Multiscale Systems Engineering (MSSE) Research...

Multiscale Systems Engineering (MSSE) Research Group

Current MembersDr. Yan Wang (faculty member)Dehao Liu (Ph.D. student)Jesse Sestito (Ph.D. student)Yanglong Lu (Ph.D. student)Michael Kempner (Ph.D. student)Luka Malashkhia (Ph.D. student)

Recent Project CollaboratorsDr. Chaitanya DeoDr. Tequila HarrisDr. Ling LiuDr. David McDowellDr. David RosenDr. Wei SunDr. Hongyuan ZhaDr. Ting ZhuDr. Fadi Abdeljawad (Clemson)Dr. Sung-Hoon Ahn (SNU)Dr. Ji-Hyeon Song (NTU)Dr. Anh Tran (Sandia)Dr. Laura Swiler (Sandia)

Multiscale Systems Engineering Research Group

VisionMultiscale Systems are systems consisting of hierarchical structures

with different sizes that recursively exhibit patterns of behaviors as the diagnostics of interactions among subsystems at lower levels.

nm μm mm m km

femto-spico-s

nsμsmssecmindayyear •Our vision is that computational

design tools for multiscale systems with sizes ranging from nanometers to kilometers will be indispensable for engineers' daily work in the near future.

“from cradle to grave”“from atoms to systems”


Research & Education MissionsTo create new modeling and simulation

mechanisms and tools with underlying scientific rigor that are suitable for multiscale systems engineering for better and faster product innovationTrain engineers of the future to gain necessary

knowledge and skills for future work in a collaborative environment as knowledge creators and integrators


Research Thrust Areas @ MSSE The overarching goal of our research is to tackle the curse-of-dimensionality

design challenge by developing new physics-based data-driven methods to enable engineers to establish comprehensive and robust process-structure-property relationships for systematic product, materials, and process design. Multiscale CAD/CAM/CAE

• Multiscale simulations (DFT, MD, KMC, PF, FEA, LBM, CFD) • Process-structure-property relationship for additive manufacturing process design• Design space exploration and global optimization

Uncertainty Quantification• Reliable multiscale simulation under uncertainty• Probabilistic design of systems of cyber-physical-social systems• Quantum scientific computing• Risk perception and risk-informed decision making

Product Lifecycle Informatics• Physics-based data-driven process monitoring• Data analytics for diagnostics & prognostics• Geometric modeling & processing, interoperability


Simulation-Based Product, Materials, & Process Design

physical model of natural or engineered system

mathematical model

numerical modelcomputer simulation

validation

Simulation-based design

scalable algorithms & solvers

data/observations

multiscale models

geometry modeling &discretization schemes

parameter inversiondata assimilationmodel/data error control

visualizationdata mining/science

optimizationuncertainty quantification

verificationapproximation error control


Multiscale Multiphysics Modeling & Simulation for Materials Design and Process Design

Thermal-Fluid Finite Element Method

Controlled Kinetic Monte Carlo

Phase Field Method

Thermal Lattice Boltzmann Method

Coarse Grained Molecular Dynamicsnm

μm

mm

length scale

ns μs ms s time scale

Substrate

Laser

Blended Ni and Ti powder

Nozzle

Built 3D SMA part

Hatch direction (y-dir.)

Build direction (x-dir.)

Built height (z-dir.)

6


Interactive Multiscale Product-Materials Modeling

Wang Y. (2007) Periodic surface modeling for computer aided nano design.Computer-Aided Design, 39(3): 179-189.


Additive Manufacturing (AM) Process-Structure Relationship

Solidification Phase field and thermal lattice

Boltzmann methods

Phase field, magnetic field, and lattice Boltzmann methods

Sintering Process simulation


Physics-Constrained Machine Learning for P-S-P Relationship

Multi-Fidelity Physics-Constrained Neural Network (MF-PCNN) Efficient training with less data

multi-physics predictions Phase transition

Temperature

9


In-situ Process and Machine Health Monitoring

Physics-Based Data-Driven Physics-Based Compressive

Sensing Monitor temperature

Monitor fluid flow

Data-Driven Statistical machine learning Machine health diagnostics and

prognostics

Measured streamwise velocity


Uncertainty Quantification in M&S

Aleatory Uncertainty: inherent randomness

Epistemic Uncertainty: lack of perfect knowledge

Interval-based stochastic simulations

Queueing Systems:

Microbial Fuel Cell:

(a) H2O in anode chamber (b) H+ in cathode chamber

Interval c.d.f.

F

x

1

0

],[ xx

],[ FF

Upper Bound

Lower Bound Enclosed real-valued c.d.f.’s

Lack ofdata Epistemic

Uncertainty in Models & Inputs

Conflictingbeliefs

Conflictinginformation

Lack ofintrospection

Measurementsystematic

errors

Lack ofinformation

aboutdependency

Truncationerrors

Round-offerrors


Bayesian OptimizationConstrained and mixed-

integer BO

Discrete BO

Multi-objective BO

Multi-fidelity BO


Probabilistic Design of Systems of Cyber-Physical-Social Systems

Internet of Things (IoT) Resilience

Trust based network design


Multiscale Uncertainty Quantification

Generalized Hidden Markov Model (GHMM)

Generalized Interval Bayes’ Rule (GIBR)

Multiscale information assimilation Single-Scale Single-Point observation Single-Scale Multi-Point observation Multi-Scale Multi-Point observation

Carbon nano-tube (CNT) composite actuator design example

Observable

1i ,x

Hidden

4i ,x3i ,x

2i ,xix

yW

xW

Scale Z

Scale Y

Scale X

k,1zkzzW

jyj ,1y

j ,2y

iX

jY

kZ

1

||

dual | duali i

i n

j jj

A E EE A

A E E

p pp

p p

microscale

z: bending strain rate of actuator(Z: observable)

y1: conductivity of composite (1%CNT)(Y1: observable)y2: conductivity of composite (2%CNT)(Y2: observable)

x: resistivity of single CNT

major contributors of epistemic uncertainty in multiscale analysis:• lack of data• inconsistent observations• measurement errors

mesoscale nanoscale


15

Cross-Scale Model Validation Example molecular dynamics model of material defects

Imprecise damage function from irradiation Imprecise probability that a stable Frenkel pair is generated at certain level

of transfer or recoil energy estimated from MD simulation

Validate unobservable nanoscale model parameters θ with observable macro-scale measurements (∆ρ/n: resistivity change)

Bayesian calibration and validation

0 50 1000

0.5

1angle=57.0

Recoil energy (T)

Dam

age

func

tion

(v)

midlowerupper

0 50 1000

0.5

1angle=0.0

Recoil energy (T)

Dam

age

func

tion

(v)

midlowerupper

0 50 1000

0.5

1angle=90.0

Recoil energy (T)

Dam

age

func

tion

(v)

midlowerupper

( ) (Δ / | ) ( | )( | Δ / )

dual (Δ / | ) ( |

( | )

( | )) ( )dm m d d m

m dm d d m

θ ρ n T T T dT dTθ ρ n

ρ n T T T

T θ

T θ d dT dθθ T

p p p

p

pp

p p p


Reliable Simulation of Stochastic Dynamics

Bi-stable stochastic resonance

Van der Pol oscillator

Prob

abili

ty d

ensi

ty

upper-boundnominal

lower-bound

Prob

abili

ty d

ensi

tyPr

obab

ility

den

sity

Prob

abili

ty d

ensi

ty

t=0

t=0.125

t=0.375

t=0.75

Simultaneous evolution of aleatory and epistemic uncertainties


Quantum Scientific ComputingQuantum walk accelerated

drift-diffusion simulations drift

bi-stable oscillation

Grover algorithm for global optimization

Earlier Projects


Objective: To investigate the feasibility of modeling and simulating nano structures based on a proposed periodic surface model from atomic to meso scales and to expand the horizon of available shapes for design engineers.

(1b) intersection of P surface and 2 Grid surfaces

(1a) Sodalite cages. Vertices are Si (Al). Edges represent Si-O-Si (Si-O-Al) bonds.

(1c) P surface and its modulation with a Grid surface

Computer-Aided Nano-Design

Reverse engineering

Model construction

(2b) Reconstructed locisurface from asynthetic Zeolite crystal(Each tetrahedronencloses a Si, eachvertex of the tetrahedralis a O, and each greensphere is a Na)

(2a) Reconstructed locisurface from a Faujasitecrystal (Eachtetradecahedron encloses aFe, each hexagonal prismencloses an Al, and eachvertex of the polygonsrepresents a Si)

Cubic

Simple

Body-Centered

Face-Centered

Orthorhombic Simple

a b c

Base-Centered

a b c

Body-Centered

a b c

Face-Centered

a b c

Tetragonal Monoclinic Simple

a c

Body-Centered

a c

Simple

90 90,a b g

Base-Centered

90 90,a b g

Triclinic Rhombohedral Hexagonal

90, ,a b g

90, ,a b g

a c

a b

c

a b

c

a b

c

a b

c

α

β

γ

α

β

γ

a a

c

a a

c

c a α

β

γ

a a

a α

β

γ

Mathematical models of Bravais Lattice


Computer-Aided Phase Transition Design

A phase transition is a geometric and topological transformation process of materials from one phase to another, each of which has a unique and homogeneous physical property.

Important to design various phase-change materials (e.g. for information storage and energy storage)

Phase transition pathway search Parallelization for efficency Robustness under uncertainty

−4715.7775

−4709.3970

Energy (eV)

(initial state)

−4715.4974

−4722.7941

−4708.5716

−4715.4812

−4717.9910(final state)

−4716.4783−4715.9963

−4717.9640 −4717.9351

Saddle-point energy level

−4715.2376

−4713.9203

−4717.1825

Coordinate linear interpolationSurface linear interpolationPotential-driven surface interpolation

Fe

TiH


Computer-Aided Nano-Manufacturing

Scanning Probe Lithography Focus Ion Beam Lithography

workpiece species

controlled species

controlled diffusion events

vaporized workpiece species

absorbent species

activated controlled

species

vacancy

workpiece species

Ga_src

Ga+

absorbent

Nanoimprint Lithography

substrate

absorbent

target

deposited metal

Ionized PVD

(a) SEM image [Zankovych 2001]

cKMC model of NIL process (b) cKMC simulation result

absorbent

resist

mold2path2mold1

path1_activepath1

mobilized_resist

(a) SEM image [Chou et al. 1996]

(b) cKMC simulation result


Surfacelets as Materials Descriptors

surface integralwith ridgelet

xz

y

αβ

μ

(a) SEM images of SiO2-MnO2 ceramics (b) wedgelet surface integral

αβ

λ

1D wavelet transform with Haar

(c) surfacelet coefficients

surface integralwith cylindricalsurfacelet

xz

y

αβ

μ

(a) SEM images of polyacrylonitrile-based nanofibers

(b) cylindrical surface integrals

αβ

λ

1D wavelet transform with Haar

(c) surfacelet coefficients Centers of circular nanofibers


Searching Feasible Design Space in Set Based Designby Solving Quantified Constraint Satisfaction Problem (QCSP)For design constraint

with input parameter a , design target b, and design variable x, we can have united solution set

tolerable solution set

controllable solution set

Vehicle chassis design example

Chassis design space searching problem Performance parameters: , , , ,

Suspension systems (Front and Rear) Immediate Variables:

, Tire systems (Front and Rear)

Immediate Variables: , , ,

Design Variables: , , p,

Design Variables: ,

Layout parameters: , , stf, str

Coil spring subsystems Immediate Variables: , , ,

Vehicle level

System level

Subsystem level

united solution set tolerable solution set controllable solution set

( , )f a x b


Reliable Molecular Dynamics

Uncertain positions and velocities as a result of imprecise interatomic potentials

Uncertainty effect is assessed on-the-fly with single run of simulation

Sensitivity Analysis results ±2%

α% = 0.1%, classical intervals

Kaucherintervals

Objective: To develop an efficient alternative to traditional sensitivity analysis for MD and KMC simulations under model form and input uncertainty.


Reliable Kinetic Monte CarloEvent type Species and reactions Rate

constant

R1: water dissociation H2O ↔ OH− + H+ 101

R2: carbonic acid dissociation CO2 + H2O ↔ HCO3− + H+ 101

R3: acetic acid dissociation AcH ↔ Ac− + H+ 101

R4: reduced thionine firstdissociation

MH3+ ↔ MH2 + H+ 101

R5: reduced thionine seconddissociation

MH42+ ↔ MH3

+ + H+ 101

R6: acetate with oxidizedmediator

Ac− + MH+ + NH4+ + H2O →

XAc + MH3+ + HCO3

− + H+101

R7: oxidation doubleprotonated mediator

MH42+ → MH+ + 3H+ + 2e− 101

R8: oxidation single protonatedmediator

MH3+ → MH+ + 2H+ + 2e− 101

R9: oxidation neutral mediator MH2 → MH+ + H+ + 2e− 101

R10: proton diffusion throughPEM

H+ → H_+ 10−2

R11: electron transport fromanode to cathode

e− → e_− 10−2

R12: reduction of oxygen withcurrent generated

2H_+ + 1/2O2_ + 2e_− →H2O_

105

R13: reduction of oxygen withcurrent generated

O2_ + 4e_− + 2H2O_ →4OH_−

103

anode chamber

cathode chamber

(a) H2O in anode chamber

(b) H+ in cathode chamber


Cybermanufacturing

“Manufacturing as Service” based on service‐oriented informationinfrastructure

B2B ConnectionRemote Client

Data Service

3D Printing Service

Cybermanufacturing Gateway /Apps

Additive Mfg Service

Machining Service

EnterpriseNetwork

FEA Service

CAD Service


Manufacturing Cost Estimation with Data Analytics

Client

Cost Analytics Service Provider

❶

Feature vector

$

$ $ $ $

❷

❸

Old jobs

vector

$❹

❺

Costs of similar jobs

Prediction by simulationFinal prediction

Geometric

Non‐geometric


Ontology-Based CAD Interoperability and Verification

Promote an open environment with no restrictive single standard of features

Protect CAD software companies’ intellectual property

(a) static mapping (b) dynamic mapping

Multiscale Systems Engineering (MSSE) Research...

Documents

Transcript of Multiscale Systems Engineering (MSSE) Research...