Dynamic Data Driven Application Systems
(DDDAS)
A new paradigm for
applications/simulations
and
measurement methodology
(Symbiotic Measurement&Simulation Systems)
Dr. Frederica Darema
Senior Science and Technology Advisor
Director, Next Generation Software Program
Director, Biological Information technology Systems
1
NSF
What is DDDAS
Experiment
Measurements
Field-Data
User
2
Experiment
Measurements
Field-Data
User
Challenges:
Application Simulations Development
Algorithms
Computing Systems Support
Examples of Applications benefiting from the new paradigm
• Engineering (Design and Control)
– aircraft design, oil exploration, semiconductor mfg, structural eng
– computing systems hardware and software design
(performance engineering)
• Crisis Management
– transportation systems (planning, accident response)
– weather, hurricanes/tornadoes, floods, fire propagation
• Medical
– customized surgery, radiation treatment, etc
– BioMechanics /BioEngineering
• Manufacturing/Business/Finance
– Supply Chain (Production Planning and Control)
– Financial Trading (Stock Mkt, Portfolio Analysis)
DDDAS has the potential to revolutionize
science, engineering, & management systems
3
Outline
* Background and New Directions
• Examples of Dynamic Data-Driven Application Systems (DDDAS)
• CurrentTechnology Trends
– Applications, Platforms ( Grids)
• Why now is the time for DDDAS
* Enabling DDDAS - Challenges and Approaches
• Systems’ Software
– Performance Engineering/ Systematic methods for building sw/hw systems
– Dynamic application composition and Run-Time support
• DDDAS application efforts
• Algorithms for DDDAS
* Agency Efforts
• Existing programs
• Future Initiatives
* Technology transfer to industry
4
NSF March 2000 Workshop on DDDAS
(Co-Chairs: Craig Douglas, UKy; Abhi Desmukh, UMass)
Invited Presentations
• New Directions on Model-Based Data Assimilation (Chemical Appl’s)
Greg McRae, Professor, MIT
• Coupled atmosphere-wildfire modeling
Janice Coen, Scientist, NCAR
• Data/Analysis Challenges in the Electronic Commerce Environment
Howard Frank, Dean, Business School, UMD
• Steered computing - A powerful new tool for molecular biology
Klaus Schulten, Professor, UIUC, Beckman Institute
• Interactive Control of Large-Scale Simulations
Dick Ewing, Professor, Texas A&M University
• Interactive Simulation and Visualization in Medicine: Applications to
Cardiology, Neuroscience and Medical Imaging
Chris Johnson, Professor, University of Utah
• Injecting Simulations into Real Life
Anita Jones, Professor, UVA
Workshop
Report: www.cise.nsf.gov/eia/dddas
5
Fire Model
• Sensible and latent heat
fluxes from ground and
canopy fire -> heat fluxes
in the atmospheric model.
• Ground heat flux used to
dry and ignite the canopy.
• Fire’s heat fluxes are
absorbed by air over a
specified extinction depth.
• 56% fuel mass -> H20 vapor
• 3% of sensible heat used
to dry ground fuel.
Kirk Complex Fire. U.S.F.S. photo
6
Slide Courtesy of Cohen/NCAR
Coupled atmospheric and wildfire models
7
Slide Courtesy of Cohen/NCAR
AMAT Centura Chemical Vapor Deposition Reactor
Operating Conditions
Reactor Pressure
Inlet Gas Temperature
Surface Temperature
Inlet Gas-Phase Velocity
1 atm
698 K
1173 K
46.6 cm/sec
Surface Reactions
Gas Phase Reactions
SiCl3H  HCl + SiCl2
SiCl2H2  SiCl2 + H2
SiCl2H2  HSiCl + HCl
H2ClSiSiCl3  SiCl4 + SiH2
H2ClSiSiCl3  SiCl3H + HSiCl
H2ClSiSiCl3  SiCl2H2 + SiCl2
Si2Cl5H  SiCl4 + HSiCl
Si2Cl5H  SiCl3H + SiCl2
Si2Cl6  SiCl4 + SiCl2
SiCl3H + 4s  Si(B) + sH + 3sCl
SiCl2H2 + 4s  Si(B) + 2sH + 2sCl
SiCl4 + 4s  Si(B) + 4sCl
HSiCl + 2s  Si(B) + sH + sCl
SiCl2 + 2s  Si(B) + 2sCl
2sCl + Si(B)  SiCl2 + 2s
H2 + 2s  2sH
2sH  2s + H2
HCl + 2s  sH + sCl
sH + sCl  2s + HCl
Slide Courtesy of McRae/MIT
8
Some Technology Challenges in
Enabling DDDAS
• Application development
– interfaces of applications with measurement systems
– dynamically select appropriate application components
– ability to switch to different algorithms/components
depending on streamed data
• Algorithms
– tolerant to perturbations of dynamic input data
– handling data uncertainties
• Systems supporting such dynamic environments
– dynamic execution support on heterogeneous
environments
– GRID Computing, and Beyond!!!
9
Why Now is the Time for DDDAS
• Technological progress prompted advances in some
of the challenges
– Computing speeds advances (uniprocessor and
multiprocessor systems), Grid Computing
– Systems Software
– Applications Advances (parallel & grid computing)
– Algorithms advances (parallel &grid computing,
numeric and non-numeric techniques)
• Examples of efforts in:
– Systems Software
– Applications
– Algorithms
10
What is Grid Computing?
coordinated problem solving
on dynamic and heterogeneous resource assemblies
DATA
ACQUISITION
ADVANCED
VISUALIZATION
,ANALYSIS
QuickTime™ and a
decompressor
are needed to see this picture.
COMPUTATIONAL
RESOURCES
IMAGING INSTRUMENTS
LARGE-SCALE DATABASES
Example: “Telescience Grid”, Courtesy of Ellisman & Berman /UCSD&NPACI
11
Application Directions
– Monolithic
– One programming language
–
–
–
–
Multi-Modular
Multi-Source Data
Multi-Language
Multiple Developers
–
–
–
–
–
–
–
– Computation Intensive
– Batch
– Hours/Days
Computation Intensive
Data Intensive
Few Minutes/hours
Real Time Turn-around
Visualization (real time)
Interactive Steering by user
and ...
DDDAS
Such characteristics require new capabilities in systems software
12
Some Examples
of Today’s Applications
13
The e-Business / (CIM, CIE)
Order Processing
Customer Service
Sales Management
Process Coordination
Management &
Monitoring
Manufacturing
Product DBs
Inventory Shipping
Enterprise Messaging
Business
to
Customer
Web
e-commerce
14
Business
to
Business
Mobile Workers
Knowledge Workers
Business Communications
Distributor
Channel
Compare with
Classical (Old) Supply Chain
Parts
Supplier
Parts
Supplier
Manufacturing
Distribution
Retail
Customer
Customer
Manufacturing
Distribution
Retail
Customer
Customer
Manufacturing
Distribution
Retail
Customer
Customer
Transportation Supplier
15
MSTAR
(DARPA)
(Moving and Stationary Target Acquisition and Recognition)
TREES
Focus of
Attention
H2O
Index Database
(created off-line)
GRASS
ROAD
...
Segmented
Terrain Map
..
.
Regions of
Interest (ROI)
Target
& Clutter
Database
SAR Image &
Collateral Data
- DTED, DFAD
- Site Models
- EOSAT imagery
Search Tree
ROI Hypothesis
y
GRASS

BMP2
Indexing
Local
Scene Map
x
Target & Scene
Model Database
(created off line)
TREES
TREES
Task Predict
Task Extract
ROI Hypothesis
Predict
Shadow
(?)
y

BMP-2
x
CAD
TREES
GRASS
TREES
Statistical
Model
Search
Extract
Local
Scene Map
Match Results
Semantic
Tree
Clutter
Database
Form Associations
Refine Pose & Score
Analyze Mismatch
Tree
Clutter
Shadow
Obscuration ?
x1,y1, 
x2,y2, 
Score = 0.75
Feature-to-Model
Traceback
16
Match
Ground
Clutter
Platform Directions
– Vector Processors, SIMD MPPs
– Distributed Memory MPs
– Shared Memory MPs
– Distributed Platforms,
Heterogeneous Computers and Networks
• Heterogeneity
– architecture (compute &network)
– node power (supernodes, PCs)
• Latencies
– variable (internode, intranode)
• Bandwidths
– different for different links
– different based on traffic
Petaflops Platform
(Grid-in-a-Box)
Distributed Platform
tac-com
alg accelerator
….
MPP
17
NOW
fire
cntl
data
base
data
base
fire
cntl
SAR
SP
TeraGrid: 13.6 TF, 6.8 TB memory, 79 TB internal disk, 576 network disk
ANL
Extreme
Blk Diamond
Caltech
0.5 TF
.4 TB Memory
86 TB disk
32
32
256p HP X-Class
24
128p HP V2500
1 TF
.25 TB Memory
25 TB disk
574p IA-32
Chiba City
32
24
128p Origin
32
24
8
32
32
92p IA-32
4
8
5
HPSS
HR Display &
VR Facilities
5
HPSS
OC-48
NTON
OC-12
Calren
OC-12 ATM
Chicago & LA DTF Core Switch/Routers
Cisco 65xx Catalyst Switch (256 Gb/s Crossbar)
Juniper M160
OC-48
OC-12
GbE
vBNS
Abilene
Calren
ESnet
SDSC
OC-12
OC-12
4.1 TF
2 TB Memory
225 TB SAN
OC-12
OC-3
4
NCSA
ESnet
HSCC
MREN/Abilene
Starlight
OC-12
6+2 TF
4 TB Memory
240 TB disk
OC-12
OC-3
vBNS
Abilene
MREN
8
HPSS
UniTree
300 TB
2
4
10
1024p IA-3
320p IA-6
1176p IBM SP
1.7 TFLOPs
Blue Horizon
Sun
Server
Myrinet
4
14
Myrinet
15xxp Origin
16
2 x Sun E10K
18
Slide Courtesy of Berman/NPACI
Grids Form the Basis of a National
Information Infrastructure
August 9, 2001: NSF
Awarded $53,000,000
to SDSC/NPACI
and NCSA/Alliance
for TeraGrid
TeraGrid will
provide in aggregate
•
•
•
•
13.6 trillion calculations per second
Over 600 trillion bytes of immediately accessible data
40 gigabit per second network speed
Provide a new paradigm for data-oriented computing
• Critical for disaster response, genomics, environmental modeling,
etc.
19
Slide Courtesy of Berman/NPACI
Examples Other Agencies Grid Efforts
DARPA
NASA’s Information Power Grid
• SF Express (Synthetic Forces Express)
– LargeScale distributed, interactive,
battle simulation
– Simulation decomposed terrain
contiguously among supercomputers
– Simulation of 50,000 entities in
8/97, 100,000 entries in 3/98
NSF and DoE
•
CACTUS/GriPhyN (ITR, NGS, SciDAC)
– Toolkit for Large-Scale Relativity
Simulations
– Largest Simulations for Colliding
Black Holes
– International Team/Grid
20
Why Now is the Time for DDDAS ?
• Technological progress prompted advances in some
of the challenges
– Computing speeds advances (uniprocessor and
multiprocessor systems), Grid Computing
– Applications Advances (parallel & grid computing)
– Algorithms advances (parallel &grid computing,
numeric and non-numeric techniques)
• Examples of efforts in:
– Systems Software
– Applications
– Algorithms
21
Agency Efforts
NSF
– NGS: The Next Generation Software Program
• develops systems software supporting dynamic resource execution
– ITR: Information Technology Research (NSF-wide)
• has been used as an opportunity to support DDDAS related efforts
• 46 DDDAS pre-proposals; many meritorious
• about 24 proposals; 8 were awarded
… more on this, next slide….
– Gearing towards a DDDAS initiative
• expect participation from all NSF Directorates:
CISE, MPS, ENG, BIO, GEO, SBE, HER
DARPA, NASA, DoE
– have related programs (NASA/IPG, DoE/SciDAC)
– and interested in DDDAS
22
“DDDAS” proposals awarded
in FY01 ITR Competition
• Biegler – Real-Time Optimization for Data Assimilation and Control
of Large Scale Dynamic Simulations
• Car – Novel Scalable Simulation Techniques for Chemistry, Materials
Science and Biology
• Knight – Data Driven design Optimization in Engineering Using
Concurrent Integrated Experiment and Simulation
• Lonsdale – The Low Frequency Array (LOFAR) – A Digital Radio
Telescope
• McLaughlin – An Ensemble Approach for Data Assimilation in the
Earth Sciences
• Patrikalakis – Poseidon – Rapid Real-Time Interdisciplinary Ocean
Forecasting: Adaptive Sampling and Adaptive Modeling in a
Distributed Environment
• Pierrehumbert- Flexible Environments for Grand-Challenge Climate
Simulation
• Wheeler- Data Intense Challenge: The Instrumented Oil Field of
the Future
23
24
25
The NGS Program develops
Performance Engineering Technology
Performance Models & Measurements
Distributed Applications
Application
Collaboration
Environments
Scalable I/O Authenication/
Data Management Authorization
Archiving/Retrieval Dependability
Services
Services
Visualization
Languages
Compilers
Libraries
Tools
..
Distributed Systems Management
Global
Management
Distributed, Heterogeneous, Dynamic, Adaptive
Computing Platforms and Networks
Computing
Engine
Memory
Technology
26
Other Services .
API
&
Runtime
Services
CPU
Technology
Device
Technology
...
Components
Technology
Enables Analysis in
Multiple views of the system
(The applications’ view)
Application
Distributed Applications
Models
...
IO / File
Models
OS
Scheduler
Models
Architecture /
Network
Models
Memory
Models
27
Languages
Compilers
Libraries
Tools
Collaboration
Visualization
Environments
Scalable I/O Authenication/
Data Management Authorization
Archiving/Retrieval Dependability
Services
Services
Other Services . . .
Distributed Systems Management
Distributed, Heterogeneous, Dynamic, Adaptive
Computing Platforms and Networks
Memory
Technology
CPU
Technology
Device
Technology
...
28
Technology Gap
Example case: Distributed Application
Platform
Dynamic Analysis
Situation
Launch
Application(s)
Programming Model
Constraint
•Network of
Workstations
(NOW)
• Message passing
• Static partition
• Inefficient
load-balancing
•Symmetric
Multiprocessor
(SMP)
•Shared queue
•Dynamic allocation
of work
•Application
“re-write”
required
•Cluster of
SMPs
•Message-passing
across SMPs
•Shared queue
within SMP
• Application cannot
be repartitioned dynamically
when problem size or
number of SMPs changes
Distributed Computing Resources
Distributed Platform
tac-com
alg accelerator
….
29
MPP
NOW
fire
cntl
data
base
data
base
fire
cntl
SAR
SP
The NGS Program developsTechnology for integrated feedback & control
Runtime Compiling System (RCS) and Dynamic Application Composition
Application
Model
Dynamic Analysis
Situation
Distributed
Programming
Model
Application
Program
Compiler
Front-End
Application
Intermediate
Representation
Compiler
Back-End
Launch
Application (s)
Dynamically
Link
&
Execute
Performance
Measuremetns
&
Models
Application
Components
&
Frameworks
Distributed Computing Resources
Distributed Platform
tac-com
alg accelerator
….
30
MPP
NOW
fire
cntl
data
base
data
base
fire
cntl
SAR
SP
Example of NGS supported effort:
• The GrADS Project (Grid
•
Application Development Software)
• Design and development of a
Grid program development and
•
execution environment
Tight coupling between program
preparation and program
execution environment
Contract-based performance
economy
Performance feedback
Perf
problem
Software
components
P
S
E
Source
application
whole
program
compiler
Config.
object
program
Realtime
perf
monitor
Scheduler/
Service
Negotiator
negotiation
Grid
runtime
System
(Globus)
Dynamic
optimizer
libraries
31
Slide Courtesy of GRADS group
NGS fosters
Employing Performance Engineering Technology for:
Application Composition and Run-Time Support
on
Dynamic, Heterogeneous Computing Platforms
so that the users
“Shouldn’t Have to be Heroes
to Achieve Grid Program Performance”
and... because heroism is not enough
32
33
Challenges
• Application development
– develop interfaces of applications with measurement
systems
– dynamically select appropriate application components
– need to switch to different algorithms/components
depending on streamed data
• Algorithms
– tolerant to perturbations of dynamic input data
– handling data uncertainties
• Systems supporting such dynamic environments
– need Performance Engineering technology
– Application Composition Frameworks
– Dynamic Run-Time Support
34
Some more Challenges on
Applications Development Issues
• Handling Data Streams in addition to Data Sets
• Handling different data structures – semantic
information
• Interfaces to Measurement Systems
- Interactive Visualization and Steering
• Standards for data exchange
• Combining Local and Global Knowledge
• Model Interactions
• Application control of measurement systems
• Dynamic Application Composition and Runtime
support
Examples from ITR supported efforts:
35
NSF ITR Project
A Data Intense Challenge:
The Instrumented Oilfield of the Future
PI: Prof. Mary Wheeler, UT Austin
Multi-Institutional/Multi-Researcher Collaboration
36
Slide Courtesy of Wheeler/UTAustin
Highlights of Instrumented Oilfield Project
I.
Motivation:
Field instrumentation for information technology
and computational science essential for
monitoring and optimizing oil and gas
production.
Integration yields:
THE INSTRUMENTED OILFIELD
II.
Field Technologies:
Time-lapse surface and borehole seismic,
permanent downhole sensors, intelligent well
completions, fiber optics, and remote control
operations
37
Slides Courtesy of Wheeler/UTAustin
Highlights of Instrumented Oilfield Proposal
III. IT Technologies:
Data management, data visualization, parallel
computing, and decision-making tools such as
new wave propagation and multiphase, multicomponent flow and transport computational
portals, reservoir production:
THE INSTRUMENTED OILFIELD
IV. Major Outcome of Research:
Computing portals which will enable reservoir
simulation and geophysical calculations to
interact dynamically with the data and with each
other and which will provide a variety of visual
and quantitative tools. Test data provided by oil
and service companies
38
Economic Modeling and Well Management
Production Forecasting
Well Management
Reservoir
Performance
Data
Analysis
Simulation Models
Multiple Realizations
Data Management and Manipulation
Data Collections from Simulations and
Field Measurements
39
Visualization
Field
Measurements
Reservoir Monitoring
Field Implementation
Highlights of Instrumented Oilfield Proposal
Simulation Framework
40
ITR Project
A Data Intense Challenge:
The Instrumented Oilfield of the Future
II.
41
Industrial Support (Data):
i.
British Petroleum (BP)
ii.
Chevron
iii. International Business Machines (IBM)
iv. Landmark
v.
Shell
vi. Schlumberger
Poseidon
Rapid Real-Time
Interdisciplinary Ocean Forecasting:
Adaptive Sampling and Adaptive Modeling
in a Distributed Environment
Nicholas M. Patrikalakis, Henrik Schmidt, MIT
Allan R. Robinson, James J. McCarthy,
Harvard
http://czms.mit.edu/poseidon
42
Ocean Science Issues
• Data driven simulations via data assimilation
• Simulation driven adaptive sampling of the
ocean
• Interdisciplinary ocean science: interactions
of physical, biological, acoustical phenomena
• Extend state-of-the-art via feedback from
acoustics to physical&biological oceanography
• Application in fisheries management, but also
in oil-slick containment
43
Interdisciplinary Ocean Science
44
System Architecture
45
Data Driven Design Optimization in Engineering
Using Concurrent Integrated Experiment and Simulation
Doyle Knight, Rutgers-The State University of New Jersey
Khaled Rasheed, University of Georgia
• Channel or enclosure with isolated heat sources
(i.e., electronic components)
• The maximum surface temperature must be
maintained below a specified level by the flow of
air or dielectric liquid
• Control flow of air or liquid for optimal heat
dissipation
46
AMAT Centura Chemical Vapor Deposition Reactor
Operating Conditions
Reactor Pressure
Inlet Gas Temperature
Surface Temperature
Inlet Gas-Phase Velocity
1 atm
698 K
1173 K
46.6 cm/sec
Surface Reactions
Gas Phase Reactions
SiCl3H  HCl + SiCl2
SiCl2H2  SiCl2 + H2
SiCl2H2  HSiCl + HCl
H2ClSiSiCl3  SiCl4 + SiH2
H2ClSiSiCl3  SiCl3H + HSiCl
H2ClSiSiCl3  SiCl2H2 + SiCl2
Si2Cl5H  SiCl4 + HSiCl
Si2Cl5H  SiCl3H + SiCl2
Si2Cl6  SiCl4 + SiCl2
SiCl3H + 4s  Si(B) + sH + 3sCl
SiCl2H2 + 4s  Si(B) + 2sH + 2sCl
SiCl4 + 4s  Si(B) + 4sCl
HSiCl + 2s  Si(B) + sH + sCl
SiCl2 + 2s  Si(B) + 2sCl
2sCl + Si(B)  SiCl2 + 2s
H2 + 2s  2sH
2sH  2s + H2
HCl + 2s  sH + sCl
sH + sCl  2s + HCl
Slide Courtesy of McRae/MIT
47
Benefits of Model Based Process Controllers
Economic Benefits
Em bedded on-board computer
Controller
Controller
l
Estimated wafer
temps. and
oxidatio n rates
l
l
Process
Process model
model
l
Cost per deposition reduced by 25%
Process cycle time cut by 20%.
Stable operations 3 times faster
3s uniformity < 4% goal
Measured Process Improvement
Chemkin Models
of Equipment
- TWAFER
- OVEND
3-Sigma Stand. Dev. (°C)
(SVG-Thermco AVP)
10
Model -B ased Contr ol
8
Standard Contr ol
6
4
2
0
10
* Controller design and implementation by Relman, Inc.
* Cray test and evaluation by SEMATECH
48
15
20
25
30
35
40
45
Stabilization time (m in.)
50
Multi-Scale Integration of Software Systems
XML As a standard for data exchange
Experimental Data
& Quantum
Chemistry
(Gaussian, DFT)
0
CFD Model of
Reactor Flow
(STARCD,
CFDRC,..)
Design
Optimization
(MINLP,
Minos)
+ NH3
20
- CH 4
60
32
+ NH
15 - CH 3
4
0
55
Close-Spa ced RD rea ctor
[Ga* ]
[GaN* ]
Horizontal
Distributed Computing Resources
Jensen - MIT
49
Interacting with “Black Box” Models
UNCERTAINTY ANALYSIS ENVIRONMENT
( Standardized Interfaces )
Uncertainty Description
•
•
•
•
Input distributions
Analysis variables
Computer environment
Input/output files
Problem
Specific
Compilers
OLE, Corba
Uncertainty Analysis
Integrated Database
Management System
BLACK BOX
SYSTEMS
•
•
•
•
Output density functions
Analysis of variance
Local sensitivity analysis
Key parameters
OLE, Corba
• Process Modeling Tools
• Chemical Modeling Tools
• Mathematical Modeling Tools
50
Problem
Specific
Output
Parsers
Some more on
need for New Algorithms
• Data Driven Application/Algorithmic Components
• e.g. need for adaptive re-meshing
• Multiple Scales and Model Reduction
• Uncertainties in Streamed Data
•
•
•
•
new algorithms for uncertainty propagation
decision-making metrics in presence of uncertainty
database design for representation of uncertainties
valuation of cost of safety factors - bounds of
uncertainty
• Optimization and Inverse Problem
51
Real Time Optimization for Data Assimilation and Control
In Large Scale Dynamic Simulations (Bielak, Ghattas, et al
Parameter Estimation: Seismic Inversion
• Forward Problem
: Given
soil material and earthquake
source parameters, find
earthquake ground motion.
• Inverse Problem : Given
earthquake observations,
estimate material and
source parameters.
52
Slides Courtesy of Ghattas/CMU
Parameter Estimation: Source Localization for
Atmospheric Release of Hazardous Materials
•
•
Atmospheric Release Advisory Capability (ARAC) project at
Lawrence Livermore National Lab (Gayle
Sukiyama et al.)
o
o
o
o
o
53
Atmospheric release of hazardous nuclear/chemical/biological
material
Collection of local meteorological data
Transport/diffusion model predicts extent of spread
Real-time inversion problem: source localization from sensor data
See http://www. llnl . gov / ees /NARAC/ arac .html
Forward vs. Inverse Problem
• PDE model
• Forward problem:
o Given “data” x (e.g. material coefficients, domain and
boundary sources, boundary and initial conditions,
geometry), find state
u
• Inverse problem:
o Given desired goal involving
u , find x
o Arise in many DDDAS scenarios
• parameter estimation
• data assimilation
• optimal control
54
Summary Remarks
•
PDE-constrained optimization key enabling technology for
DDDAS problems
o
o
•
Fast algorithms can be designed for many problems classes
o
•
Turnaround time small constant multiple of PDE solve
Numerous algorithmic challenges remain
o
o
o
o
o
o
o
o
o
55
Data assimilation & parameter estimation
Optimal control
Memory vs. work tradeoff for time-dependent
adjoint PDEs
Non-smooth problems
Inexact Jacobians
nd derivatives
Approximation of 2
Robustness to noise
Ill-conditioning and illposedness , regularizaton
Real time aspects (bootstrapping, early termination)
Pointwise inequality constraints
Physics-based globalizations
Modeling Uncertainty
Irreducible versus epistemic uncertainty
•
•
•
•
•
•
Stochastically-excited structures
Boundary conditions, geometry, properties
Sensitivity/failure analysis
Gaussian and non-Gaussian processes
Polynomial Chaos vs. Monte Carlo
Stochastic spectral/hp element methods
“…Because I had worked in the closest possible ways with
physicists and engineers, I knew that our data can never be precise…”
Norbert Wiener
Slides Courtesy of Karniadakis/Brown
56
Partially Correlated non-Uniform Random Inflow
•Deterministic
•Stochastic
•Pressure
Vorticity: Regions of Uncertainty
57
Non-uniform Gaussian Random BC
• Exponential correlation
C( x 1 , x 2 )  s 2 e  x  x
1
• Stochastic input: s  0.1
Umean along centerline
58
• 2D K-L expansion
2
/b
• 4th-order Hermite-Chaos expansion
• 15-term expansion
Vmean along centerline
Non-uniform Exponential Random BC
• Exponential correlation
C( x 1 , x 2 )  s 2 e  x  x
1
• Stochastic input: s  0.1
Umean along centerline
59
• 2D K-L expansion
2
/b
• 4th-order Laguerre-Chaos expansion
• 15-term expansion
Vmean along centerline
Research Opportunities in Uncertainty
lUUncertainty
analysis is a fertile and much needed area
for inter-disciplinary research
lEEstimates of uncertainties in model inputs are
desperately needed
Uncertainty  Ignorance
60
Additional Considerations/Requirements
on Hardware and Software Systems
• Extended Spectrum of platforms
– Assemblies of Sensor Networks and Computational Grid
platforms
• Systems Architectures including Measurement
Systems
• Programming Environments
• Application, System, and Resource Management
• Models of the Computational Infrastructure
• Security and Fault Tolerance
• DDDAS will accentuate and create the need for
advances in such areas
61
Programming Environments
• Procedural - > Model Based
• Programming -> Composition
• Custom Structures -> Customizable Structures
(patterns, templates)
• Libraries -> Frameworks ->
Compositional Systems
(Knowledge Based Systems)
• Application Composition Frameworks and
Interoperability extended to include measurements
• Data Models and Data Management
– Extend the notion of Data Exchange Standards
(Applications and Measurements)
62
Research and Technology Roadmap
(emphasis on multidisciplinary research)
Application Composition System
}
•Distributed programming models
•Application performance Interfaces
•Compilers optimizing mappings on complex
systems
..
.
}
Application RunTime System
•Automatic selection of solution methods
•Interfaces, data representation & exchange
•Debugging tools
..
.
Measurement System
}
•Application/system multi-resolution models
•Modeling languages
•Measurement and instrumentation
..
.
Y1
i
n
t
e
g
r
a
t
i
o
n
Y2
Exploratory
i
n
t
D E
Eg
r
a
t
i
o
n
Y3
Y4
M
O
Y5
Development
Integration & Demos
63
S
Providing
enhanced
capabilities
for
Applications
Agency Efforts
• NSF
– NGS, SES, ITR
ITR broad, NSF-wide
– Gearing for DDDAS initiative
• will provide a focus for new exciting work in
applications, algorithms and systems’ areas
NSF report www.cise.nsf.gov/eia/dddas
• Also DARPA, NASA, DoE interested
64
What about Industry &DDDAS
• Industry has history of
– forging new research and technology directions and
– adapting and productizing technology which has demonstrated promise
• Need to strengthen the joint academe/industry
research collaborations; joint projects / early stages
• Technology transfer
– establish path for tech transfer from academic research to industry
– joint projects, students, sabbaticals (academe <----> industry)
• Initiatives from the Federal Agencies / PITAC
• Cross-agency co-ordination
• Effort analogous to VLSI, Networking, and
Parallel and Scalable computing
• Industry is interested in DDDAS
65
DDDAS has potential
for significant impact to
science, engineering, and commercial world,
akin to the transformation effected
since the ‘50s
by the advent of computers
http://www.cise.nsf.gov/eia/dddas
66