Towards CI-enabled, high-fidelity, simulation-based engineering sciences and decision making

Omar Ghattas Ultrascale Computing Lab Biomedical Engineering Civil & Environmental Engineering Carnegie Mellon University

From physical models to simulation-based engineering sciences and decision making
physical model of natural or engineered system visualization data mining/science multiscale models

validation

optimization stochastic models uncertainty quantification

mathematical model

parameter inversion data assimilation model/data error control

e.g. design, control, operations, disaster response, manufacturing, hazard assessment, planning, treatment

Simulation-based decision making

advanced geometry & discretization schemes

data/observations computer simulation

numerical model scalable algorithms & solvers approximation error control verification

From physical models to simulation-based engineering sciences and decision making
Cy
physical model of natural or engineered system

be

rE

ng

visualization data mining/science

in ee ri

multiscale models

validation

optimization stochastic models uncertainty quantification

ng

mathematical model

parameter inversion data assimilation model/data error control

e.g. design, control, operations, disaster response, manufacturing, hazard assessment, planning, treatment

Simulation-based decision making

advanced geometry & discretization schemes

data/observations

Cy computer be simulation rIn fra s

numerical model scalable algorithms & solvers

tapproximation ru control errorc tur e

verification

Earthquake Modeling for Seismic Hazard Assessment
Volkan Akcelik , Jacobo Bielak, George Biros (UPenn), Steven Day (SDSU), Omar Ghattas, Loukas Kallivokas (Texas), Harold Magistrale (SDSU), David O’Hallaron

Region of interest for 1994 Northridge earthquake simulation

SCEC geological model provides 3D soil properties in Greater LA Basin

Adaptive grid resolves up to 1Hz freq. w/100 million grid pts; uniform grid would require 2000x more points

Snapshot of simulated ground motion (simulation requires 3hr on 6Tflops PSC machine, running at >80% parallel eff)

Comparison of observation with simulation (improved prediction requires petaflops capability)

Inversion of surface observations for 17 million elastic parameters (right: target; left: inversion result)

Multiscale Blood Flow Modeling for Artificial Heart Device Design
James Antaki, Guy Blelloch, Omar Ghattas, Marina Kameneva (Pitt), Robert Kormos (Pitt), Gary Miller, K. Rajagopal (Texas A&M), George Turkiyyah (Washington), Noel Walkington At • macroscopic (device) scales: Development of artificial heart assist device at Univ Pitt Med Center (Antaki) • Numerous advantages (size, power, reliability, non-invasiveness) • Design challenge: overcome tendency to damage red blood cells • Need macroscopic blood flow theory that accounts for blood (cell) microstructure At microscopic (cell) scales: • Macroscopic model fails in small-lengthscale regions (blade tip, rotor bearing) • Need modeling at cell scales to account for blood damage • Our mesoscopic simulations resolve interaction of RBCs elastic membrane with plasma fluid dynamics • Prospects for 3D simulation of blade-tip region: 1 week at sustained 1 petaflops/s

Real time optimization for dynamic inversion & control
Volkan Akcelik, Roscoe Bartlett (Sandia), Lorenz Biegler, George Biros (UPenn), Frank Fendell (TRW), Omar Ghattas, Matthias Heinkenshloss (Rice), David Keyes (Columbia), John Shadid (Sandia), Bart van Bloemen Waanders (Sandia), Andreas Wachter (IBM), David Young (Boeing)

Inversion and control for airborne contaminant transport
• • • sensor data provides concentrations of hazardous agents inverse problem solved to reconstruct initial conditions control problem solved to find optimal remediation strategy

Water network contaminant inversion
• • • • Nonlinear optimization problem with >300K variables and >100k controls Solution time < 2 CPU minutes  real time source detection Algorithm successful on thousands of numerical tests on several municipal water networks Formulation tool links to existing modeling software (EPANET) and powerful NLP solver (IPOPT)

Workshops on key drivers/technologies for CI
• Simulation-based engineering sciences (SBES)
o NSF-sponsored workshop held April 2004 o Report identifies key enabling technologies and challenges (multiscale modeling, V&V, large-scale simulation) o Identifies strategic applications poised to benefit from SBES/CI (medicine, energy & environment, national security, manufacturing, materials)

• Dynamic data-driven application systems (DDDAS)
o NSF-sponsored workshop in 2000 o Focuses on hardware/software/algorithms/models for integration of dynamic data with online simulations o ~20 ITR medium projects address aspects of DDDAS

• Science Case for Large Scale Simulation (SCaLeS)
o DOE-sponsored workshop June 2003 o Identifies needs and opportunities for large-scale simulation across DOE Science areas o Vol 1 report available online; Vol 2 to be published by SIAM

SBES Report: Example applications
Simulation-based planning for vascular bypass surgery. From left: MRI image data, preoperative geometric solid model, operative plan, computed blood flow velocity in aorta and proximal end of bypass, and postoperative image data used to validate predictions.

Simulation-based medicine (C. Taylor, Stanford)

Biomimetic Devices: functionalized carbon nanotube to mimic biological ion channels (N. Aluru, UIUC)

Multi-scale design of a composite component of an aircraft (SCOREC, RPI)

Summary
• CI will help catalyze a transformation to highfidelity simulation-based engineering science and decision-making • But hardware/middleware infrastructure alone is insufficient to achieve this goal • Simultaneous advances on the models, methods, and algorithms that underpin the components – and on their systematic integration to target strategic applications – are crucial for realizing the potential of CI

Final editorial comment
• Advances in models and algorithms have often to led to greater improvements in simulation capability than improvements in hardware
Example from magnetohydrodynamics: 2.5 orders of magnitude from hardware improvements; 3.5 orders of magnitude from modeling and algorithmic advances (From SCaLeS report, Vol 2 & SBES rpt)

Master your semester with Scribd & The New York Times

Special offer for students: Only $4.99/month.

Master your semester with Scribd & The New York Times

Cancel anytime.