A nonlinear model predictive control algorithm is implemented on-line to co
ntrol the liquid level and temperature in a pilot plant CSTR, where an irre
versible exothermic chemical reaction is simulated experimentally by steam
injection. The dynamic behavior of the pilot plant reactor is represented u
sing a mechanistic, first principle model and a comparison between off-line
simulation and experimental data is presented. Several sources of model mi
smatch and unmeasured disturbances are identified that affect the quality o
f the model in representing the reactor dynamics. Despite these mismatches
and disturbances, the closed loop system is able to track setpoint changes
and reject disturbances quite well. In particular, the NMPC controller is d
emonstrated for different tuning parameters and under conditions of constra
int saturation at unstable points. (C) 2001 Elsevier Science Ltd. All right
s reserved.