Ateliers "sujets de recherche" - 17 may 2011
Disussions day on research themes :
17 may 2011, 9h30-17h00 amphithéatre Hermite (IHP) Institut Henri Poincaré , Paris
Morning: Stochastic and noisy simulators
Organizers : H. Monod, V. Picheny & B. Iooss
Speakers
/* Victor Picheny (ECP), Modeling and optimization of noisy functions with Gaussian Processes */
- 9h30 Hervé Monod (INRA) & Bertrand Iooss (EDF R&D) - Introduction - Slides
- 9h45 Jack P.C. Kleijnen (Tilburg University), Simulation Optimization via Bootstrapped Kriging: Survey
- 10h30 Yann Richet (IRSN), Overview of stochastic global optimization practice in nuclear criticality safety assessment - Slides
- Abstract: Nuclear criticality safety assessment aim at avoiding criticality accident within the range of operational conditions of a given system including fissile materials, described by several physical and chemical parameters. In order to check sub-criticality of the studied system, the neutrons multiplication factor (also known as k-effective) is estimated using a Monte Carlo code. The traditional “expert driven” search for penalizing safety conditions - say maximum k-effective within hypothesis range - may be supplemented by an optimization algorithm in order to enhance the whole assessment robustness. This new safety practice relies on a grid computing workbench (named Promethee) linked to the suite of R packages that implement the EGO (efficient global optimization) algorithm over noisy kriging. Following an overall presentation of this computer experiments workbench and its users feedback, we will focus on the R packages features and integration in the Promethee framework. The raising question of noise support and integration in both kriging/optimization algorithm and criticality Monte Carlo code will be openly discussed.
- 11h Pause
- 11h15 Régis Duvigneau (INRIA), Sensitivity analysis and metamodeling techniques for uncertainty quantification and robust design in aerodynamics - Slides
- Abstract: This presentation deals with the delicate issue of estimating physical uncertainties in aerodynamics. Usually, flow simulations are performed in a fully deterministic approach, although in real life operational uncertainty arises due to unpredictable factors that alter the flow conditions. We present and compare two approaches to account for uncertainty in aerodynamic simulation. Firstly, sensitivity analysis tools, at discrete or continuous levels, are used to estimate first- or second-order derivatives of aerodynamic coefficients, or entire flow fields, with respect to uncertain variables, yielding an estimate of expectation and variance values. Secondly, metamodelling techniques (radial basis functions, kriging) are employed in conjunction with Monte-Carlo simulations to derive statistical information. These methods are demonstrated for 3D Eulerian flows around the wing of a business aircraft at different regimes subject to uncertain Mach number and angle of attack. Finally, an example of robust design is presented.
- 12h00 Laurent Grosset (Renault), Some noisy simulators in automotive industry
- For more information you can see the poster presented at mascot11 by Nicolas Rosenblatt.
- 12h30 Lunch & coffee
Afternoon: Statistical learning for computer experiments
Organizer : F. Mangeant
Speakers
- 14h15 Fabien Mangeant (EADS) - Introduction - Slides
- 14h30 Guillaume Obozinski (INRIA) - Introduction to Statistical learning
- 15h15 Nabil Rachdi (EADS) Application to computer experiments in aeronautics - Slides
- 16h00 Everybody Discussions
- 16h30 End
/*
BEGIN:VCALENDAR
BEGIN:VEVENT
DTSTART;TZID=Europe/Paris:20110517T093000
DTEND;TZID=Europe/Paris:20110517T170000
SUMMARY:Atelier "Sujets de recherche" GDR MASCOT NUM
LOCATION:Amphithéâtre Hermite, Institut Henri Poincaré, 11, rue Pierre et Marie Curie, 75005 Paris
OSMLOCATION:institut henri poincaré, paris
END:VEVENT
END:VCALENDAR
*/