Super-Twisting Observer-Based Output Feedback Control of a Class of Continuous Exothermic Chemical Reactors Jaime A. Moreno * Jesus Alvarez ** EdmundoRocha-C´ozatl *** Jorge Diaz-Salgado * * Instituto de Ingenier´ ıa, Universidad Nacional Aut´ onoma de M´ exico (UNAM). M´ exico D.F., Mexico. JMorenoP{JDiazS}@ii.unam.mx ** Universidad Aut´ onoma Metropolitana-Iztapalapa (UAM-I), M´ exico DF., Mexico. jac@xanum.uam.mx *** Facultad de Ingenier´ ıa, Universidad Nacional Aut´ onoma de M´ exico (UNAM). M´ exico D.F., Mexico. edmundor@dctrl.fi-b.unam.mx **** Tecnol´ ogico de Estudios Superiores de Ecatepec, Av. Tecnol´ ogico S/N. C.P. 55210. Ecatepec de Morelos, Edo. de M´ exico, Mexico. Abstract: In this paper, the problem of designing an output feedback (OF) stabilizing control scheme for (possibly open-loop unstable) exothermic reactors with temperature measurements is addressed. The proposed OF controller consists of the combination of a nonlinear state- feedback (SF) passive controller (built by passivation by backstepping) with a finite-time robustly convergent Supertwisting Observer (STO). The approach is tested with a representative example through simulations finding that the proposed controller behavior outperforms the one of its counterpart implemented with an asymptotic (infinite-time convergent) Extended Kalman Filter (EKF) 1 . Copyright c 2009 IFAC Keywords: Reactor Control, Sliding Mode Observers, Output Feedback Controllers, Constructive Control 1. INTRODUCTION An important class of chemical materials are produced in continuous exothermic jacketed reactors with reaction rate which grows with reactant concentration and temper- ature. (Lapidus et al. , 1977). In exothermic reactors, the interplay between heat generation-removal and reaction kinetics manifests itself as strongly nonlinear behavior, with asymmetric input-output coupling, steady-state (SS) multiplicity, and parametric sensitivity (Aris , 1969). In in- dustry (Shinskey , 1988), volume and cascade temperature linear PI loops are employed, and the concentration is reg- ulated by adjusting the reactant dosage via supervisory or advisory control. Thus, the objective of a process-control design scheme is to attain a closed-loop operation with an adequate compromise between safety, operability, pro- ductivity, and quality in the light of investment operation costs. The robust Output Feedback control of continuous stirred- tank reactors with monotonic kinetics has been success- fully addressed recently (Alvarez et al. , 2007; Diaz- Salgado et al , 2006) using PI controllers. The under- lying idea consists in designing a robust state feedback controller that renders the closed loop system passive, is a consequence of the (robust) relative degree structure of the system. In order to construct a (robust) output feedback controller that recovers the performance of the 1 This work has been supported by DGAPA-UNAM, project IN117610, and Conacyt (51244), Mexico. SF, reduced order observers are designed, that are able to (asymptotically and approximately) reconstruct the state and uncertain functions, such as the kinetics, the heat transfer coefficient or the feed concentration. These esti- mated values are used in the SF instead of the true ones, a structure that can be implemented as a PI controller, leading to a classical structure that recovers quite well the behavior of the SF, and is considerably less model dependent due to the reconstruction of unknown and/or time-varying parameters. This state/uncertainties recon- struction is possible because of the strong observability properties of the uncertain model. One of the limiting factors to achieve a perfect recovery of the SF performance by the OF controller is the attainable characteristics of the observer: since only an approximate estimation of the states is obtained asymptotically, due to the uncertainties and the convergence properties of a smooth observer, the control action of the OF will be only an acceptable approximation of the SF after an es- timation transient time. In principle, if the states were known perfectly in a (shorter) finite time, in spite of the uncertainties/perturbations, a much better recovery of the SF properties should be obtained. This considera- tion constitutes the first motivation of the present work: the improvement of the reactor close-loop behavior with temperature-measurement driven control. Sliding modes are well-known for their robustness against bounded perturbations and finite time convergence to the 9th International Symposium on Dynamics and Control of Process Systems (DYCOPS 2010) Leuven, Belgium, July 5-7, 2010 978-3-902661-69-2/10/$20.00 © 2009 IFAC 727 10.3182/20100705-3-BE-2011.0173