Report on Predictive Science Courses

advertisement
CRASH Students and
Courses
Ken Powell and Ryan McClarren
CRASH Review, October 2010
About Our Students
 Each UM and TAMU student has a home department
 Current students from







Atmospheric, Oceanic and Space Sciences (UM)
Aerospace Engineering (UM)
Applied Physics (UM)
Computer Science (UM, TAMU)
Mathematics (UM)
Statistics (UM, TAMU)
Nuclear Engineering (UM, TAMU)
 Many CRASH students co-advised
About Our Students
 Students funded from several sources
 CRASH center funds
 CRASH fellowship cost-sharing funds
 Non-CRASH fellowship or RA funds
 Students engaged in CRASH community
 Student involvement in 22 of the posters
 Strong connections to NNSA labs
 Several students visited labs in 2010
 Maginot (LANL), Pandya (LLNL), Stripling (LLNL), Zaide (LANL),
Huntington (LLNL), Starinshak (LLNL)
 Several possibilities for visits to the labs in 2011/12
 Ongoing effort to encourage this and make connections
Selected Student Research
Topics
 Modeling and Theory




Discontinous Galerkin methods for hydrodynamics
Coupling methods for rad-hydro
Radtran and turbulence effects in blast waves
Time discretization methods for radtran
 Experiments




Structure in radiative shocks
Radiative shock experiments at the OMEGA facility
Reverse radiative shocks
Hydrodynamic shock experiments at the OMEGA facility
 UQ
 Unsteady adjoints for error estimation and AMR
 Bayesian and traditional regression methods for analysis of data from highdimensional computational experiments
CRASH Courses
 Predictive Science course at TAMU




First offered Fall 2009
Taught by Ryan McClarren
Covered verification, validation, sensitivity analysis and UQ
9 students
 Uncertainty Quantification course at UM
 First offered Winter 2010
 Team-taught by James Holloway, Vijay Nair, Ken Powell
 Focused on input/output modeling, screening and sensitivity
analysis, UQ
 24 students
TAMU Predictive Science
Course

Verification






Numerical analysis preliminaries
Verification with exact solutions
Manufactured Solutions
Designing a test suite
Analyzing the results
Validation
 Validation using experimental data
 Model drill-down

Uncertainty quantification and
sensitivity analysis
 Statistics preliminaries
 Sensitivity analysis
 Stochastic uncertainty
quantification




Reliability methods
Polynomial chaos
Bayesian inference / Calibration
Dealing with epistemic uncertainty
Structure of TAMU Course
 The course had students from nuclear engineering,
geophysics, and statistics.
 Course was lectured-based with graded homework.
 Final project covered a topic of the student’s choice.
 In final project the students had to include
 V&V of the code/model they were using
 Uncertain inputs
 A prediction with quantified uncertainty
Sample Final Projects – TAMU
course
 Several used polynomial chaos techniques to
 Compute the uncertainty in dose for a radiation shielding
calculation.
 Predict maximum temperature / flux in coupled neutronics
heat conduction simulation.
 Compute sensitivity to coupling schemes in multiphysics
problems.
 Predict the spectral radius of a transport solve using a
Kennedy-O’Hagan model
UM Uncertainty Quantification
Course
 Introduction
 Sources and types of

 Sampling the input space
(Monte Carlo, Latin
Hypercube, design of
experiments)
uncertainty
 Key probability concepts used
in UQ
 Uncertainty quantification
 Overview of UQ process
 Estimating output
uncertainties
Input/Output Modeling
 Reducing input uncertainties
 Emulators and response
for a new prediction
surfaces
 Estimating model discrepancy
 Parametric regression
function
 Semi-parametric modeling
 Building a predictive model
(MARS, MART)
 Gaussian process modeling
Structure of UM course
 Lectures
 Lab sessions
 Introduced background probability and statistics information
 Introduced software (MATLAB, R)
 Worked examples related to lectures
 Homeworks introduced key concepts, based on a simple
simulation code written by each student (trajectory of a
ball)
 Final projects, based on their own research, presented in
final weeks of class
Sample Final Projects – UM
course
 MARS/MART analysis of drag in a Mars Re-entry system
 Gaussian process modeling and Markov-Chain Monte
Carlo for turbulence model calibration
 UQ analysis of a Fischer-Tropsch synthesis process
 Constructing and sampling a response surface for
radiative heat transfer in a scramjet
 UQ in military ground vehicle blastworthiness simulations
Blastworthiness Project
Concluding Remarks
 Both classes drew a diverse student group – there is real
interest in the topic among grad students
 The combination of simpler homework sets to teach the
concepts and final projects driven by the students’
research worked well
Download