Réunions GTs AA, Identif, MEA (3 réunions )

Mardi 15/11 de 14h45 à 17h15

14h45

Titre : Techniques d’estimation d’état ensembliste pour systèmes incertains

Auteurs : C. Stoica Maniu, S. Ben Chabane, T. Alamo, E.F. Camacho, D. Dumur, L2S.

Résumé : Cette présentation porte sur des techniques d’estimation d’état ensembliste des systèmes linéaires, soumis aux perturbations et bruits de mesures bornés. Le problème d’estimation est mis sous forme d’un problème d’optimisation du type Inégalités Matricielles Linéaires. La technique d’estimation ensembliste proposée est également utilisée pour la détection des défauts et la commande tolérante aux défauts.

15h20

Title: State-of-charge and state-of-health estimation for Li-ion battery based on an electrochemical model

Authors: Luis D. Couto, Julien Schorsch and Michel Kinnaert

Abstract: The reliability and the durability of lithium-ion batteries can be strongly affected by their operating conditions. Therefore, there is a need for an on-line monitoring system that constantly provides an estimate of their state-of-charge (SOC) and their state-of-health

(SOH). This is one of the roles of a battery-management system (BMS). Many kinds of battery models have been proposed in the literature. Among them, the electrochemical model is the most suitable to control the electrochemical states in a wide operating range. Although more accurate and physically meaningful, this model comes associated with a computational burden due to its complexity. Therefore, reduced-order models must be considered for the BMS design. In this context, the equivalent-hydraulic model (EHM) is a simple representation of the lithium diffusion process that takes place within the electrode solid particles. It is based on two relevant electrochemical states, namely SOC and surface concentration. Although the EHM is originally a linear model, it depends on the diffusion coefficient, which is unknown. To tackle this problem, an extra state was included to account for the uncertain parameter in state estimation. This EHM was used to design an extended Kalman filter for the SOC estimation from available measurements (current and voltage). The SOH which is related to the unknown diffusion coefficient that control the electrochemical processes inside the battery is next identified. The estimation procedure is done from the estimated surface concentration and the current input. The proposed method relies on a three steps procedure. A temporal transfer function (TF) is first highlighted from the partial differential equation describing the evolution of the lithium concentration inside the particles. Because the obtained TF is unstable, a solution based on the pre-filtering of the data is then developed. Next, a nonlinear parameter estimation problem for the TF identification is solved by resorting to the so-called simplified refined instrumental variable method. From this first estimation, the diffusion coefficients are then computed by employing the weighted least-squares method.

15h55

Title: Secure a zone with robots using a set-membership state observer.

Author : Luc Jaulin, ENSTA Bretagne

Abstract: Last January, a Russian submarine entered inside the Bay of Biscay, for the first time since the cold war. This event motivated us to propose an approach to secure a zone again submarine intruders using a group of underwater robots. We assume that each robot is able to detect any intruder inside a disk of a known radius. Moreover, the speed of the intruders is assumed to be bounded. Our goal is twofold: (1) to characterize a set for which we can guarantee that there is no intruder (this corresponds to the secure zone) and (2) to find a control strategy for the group of robots in order to extend the secure zone as much as possible. Our contribution is to show that the problem of securing a zone can be solved using a set-membership observer shared between all robots. The approach can also be used for other situations, as for instance to guarantee that there is no intruder (cars, pedestrians, etc), inside a prohibited area.

16h30

Title: Kernel-based system identification in the frequency domain

Authors: John Lataire and Tianshi Chen, Vrije Universiteit Brussel, Belgium

Abstract: In the recent years, the system identification community has developed an interest in kernel-based methods. One particularity is that these methods allow handling model structure and complexity selection in a convenient manner. This presentation consists of two parts. The first part demonstrates how kernel-based regression allows for a continuous tuning of the model complexity, which is an intrinsic difference with classical regression. The second part applies kernel-based methods to non-parametric LTI system identification. In this context, most existing methods have been implemented in the time domain. On the contrary, in this presentation the peculiarities and advantages of formulating kernel-based system identification in the frequency domain will be highlighted, applied to the non-parametric estimation of transfer functions. These include 

i) the use of complex data, 

ii) the possibility to work in a local frequency band, and

iii) the use of non-parametric noise models

Mercredi 16/11 de 9h15 à 11h45

9h15

Title: Optimal experiment design for local LPV identification 

Authors: D. Ghosh, X. Bombois, J. Huillery, G. Scorletti, G. Mercère

Abstract. In local LPV identification, a number of local LTI models of a Linear Parameter Varying (LPV) system are identified by performing identification experiments for fixed values of the scheduling variable. These local LTI models are then interpolated in order to yield a LPV model of the LPV system. In this talk, we are looking at ways to optimally choose the values of the scheduling variables at which the local identification experiments have to be performed  as well as the excitation signals that have to be applied to the system during these local identification experiments. The objective of this optimal experiment design problem is to obtain a sufficiently accurate LPV model with the least injected power.

9h50

Titre : 3rd generation CRONE controller for ABS function of an electric-mechanical brake system.

Authors : André BENINE-NETO and Xavier MOREAU, Laboratoire IMS, UMR CNRS 5218

Abstract : Anti-Lock Braking (ABS) system is an essential functionality for the safety of road vehicles. Recent advances in electronics and electro-mechanical actuators have broadened the design possibilities of ABS. This work addresses firstly an analysis of the influence of vehicle load and road-tyre adhesion on the vehicle braking dynamics, since these parametric variations affect significantly the tyre braking forces. Then, taking into account the parametric variations, the synthesis of a 3rd generation CRONE controller (French acronym for fractional order robust control) is presented. Finally, the performance of the proposed controller is compared to others through simulations of realistic scenarios.

10h25

Title : Solving inverse optimal control problems via bounded-error parameter identification

Author : Nacim Ramdani, Univ. Orléans

I will address inverse optimal control, i.e. how to recover the cost function that can explain given observations, for the class of constrained optimization problems. Using parametrized Lagrangians, then applying the residual techniques for inverse parametric optimisation to the Karush-Kuhn-Tucker (KKT) optimality conditions of the nonlinear program underlying a direct optimal control or optimal planning problem, one can show that solving the inverse problem boils down to a constraint satisfaction problem. I will show how to come with a robust solution to solve the inverse optimal control problem in the bounded-error framework, while explicitly considering the presence of uncertainty in the measured observations, in order to derive uncertainty intervals on the imputed composite objective cost function.

11h00

Titre : Dynamic sensitivity analysis of a suspension model

Auteurs : S. Hamza & F. Anstett-Collin, CRAN Université de Lorraine,

Q. Li & L. Denis-Vidal, LMAC Sorbonne Universites, UTC

A. Birouche & M. Basset, MIPS Universite de Haute-Alsace, France.

Résumé : A sensitivity analysis of a suspension model has been performed in order to highlight the most influential parameters on the sprung mass displacement. To analyse this dynamical model, a new global and bounded dynamic method is investigated. This method, based on the interval analysis, consists in determining lower and upper bounds including the dynamic sensitivity indices. It requires only the knowledge of the parameter variation ranges and not the joint probability density function of the parameters which is hard to estimate. The advantage of the proposed approach is that it takes into account the recursive behavior of the system dynamics.

Mercredi 16/11 de 14h15 à 17h15

14h15

Titre : Identification du conducteur pour la conception des systèmes d’aide à la conduite.

Auteur : Sergey ABRASHOV, PSA/Laboratoire IMS, UMR CNRS 5218

Résumé : Le confort et la sécurité de conduite sont les principaux critères de vente de l’industrie automobile actuelle. De nombreux projets de recherche sont mis en place afin de les améliorer et pour faire face aux mesures de législation et de contrôle mises en place pour réduire le nombre d’accidents routiers. D’après de récentes études, 90% de ces  accidents ont pour cause le facteur humain et il devient nécessaire de prendre en compte le conducteur pendant la phase de conception des systèmes de sécurité et d’aide à la conduite.  L’identification du modèle conducteur correspond dans la majorité des cas à un problème d’identification en boucle fermée, puisse que le conducteur forme une boucle naturelle avec le véhicule. Les difficultés liées à l’identification en boucle fermée, ainsi que des résultats d’estimation paramétrique  sont présentées.

14h50

Title : Towards a viability kernel computation in higher dimensions with interval analysis

Authors : Olivier Mullier, Alexandre Chapoutot, Julien Alexandre dit Sandretto, ENSTA ParisTech

Abstract : In that work, controlled continuous dynamical systems S are considered and they are represented by non-linear systems of ordinary differential equations. From a given closed set K, a viable state for S is a state x ∈ K such that their exists a control u allowing the trajectory from x that is solution of S to remain in K. In this presentation, we focus on the computation of the set of all viable states, namely the viability kernel. The algorithm used to compute this set is an improvement of the one described in [1] which is generalized to state-space of dimension greater than 2. It is a two step method. Step one is computing an a priori inner approximation of the viability kernel, and step two is focusing on the expansion of the inner approximation. More precisely, step one uses Lyapunov theory to look for regions of the state-space where a linearized version of system S is stable, resulting on a set of ellipsoids proved to be included in the viability kernel. Step two makes use of latest improvements on guaranteed integration [2] (guaranteed Runge-Kutta integration methods using interval analysis and affine arithmetic) to enhance the quality of the inner approximation provided by step one. It results on a sub-paving of the viability kernel using boxes.

15h25

Titre: Parameter-varying modeling for the efficient design cycle of complex systems

Authors: Francesco  Ferranti , Telecom Bretagne

Abstract:  Design tasks, such as design space exploration, design optimization and uncertainty quantification, become computationally expensive when they are performed using only physics-based solvers (e.g., electromagnetic, mechanical and thermal solvers). Linear parameter-varying (LPV) models are able to efficiently and accurately describe the behavior of systems as a function of frequency/time and additional design parameters (e.g., layout and material features). These models are very useful to speed up the design cycle of complex systems in multiple application domains. In addition to accuracy and efficiency, the preservation of some properties, such as stability and passivity, in these models is also very important in order to perform meaningful simulations. The generation of LPV models can be automated by using adaptive sampling schemes.

Different application examples will be discussed in order to show the applicability of LPV models to complex design tasks.

16h00

Titre: Observateurs Intervalles pour des Systèmes Linéaires à Commutation

Auteurs: Djahid Rabehi ; Denis Efimov ; Jean-Pierre Richard ;

Résumé: Dans ce travail, le problème de l'observation de l'état est considéré pour des systèmes linéaires à commutation, une sous-classe des systèmes dynamique hybrides. On montrera qu’il est possible de concevoir un observateur par intervalles sous des conditions modérées, au moins à des instants en temps discret à partir de mesures en temps continu. Des conditions de coopérativité pour les systèmes à commutation  sont proposées. L'efficacité de l’observateur par intervalles proposé est illustrée par quelques exemples en simulation.