Aiming for Zero Tracking Error in Control System by Iterating in Hardware Instead of Software

1. Speaker : Richard W. Longman

   Columbia University

   New York, NY USA

2. Abstract

   In engineering one usually develops mathematical models of the world and uses them to design and optimize system performance. Iterative learning control (ILC) is a relatively new field that aims to optimize control system tracking performance by iterating with the real world -- instead of iterating on a computer with a model. Experiments demonstrate that such methods can reach particularly small error levels that could never be reached by use of a mathematical model. To make ILC effective it must do a delicate balance – it must use information from a model to help it to quickly decrease the error, but it must not rely so heavily on the model that model inaccuracy is allowed to interfere with the ability to decrease toward zero error in the world, decreasing beyond the accuracy of one’s model. Iterative learning control methods are presented for linear systems. Then techniques are presented to extend the use of such methods to apply to multi-input, multi-output nonlinear systems.

Robust Optimization Issues in Parameter Estimation and Optimum Experimental Design for DAE models

1. Speaker : Ekaterina Kostina

   Faculty of Mathematics and Computer Science

   University of Marburg, German

   kostina@Mathematik.Uni-Marburg.de

2. Abstract

   Estimating model parameters from experimental data is crucial to reliably simulate dynamic processes.

   The identification problem can be described as follows. Let the dynamics of the model be described by a system of differential algebraic equations where the right-hand side depends on an unknown vector of parameters. It is assumed that there is a possibility to measure a signal of an output device that writes at given time points the output signal of the dynamic system with some errors. According to the common approach, in order to determine the unknown parameters the optimization problem is solved in which the special functional is minimized under constraints that describe the specifics of the model. Any norm of the measurement errors may be used as the functional in the optimization problem. The choice of an adequate norm depends on the statistical properties of the measurement errors. The traditional choice is weighted l2-norm. In practical applications, however, it often appears that the data contains outliers. Thus, a reliable parameter estimation procedure (e.g. based on l1-norm) is necessary that deliver parameter estimates less sensitive (robust) to errors in measurements.

   Another difficulty that occurs in practical applications is that the experiments performed to obtain necessary measurements are expensive, but nevertheless do not guarantee sufficient identifiability. The optimization of one or more dynamic experiments in order to maximize the accuracy of the results of a parameter estimation subject to cost and other technical inequality constraints leads to very complex non-standard optimal control problems. One of the difficulties is that the objective function is a function of a covariance matrix and therefore already depends on a generalized inverse of the Jacobian of the underlying parameter estimation problem. Another difficulty is that the optimization results depend strongly on the assumed values of parameters which are only known to lie in a - possibly large – confidence region. Hence, robust optimal experiments are required that solve worst-case (min-max) optimization problems.

   The talk presents new effective algorithms for robust parameter estimation and design of robust optimal experiments in dynamic systems. Numerical results for real-life applications from chemistry and chemical engineering will be presented.

   This talk is based on joint work with H. G. Bock, S. Koerkel and J. P. Schloeder.

2D and 3D non-photorealsitic rendering for the humanities

1. Speaker : Michael J. Winckler

   INF 368, 69120 Heidelberg, Germany

   Michael.Winckler@iwr.uni-heidelberg.de

   http://www.iwr.uni-heidelberg.de/%7EMichael.Winckler

2. Abstract

   The production of photorealistic images is a prosperous area of research in computer graphics. The aim is to produce pictures that are virually indistinguishable from photographic imapes. In contrast, non-photorealistic rendering (NPR) focuses on image generation techniques to produce pictures resembling artistic or graphic rendering styles which are clearely discernible from photographies. Application areas of NPR are architectural drawings, imitation of artistic drawings and cartoon generation. While the use of NPR for artistic purposes is widely recognized, applications to the humanities only lately aim at the possibility to produce images that show possible results of research through the execution of adapted rendering styles.

GREEN-* : towards energy efficient solutions for next generation large scale distributed systems

1. Speaker : Laurent Lefevre

   INRIA RESO, University of Lyon, Ecole Normale Superieure de Lyon,

   46, allee d'Italie - 69364 LYON Cedex 07 - FRANCE

   laurent.lefevre@inria.fr

   http://perso.ens-lyon.fr/laurent.lefevre

2. Abstract

   With the emergence of large scale Grids and data cen! ters, the amount of energy (watts) required for high performance distributed computing and networking becomes a real challenge. Taking into account of an efficient energy usage will have an impact on how we design architectures, services and protocols. Various "green" approaches (like "Green Grid", "Green500", "Green Internet"..) are currently proposed by academics and industrial consortiums. This talk will review current challenges and solutions associated to power aware approaches in distributed systems.

Computational Fluid Dynamics and Its Applications Using PC Clusters

1. Speaker : Jang-Hyuk Kwon

   Department of Aerospace Engineering, KAIST

   373-1 Guseong-dong, Yuseong-gu, Daejeon, 305-701 South Korea

   jhwkon@kaist.ac.kr

2. Abstract

   These days, the supercomputing is popular everywhere including computational fluid dynamics. Still, the cost for supercomputing is high in cost for researchers in academia, hence a PC cluster is one of the solutions for high performance computing.

   In this presentation, some of tools and CFD algorithms developed in my lab are explained. Firstly, a grid generator, KGRID is introduced. This is a structured grid generation and visualization tool which can handle complex configurations with multi-block grid and Chimera grid generations. Secondly, the algorithms developed for Euler and Navier-Stokes equations are explained. The flow solver, KFLOW has all Mach number flow calculation capability with high order hybrid schemes and convergence acceleration techniques.

   The application problems are; flow computation for a wing-body configuration and aircrafts, prediction of store separation with Chimera grid technique, hyperbolic flow calculation with HLLE+ scheme, and prediction of dynamic damping coefficients of missiles and rockets. Prediction of flutter speed in transonic flow of an aircraft is one of important but time consuming applications using CFD.

   Another popular application of CFD these days is the aerodynamic shape optimization. The design optimizations using the adjoint method, genetic algorithms, and reliability based method are explained briefly with some applications, such as the drag minimizations for an aircraft wing, an aircraft and a ship. Also, the multidisciplinary design optimization (MDO) is introduced briefly with applications.