Recent and future developments in finite element metal ... · ICTP'2014 - JL Chenot et al 1 Recent...
Transcript of Recent and future developments in finite element metal ... · ICTP'2014 - JL Chenot et al 1 Recent...
ICTP'2014 - JL Chenot et al 1
Recent and future developments in finite element metal forming
simulationJ.-L. Chenota,b, M. Bernackib, P-O Bouchardb,
L. Fourmentb, E. Hachemb, E. PerchataaTransvalor, 694 Avenue du Dr. Maurice Donat, 06255 Mougins Cedex, France
bCEMEF, Mines Paristech, B.P. 207, 06904 Sophia Antipolis Cedex, France
ICTP'2014 - JL Chenot et al 2
Short historical introduction
FE simulation of metal forming processes: in 2D : end of 60’s and 70’s (Japan, USA, UK). in 3D : 80’s (France, Japan, USA).
Main steps 1995: automatic remeshing, 1996: element quality control, 1997: parallel computing, 2002: heat treatment, quenching, 2003: deformable tools, 2005: coupling with metallurgy , 2007: thermal regime in tools, multi body forming, 2009: process optimization, 2014: three-dimensional modeling of induction heating.
ICTP'2014 - JL Chenot et al 3
Summary
Mechanical and thermal formulations
Space and time discretization
Resolution procedures
Coupled approaches
Optimization and identification
Computation at the Micro Scale
Future challenges
ICTP'2014 - JL Chenot et al 4
Mechanical and thermal formulations
Updated Lagrangian, Eulerian or ALE
Constitutive modeling
Contact, friction and wear
Damage
Heat equation
Updated Lagrangian, Eulerian or ALE
Classical integral formulation:
For steady state processes (e. g. rolling, extrusion, wire drawing, etc.) A fixed domain Ω is considered, Free surface must be determined, Stress and memory variables are computed along
streamlines, with SUPG, etc.
ICTP'2014 - JL Chenot et al 5
c
v*dV dV v dS 0ργ σ ε τΩ Ω ∂Ω
+ − =∫ ∫ ∫: * *ɺ
Inertia terms can often be neglected:
For non steady state processes, Updated Lagrangian:
Parameters update :
ALE is a compromise Numerical velocity material velocity
ALE derivative:
ICTP'2014 - JL Chenot et al 6
t tc
t t t tdV v dS 0σ ε τ+∆ +∆
Ω ∂Ω
− =∫ ∫: * *ɺ
t t t t
t t t t
t t t t
x x tv
t
t
σ σ σε ε ε
+∆
+∆
+∆
= + ∆
= + ∆
= + ∆
ɺ
ɺ
ALEv v≠
( ) ( )ALEALE
dv v grad
dt
ε ε ε= + −ɺ
Constitutive modeling
Viscoplasticity (Perzyna):
Elastoviscoplasticity: Additive decomposition:
Elastic law:
Anisotropic behaviour: Hill, Barlat, etc.
Compressible materials: Gurson, Abouaf et al.…
ICTP'2014 - JL Chenot et al 7
11
ε 1 σ σp
mK R K− ′= −/ ( ) /ɺ
e pε ε ε= +ɺ ɺ ɺ
2e eJdtr( )
dt
σ λ ε µε= +ɺ ɺ
Contact, friction and wear
Contact condition between bodies a and b:
treated by penalty
Friction: example of Coulomb Norton law:
Wear: generalized Archard law:
ICTP'2014 - JL Chenot et al 8
( - ). . 0a bv v n v n∆= ≤
1 fpf f n v vτ µ σ= ∆ ∆ -
- ( ) /
W
nW W m
V
VK dt
H
σδ ⋅ ∆= ⋅ ⋅∫
Modelling Ductile fracture
Phenomenological approaches: Failure criteria maximum stress : Latham & Cockcroft,
triaxiality ratio : Oyane,
& Lode angle : Bao & Wierzbicki …
Continuum Damage Mechanics:
Kachanov, Lemaitre …
Micromechanical approaches: Void growth models:,
Porous plasticity modelsICTP'2014 - JL Chenot et al 9
Cup and cone defect
Heat equation
ICTP'2014 - JL Chenot et al 10
0t t t t
V ncTwdV kgrad T grad w dV q wdV wdSΩ Ω Ω Ω
ρ φ+ − + =∫ ∫ ∫ ∫ɺ ɺ( ) ( )
Non-linear and fully coupled with mechanical equations:
• ρ, c, k depend on T,
• Non-linear boundary conditions,
• is the plastic work,
• Constitutive equation depends on T.
• Possibility of strain localization: e. g. adiabatic shear band
Vqɺ
Space and time discretization
Tetrahedral elements
Time discretization
Space-time finite element
Remeshing and adaptive remeshing
Anisotropic remeshing
Remapping
ICTP'2014 - JL Chenot et al 11
ICTP'2014 - JL Chenot et al 12
Tetrahedral elements
P1+P1 tetrahedral elements
Discretization in term of:
velocity v, or displacement increment ∆u,
Pressure p,
and temperature T
Non linear equations:R V P T 0
S T V 0
( , , )
( , )
==
pressure pvelocity v
Mechanical equation
Heat equation
Time discretization Classical incremental approach: First order Second order :
Runge et Kutta, “Crank and Nicholson” scheme:
“Acceleration method” for quasi static formulation Time differentiatation of the integral equation
-> linear equation in γ
Space-time elements Slices of space-time are discretized Possibility of local time steps
Necessitated 3D or 4D elements
ICTP'2014 - JL Chenot et al 13
2t t t t t tx x t( v v ) /+∆ +∆= + ∆ +
Remeshing and adaptive remeshing
“Geometric” remeshing by iterative method (T. Coupez et al.)
Iterative local improvements, Restore possible degeneracy and maximize element quality, Control contact or penetration / tools, Parallel implementation.
Adaptive remeshing (L. Fourment) : Error estimation: modified Zienkiewicz energy norm method:
Map of element size, Remeshing respecting size and quality, Use of iterative remeshing method with weighting factors, Prescribed accuracy or maximum number of nodes.
ICTP'2014 - JL Chenot et al 14
1/2
( ) : ( )h h h h dη σ σ ε εΩ
= − − Ω ∫ ɶɶ
Application to sheet blanking (L.Fourment et al.)
ICTP'2014 - JL Chenot et al 15
ICTP'2014 - JL Chenot et al 16
Application to sheet blanking (L.Fourment et al.)
Anisotropic remeshing Basic principle:
Introduce a local metric defined by the tensor , Use the same remeshing algorithm.
Computation of the metric tensor: Error estimation contribution:
Skin adaptation:• Compute normal vectors,• define mesh size along• normals:
ICTP'2014 - JL Chenot et al 17
M
21
22
23
1 0 0
0 1 0
0 0 1
e
/ h
M / h
/ h
′ =
2
1s
s
M n nh
′ = ⊗
Curvatures adaptation:
Metric tensor:
ICTP'2014 - JL Chenot et al 18
212
22
0 0 01
0 1 0
0 0 1
cM / R
/ Rα
′ =
e s cM M M M= + +
Isotropic case: 138 840 nodes
Anisotropic case: 18 770 nodes=> divided by 7
ICTP'2014 - JL Chenot et al 19
Isotropic: 4 procs - 1h 20mn Anisotropic: 4 procs – 40 mn
=> divided by 2 Equivalent or better accuracy
Thickness without remeshing
Thickness with remeshing
Application to deep drawing
ICTP'2014 - JL Chenot et al 20
Application to rolling
Comparison: anisotropic 8 786 nodes / isotropic 61 474 nodes, => divided by 7,Equivalent accuracy close to the cylinders,Higher thickness accuracy.
21
( ) ( )k
f f Nk k
=∑ ɶx x
M
P0
f
M
P1
M2
P0
fɶ f
Mapping SPR
( ) kfk
=ɶ x P.a
Interpolation
Nodecentered
patch
( )
( )( )2
4,
( )
( )
, ( )
( )
kk k k
g k k g g
k g g
f fk k g k g
kfk
f f
P f f
f fMIN
∇
=
= +∇ −
∀ ∈ =
− ∑ −
ɶ ɶ
ɶ
ɶ ɶ ɶ
ɶ
ɶ
x
x P.a
x f x x
x x
x
x x
k
Use of Super convergent patch recovery method
Remapping (L. Fourment et al.)
ICTP'2014 - JL Chenot et al
Initialmesh
finalmesh
Resolution procedures
Linearization of the equations Localized contact with the tools Parallel computing Multi mesh Multigrid
ICTP'2014 - JL Chenot et al 22
Linearization of the equations
Integral equations , with
Newton-Raphson iterative procedure:
Several linear systems to solve with unknown
Possibility of minimization of the residual / α :
ICTP'2014 - JL Chenot et al 23
Z (U ,P)=0R( Z )=1n n nZ Z Zδ+ = +
( )1 0 0n n n nR( Z ) R( Z ) R( Z ) / Z Zδ+ = ⇒ + ∂ ∂ ≅nZδ
n nR( Z Z )αδ+
Localized contact with the tools
The distance to the tool must be positive or null:
Penalty form:
Improvements: Penalty can be imposed at each Newton-Raphson iteration, Second order smoothing of the surface : Nagata
Formulations: Master and slave, Quasi symmetric (L. Fourment et al)
ICTP'2014 - JL Chenot et al 24
( ) ( ) ( )2( ) ( ) .t t t t t t t t t tn n n n n tool n n
dM M M t O t v V n t
dt∆ ∆ δδ δ δ ∆ ∆ δ ∆+ += = + + ≈ + −
( ) ( )2
1.
2c
n pencontact n tool n n
n
V V v n StΩ
δ δϕ ρ
∆
+−
∈∂
+= − −
∑
25
Upper tool
Ω3Ω2
Ω1
Lower tool
Forge
Forge
Forge
P2
P3
P1
Forge Parallel
Interface data files
Parallel computing
Iterative solverLU Preconditionning Domain decomposition
26
Algorithm for parallel solving
kxA
Np
kk Ω
=∑=
1
xA .Product Matrix / Vector in parallel :
Computation on each domainCommunication for updating interfaces
Scalar Product in Parallel : ∑ ∑
⋅⋅=
DomainSub NodesnodesNodesNodes
wvuvu ),(
Heavy computation distributed on sub domainsLimited communication
Np : number of processors
ICTP'2014 - JL Chenot et al 27
Computation speed-up from 4 to 24 cores
Multi mesh (L. Fourment et al.)
Basic principle: - refined mesh for heat equation: TM, - less refined mesh for mechanical equation: MM, - mesh for other parameters (equivalent strain, etc.). - transfer data between meshes for coupling.
Benefit: Adapt mesh refinement to problems, Save computer time. Compatibility with parallel computing.
ICTP'2014 - JL Chenot et al 28
29
Application to Cogging
Thermal Mesh
Mechanical Mesh
29
Configuration Nb elementsThermal Mesh
Nb elementsMechanical Mesh
Derefinement rate Speed-up
Beginning 53 500 13 800 3.9 9.2
End 66 300 13 900 4.8 18.1
Test of efficiency (single processor)
ICTP'2014 - JL Chenot et al
6th ISPF – J-L Chenot 30
Multigrid method (K. Mocellin, L. Fourment)
2 or 3 levels of grids: Automatic de-refinement by a local operator.
Data transfer between grids. For ring rolling: speed up about 7.
61 987 nodes 6 061 nodes 790 nodes10,2τ = 7,6τ =
Multigrid new developments (in progress): More general algorithm, Adaptation to parallel computation.
6th ISPF – J-L Chenot 31
Coupled approaches
Thermal and mechanical coupling Multi material coupling Induction heating Coupling with micro structure evolution
ICTP'2014 - JL Chenot et al 32
Thermal and mechanical coupling Mechanical and heat equations :
Solve mechanical and thermal eqs. separately;
First :
Then :
Possibility of iteration : fix point.
ICTP'2014 - JL Chenot et al 33
t t t t t t
t t t t
R V P T 0
S T V 0
+∆ +∆ +∆
+∆ +∆
=
=
( , , )
( , )
t t t t t t t t tR V P T 0 V P+∆ +∆ +∆ +∆= →( , , ) ,
t t t t t tS T V 0 T+∆ +∆ +∆= →( , )
For strong coupling (e. g. Adiabatic Shear Band)
iterative scheme may not converge,
solve a global system - Newton-Raphson method
on V, P, T.
Example of Cutting
Equivalent strain
ICTP'2014 - JL Chenot et al 34
Multi material coupling
See presentation:
Finite Element simulation of multi material metal forming
By J.-L. Chenot, C. Béraudo, M. Bernacki, L. Fourment
ICTP'2014 - JL Chenot et al 35
Induction heating (F. Bay, J. Barlier et al.)
Maxwell equations are simplified:
ρ density of charge
V potential
A : magnetic vector potential with magnetic field
µ : magnetic permeability
Heating: with electric current density
ICTP'2014 - JL Chenot et al 36
10
1 1 1
( V )
A( A) V
t
ρ
ρ µ ρ
∇ ∇ =
∂ + ∇× ∇× = ∇∂
B A= ∇×
2EM elecw jσ=ɺ j V= ∇
Induction heating FE solution of the electromagnetic equations:
Classical discretisation for potential V,
Special tetrahedral edge elements for A,
Necessity of coupling with heat equation,
And metallurgical evolution.
Global mesh for:
Inductor,
Workpiece,
Air.
ICTP'2014 - JL Chenot et al 37
Application to heating of a crank shaft
ICTP'2014 - JL Chenot et al 38
Kinematics Austenite
Quenching
ICTP'2014 - JL Chenot et al 39
% martensite
Hardness
ICTP'2014 - JL Chenot et al 40
Dis
loca
tion
Den
sity
Gra
in R
adiu
sR
ecry
stal
lizat
ion
ratio
Flo
w S
tres
s
Determination of the flow stress as afunction of the mean dislocation density
⋅
⋅⋅⋅⋅
+⋅⋅⋅⋅=m
mbmp
AArg
bc
TKMbM
ρρ
ρµασ 52
9
sinh
( ) XX xdm ⋅+−⋅= ρρρ 1
⋅⋅−⋅−⋅−⋅+=
d
ddddd A
ArgAAAAAdt
d
ρρρρρρ 52/5
42/5
2210 sinh
Dislocation evolution model
Temperature = 1100 °C – Strain-Rate = 1 s-1
Recrystallization ratio and grain sizes
Developed at Aachen University for microalloyed steels
Precipitation model for V(C,N)
Coupling with micro structure evolutionExample of the mean-field approach: AFP model
Example : Reducer-rolling simulation
FORGE simulation
MEAN-FIELD APPROACH – AFP MODELX dm
ρm σ
Optimization and identification
Process optimization
Identification of material parameters
ICTP'2014 - JL Chenot et al 42
Process optimization (L. Fourment et al.)
Deterministic optimization Genetic algorithm + Surface response
ICTP'2014 - JL Chenot et al 43
1st operation with optimized die
Second operation
(3)
no defect
piping defect
Identification of material parameters
Data is obtained obtained from: Laboratory experiments, Industrial measurements, And define the set of observables.
Parameters to be identified: Constitutive equations, Friction laws, Damage, Metallurgical laws,…
Minimization of the mean square difference: Genetic algorithm, Kriging meta-model.
ICTP'2014 - JL Chenot et al 44
Sensitivity Analysis Sensitivity of observables wrt input parameters
PARAMETERS
OBSERVAB L ES
Input Output
Layer 1
MOdular software dedicated to Optimisationand Parameters Identification (P-O Bouchard et al.)
M IPOO
Layer 0
Direct model – FE software
Software
Postprocessing : Computation of the cost function
OptimizationLayer 2
Optimal parameters
Inverse Analysis Identified set of parameters
Layer 3
ICTP'2014 - JL Chenot et al 45
( )0n
y Kσ ε σ ε= + If
elseTension test6 Parameters: Observables from a tension test.
Necessity of enrichment of observables: Full field measurement
Load vs. Displacement Necking vs. Displacement
Example of parameters identification / MOOPIStrain hardening Damage
Digital Image Correlation
Displacementfield
0y D,K ,n,b,S ,σ ε
ICTP'2014 - JL Chenot et al 46
Computation at the micro-scale
General approach
Identification of macro laws using finite element micro modeling
ICTP'2014 - JL Chenot et al 47
Principle of micro scale computation(M. Bernacki et al.)
Definition of a representative elementary volume.
F. E. mesh of the physical entities.
Initialization of the Level Set describing interfaces
between entities.
Introduce physical laws at the micro scale.
Simulation of the evolution of the entities.
ICTP'2014 - JL Chenot et al 49
Example: recrystallization modeling
θ3
G1 G3
G2
Γ23θ2
θ1
v23 = c23κ23 + e3 − e2
Interface mobilityGi / Gj
Grain boundaryenergy
between Gi & Gj
Curvatureof Γij
PrimarySecondary
Difference of bulk energies
Gi / Gj
6th ISPF – J-L Chenot 50
• GG modelling
(200 grains)
• Uniform mobility
and grain boundary
energy,
2D example of secondary recrystallization
2D grain growth with germs
ICTP'2014 - JL Chenot et al 51
New developments
Zener pinning, Dynamic Recrystallization, Crystal plasticity, Twinning, GB anisotropy, Phase transformation,…
Identification of mean field models.
ICTP'2014 - JL Chenot et al 52
Future challenges
Process stability
Introduction of stochastic phenomena
Model reduction
Toward simulation of the whole process chain
ICTP'2014 - JL Chenot et al 53
Process stability and introduction of stochastic phenomena: application to robust optimization
Various sources of uncertainties: on process parameters, … on material data: behaviour,
friction, boundary conditions, on function evaluation (numerical noise)
ICTP'2014 - JL Chenot et al 54
54
Φ(p)
p
noisy functionΦ(p)
p
noisy function
smoothed
functionfalseoptimum reliable
optimum
piping defect
55
Smooth metamodel = minimize Hessian of metamodel
( ) ( ) ( )= +Φ δ Φ δ ωϕ δ
( )( ) ( ) 22min , f d
δΦ δ Φ δ ω
Ω
= ∇∫ ɶ
( )( ) ( )2
1
jmaxj
mj ,n
min , e
δδ
δϕ δ ϕ δ
=
=∑
+ among possible solutions, take the most probable
balanced average:
1, , m j j jj n f f δ∀ = = −
Metamodel is smoothed by correcting master points values:
ICTP'2014 - JL Chenot et al
Model reduction
Basic principle:
Computation of the more important eigen modes,
Approximation of the solution on these modes,
as linear combinations -> smaller system to solve
New solutions with different parameters are less
expensive in principle,
Remark: the evolution of the domain must not depend
too much on the parameters.
ICTP'2014 - JL Chenot et al 56
Preliminary work on sheet bulging (L. Fourment)
ICTP'2014 - JL Chenot et al 57
Imposed pressure
Rigid tool
sheet
Classical simulation (vertical velocity distribution)
Computation of eigen modes
ICTP'2014 - JL Chenot et al 58
mesh Mode 1
Mode 5Mode 4Mode 3
Mode 2
Toward simulation of the whole process chain
Next objective of the simulation:
ICTP'2014 - JL Chenot et al 59
HeatingPlastic deformation:Rolling, forging, etc.
Heat treatments In servicebehavior
Evolution of shape, microstructure, mechanical properties
GLOBAL OPTIMIZATION
First approach:Optimization of the strength of a clinched joint
(P-O Bouchard et al.)
Initial design
Forming Tension test
Analysis of the influence of the geometry of the tools:→ 2 parameters are important
AMPT Octobre 2010 - 61
Optimization / 2 parametersPunch radius Rpand die depth Pm
Configuration Strength (N)
Reference 737
Optimized 894
Improvement of 21%
on strength in tension
(b) Die
Pm
Lm
(a) Punch
w1
Rp
Rcp
w2
AMPT Octobre 2010 -62
After optimization
Forming Tension test
ICTP'2014 - JL Chenot et al 63
Conclusions Metal forming simulation is based on a complex
balance of developments between : Mechanics, Physics of materials, Numerical analysis, Computer science, Knowledge of the technology of industrial processes,…
New requests from industry in development: Realistic computation of µ structure evolutions, Prediction of the final properties, Introduction of stochastic phenomena, etc.
ICTP'2014 - JL Chenot et al 64
Thank you for
your kind attention