local optimal polarization of piezoelectric material
DESCRIPTION
My presentation at the 9th Int. Workshop on Direct and Inverse Problems in Piezoelectricity in Weimar (30.09-02.10.2013).TRANSCRIPT
Introduction Local Optimal Polarization Numerical Examples Summary
Local Optimal Polarization of Piezoelectric Material
Fabian Wein, M. Stingl
9th Int. Workshop on Direct and Inverse Problems in Piezoelectricity30.09-02.10.2013
Introduction Local Optimal Polarization Numerical Examples Summary
Overview
General
linear continuum model
numerical approach based on finite element method
PDE based optimization with high number of design variables
Optimization
optimization helps to understand systems better
manufacturability in mind
no real prototypes
Introduction Local Optimal Polarization Numerical Examples Summary
Structural Optimization = Topology Optimization + Material Design
Topology optimization
“where to put holes”/ material distribution
design of (piezoelectric) devices
macroscopic view
Material design
“assume you could have arbitrary material, what do you want?”
realization might be another process
realizations might be metamaterials
Introduction Local Optimal Polarization Numerical Examples Summary
Motivation
stochastic orientationJayachandran, Guedes,Rodrigues; 2011
Material Design
common homogeneous material seems tobe not optimal
Free Material Optimization → why it doesnot work
local optimal material → new approach
Introduction Local Optimal Polarization Numerical Examples Summary
Standard Topology Optimization
distributes uniform polarized material/holes
“macroscopic view”
established in 2 1/2 dimensions (single layer)
scalar variable ρe for each design element (= finite element cell)
SIMP (solid isotropic material with penalization)
piezoelectric topology optimization Kogel, Silva; 2005
[ cEe ] = ρe [cE ] [ ee ] = ρe [e ] [ εSe ] = ρe [ε
S ], ρe ∈ [ρmin,1]
Introduction Local Optimal Polarization Numerical Examples Summary
Piezoelectric Free Material Optimization (FMO)
all tensor coefficients of every finite element cell are design variable
[c ] =
c11 c12 c13− c22 c23− − c33
, [e ] =
(e11 e13 e15e31 e33 e35
)>, [ε] =
(ε11 ε12
− ε22
)
properties
[c ] and [ε] need to be symmetric positive definite
[ε] only for sensor case (mechanical excitation) relevant
questions to be answered
[c ] orthotropic?
[e ] with only standard coefficients?
orientation of [c ] and [e ] coincides?
something like an optimal oriented polarization?
Introduction Local Optimal Polarization Numerical Examples Summary
FMO Problem Formulation (Actor)
min l>u maximize compression
s.th. K u = f, coupled state equation
Tr([c ]e) ≤ νc, 1≤ e ≤ N, bound stiffness
Tr([c ]e) ≥ νc, 1≤ e ≤ N, enforce material
(‖[e ]e‖2)2 ≤ νe, 1≤ e ≤ N, bound coupling
[c ]e −ν I � 0, 1≤ e ≤ N. positive definiteness
realize positive definiteness by feasibility constraints
c11e −ν ≤ ε, 1≤ e ≤ N,
det2([c ]e −νI) ≤ ε, 1≤ e ≤ N,
det3([c ]e −νI) ≤ ε, 1≤ e ≤ N.
Introduction Local Optimal Polarization Numerical Examples Summary
Tensor Visualization similar to [Marmier et al.; 2010]
[c ] =
12.6 8.41 08.41 11.7 0
0 0 4.6
, [e ] =
0 −6.50 23.3
17 0
, [ε] =
(1.51 0
0 1.27
)
[c ] [e ] [ε] ‖[c ]‖“ortho” ‖[e ]‖“zeros” ‖[ε]‖“ε12”
orientational stiffness
σ[c ]x (θ) =
100
> [c ](θ)
100
, σ[e ]x (θ) =
100
> [e ](θ)
(10
), D
[ε]x . . .
Introduction Local Optimal Polarization Numerical Examples Summary
Actuator Model Problem
Introduction Local Optimal Polarization Numerical Examples Summary
FMO Results - Elasticity Tensor [c ]
orientational stiffness
orientational orthotropy norm
Introduction Local Optimal Polarization Numerical Examples Summary
FMO Results - Piezoelectric Coupling Tensor [e ]
orientational stress coupling
orientational “zero norm”
Introduction Local Optimal Polarization Numerical Examples Summary
Discussion of the FMO Results
objective
maximize vertical displacement of top electrode
observations
less vertical stiffness to support compression
in coupling tensor e33 is dominant
characteristic orientational polarization
standard material classes (orthotropic)
coinciding orientation for [c ] and [e ]
ill-posed problem (stiffness minimization)
inhomogeneity due to boundary conditions
boundary conditions
deformation
elasticity
coupling
Introduction Local Optimal Polarization Numerical Examples Summary
Electrode Design vs. Optimal Polarization
Electrode Design
pseudo polarization Kogel, Silva; 2005
[ cEe ] = [cE ], [ ee ] = [e ], [ εSe ] = ρp[ε
S ] ρp ∈ [−1,1]
(continuous) flipping of polarization (+ topology optimization)
applied on single layer piezoelectric plates
only scales polarization, does not change angle
known to result in -1 and 1 full polarization (static)
erroneously called “optimal polarization”
Introduction Local Optimal Polarization Numerical Examples Summary
Optimal Orientation
parametrization by design angle θ
[ cE ] = Q(θ)>[c ]Q(θ) [ e ] = R(θ)>[e ]Q(θ) [ εS ] = R(θ)>[ε]R(θ)
R =
(cosθ sinθ
−sinθ cosθ
)Q =
R211 R2
12 2R11R12
R221 R2
22 2R21R22
R11R21 R12R22 R11R22 +R12R21
concurrent orientation of all tensors
corresponds to local polarization
Introduction Local Optimal Polarization Numerical Examples Summary
Numerical System
linear FEM system (static)(Kuu Kuφ
K>uφ−Kφφ
)(uφ
)=
(fq
), short Ku = f
K∗ assembled by local finite element matrices K∗e
K∗e constructed by [ cEe ](θ), [ ee ](θ) and [ εSe ](θ)
f is discrete force vector, corresponding to mesh nodes.
q from applied electric potential (inhomogeneous Dirichlet B.C.)
f = 0 for sensor, q = 0 for actuator
Introduction Local Optimal Polarization Numerical Examples Summary
Function
discrete solution vector u =(u1x u1y u2x u2y . . .φ1 φ2 . . .
)>displacement (each direction) and electric potential at mesh nodes
generic function f identifying solution
f = u>l
scalar product of solution with selection vector l = (0 . . . 1 . . .0)>
f can be maximized or used to specify a restriction
vertical displacement of all upper electrode nodeshorizontal displacement of a cornerdiagonal displacement of a given regionselection of electric potential at electrode. . .
Introduction Local Optimal Polarization Numerical Examples Summary
Sensitivity Analysis
the gradient vector ∂ f∂θ
determines for every θe the impact on f
sensitivity analysis based on adjoint approach
f = uT l,∂ f
∂θe= λ>e
∂ Ke
∂ρeue with λ solving Kλ =−l
one adjoint system Kλ =−l to be solved for every function f
∂ Ke∂ρe
easily found by product rule
numerically very efficient, independent of number of design variables
iteratively problem solution by first order optimizer (SNOPT, MMA)
Introduction Local Optimal Polarization Numerical Examples Summary
Problem Formulation
generic problem formulation
minθ
l>u objective function
s.th. K u = f, coupled state equation
l>k u ≤ ck , 0≤ k ≤M, arbitrary constraints
θe ∈ [−π
2,π
2], 1≤ θe ≤ N, box constraints
for sensor and actuator problem
full material everywhere
individual polarization angle in every cell
Introduction Local Optimal Polarization Numerical Examples Summary
Regularization
orientational optimization in elasticity known to have local optimima
restricts local change of angle
filtering Bruns, Tortorelli; 2001
θe =∑Nei=1w(xi )θi
∑Nei=1w(xi )
w(xi ) = max(0,R−|xe −xi |)
local slope constraints Petersson, Sigmund; 1998
gslope(θ) = |< ei ,∇θ(x) >| ≤ cs i ∈ {1, . . . ,DIM}gslope(θe , i) = |θe −θi | ≤ c ,
Introduction Local Optimal Polarization Numerical Examples Summary
Example Problems
A BC
actuator problems
maximize compression C ↓maximize compression C ↓ and limit A ← and B →twist A ↓ and B ↑
sensor problem
maximize electric potential at C
Introduction Local Optimal Polarization Numerical Examples Summary
maximize compression C ↓
initial |u| optimized |u|
gain: 6.1% of integrated y -displacement of C nodes
C is flattened
probably no global optimum reached
Introduction Local Optimal Polarization Numerical Examples Summary
maximize compression C ↓ and limit A ← and B →
loss: 4.9% of integrated y -displacement of C nodes
but A and C bounded to 50 % of initial x-displacement
Introduction Local Optimal Polarization Numerical Examples Summary
twist A ↓ and B ↑
note θ ∈ [−π
2 ,π
2 ]
electrode design might be more effective for this case
Introduction Local Optimal Polarization Numerical Examples Summary
maximize electric potential at C
gain: 0.6 % in difference of potential
possibly due to poor local optima
Introduction Local Optimal Polarization Numerical Examples Summary
Coupling Tensor vs. Stiffness Tensor
what is the impact of the transversal isotropic stiffness tensor?
assume isotropic stiffness tensor
gain: 4.7 % vs. 6.1 % with PZT-5A tensors
Introduction Local Optimal Polarization Numerical Examples Summary
Conclusion
General
local polarization works in principle
solutions might be far from global optimium
more feasible than piezoelectric Free Material Optimization
simple support would change everything
Applications
not to improve performance
exact tuning of devices
metamaterial not yet possible (e.g. auxetic material)
Introduction Local Optimal Polarization Numerical Examples Summary
Future Work
Examples
dynamic problems, shift of resonance frequencies possible?
metamaterials (e.g. auxetic material)
Mathematical
novel tensor based solver
very promising for elasticity
Technical Realization
polarization by local electric field
piezoelectric building blocks
. . . any suggestions?
Introduction Local Optimal Polarization Numerical Examples Summary
End
thanks for your patience :)