non-equilibrium, finite temperature …...upscaling in time non-equilibrium statistical...
TRANSCRIPT
Non-Equilibrium, Finite Temperature Quasicontinuum Analysis of
Plastic Deformations in Metals
Kevin G. Wang, Mauricio Ponga, and Michael Ortiz California Institute of Technology
USNCCM 12 Raleigh, NC, July 24th, 2013
MOTIVATION
Non-equilibrium thermomechanics in extreme environments
- challenge in modeling and simulation: vastly disparate scales
• atomic level rate-limiting processes: thermal vibrations, lattice defects,
atom hopping, …
• macroscopic processes of interest: ductile fracture, grain boundary
embrittlement, fatigue, irradiation damage, …
hypervelocity perforation pipeline explosion
multiscale models: bridging the scales
OVERVIEW
Non-equilibrium, finite-temperature quasicontinuum (“Hot QC”)
- provides a time-averaged and coarse-grained description of non
equilibrium thermodynamic systems
- can be coupled with (discrete) kinetic laws to solve
thermo/chemo/mechanical problems
- capable of non-equilibrium processes, large samples, at atomic scale
iT QCiT
time-averaged position and momentum, atomic temperature, atomic frequency
references: Y. Kulkarni et al. (2008), M.P. Ariza et al. (2012)
UPSCALING IN TIME
Non-equilibrium statistical thermodynamics
- assumption: separation of time scales
• fine scale: thermal vibration (~fs) “modeled”
• coarse scale: global relaxation (>ps) “solved for”
probability: 𝒒 , 𝒑 ~ 𝑝 𝒒 , *𝒑+
- maximum entropy principle: the least biased distribution maximizes
system entropy, subject to all imposed constraints
max𝑝
𝑆 𝑝 = −𝑘𝐵⟨log 𝑝⟩ (information entropy)
𝑓 = 𝑓 𝒒 , 𝒑 𝑝 𝒒 , 𝒑 𝑑𝒒𝑑𝒑Γ
with
subject to: 1 = 1
𝐻 = 𝐸
= 𝑖( 𝒒 , 𝒑 )𝑁
𝑖<1 Hamiltonian: 𝐻 = 𝐻 𝒒 , 𝒑
𝑖 = 𝑒𝑖 𝑖 = 1, … , 𝑁 (atomic internal energy)
- solve the constrained optimization problem using Lagrange multipliers:
𝐿 𝑝, 𝛼, 𝛽 = 𝑆 𝑝 − 𝑘𝐵𝛼 1 − 1 − 𝑘𝐵 𝛽 𝑇( − 𝑒 ) (free entropy)
- enforcing stationarity with respect to 𝑝 yields
NON-EQUILIBRIUM THERMODYNAMICS
Maximum entropy principle
𝑝 𝒒, 𝒑 =1
Ξe; 𝛽 𝑇 ℎ( 𝒒 , 𝒑 ) with Ξ( 𝛽 ) =
1
h3NN! 𝑒; 𝛽 𝑇 ℎ 𝒒 , 𝒑 𝑑𝒒𝑑𝒑𝑅3𝑁×𝑅3𝑁
- thermodynamic potentials
𝑇𝑖 =1
𝑘𝐵𝛽𝑖
𝑆 = −𝑘𝐵 log 𝑝 = 𝑘𝐵 log Ξ + 𝛽 𝑇 = 𝑘𝐵(log Ξ + 𝛽 𝑇 𝑒 )
Φ = sup𝑝
𝐿 𝑝, 𝛽 = 𝑘𝐵 log Ξ (free entropy)
𝛽𝑖 =1
𝑘𝐵
𝜕𝑆
𝜕𝑒𝑖
atomic temperature:
(entropy)
- 𝑝, 𝑆, and Φ all depend on the explicit form of 𝑖 = 𝑖( 𝒑 , 𝒒 ). In general, the closed-form formulation of thermodynamic potentials is intractable
𝑖 = −
𝒑 − 𝒑𝟎2
2𝑚+ 𝑉𝑖(*𝒒+) (e.g. )
NON-EQUILIBRIUM THERMODYNAMICS
Meanfield approximation (Yeomans 1992)
- given a class of trial Hamiltonian parameterized by 𝝅 : 0𝑖( 𝒑 , 𝒒 ; 𝝅 ), the trial probability density function can be defined as
- theorem (bounding principle for the free entropy):
- for fixed Lagrange multipliers 𝛽 , the optimal probability of the trial
space *𝑝0∗+ is obtained by maximizing the constrained entropy with
respect to 𝝅 , i.e.
𝑝0∗ = argmax
*𝝅+𝐿, 𝑝0 𝜋 , 𝛽 -
Φ *𝛽+ ≥ maxℎ0
Φ( 0 , 𝛽 )
the maximum free entropy of the trial space provides a lower bound
of the actual free entropy
𝑝0 𝒒, 𝒑 =1
Ξ0e; 𝛽 𝑇 ℎ0( 𝒒 , 𝒑 ;*𝝅+)
𝑝0∗ converges to 𝑝∗ as 0 →
NON-EQUILIBRIUM THERMODYNAMICS
Meanfield model: an example
- in practice there is always a trade-off between accuracy and
computability
- we define the trial Hamiltionian as
• meanfield parameters: 𝒒 𝑖 (mean atomic position), 𝒑 𝒊 (mean atomic
momentum), 𝜔𝑖 (mean atomic frequency)
𝑝0∗ = argmax
𝒑 𝑖 , 𝒒 𝒊 , *𝜔𝑖+𝐿,𝑝0, 𝛽 -
Gaussian distribution
“uncoupled harmonic oscillators”
0𝑖 𝒒 , 𝒑 ; 𝒒 , 𝒑 , 𝜔 =1
2𝑚𝒑𝑖 − 𝒑 𝑖
2 +𝑚𝜔𝑖
2
2𝒒𝑖 − 𝒒 𝑖
2
*𝝅+
𝑝0(*𝒒+, *𝒑+) =1
Ξ𝑒𝑥𝑝 − 𝛽𝑖
1
2𝑚𝒑𝑖 − 𝒑 𝑖
2 +𝑚𝜔𝑖
2
2𝒒𝑖 − 𝒒 𝑖
2
𝑁
𝑖<1
𝑖 = −𝒑 − 𝒑𝟎
2
2𝑚+ 𝑉𝑖(*𝒒+) explicit formulations of 𝐸, 𝑆, Φ, 𝐹, 𝑒𝑡𝑐.
- motivation
COARSE-GRAINING IN SPACE
Quasicontinuum reduction
crack propagation in Al 6061 (courtesy of K. Ravi-Chandar, UT Austin)
• plastic core surrounded by large elastic zone
• multiple plastic cores interact through elastic
region
adaptive mesh refinement
- key ideas
- reference:
atom
represent -ative atom
summation cluster
• finite element approximation
• cluster summation rule
• adaptive mesh refinement
• qcmethod.org
THERMO-MECHANICAL COUPLING
Governing equations
- local equilibrium conditions
- atomic heat conduction equation
Staggered solution procedure from 𝑇𝑛 to 𝑇𝑛:1
𝑆 𝑖 −
𝜅
𝑇𝑖∆𝑙𝑜𝑔
𝑇𝑖
𝑇0= 0
- dynamic equation of motion (for dynamic analysis)
𝑚𝒒 𝑖 + 𝐹𝑖𝑖𝑛𝑡 𝒒 i, 𝜔𝑖 , 𝑇𝑖 = 𝐹𝑒𝑥𝑡(𝒒 𝑖), where 𝐹𝑖
𝑖𝑛𝑡 𝒒 i, 𝜔𝑖 , 𝑇𝑖 =𝜕Φ
𝜕𝒒 𝑖
𝒕 = 𝑻𝒏
given:
𝒒 𝑛, 𝑻𝑛, 𝝎𝑛
solve heat
conduction equation
𝒒 𝑛
𝑻𝑛:1
𝝎𝑛
solve
equation of motion
𝒒 𝑛:1
𝑻𝑛:1
𝝎𝑛
enforce
equilibrium
conditions
𝒕 = 𝑻𝒏:𝟏
𝒒 𝑛:1,
𝑻𝑛:1, 𝝎𝑛:1
𝜕Φ
𝜕𝜔𝑖= 0;
𝜕Φ
𝜕𝒒 𝑖= 0;
more details on thermo-mechanical coupling
Thermodynamics and Quasicontinuum
Models of Equilibrium and Non-
Equilibrium Thermomechanics of Alloys
Speaker: Ignacio Romero
July 25, Friday, 10:00AM
MS 8.2
MODAL VALIDATION
Specific heat at constant volume (𝐶𝑣)
Max-ent (EAM)
Classical (Dulong-
Petit Law): Cv = 3kB
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0 200 400 600 800 1000
temperature (Kelvin)
ch
an
ge i
n i
nte
rnal
en
erg
y
(eV
)
- minimize free entropy (Φ) with respect to atomic
frequency (𝜔𝑖)
- validates the thermodynamic models
- also validates the empirical interatomic potentials
𝐶𝑣 =1
𝑁
𝜕𝐸
𝜕𝑇 𝑉<𝑐𝑜𝑛𝑠𝑡
≈1
𝑁 𝐸 𝑞 0, 𝑇1, 𝜔1 − 𝐸 𝑞 0, 𝑇0, 𝜔0
𝑇1 − 𝑇0
computational domain for Mg (hcp lattice), NVT
Mg, EAM potential by X-Y Liu et al. (1996)
Cu, EAM-FS potential by Sutton and Chen (1990)
Thermal expansion
- solution approach
for 𝑇 = 100𝐾, 200𝐾, 300𝐾, 𝑒𝑡𝑐;
for 𝑎 = 𝑎1, 𝑎2, 𝑎3, 𝑒𝑡𝑐 (pre-selected sample points)
find 𝛷𝑎 = 𝛷 𝑎, 𝑇 = 𝑚𝑖𝑛 𝜔 𝛷(𝑎, 𝜔 , 𝑇)
fit *(𝑎, 𝐹𝑎)+ by a polynomial function 𝑓(𝑎)
find 𝑎𝑇 = 𝑎𝑟𝑔𝑚𝑖𝑛 𝑓(𝑎)
approximate 𝛼𝑉(𝑇) by finite difference
MODAL VALIDATION
- minimize free entropy (Φ) with respect to atomic frequency (𝜔𝑖) and
mean atomic position (𝒒 𝑖)
𝒒 = argmin𝒒 ,*𝜔+
Φ( 𝒒 , 𝜔 , 𝑇)
𝛼𝑉 𝑇 =1
𝑉0
𝜕𝑉
𝜕𝑇≈
1
𝑉0
𝑉1 − 𝑉0
𝑇1 − 𝑇0
(volumetric thermal expansion coefficient)
Thermal expansion coefficient (volumetric): Mg single crystal
THERMAL EXPANSION
temperature
(K) simulation
(× 10;6) experimental
(× 10;6) data source
rel. error
(%)
90 - 190 64.5 57.9 Goens & Schmid (1936) 10.23
270 - 430 81.7 81.9 Raynor & Hume-Rothery (1939) 0.24
290 - 470 81.3 82.2 Goens & Schmid (1936) 1.11
320 - 520 82.4 84.7 Hanawalt & Frevel (1938) 2.79
430 - 580 88.2 82.8 Raynor & Hume-Rothery (1939) 6.12
580 - 725 81.2 93 Raynor & Hume-Rothery (1939) 14.53
APPLICATION
Nanovoid growth under tension at finite temperature
- problem setup
• objective: study the evolution of nanovoid in a single FCC crystal of Cu
• representative volume: 26nm x 26nm x 26nm (~ 106 atoms)
• initial void diameter: 2nm
• initial atomic zone: 5nm x 5nm x 5nm (12a0 x 12a0 x 12a0)
• interatomic potential: EAM-Mishin (Y. Mishin et al. 2001)
• uniaxial and triaxial loading with prescribed boundary displacement - temperature (K): 150, 300, 450, 600
- strain rate (1/s): 105, 107, 108, 109, 1010
cut-view of initial FE mesh QC approximation
atomic zone
void
26nm
5nm
NANOVOID GROWTH
Uniaxial loading, T = 300K, 𝜖 = 10101/𝑠
* centro-symmetric deviation:
𝐶𝑆𝐷 =
𝑞𝑖 + 𝑞𝑖:
𝑁2
2
𝑎0
𝑁2
𝑖<1
𝜖 = 6.6% (emission of dislocation)
𝜖 = 6.8% (leading
Shockley partials)
𝜖 = 7.0% (trailing
Shockley partials)
1/6,1 1 2-
dislocation core identified by CSD*
1/6,11 2-
1/6,112-
1/6,1 12- *111+
atomic temperature
before dislocation emission after dislocation emission
NANOVOID GROWTH
Triaxial loading, T = 300K, 𝜖 = 105, 108,109 , 1010 1/𝑠
dislocation core
colored by CSD
(𝜖 = 10101/𝑠) engineering strain
vir
ial str
ess (
GP
a)
0
2
4
6
8
10
12
14
0 2 4 6 8 10
𝜖 = 1010, MD
- virial stress (Allen and Tildesley, 1989)
0
2
4
6
8
10
12
14
0 2 4 6 8 10
𝜖 = 1010, MD
𝜖 = 1010, HotQC, dynamic
0
2
4
6
8
10
12
14
0 2 4 6 8 10
𝜖 = 1010, MD
𝜖 = 1010, HotQC, dynamic
𝜖 = 109, HotQC, dynamic
𝜖 = 108, HotQC, dynamic
𝜖 = 105, HotQC, quasi-static
stress-strain response
𝜎𝑖𝑗 =1
𝑉Ω −𝑚𝑘 𝑞 𝑖
𝑘 − 𝑞 𝑖𝑘 𝑞 𝑗
𝑘 − 𝑞 𝑗𝑘 +
1
2𝑞𝑖
𝑙 − 𝑞𝑖𝑘 𝑓𝑗
𝑘𝑙
𝑘∈Ω
DEFORMATION TWINNING
Mobility of twin boundary in Magnesium
- objective of study: twin boundary mobility
• intermediate temperatures and
strain rates
• velocity vs. shear stress, strain rate,
temperature, etc.
- challenge: multiple scales in time/space
• MD is limited to extreme strain rate
(>108 1/s) and small sample size
- why Mg?
• low density (“lightest useful metal”)
• high strength-to-weight ratio
c
a3 a1
a2
(101 2)
,1 011-
migration of *101 2+ twin boundary
(Serra, A. et al, 2007)
- mechanical applications
• automobile engine blocks
• helmet and body armor
• etc.
- future goals: stronger; more ductile; corrosion resist; better formability
*101 2+ TWIN BOUNDARY MIGRATION
Problem setup
- quasi-static analysis with heat conduction (for Mg, 𝜅 = 156 𝑊𝑚;1𝐾;1)
- twin boundary introduced in initial sample
- prescribed displacement at each loading step: ∆𝑞 = 0.025𝑎0
- strain rate 𝜖 = 1.0 × 1081/𝑠, 1.0 × 106 1/𝑠
- using the EAM potential of X-Y Liu et al. (1996)
y
z basal plane
twin boundary
unit cell
2D projection onto the shear plane
twin crystal
matrix crystal
twin boundary
basal plane
periodic
periodic
fixed
prescribed displacement
x y
z
,1 011- ,12 10-
,𝟏 𝟎𝟏𝟏-
SIMULATION RESULT
Simulation results (𝜖 = 108 1/𝑠)
2D projection on shear plane
prescribed displacement
-0.9160
-0.9158
-0.9156
-0.9154
-0.9152
-0.9150
-0.9148
-0.9146
-0.9144
-0.9142
0 10 20 30 40 50 60 70
evolution of internal energy
loading step no.
inte
rnal energ
y (
E)
per
ato
m (
eV
)
3D view
-0.9160
-0.9158
-0.9156
-0.9154
-0.9152
-0.9150
-0.9148
-0.9146
-0.9144
-0.9142
0 10 20 30 40 50 60 70
twin boundary migration
ONGOING WORK
Ongoing work
- verification and validation strain rate (1/s)
dynamic HotQC
compare with MD
quasi-static HotQC
compare with
dynamic HotQC and MD
quasi-static HotQC
compare with
experiments
1010 108 107 105 (or less)
- extension to Mg alloys with mass transport
Thermodynamics and Quasicontinuum
Models of Equilibrium and Non-
Equilibrium Thermomechanics of Alloys
Speaker: Ignacio Romero
July 25, Friday, 10:00AM
MS 8.2