June 2015 review bhattacharya - Carnegie Mellon...
Transcript of June 2015 review bhattacharya - Carnegie Mellon...
K. Bha'acharya Review 2015: #1
Compact representa=on of micromechanical fields
Kaushik Bha'acharya
Gal Shmuel Jin Yang
Chun-‐Jen Hsueh Md. Zubaer Hossain Dingyi Sun
Acknowledge: Mary Comer, Charles Bouman, G. Ravichandran
K. Bha'acharya Review 2015: #2
Mul=scale understanding of materials
Brinson et al.
James and Chu
Schryvers
Sun
Rapid growth of digital experimental techniques and computa=onal power has given unprecedented level of data about materials
How can we harness this for understanding and ul=mately designing?
How do we represent micro-‐mechanical fields?
Mabe
K. Bha'acharya Review 2015: #3
Topics Mul=scale Phase-‐field Simula=on Digital Image Correla=on
Fracture of heterogeneous materials Coarse-‐grained Density Func=onal Theory
Axial strain using 10% of Symlet5 wavelet detail coefficients
0
0.005
0.01
0.015
0.02
0.025
0.03
0.035
0.04
0.045
0.05Exact axial strain
0
0.005
0.01
0.015
0.02
0.025
0.03
0.035
0.04
0.045
0.05
K. Bha'acharya Review 2015: #4
Phase transforma=on
K. Bha'acharya Review 2015: #5
Mul=scale phase field method
= min⌫,µ
Z
⌦1
✓Z
RW (�, x)d⌫
x
(�) +1
2
Z
RT (', x) · LT (', x)d⌫
x
(�)
�1
2
Z
S
2
A
�1pi
(↵)↵q
L
pqrs
↵
b
L
ibcd
dµ
rscd
x
(↵)
◆dx.
lim⌘!0
min�,u
E
⌘(�, u) = inf�,u
Z
⌦1
✓W (', x) +
1
2(ru� T (', x)) · L(ru� T (', x))
◆dx
E
↵L
3=
Z
⌦1
✓1
2
�
↵L
2|r'|2 +W (', x) +
1
2(ru� T (', x)) · L(ru� T (', x))
◆dx
E =
Z
⌦
✓�
2|r'|2 +W (', x) +
1
2(ru� T (', x)) · L(ru� T (', x))
◆dx
Energy
Scaling: Length scale L, Energy scale α x 7! 1
L
x, W 7! 1
↵
W, L 7! 1
↵
L
E⌘ ⌘2
Large body limit
= E
K. Bha'acharya Review 2015: #6
Measures and elas=c energy
Young measure (Ball, Tartar) νx describes the one-‐point sta=s=cs near the point x
• Local volume frac=on • Pole figures
H measure (Tartar) µx describes the part of the two-‐point sta=s=cs that is related to the induced elas=c energy. Closely related to
• Localiza=on tensor • Eshelby tensor • (Structure factor)
E = min⌫
✓minµ
. . .
◆
⇡ minu
min�i
Z
⌦
0
@X
wi�i +X
i,j
�imij�j +1
2
e(u)�
X
i
�iTi(x)
!· L
e(u)�
X
i
�iTi(x)
!1
Adx
New transla=on bounds
Postulate
�̇ = K(��E)Phase field model for volume frac=ons
K. Bha'acharya Review 2015: #7
Superelasticity"
Initiation. Well oriented grains Stress concentrations
Progress. Autocatalyic strips Some reorienta=on
Saturation. Incomplete transformation Reorientation
Reversal. Last on, first off Slight differences from (B)
Richards et al. (2013)
Is there a compact representa=on for these
transforma=on strain fields that retains the essen=al physics
K. Bha'acharya Review 2015: #8
Wavelets Generate a nested sequence of subspaces {0} ⇢ ...Vj ⇢ Vj+1 ⇢ ... ⇢ L2
(R):
Define Vj = {�j,k (x) , k 2 Z},and note Vj ⇢ Vj+1 = Vj �Wj where Wj = { j,k (x) , k 2 Z}
Approximate a function at scale j0 as
f(x) =
X
k2Zak'j0,k(x) +
X
j>j0,k2Zdj,k j,k(x)
• Scaling function: '(x) satisfying
Z|'|dx = 1
• Mother wavelet: (x) satisfying
Z dx = 0
• (x) =X
n2Zbn'(2x� n)
Introduce
Space
Freq
uency
-1.0-0.5 0.5 1.0 1.5 2.0
-1.0-0.5
0.51.0
-1.0-0.5 0.5 1.0 1.5 2.0
-1.0-0.5
0.51.0
Haar
-1.0-0.5 0.5 1.0 1.5 2.0-1.0-0.5
0.51.01.5
-1.0-0.5 0.5 1.0 1.5 2.0-1.0-0.5
0.51.01.5
Daubechies
Consider their translations and dilations
j,k (x) := 2j/2 �2jx� k
�
'j,k (x) := 2j/2'�2jx� k
�j, k 2 Z
Compression Adaptively choose basis
K. Bha'acharya Review 2015: #9
Compressing the results using wavelets
0 1 2 3 40
100
200
300
400
500
600
700
!̄x (%)
!̄x(M
Pa)
A B
C
ED
• Simulate using FFT • At each =me step, take a (Symlet 5) wavelet transform of the result • Keep only 1 % of the coefficients • Compute the macroscopic stress-‐strain curve1
K. Bha'acharya Review 2015: #10
Wavelets also provide new insight
0 1 2 3 40
100
200
300
400
500
600
700
800
900
1000
!̄x(%)
!̄x(M
Pa)
• Rela=ve change in # of ac=ve wavelets: ~4% • Ac=ve wavelets are predictors of stress and strain
intensi=es • Dominant wavelets are basis for model reduc=on • Current ac=ve set are predictors of transforma=on • Ac=ve wavelets carry informa=on about inter-‐
granular compa=bility
Ac=ve
Inac=ve
Total
K. Bha'acharya Review 2015: #11
Experiments with full-‐field strain
Daly, Ravichandran and Bha'acharya, Acta Mater. (2007)
K. Bha'acharya Review 2015: #12
Axial strain using 10% of Symlet5 wavelet detail coefficients
0
0.005
0.01
0.015
0.02
0.025
0.03
0.035
0.04
0.045
0.05Exact axial strain
0
0.005
0.01
0.015
0.02
0.025
0.03
0.035
0.04
0.045
0.05Exact axial strain
0
0.005
0.01
0.015
0.02
0.025
0.03
0.035
0.04
0.045
0.05Axial strain using 10% of Symlet5 wavelet detail coefficients
0
0.005
0.01
0.015
0.02
0.025
0.03
0.035
0.04
0.045
0.05Exact axial strain
0
0.005
0.01
0.015
0.02
0.025
0.03
0.035
0.04
0.045
0.05Axial strain using 10% of Symlet5 wavelet detail coefficients
0
0.005
0.01
0.015
0.02
0.025
0.03
0.035
0.04
0.045
0.05
Representa=on of Daly et al. Experiments
99.3% 99.9% 99.8% ||"approxyy ||||"yy||
=
Compression using 1% coefficients
1.93% 2.94% 1.34% Total strain
K. Bha'acharya Review 2015: #13
Digital image correla=on
K. Bha'acharya Review 2015: #14
Digital image correla=on
Su'on et al., Image Correla=on …, Springer Hild and Roux, in Op#cal Methods for Solid Mechanics, Wiley
y x
y(x)
f(x) g(y) miny:⌦!Rn
Z
⌦|f(x)� g(y(x))|2 dx
() max
y:⌦!Rn
Z
⌦f(x)g(y(x))dx
Can be done locally • Very computa=onally efficient
But, • Need high frequency random pa'ern … large data sets
• Results can be very noisy
Can we use filtered/ compressed images for DIC?
K. Bha'acharya Review 2015: #15
DIC with compressed images
Original 20 % DCT min
y:⌦!Rn
Z
⌦|f(x)� g(y(x))|2 dx
() max
y:⌦!Rn
Z
⌦f(x)g(y(x))dx
Since the method is local, • we need high frequency random images • filtered and lost during compression
Generally gives significant error
K. Bha'acharya Review 2015: #16
Need to introduce global informa=on
miny:⌦!Rn
Z
⌦
⇣|f(x)� g(y(x))|2 + ↵ |ry|2
⌘dx
• Global Method and Regulariza=on (Hild and Roux)
Reduces noise, but expensive
Use finite element basis for y
• Local method with global constraints
max
�,µmin
{yi},{Fi},y
Z
⌦
8<
:X
i
�i|f(x)� g(yi + Fix)|2 + �
�����y �X
i
�iyi
�����
2
+ µ
�����ry �X
i
�iFi
�����
29=
; dx
max
�,µmin
{yi},{Fi},y
Z
⌦
8<
:X
i
�i|f(x)� g(yi + Fix)|2 + �
�����y �X
i
�iyi
�����
2
+ µ
�����ry �X
i
�iFi
�����
29=
; dx
Par==on of unity Related interpola=on
- Computa=onally efficient (mul=scale approach)
- Consistent with compression
K. Bha'acharya Review 2015: #17
Results with global + compression
100%
20%
Original
Global Local
20% compressed
Avg Eyy = 1.89 % Avg Eyy = -‐1.09 %
Global 0.0071 % 0.044 %
Local 0.49 % 4.61 %
K. Bha'acharya Review 2015: #18
Use of priors: Fracture
Seek the stress intensity, but the displacements are extremely small.
minr,R,c,KI ,x0
Z
⌦r,R
|f(x)� g(KI
U(x� x0) + c)|2 dx
K. Bha'acharya Review 2015: #19
Impact/Future • Personnel
– Gal Shmuel, Post-‐doc (currently Asst. Prof. Technion, Israel) – Md. Zubaer Hossain, Post-‐doc (par=al, soon Asst. Prof. U Delaware) – Jin Yang, Graduate student – Chun-‐Jen Hsueh, Graduate student (par=al) – Dingyi Sun, Graduate student (NDSEG)
• Publica=ons – G. Shmuel, A.T. Thorgeirsson, and K.Bha'acharya. 2014. “Wavelet Analysis of
Microscale Strains.” Acta Materialia 76: 118–126. – J. Yang, G. Ravichandran and K. Bha'acharya. 2015. “Data compression for digital
image correla=on”. In prepara=on for submission to Experimental Mechanics – 4 others
• Provisional Patent – C-‐J Hsueh, G. Ravichandran, K. Bha'acharya. 2015. “A Novel device of measuring
the fracture toughness of heterogeneous materials”
• Future direc=ons – Digital Image Correla=on with compression (Comer) – Scale-‐dependent Young measure, H-‐measures (James) – Mechanical proper=es of Al-‐Si (Voorhees/Kalidinidi)