Seminars Spring 2008

Le Laboratoire d’Automatique de l’EPFL a le plaisir de vous inviter aux séminaires selon la liste ci-après. Une mise à jour régulière des informations concernant ces séminaires est disponible à l’adresse sur cette page. En particulier, il est conseillé aux visiteurs externes de vérifier que les séminaires soient dispensés comme prévu ci-dessous.

Where: Salle de séminaire LA-EPFL, ME C2 405 (2è étage), 1015 Lausanne

When: Friday at 10.15am (Except seminar by Prof. R. Bitmead on Tuesday at 10.15am)

Spring 2008 seminars

Experimental Certification of Jet Engine Controllers

01.04.2008 (Tuesday 10.15am) Prof. R. Bitmead – Department of Mechanical and Aerospace Engineering, University of California, San Diego, USA.

Modern jet engine controllers are necessarily multi-input/multi-output (MIMO) and this requires new approaches to guaranteeing that the controllers work in practice, as compared to SISO controllers which were evaluated by gain or phase margin. The new engines are designed to perform well at many; operating points, constrained conditions, and state of degradation or fleet variability. Our aim is to consider the application of system identification using experimental data to ensure the stability of perhaps hundreds or thousands of different controllers through the testing of a small number of candidate systems via experiment. We shall pose the following problem: Given an unknown MIMO system with a linearized model and a large set of controllers, each of which stabilizes the model, develop an experimental test procedure for a small number of controllers operating on the real system to guarantee that all of the controllers will stabilize the real system. Ideas covered will include; generalized stability margins, the Vinnicombe nu-gap metric, empirical transfer function estimation, and MIMO experiment design. These methods are being developed and implemented as part of a DARPA-funded program to facilitate certification of actual MIMO engine controller designs.




Implantation Industrielle d’Algorithmes de Commande pour Presses Plieuses Synchronisées

11.04.2008 – Dr. D. Garcia – Cybelec S.A., Yverdon-les-Bains, Switzerland.

Le domaine du pliage de la tôle fait appel, même parfois malgré les apparences, à des méthodes de fabrication relativement complexes. Après un rapide survol de la problématique ainsi que des techniques de pliage industrielles, l’attention se portera principalement sur le système hydraulique permettant la réalisation des plis. En particulier, la commande de l’élément principal d’une presse plieuse, le coulisseau, sera abordée plus en détail. Cet outil, pesant généralement plusieurs tonnes et piloté par deux vérins hydrauliques capables de générer des forces colossales, est descendu verticalement à l’intérieur d’une matrice, pour déformer la tôle métallique. En outre, la précision de son positionnement ainsi que la répétitivité de ses mouvements revêt une importance capitale pour garantir la qualité du résultat final. Un modèle de connaissance des éléments constitutifs du système sera présenté. L’utilisation de celui-ci, associé aux méthodes classiques de synthèse de commande, permet l’obtention d’une commande simple et robuste, conférant au système résultant les performances souhaitées.




Robust Global Stabilization of Continuous Bioreactors

18.04.2008 – Prof. C. Kravaris – Process Control Laboratory, Department of Chemical Engineering, University of Patras, Greece.

This work concerns problems of robust global stabilization of continuous stirred microbial bioreactors. A specific application is in anaerobic digestion processes, which are characterized by narrow stability regions and extreme sensitivity to external perturbations, if operated under optimal steady-state conditions. A control Lyapunov function approach is used to derive a globally stabilizing state feedback control law, which turns out to be a proportional controller with respect to the microbial growth rate. The robustness properties of the feedback controller are investigated with respect to both constant and time-varying perturbations. In the latter case, precise robustness margins are derived via a vector Lyapunov function approach. The theoretical derivations and results are extended to the case where a feedforward measurement of the organic load is incorporated in the control law; it is shown that this leads to improved robustness margins.




Design and Control of Autonomous Systems

25.04.2008 – Prof. R. D’Andrea – Chemical Department of Mechanical and Process Engineering, ETH Zürich, Switzerland.

The commoditization of computation, sensing technology, and communication has enabled the conceptualization of new physical systems with large levels of autonomy. In many instances, we are no longer limited by physical hardware, but rather our ability to design and deploy reliable systems that take advantage of these new capabilities. In this talk I will discuss some of our contributions in the area of control and system design, and our attempts to tame and manage the complexity inherent in high performance autonomous systems.




Probabilistic Methods in Machine Learning

09.05.2008 – Dr. D. Barber – Department of Computer Science, University College London, UK.

Probability and, in particular, Bayesian methods, allow a principled treatment of uncertainty in Machine Learning. In this talk I will cover some of the rationale behind using probabilistic techniques and discuss how Graphical Models, a marriage of Graph and Probability Theory are a useful framework for developing models and approximations in Machine Learning and related areas. A particular benefit of this framework has been the unification of techniques from diverse areas such as Statistical Physics, Information Theory, Statistics and Computer Science. I will cover some of the basics of exact inference using message passing and related approximation methods. Specific models I will discuss will include Bayesian Networks, Gaussian Processes and Linear Dynamical Systems, all of which can be treated within the same framework. The aim of the talk is to impart some of the key ideas, motivations and methods behind recent advances in this fast moving field.




Generation of Accurate Translational Motion for Testing Inertial Sensors

16.05.2008 – Prof. S. Spiewak – Department of Mechanical and Manufacturing Engineering, University of Calgary, Canada.

This presentation is concerned with the difficulties posed by evaluating high performance inertial sensors. They feature the frequency bandwidth from DC to several kilohertz, nonlinear distortions down to 0.0001%, and “on-board” 24 bit analog-to-digital converters, which define their best case resolution as 0.06 parts per million (ppm). Sensor weight ranges from a few to over 500 grams. Such specifications define a challenging performance envelope for generating accurate motion. By comparing various means, we choose precision air bearing stages driven by brushless DC motors. To optimize the stages we perform their comprehensive evaluation to detect, characterize, and prioritize various motion errors. The gathered knowledge leads to (1) eliminating some sources of disturbances and (2) implementing effective control algorithms to suppress an impact of the remaining disturbances.




Historical, Generic and Current Challenges of Adaptive Control

20.06.2008 – Prof. B. D. O. Anderson – Research School of Information Sciences and Engineering, Australian National University, Canberra, Australia.

This talk reviews three different types of challenges to adaptive control. The first group comprises challenges met in the subject’s development. They include difficulties unforeseen by overconfident theorists, associated with a number of situations in which there could be massive misbehavior, including instability, in adaptive systems. An understanding of these phenomena and mitigating strategies are now available. The second group comprises difficulties that are intrinsic to virtually any adaptive control algorithm, and that have frequently been overlooked. For example, if a plant is unknown, and a control objective is set, the objective may in practical terms be unachievable, and any adaptive control algorithm needs to deal with that possibility. The third group comprises some issues to which researchers are currently devoting significant attention, including multiple model adaptive control and model free design.