 |
|
Publications, papers published in December 1999/communications publiées en décembre 1999
- 1
-
J.-P. Chrétien.
Optimal filtering of wind and temperature balloon sounding.
In CNES, editor, Launch Vehicles Vibrations Proceedings, pages
233-242, Toulouse (France), 14 - 16 December 1999. CNES.
The determination of extreme wind-shear conditions during the
atmospheric phase of the trajectory of a launcher is a critical point as it
is used to specify extremal loads on the launcher itself and on its payload.
Atmospheric models used to predict extremal shears rely on parameters which
must ne numerically known with a sufficient precision : the in situ
simultaneous measurement of the horizontal wind velocity (two components) and
the atmospheric temperature, using an instrumented balloon, is one of the
ways offered. In spite of the fact that the in situ measurement is non
stationary, spectral analysis has been shown to be an efficient tool to
separate different contributions to the measurement. In a normalized spectral
diagram (power spectral density as a function of spatial frequency), 3 zones
can be recognized:
- in low frequency (over a spatial
resolution of 100 m), a meso-scale stochastic atmospheric model,
characterized by a linear decrease of the spectrum, following a slope around
-3 (``spectral index''),
- in mean frequency, a dominant effect of the
balloon dynamics, excited by a source of noise independent of the meso-scale
model,
- in high frequency (up to the Nyquist spatial frequency). a
dominant effect of the location device noise, particularly in the case of
radar location of the balloon ; use of the Global Positioning System (GPS)
reduces significantly this source of error, which lies anyway in a frequency
range in which the meso-scale atmospheric model is no more valid
The main filtering problem to improve the wind velocity
profiles as a function of altitude is then to extract the correct atmospheric
model in the frequency range of interest (from 100 m to 1000 m resolution) in
spite of the overlap of the spurious balloon motions in this frequency range.
The temperature profiles require less filtering in this frequency range as
they are less depending upon the balloon motions. The paper is dedicated to
the application of optimal filtering to this problem : visualizing the
atmospheric model as a reference signal and balloon motion as a measurement
noise lead naturally to apply Wiener theory, in the simplified case where the
signal and noise are uncorrelated. The filter minimizing the estimation error
in the root mean square sense is simply obtained in frequency domain by
dividing the reference model spectrum by the measurement spectrum. This can
be done in two ways :
- pointwise in frequency, using the
estimated measurement spectrum,
- analytically, using representations of
both signal and noise as functions of the operational variable z (as
the data are sampled by the location and wind measurement .
In
order to be able to apply analytically the theory, we consider first the
possible analytical models for the noise (combining balloon dynamics and
measurement device). A second section is devoted to the synthesis of optimal
filters in the simplest canonical case of a second order atmospheric spectrum
(Wiener-Levy process for the signal) and a discrete white noise for the
balloon. The data being processed a posteriori, a non causal filter can be
used, which theoretical performances are calculated and verified by
statistics on a set of realizations. The robustness to mismatch in signal or
noise model is also discussed. It is thus shown that optimal filtering
extracts the meso-scale model in spite of spurious motions of the balloon in
the frequency range of interest, and that is is possible to deal with more
complicated balloon models (pendular motion for example). In the frame of the
canonical model, a comparison is then.
- 2
-
H. D. Tuan and P. Apkarian.
Low nonconvex rank bilinear matrix inequalities: Algorithms and
applications.
In CDC 1999 - 38th IEEE Conference on Decision and Control,
Phoenix (USA), 8 - 11 December 1999. IEEE.
A new Branch and Bound (BB) algorithm for solving a general class of
Bilinear Matrix Inequality (BMI) problem is proposed. First, Linear Matrix
Inequality (LMI) constraints are incorporated into BMI constraints in a
special way to take advantage of useful informations on nonconvex terms.
Then, the nonconvexity of BMI is centralized in coupling constraints so that
when the latter are omitted, we get a relaxed LMI problem for computing lower
bounds. As in our previous developments, the branching is performed in a
reduced dimensional space of complicating variables. This makes the approach
practical even with a large dimension of overall variables. Applications of
the algorithm to several test problems of robust control are discussed.
- 3
-
P. Apkarian, H. Kajiwara, and P. Gahinet.
LPV techniques for control of an inverted pendulum.
IEEE Control System Magazine, 1999.
The purpose of the paper is to demonstrate the ability of LPV (Linear
Parameter Varying) control techniques to handle difficult nonlinear control
problems. The focus in this paper is on the wide range stabilization of an
arm-driven inverted pendulum. Two different LPV control techniques are used
to design nonlinear controllers that achieve stabilization of the pendulum
over the maximum range of operating conditions while providing time- and
frequency-domain performances. The merits of each of these techniques are
investigated and the improvements over more classical LTI (Linear
Time-Invariant) control schemes such as Hinfinite or controllers are
discussed. A particular emphasis is put on the real-time implementation of
these controllers for the inverted pendulum experiment. It is shown in the
application that suitable multi-objective extensions of the standard
characterization of LPV controllers allow us to cope with sampling rate
implementation constraints. Finally, a complete validation of the proposed
LPV controller structures is carried out through a set of realistic nonlinear
simulations but also by means of physical experiment records.
- 4
-
C. Cossart and C. Tessier.
Filtre de kalman et processus de révision et de mise jour : une étude
comparée.
Revue d'Intelligence Artificielle, 1999.
Partant du constat qu'il existe des similitudes dans les objectifs de
l'estimation numérique et de la révision et de la mise à jour, nous proposons
d'effectuer un rapprochement et une comparaison des différents processus.
Nous étudions ainsi en détail les points communs et différences de
l'estimation numérique d'une part - à travers le filtre de Kalman mono et
multimodèle - et des mécanismes de révision et de mise à jour d'autre part -
depuis les postulats d'AGM à la mise à jour généralisée de Boutilier. Nous
soulignons l'évolution progressive des processus de changement de croyances
vers les principes du filtrage de Kalman et identifions les ingrédients d'un
estimateur symbolique dans le cas particulier du suivi de situation.
- 5
-
M. Chilali, P. Gahinet, and P. Apkarian.
Robust pole placement in LMI regions.
IEEE TAC, 44(12):2257-2270, December 1999.
This paper discusses analysis and synthesis techniques for robust
pole placement in LMI regions, a class of convex regions of the complex plane
that embraces most practically useful stability regions. The focus is on
linear systems with static uncertainty on the state matrix. For this class of
uncertain systems, the notion of quadratic stability and the related
robustness analysis tests are generalized to arbitrary LMI regions. The
resulting tests for robust pole clustering are all numerically tractable
since they involve solving linear matrix inequalities (LMIs), and cover both
unstructured and parameter uncertainty. These analysis results are then
applied to the synthesis of dynamic output-feedback controllers that robustly
assign the closed-loop poles in a prescribed LMI region. With some
conservatism, this problem is again tractable via LMI optimization. In
addition, robust pole placement can be combined with other control objectives
such as H_2 or Hinf performance to capture realistic sets of design
specifications. Physically-motivated examples demonstrate the effectiveness
of this robust pole clustering technique.
- 6
-
P. Coton, J. C. Monnier, R. Stuff, L. Dieterle, and G. Schneider.
Characterization of the wake far field from high lift configurations
of the airbus A300 by PIV measurements in the onera lille flight analysis
laboratory.
Aerospace Science and Technology, 1999.
Une nouvelle méthodologie a récemment été développée pour l'étude du
sillage des avions. Les essais se déroulent dans le Laboratoire d'Analyse du
Vol de l'ONERA/Lille. Une maquette d'Airbus A 300 au 22ème est catapultée
puis évolue dans un espace d'observation de 30 m de long, 9 m de large et 10
m de haut. Tra-jectoire et attitude de la maquette sont déterminées au moyen
d'une instrumenta-tion embarquée et de mesures faites au sol. L'écoulement
est visualisé par de la fumée et mesuré par PIV (Vélocimétrie par Images de
Particules) dans un réfé-rentiel fixe par rapport au sol. Ces mesures ont été
effectuées en partie dans le cadre du programme de coopération ONERA/DLR et
en partie dans le cadre du projet européen de recherche WAVENC (Wake Vortex
Evolution in Far-Wake Region and Wake Vortex Encounter). Deux configurations
hyper-sustentées, sans nacelles, ont été testées. Dans la première
configuration seuls les volets.
- 7
-
H. D. Tuan, P. Apkarian, and M. James.
Parameterized linear matrix inequalities for nonlinear discrete
H_infty control.
In CDC 1999 - 38th IEEE Conference on Decision and Control,
Phoenix (USA), 8 - 11 December 1999. IEEE.
A new approach based on parameterized linear matrix inequalities
(PLMIs) for solving nonlinear discrete  control is proposed.
The ``curse of dimensionality'' is one of the main obstacles preventing
nonlinear  control theory to be used in practice. However, in most
applications, nonlinearity of a nonlinear system is caused by a few variables
with dimension much lower than the plant dimension. Considering such
nonlinear variables as parameters, the system can be handled as a family of
linear systems depending on parameters in a reduced-dimension space. Then,
PLMI tools can be used to render the nonlinear control
problem computationally tractable. Moreover, this approach can work well for
output feedback problem giving a finitely dimensional control instead of a
customized infinitely dimensional one.
- 8
-
C. Cossart.
Un estimateur symbolique pour le suivi de situation.
Thèse de doctorat Sup'Aéro, (20), décembre 1999.
Le thème des travaux est l'élaboration et le suivi de situation dans
un monde dynamique pouvant évoluer librement, dans le cadre de missions de
surveillance, de reconnaissance ou de renseignement. Le problème central de
la reconnaissance de situations est la mise en correspondance des données
visuelles avec des modèles d'actions et d'enchaînements temporels d'actions,
définis a priori et censés décrire au mieux les situations susceptibles
d'être observées dans l'environnement considéré. Il est renforcé par un
problème de rigidité des modèles et de l'algorithme de mise en
correspondance. Dans ce contexte, l'objectif de la thèse est de proposer un
système d'élaboration et de suivi de situation, au niveau symbolique, capable
de prendre en compte, d'une part, les imperfections de données et, d'autre
part, les écarts existant entre les comportements réels et les comportements
prédéfinis dans les modèles. La construction d'un tel système, désigné sous
le nom d'estimateur symbolique, a été réalisée en deux étapes : spécification
et formalisation. Un premier ensemble de spécifications a été déduit de
l'étude du système conçu dans le cadre du projet Perception et d'approches
récentes de la reconnaissance d'actions. Des spécifications plus précises ont
été déduites de la confrontation du filtre de Kalman et des processus de
révision et de mise à jour d'ensembles de croyances, les deux types de
démarches présentant des points communs avec le problème du suivi de
situation. L'estimateur symbolique construit en réponse aux spécifications
est un processus en deux étapes, prédiction et recalage, qui autorise des
mises en correspondance données-modèles imparfaites. Il manipule des
relations de préférence qualitatives, qui représentent l'incertitude sur la
situation en cours dans l'environnement. Le principe de l'estimateur ainsi
défini est appliqué à la surveillance d'un parking. This thesis deals with
automatic scene recognition in dynamic environments, for surveillance and
intelligent systems, autonomous vehicles or decision aid systems. The
recognition process is based on the matching of visual data with action
prototypes and temporal plan prototypes. Such a process is characterized by
uncertainty issues linked both to the models and to the matching algorithm:
on the one hand, data may be uncertain and, on the other hand, the matching
quality depends on the accuracy and the relevance of the models. In such a
context, our work aims at designing a symbolic estimation tool, in which
imperfect data are taken into account and imperfect matchings are allowed.
The design of such an estimator is achieved through two steps: specification
and formalization. In order to deduce specifications, the situation
assessment system designed in the framework of Perception project is analysed
and recent approaches of action recognition are studied. More precise
specifications are deduced from a confrontation of Kalman filtering and
belief revision and update processes as both types of processes have common
characteristics with situation assessment problem. The symbolic estimation
tool for situation assessment is a two-step process, prediction and
correction, in which imperfect matchings are defined. It handles qualitative
preference relations that encode the uncertainty on the situation in progress
in the observed environment. The principle of the symbolic estimator is
illustrated on the surveillance of a parking-lot.
- 9
-
H. D. Tuan, P. Apkarian, and H. Tuy.
Advanced global optimization algorithms for parameterized LMIs.
In CDC 199 - 38th IEEE Conference on Decision and Control,
Phoenix - Arizona (USA), 7 - 10 December 1999. CDC.
A new approach based on parameterized linear matrix inequalities
(PLMIs) for solving nonlinear discrete Hinfinity control is proposed. The
``curse of dimensionality'' is one of the main obstacles preventing nonlinear
Hinfinity control theory to be used in practice. However, in most
applications, nonlinearity of a nonlinear system is caused by a few variables
with dimension much lower than the plant dimension. Considering such
nonlinear variables as parameters, the system can be handled as a family of
linear systems depending on parameters in a reduced-dimension space. Then,
PLMI tools can be used to render the nonlinear Hinfinity control problem
problem computationally tractable. Moreover, this approach can work well for
output feedback problem giving a finitely dimensional control instead of a
customized infinitely dimensional one.
- 10
-
Y. Meiller and P. Fabiani.
Perception-action dilemma at planning time : Getting the best out of
game theory and classical planning.
In 18th Workshop of the UK Planning and Scheduling Special
Interest Group, Salford (UK), 15 - 16 December 1999. University of Salford.
Dealing with uncertainty in motion planning combines a number of
problems of estimation, planning and execution control. Classical methods in
Planning can take advantage of the structure of the workspace, but handling
uncertainty within these frameworks generally raises a question of
representation compatibility. Game theory offers efficient tools for decision
making in the presence of uncertainty, but often leads to a combinatorial
exploration of the decision tree. The association of both frameworks seems to
be a promising approach in order to combine symbolic reasoning tools with
representations of numerical uncertainty coming from sensory information. We
propose to introduce Graphplan-like techniques in game-theoretic planning.
Colored Petri nets allow us to present both standard forward state space
search and Graphplan like approaches within the same formalism. Expected
benefits of this proposal are discussed together with the example of a
possible application to the perception-action dilemma.
- 11
-
L. Chaudron and N. Maille.
The continuous inference model based on the analysis of mathematical
reasoning.
AI Communications, The European Journal on Artificial
Intelligence, 1999.
In this paper, we aim at designing a flexible model for the reasoning
of cooperating inferential agents.The context is mathematical problem solving
and the model is based on an experimental analysis of oral tests. Our
contribution consists in: i) the analysis of real cases and the induction of
empirical laws, ii) the definition of a general model, called "continuous
inference", dedicated to problem representation, iii) the logical and
algebraic description of the model thanks to the cube lattice structure (and
its implementation).
- 12
-
L. Chaudron and N. Maille.
Le modèle des cubes : représentation algébrique des conjonctions de
propriétés.
Revue d'Intelligence Artificielle, 1999.
Dans cet article, nous présentons un modèle, dit des « Cubes »,
destiné à formaliser tout type de connaissance qui soit fondée sur des
conjonctions de propriétés. Ce modèle est utile pour toute approche dans
laquelle il est nécessaire de mathématiser fidèlement la connaissance ainsi
que les moyens intrinsèques d'en comparer les éléments : modélisation de la
connaissance de perception, représentation d'activités, fusion symbolique,
etc.
- 13
-
J.-L. Farges.
Neural networks and dynamic programming for robust landing.
International JournalEngineering Applications of Artificial
Intelligence, 1999.
This paper deals with the optimization of non linear state feedback
for systems submitted to bounded perturbations and assessed through the value
of the final state of their trajectory. Neural Networks are used to
approximate either the optimal cost to go or the optimal control and the
worst perturbation during the search of the solution of minimum of maximum by
Dynamic Programming. In function of the quantity approximated, two kind of
algorithms are proposed. The methods are applied to a landing problem in
presence of unknown but bounded wind. For this problem, we show that the
initialization of the Neural Networks approximating the functions can be
performed through the approximation of elementary logical functions. The
study and the associated numerical experiments indicate that it seems more
adequate to approximate only the value function. Nevertheless, in this
context, some problems remain open at the level of the convergence of the
on-line retropropagation and the adaptation of only some given network
weights may induce improvements.
|
