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 (2nd floor), 1015 Lausanne
When: Friday at 10.15am
27.02.2009 – Dr Filip Logist – Dept. Chemische Ingenieurstechnieken (Leuven, Belgium)
In practical optimal control problems multiple and conflicting objectives are often present, giving rise to a set of Pareto optimal solutions. Although combining the different objectives into a convex weighted sum and varying the weights is the most common approach to generate the Pareto front (when deterministic optimization routines are exploited), it suffers from several intrinsic drawbacks. A uniform variation of the weights does not necessarily lead to an even spread on the Pareto front, and points in non-convex parts of the Pareto front cannot be obtained (Das and Dennis, 1997). Therefore, this paper investigates alternative approaches based on novel methods as Normal Boundary Intersection (Das and Dennis, 1998) and Normalized Normal Constraint (Messac et al., 2003) to mitigate these drawbacks. The resulting multiple objective optimal control procedures are successfully used in (i) the design of a chemical reactor with conflicting conversion and energy costs, and (ii) the control of a bioreactor with a conflict between yield and productivity.
06.03.2009 – Prof. H. Geering – Measurement and Control Laboratory (ETHZ, Zürich)
Many nonlinear dynamic systems can easily be controlled by robust linear output feedback control while the system is operating at a “constant” operating point.
However, in some interesting applications, the operating point of the plant is not constant. The method of linear parameter-varying control takes this into account by continuously updating the linearized dynamics of the plant, modifiying the bandwith of the desired control loop, and adapting the control gains.
The method of linear parameter-varying control will be demonstrated with an application in the fuel control of an automotive gasoline engine and an application in controlling the pitch angle of an aircraft which flies at variable speeds and altitudes.
13.03.2009 – Prof. M. Steinbuch – Dept of Mechanical Engineering, Control Systems Technology Group, Eindhoven University of Technology, Netherlands.
In motion systems plant models are relatively easy to acquire with high accuracy, and feedforward is always used for servo tasks. The primary role of the feedback is to suppress disturbances, however, here the fundamental limitations with respect to controlled performance are felt, which are primarily due to causality. The Bode Sensitivity integral expresses the causality in the form of reduction of low-frequent disturbances, on the cost of the amplification of (typically) high frequent signals (Water bed effect), e.g. measurement noise. It is this fundamental limitation which is the driving force for a few lines of research: (i) further exploring feedforward, including iterative learning control, (ii) disturbance- and data- based control, i.e. using the internal model principle and also the principle of ‘machine-in-the-loop’ for adjusting the controller parameters on the basis of on-line measurement, (iii) non-linear control of linear motion systems, and finally (iv) multivariable robust control of mechanical systems.
20.03.2009 – Prof. A. Ijspeert – Biologically Inspired Robotics Group (EPFL, Lausanne).
Animal locomotion control is in a large part based on central pattern generators (CPGs), which are neural networks capable of producing complex rhythmic patterns while being activated and modulated by relatively simple control signals. These networks are located in the spinal cord for vertebrate animals. In this talk, I will present how we model CPGs of lower vertebrates (lamprey and salamander) using systems of coupled oscillators, and how we test the CPG models on board of amphibious robots, in particular a new salamander-like robot capable of swimming and walking. I will also present how abstract CPG models can be used to control the locomotion of various robots (quadruped, humanoid and reconfigurable modular robots), and serve as useful substrates for doing online learning (i.e. learning while moving).
03.04.2009 – Prof. B. Ogunnaike – Chemical Engineering Department, University of Delaware (USA).
The key attributes of the overall performance of any controller are robustness, setpoint tracking, and disturbance rejection. In the specific case of PID controllers, the tuning parameters are not related to these critical attributes directly; designing the controller to achieve desired performance in each of these attributes is therefore not straightforward. We have recently argued for the need to develop an alternative regulatory controller and proposed the RTDA controller. This new control scheme combines the simplicity of the PID controller with the versatility of Model Predictive Control (MPC). The tuning parameters are related directly to the performance attributes and are all naturally scaled between 0 and 1, leading to a more direct and transparent design. We have also developed tuning rules, based on robust stability analysis, in which stability regions can be characterized as functions of the controller parameters. We will present in this seminar a derivation of the controller technique and its tuning rules; we will then illustrate its performance with a simulated nonlinear polymerization reactor along with actual implementation on a laboratory-scale water tank system and on a pilot-scale physical vapor deposition process for manufacturing flexible materials for photovoltaic cells.
NEW DATE 01.05.2009 – Prof. K. Schlacher – Institute for Cybernetics and Process Automation, Johannes Kepler University (Linz, Austria)
Hydraulic devices are used where high forces must be generated by small devices, but the nonlinear behavior of these actuators significantly deteriorates the dynamic behavior in the large signal scenario like in steel industries. Steel rolling simulators are tools for material scientists for the study of the mechanical deformations, which occur in rolling mills, at a laboratory level. Special hydraulic pistons about 3m long are used to simulate the neighbor stands, hydraulic devices are used to control the rolling gap. High standards of reproducibility and precision are essential, but both are obtainable only by advanced control. Two methods for the nonlinear control of such devices will be presented, the first one uses energy considerations and a generalized Hamiltonian approach, whereas the second one is based on state feedback combined with a state, disturbance observer, also called internal model approach. The stability considerations for both methods will be accompanied by simulations and measurements of a cold rolling simulator, which belongs to one of the leading European steel companies.
08.05.2009 – Prof. T. Söderström – Dept. of Systems and Control, Uppsala University (Uppsala, Sweden).
The lecture gives a tutorial survey of errors-in-variables methods in system identification, where both input and output measurements are noise-corrupted. Background and motivation are given, and examples illustrate why such an identification problem can be difficult. Under general weak assumptions, the systems are not identifiable, but can be parameterized using one degree of freedom. Examples where identifiability is achieved under additional assumptions are also provided. A number of typical approaches for parameter estimation of errors-in-variables models are presented.The underlying assumptions and principles as well as key properties for each approach are highlighted.
15.05.2009 – Prof. M. Perrier – Département de Génie Chimique (Ecole Polytechnique de Montréal, Canada)
In this talk, our recent work on extremum seeking control and its application to chemical and biochemical systems will be presented. Two important classes of approaches are considered: (i) perturbation based methods, for which an excitation dither signal is added to the input and the gradient is forced to zero, and (ii) model-based methods, for which a dynamical model with unknown parameters is typically used. The main drawback of the perturbation approach lies in its potentially slow speed of convergence mainly due to the several time-scale separations performed. However, our results show that when the dither frequency is increased, which can accelerate convergence, the system might converge away from the optimum even when the amplitude of the excitation goes to zero. One the other hand, for systems described by Wiener/Hammerstein approximations, the error will indeed go to zero with the excitation amplitude. Hence it is possible to accelerate convergence, but it may lead to instability due to the phase shift introduced. A phase compensation scheme has been developed to ensure stability. For the model-based approach, optimum search and adaptive control are combined to guarantee convergence of the closed-loop system to the optimum operating point. The approach will be illustrated through a simple example of a bioreactor with two different scenarios. In the first case, the kinetics model structure is assumed to be available with unknown parameters. For the second case, the kinetic model is assumed to be unknown and a universal approximation is considered to update the model kinetics and drive the system to its optimum.
Bernoulli Center 2009 program on Advances in the Theory of Control Signals and Systems with Physical Modeling
Dr Philippe Mullhaupt and Prof Jean Levine are organizing of a Bernoulli program at the Centre Interfacullatire Bernoulli, which is attached to the Maths department. The topic deals wilth advances in signal, systems and control with an emphasis on physical modelling as a unifying trend. There will be three main workshops: The first one will address electrical and mechatronical systems; the second will deal with the mathematical tools and the third concerns chemical and life sciences.
Lectures will be given by renown mathematicians such as Claude Lobry or Alexander Shoshitaishvilii, who have strongly contributed to the theory of dynamical systems, for instance, or by well-known specialists of control theory such as Richard Murray, Georges Bastin, Arjan van der Schaft and Frank Allgöwer, also by chemists and biologists (Didier Gonze, Philip Crooke, Babatunde Ogunnaike, Jean-Philippe Vert, Jean-Luc Gouzé), biomechanicians and physicists (Dominique Pioletti, Felix Naef), and by a theoretical ecologist (Alan Hastings).
You are all freely welcome to both attend all, or part, of the workshops and also to participate in the program in general.
For more information concerning the program: http://bernoulli.epfl.ch/advances/web-content/
Feb 18-20, electrical and mechatronical systems, speakers list and lectures titles
Apr 15-17, mathematical tools, speakers list and lectures titles
June 3-5, chemical and life sciences, speakers list and lectures titles
For those interested DO NOT FORGET TO REGISTER (small red link appearing on each of the web-pages concerning the workshops) so that we can have a good estimate of the people attending the workshops.