Identification non lineaire du modele de frottement GMS pour l'amelioration de la commande des systemes mecaniques

Friction causes important undesirable phenomena in a large class of control systems. This is particularly the case in systems that require a high degree of accuracy. Friction can deteriorate control performance by introducing tracking errors, limit cycles and possibly shattering.;While dynamic models represent many aspects of friction phenomena, they do not render both stiction in pre-sliding regime. Moreover, they represent a steady drift in position. For this reason, more recent models have been introduced to better illustrate friction behaviour.;More recently, the Generalized Maxwell Slip friction model known as the GMS model was introduced. It was compared to the other recently developed models and showed that it illustrates the majority of friction behaviours as well as their static and dynamic aspects. Moreover, a reduced friction model which is appropriate for simulation and control purposes was introduced.;Modelling friction in control systems is very important. I however, parameter identification is necessary in order to compensate for that friction. A great deal of work has been reported in the literature.;The goal of friction modelling is to quantify the various facets of friction behaviour. Many models, from simple to complex, designed to compensate for the unwanted friction effect have been introduced in the literature, In fact, the first friction models that were developed were static. The disadvantage of such models is that they do not represent all friction behaviours. It is thus that dynamic models were introduced.;Until now, the GMS model has not yet been adequately identified. Identification was based only on the pre-sliding regime or the pre-sliding and sliding regimes, but these identifications were based on off-line approaches like linear regression, dynamic linear regression, nonlinear regression and Monte Carlo approach.;In this thesis, our objective is to estimate the GMS friction model in both the pre-sliding and sliding regimes in order to improve the control strategy of mechanical systems by using a simple controller. The research results that led to reach our objective are: (1) The identification of the Stribeck friction using a new approach based on non linear identification using min-max algorithm. The identification of the non linear function of Stribeck was successful and was validated by numerical simulations. (2) The identification of the GMS model of friction based on using the measured friction force. This approach is based on a linear approximation of the GMS model over the unknown parameters. A robust observer of Marino is then used to estimate the unknown parameters in spite of the perturbation introduced by the approximation error. In order to apply this observer, the discontinuities in signals are eliminated by a filtering approach. A particular filter implementation is proposed to take into account the signal commutations introduced by the switching between sliding and pre-sliding regimes. (3) The identification of the GMS friction model without using force measurement. In fact, by constructing the filtered friction force from measurement signals, we develop a new strategy for the estimation of unknown parameters using an appropriate formulation for the robust observer of Marino. (4) The validation of identification approach using a specific experimental test in laboratory. In fact, the experimental tests prove the efficiency of our identification approach in real environment.;Key words: Identification, GMS friction model, Stribeck friction model, Adaptatif observer.