Commande non-lineaire d'un systeme decrit par des equations paraboliques: Application au procede de blanchiment (French text)

Dynamical isothermal tubular reactor modeling by mass balance leads to a set, of partial differential equations. This model is not often used for the design of control laws. Is it possible to enhance process performance by using a control law based on a distributed parameter model? This problem is addressed here considering a bleaching reactor application.; The first part of this work presents the modeling of a tubular reactor and simulation algorithms. The analysis of classical numerical approaches identifies their limits in the case of the simulation of dominant convection systems in transiant modes. A sequencing method is proposed to solve numerical diffusion and oscillation problems. This algorithm solves at each time step and successively each phenomena in the reactor, i.e. convection, diffusion and reaction. This simple approach reduces the simulation time and is stable. It is well suited for on-line control application.; The second part of the thesis presents three different control algorithms applied to the bleaching reactor. Those strategies differ by the complexity of the model used for the design. Hence, the predictive control uses an input-output model, the global differences algorithm uses an early lumping of the partial differential equation system and finally, the late lumping approach uses directly the distributed parameter model. The performance analysis of these three approaches shows that using a distributed parameter model is more efficient by taking into account the nonlinearities of the kinetic model. The late lumping control law proposed in this thesis, shows good performance with regard to kinetic parameters variations while maintaining good robustness properties to model mismatch.