT / PARAVIEW
34
Shell element = membrane element + plate element
Locking problemsin 3-node shell elements
Membrane locking
Transverse shear locking
Transverse shear locking
• In-plane shear locking• Trapezoidal locking• Volumetric locking
In-plane shear locking Trapezoidal locking
Introduction to locking phenomenon – 3-node triangular shell element
35
Newly developed shell elements – MITC3+ and enriched MITC3+ shell elements
Two types of newly developed triangular shell elements
Membrane locking
Transverse shear locking
In-plane shear / Trapezoidal / Volumetric locking
MITC3+PU based MITC3+(Enriched MITC3+)
MITC3+ Enriched MITC3+
1. BASIS
Original MITC3 shell element
MITC3+ shell element
2. METHODS
Cubic bubble function
Cubic bubble function
Partition of unity approximation
4. FEATURE
Improved bending behaviors
Improved bending / membrane behaviors
36
MITC3+ shell element – Key concepts
Lee Y et al., The MITC3+ shell element and its performance, Comput Struct 2014;138:12-23 Excellent convergence behavior Only linear analysis conditions have been considered
3 3
2 11 1
( , , ) ( , ) ( , )( )2
i ii i i i i i
i i
r s h r s a h r s
u u V V-∑ ∑
3 4
2 11 1
( , , ) ( , ) ( , )( )2
i ii i i i i i
i i
r s h r s a f r s
u u V V-∑ ∑
Original 3-node shell element
3-node shell element with the bubble function
Suggested new assumed strain fields MITC3+ shell element
37
MITC3+ shell element – Nonlinear formulation
Incremental Green-Lagrange strain tensor components :
Displacement interpolation with finite rotations :
3 42 2
2 11 1
1( , ) ( )2 2
t i t i t ii i i i i i i i n
i i
th r s a f
u u V V V
0 , , , ,1 1( ) ( )2 2
t t t t t t t tij i j i j i j i j i jε g g g g u g g u u u
Kinematics
nonlinear part in the GL strain
Iterative solution procedures
( 1) ( ) ( 1)0 0t i i t t t i K U F
)()1()( iittitt UUU
With the full Newtorn-Raphson iteration scheme, the equation for the i-th iteration in a finite element model are
for the displacement,
3 4
2 11 1 2
i ii i i i i i
i i
th a f
u u V Vcf. linear analysis
38
Nonlinear performances of the MITC3+ element #1 – Slit annular plate problem
Slit annular plate problem
6
6, 10, 0.03, 3.2
21 10 , 0i eR R h P
E v
Element Description
MITC4 4-node quadrilateral shell element(Bathe, 1986)
MITC3 3-node triangular shell element(Lee, 2004)
MITC3+ 3-node triangular shell elementwith a cubic bubble function (present)
Reference 9-node quadrilateral shell element
39
Hemisphere shell problem
Nonlinear performances of the MITC3+ element #2 – Hemisphere shell
7
10, 0.04, 4006.825 10 , 0.3
R h PE v
Element Description
MITC4 4-node quadrilateral shell element(Bathe, 1986)
MITC3 3-node triangular shell element(Lee, 2004)
MITC3+ 3-node triangular shell elementwith a cubic bubble function (present)
Reference 9-node quadrilateral shell element
40
Plate under end shear force Plate subjected to end moment Clamped semi-cylindrical shell
Nonlinear performances of the MITC3+ element #3 – Other problems
Other benchmark problems solved
41
Newly developed shell elements – MITC3+ and enriched MITC3+ shell elements
Two types of newly developed triangular shell elements
Membrane locking
Transverse shear locking
In-plane shear / Trapezoidal / Volumetric locking
MITC3+PU based MITC3+(Enriched MITC3+)
MITC3+ Enriched MITC3+
1. BASIS
Original MITC3 shell element
MITC3+ shell element
2. METHODS
Cubic bubble function
Cubic bubble function
Partition of unity approximation
4. FEATURE
Improved bending behaviors
Improved bending / membrane behaviors
42
In plane shear locking – Bilinear quadrilateral finite element
2 2 2 2
2 2( , ) 1 12 2Ma r Mb sv r s
EI EIa b
( , ) Mu r s rsEI
( , )u r s urs
0v
0xyu vy x
xy ur
Parasitic in-plane shear strain
In-plane shear locking : The element has an excess of shear strain which contribute to the poor ability of the element to reproduce bending modes.
Pure bending 2D problem
Correct beam deformation
Finite element deformation
43
Comparison with other methods #1 – Introductions
The methods to improve membrane behaviors
Reduced Integration method Incompatible mode Additional bubble node Assumed Natural Strain(ANS) method Drilling degrees of freedom Enhanced Assumed Strain(EAS) method Discrete Shear Gap (DSG) method ….
Partition of unity approximations
44
1 1
1 11 1
( ) ( , )M M
ij i ji j
I f r,s dsdt W f r s
Gauss quadrature
Full integration
( , )u r s urs
0v
xy ur Reduced integration
(Zienkiewicz OC et al., 1971)
r
s
r
s
2 hour–glass modes
Hourglass control (Flanagan & Belytschko, 1981)
Propagation of hour-glass modes through a mesh
Geometric-invariantQuadrilateral element
Comparison with other methods #2 – Reduced Integration method
45
2 24
5 62 21
1 1i ii
r su N u u ua b
2 24
5 62 21
1 1i ii
r sv N v v va b
Incompatible modes (Wilson(1971), Taylor(1976))
Incompatible modes
Only 4-node quadrilateral element Spurious energy in nonlinear analysis
(Sussman T and Bathe KJ, Comput Struct, 2014)
Comparison with other methods #3 – Incompatible modes
46
Distortion sensitivity test
Partition of unity approximation
3 3
2 31 1
( ) ( ) ( )i i i i i i i ii i
u h u h x x a h y y a
x
Standard term Additional term
3-node triangular element
Geometric invariant
Pass basic tests
Accurate results than other methods
)( )) ˆ((u x xu u x3
1
( ) i ii
u x h u
3
2 31
ˆ( ) ( ) ( )i i i i i ii
u x h x x a h y y a
Comparison with other methods #4 – Partition of unity approximation
Advantages
47
Proposed remedy to improved membrane response – Partition of unity approximation
12
xx
yy
xy
yz
zx
u ux xv vey y
eu v
ey x
ev w
e z yw ux z
3 4
2 11 1
ˆ( )2
i ii i i i i i
i i
tu h u a f u
V V-∑ ∑3 4
2 11 1
ˆ( )2
i ii i i i i i
i i
tv h v a f v
V V-∑ ∑3 4
2 11 1
( )2
i ii i i i i i
i i
tw h w a f
V V-∑ ∑
3 3
T2 1
1 1
( ), with2
i ii i i i i i
i i
th a h u v w
u u V V u-∑ ∑
Original 3-node displacement interpolation
PU based MITC3+(Enriched MITC3+)
MITC method(transverse shear locking)
MITC3+ shell element
High order interpolation(enhanced membrane)
PU approximation
PU based MITC3+ shell element(Enriched MITC3+)
48
Proposed remedy to improved membrane response – Partition of unity approximation
Incremental Green-Lagrange strain tensor components :
Finite rotations :
3 42 2
2 11 1
1 ˆ( , ) ( )2 2
t i t i t ii i i i i i i i n
i i
th r s a f
u u V V V u
3 4
2 11 1
ˆ2
i ii i i i i i
i i
th a f
u u V V u
0 , , , ,1 1( ) ( )2 2
t t t t t t t tij i j i j i j i j i jε g g g g u g g u u u
Kinematics
nonlinear part in the GL strain
Iterative solution procedures
( 1) ( ) ( 1)0 0t i i t t t i K U F
)()1()( iittitt UUU
With the full Newtorn-Raphson iteration scheme, the equation for the i-th iteration in a finite element model are
for the displacement,
Linear
Nonlinear
49
Performances of the PU based MITC3+ element #1
Cantilever beam subjected a tip moment
20, 1, 101200, 0.2
l b h ME v
50
Performances of the PU based MITC3+ element #2
Curved cantilever beam
1 27
4.12, 4.32, 0.1
1 10 , 0.25
R R h
E v
Element Description
MITC4 4-node quadrilateral shell element(Bathe, 1986)
MITC3 3-node triangular shell element(Lee, 2004)
MITC3+ 3-node triangular shell elementwith a cubic bubble function (Jeon, 2014)
Enriched MITC3+ PU based MITC3+ (present)
Reference 9-node quadrilateral shell element
51
Performances of the PU based MITC3+ element #3
Hemisphere shell problem
8
50, 25, 0.25, 90
4.32 10 , 0zL R t f
E v
Element Description
MITC4 4-node quadrilateral shell element(Bathe, 1986)
MITC3 3-node triangular shell element(Lee, 2004)
MITC3+ 3-node triangular shell elementwith a cubic bubble function (Jeon, 2014)
Enriched MITC3+ PU based MITC3+ (present)
Reference 9-node quadrilateral shell element
52
Papers – Nonlinear MITC3+ / Enriched MITC3+ shell elements
The Enriched MITC3+ shell elementwill be submitted in CMAMENonlinear analysis of the MITC3+ shell element
published in Computers and Structures, Jan 2015
53
Future works #1 – PU based finite element method with the FE error estimation
Integrating the PU based FEM with the error estimator
Mesh generation Adaptive mesh refinement algorithm Error estimates Coupling method or indicator
54
4-node partition of unity based finite shell element : Linear dependency problem
Future works #2 – PU based finite element method with reduced integration
2D solid element : No locking treatments due to the high order interpolation
Shell element :Absolutely needs to reduce the transverse shear locking even using the high order interpolation• MITC4-MITC9-MITC16 – Too much computational cost• MITC4-Reduced Integration – Effectively decrease in computational time• Reduced Integration – Reduced Integration – Much faster than others
8-node serendipity element
Hourglass mode
ˆe e e
MITC4Reduced Integration
MITC9 / MITC16Reduced Integration
55
Future works #3 – PU based shell element with high performance computing
“Parallel programming is not a trivial task in most programming languages, and demands a great theoretical knowledge about the hardware architecture and good programming skills” - Neto DP-
Windows Cluster
GPU Cluster
WCU Cluster
Linux Cluster
Partition of unity based shell finite element
( ) ˆu uu x
3
2 31
( ) ( )ˆ i i i i i ii
h x x a h y y au
3
1i i
i
h uu
“The computing of each local matrix is totally independent and so parallelization of these computations is straightforward and can be naturally explored”
56
We proposed two possibilities of using the partition of unity approximation
Increase solution accuracy without local mesh refinements (Part I) Improvement membrane behaviors (part II)
I. PU based MITC3 shell finite element (Enriched MITC3) is reviewed.
This element is obtained by applying linear displacement interpolation covers to the standard 3-node shell element and MITC procedures are used.
The method increases the solution accuracy without any local mesh refinement. We can provide convenience for user by combining PU shell element and FE error
estimates.
II. PU based MITC3+ shell finite element (Enriched MITC3+) is proposed. This shell element is based on the MITC3+ shell element and partition of unity
approximations to improve membrane response. Membrane behaviors of the triangular shell element efficiently is improved by
partition of unity approximation.
Conclusions
57
Conclusion remarks – Most downloaded papers
58
Conclusion remarks – Frontier in the analysis of shells
59
Previous research (2003 ~ 2012)
Virtual heart (NRL project)
Virtual cancer system (NRL project) Angiogenesis and vascular tumor growth
Arterial network with the hemodynamic modelCoronary artery generationK
angw
onN
atio
nal U
nive
rsity
KA
IST,
SB
IE
60
Education & Publications
Education• Ph.D. Student : Korea advanced Institute of Science and Technology (Sep 2010 - present)• MS : Kangwon National University (Sep 2006 – Aug 2008)
(Thesis : Development of a cell-system coupled model of cardiovascular hemodynamics)• BS : Kangwon National University (Mar 2000 – Aug 2006) (Summa Cum Laude & Early Graduation)
International Journals (published : 4)1. Jeon HM, Yoon K, Lee PS. A partition of unity based triangular shell element with improved membrane response, In preparation2. Lee Y, Jeon HM, Lee PS, Bathe KJ. On the behavior of the 3-node MITC triangular shell elements, In preparation.3. Jeon HM, Lee Y, Lee PS, Bathe KJ. The MITC3+ shell element in geometric nonlinear analysis, Computers and Structures, 146, 91-
104, Jan 2015.4. Jeon HM, Lee PS, Bathe KJ. The MITC3 shell finite element enriched by interpolation covers, Computers and Structures, 134, 128-
142, Apr 2014.5. Shim EB, Jun HM, Leem CH, Matusuoka S, Noma A. A new integrated method using a cell-hemodynamics-autonomic nerve control
coupled model of the cardiovascular system. Progress in Biophysics and Molecular Biology, 96(1-3), 44-59, Jan 2008.6. Jun HM, Shim EB. Theoretical analysis of the cross-bridge sliding rate in modulating heart mechanics, International Journal of
Vascular Biomedical Engineering, 5(2), 34-45, Oct 2007.
Presentation (27)• Jeon HM, Yoon K, Lee PS. Development of the enriched MITC3 shell element. KSME Annual Conference, 192-193, Apr 2014.• …..• Computational study on the arterial tree generation based on blood volume optimization. KSME Annual Conference, Oct 2005.
Experience• URP program, Kangon National University (Sep. 2003 ~ Aug 2006), Dep. of Mechanical Engineering• Visiting Researcher, Kyoto University (Dec 2007- Mar 2008 and Apr 2009- Jun 2009), Graduate School of Medicine• Visiting Researcher, MIT (Aug 2008 – Oct 2008), Dep. of Health Sciences and Technology • Visiting Researcher, Oxford (Jul 2010- Aug 2010), Centre for Mathematical Biology• Research Assistant, KAIST (Jan 2010 – Aug 2012), Dep. of Bio and Brain Engineering, Advisor : Prof. Kwang-Hyun Cho.