Transcript Training
Decomposition and Visualization of
Fourth-Order Elastic-Plastic Tensors
Alisa G. NeemanSC, Rebecca BrannonU,
Boris JeremicD, Allen Van GelderSC,
and Alex PangSC
7/6/2015
1
Driving Problem
How to visualize fourth-order tensor fields
representing a solid’s time-varying stiffness
(such as soil during an earthquake).
Two Stages:
1.
Meaningful decomposition to reduce number of
components per field location:
3x3x3x3
2.
6x6
3x3
1
Meaningful visualization to find locations of
softening and mode of stress to which the
material is vulnerable.
2
What is a Fourth-Order Tensor?
Order
Notation
Zero
Represented
by
Scalar
First
1D array
vi
Second
2D array
Aij
Fourth
4D array
Eijkl
c
Symmetry: Aij = Aji
Eijkl = Ejikl = Eijlk = Ejilk = Eklij
3
Solid Mechanics Vocabulary
Elastic:
recoverable deformation
Elastic-Plastic:
recoverable + permanent
deformation
Localized failure
4
Basic Constitutive Equation
Hooke’s Law generalized to 3D
(what gets solved by the simulator)
: stress increment
: 4th-order elastic-plastic tangent stiffness
: strain increment
i, j, k, l range from 1 to 3, and represent orthogonal spatial
5
axes x, y, z
Stiffness Changes With
Increasing Stress
Material Behavior
Elastic
Elastic-plastic
Failure, localized
Eijkl Properties
Positive-definite,
fully symmetric
Eigenvalues get smaller,
non-singular, may lose
major symmetry
Non-positive-definite,
possibly singular
Non-trivial to decompose (or visualize!)
6
Approach
1.
Unroll the 3x3x3x3 tensor into a 6x6 matrix.
»
2.
Polar decomposition on the matrix produces
»
3.
5.
two 6 x 6 matrices; the rotation part and the
symmetric stretch part.
Eigen-decomposition on the stretch yielding
»
4.
Numerical methods only exist for matrices
6 (6-element) eigenvectors and 6 real eigenvalues.
Select a single eigenvector and compose it into a
symmetric 3 x 3 tensor.
Visualize the second-order eigentensor with a glyph
that shows its structure and also reflects its
eigenvalue.
7
Application Constraints
Small Deformation Theory
Small displacement gradients
Stiffness cast with respect to reference
configuration; changes associated with rigid
material rotation eliminated from
consideration
Second-order tensors must be symmetric
Constrains fourth-order tensors to exhibit
minor symmetry, i.e. Eijkl = Ejikl = Ejilk = Eijlk
Natural materials exhibit minor symmetry
8
Polar Decomposition
Uniquely separates a matrix into two components:
E = QS
where Q is a pure rotation matrix (orthonormal, positive
determinant) and S is a stretch (symmetric,
positive-semi-definite).
9
Polar Stretch and
Mode of Vulnerability
Eigen-decomposition applied to
stretch S yields 6 second-order
eigentensors and eigenvalues.
A reduced eigenvalue means the solid
is less stiff* and the associated
eigentensor is the mode of stress for
which the solid has the most
reduction in stiffness.
* terminology may vary
10
Eigentensor Glyphs
Eigentensor glyphs drawn by stretching a
unit sphere according to the formula
n
where n is a unit length direction
vector from the center of the
sphere to a point on the surface
11
Glyphs and Finite Elements
As the solid deforms,
stress changes induced at
Gauss integration points
z
Irregular layout on
X, Y, and Z
1-2 orders of magnitude
difference in element sizes
Glyph scale factor:
distance between Gauss
points in single element
node
y
Gauss
point
x
Finite Element
(8 node brick) 12
Physical Meaning
Stress Modes
Isotropic, 3 equal
eigenvalues
Two equal eigenvalues One zero eigenvalue,
+ distinct eigenvalue with 2 others equal but
more compressive
opposite in sign
than the others
13
Experiments
Stage 1:
Soil self-weight compression (–Z), 25 steps
Induces deformation and may change stiffness
Stage 2:
Two point loads, 100 steps
Small –Z component (0.9659 kN) and
Large +X component (1294 kN)
14
Deformation From
Experiments
Black arrows
indicate point
load locations
Variation
due to self
weight
15
Material 1: Drucker-Prager
Soil Model
Fails under
tension
Non-hardening
(no change with
compression)
Stage 1:
No change
Stage 2:
Experienced
compression in
front of point loads
and tension
behind
16
Material 2: Dafalias-Manzari
Soil Model
Pressuredependent
Non-associated
flow
Stage 1:
induced hardening
Stage 2:
softening and
singularity
Difficult to correlate
stress to soil’s
behavior
17
What About the Polar Rotation?
We need a little more solid
mechanics vocabulary
18
Material Model Vocabulary
Yield Function F(σ):
delineates stress that
causes elastic versus
elastic-plastic
deformation.
Yield Surface:
is a convex isosurface
in stress space where
F(σ)= 0.
19
Associated vs. Non-Associated
Plastic Flow G(σ): determines
plastic strain increment
vectors. If the material is
associated, the plastic flow is
along the normal to the yield
surface.
Non-associated: the plastic
flow can diverge from the
yield surface normal.
This is where the stiffness
tensor loses symmetry!*
* In the absence of elastic-plastic coupling
20
Polar Rotation and
Elastic-Plastic Materials
We conjecture that rotation Q quantifies
the misalignment between yield
surface normal and the plastic flow in
non-associated materials.
Early results look promising on
simulated and measured data.
21
Summary
Original Goals:
Meaningful decomposition method to
facilitate visualization of 4th-order tensors
Meaningful visualization technique
Results:
First application of polar decomposition to
analyze stiffness
Technique meaningful within subset of
solid mechanics simulations
22
Where to Get More
Information
Supplementary Materials:
How to unroll a 3x3x3x3 tensor into 6x6
Algorithm for polar decomposition in the face of
a near-zero eigenvalue
NEESforge source code repository for
VEES visualization application
http://neesforge.nees.org/projects/vees/
23
Thanks!!
This work funded by a GAANN fellowship
and the UCSC/Los Alamos Institute for
Scalable Scientific Data Management
(ISSDM)
Thanks for your attention!
24
Plastic Flow Along the Normal
to the Yield Surface
Mathematically, it means that the plastic
strain increment tensor is a positive scalar
multiple of the yield surface normal
where the yield surface normal is itself a
positive scalar multiple of the derivative of
the yield function with respect to stress
holding internal variables constant.
25
Polar Decomposition With
Zero/Near-Zero Eigenvalue
E = QS for
Square Matrix
Non-negative determinant
If one zero eigenvalue, decomposition
is unique with specification that
det(Q) = +1
26
Stretch
Define M = ETE = STS = S2
M = T J TT
Eigen-decomposition
J diagonal, ascending order
S = √S2 = T√J T
27
Rotation with one zero column
(find C!)
Define C = QT
B = QST= C√J
Cj = Bj/√Jjj for j = 2,…,n
QST = QT √JTTT
but TTT = T-1T = I, since T is orthogonal
Remove √J from non-zero columns
C1: to calculate, use Gramm-Schidt
completion of 6-D orthonormal basis
Q = CTT
TT = T-1
28
More accurate stretch
S = sym(Q-1 E)
29
Unrolling with Mandel Constants
E1111 E1122 E1133 √2E1112 √2E1123 √2E1131
E2211 E2222 E2233 √2E2212 √2E2223 √2E2231
E3311 E3322 E3333 √2E3312 √2E3323 √2E3331
√2E1211 √2E1222 √2E1233 √2E1212 √2E1223 √2E1231
√2E2311 √2E2322 √2E2333 √2E2312 √2E2323 √2E2331
√2E3111 √2E3122 √2E3133 √2E3112 √2E3123 √2E3131
30