Modeling and nonlinear predictive control of a fixed-bed water-gas shift reactor

Successful control of a severely nonlinear system depends upon knowledge of the process dynamics. Therefore, nonlinear dynamic models, formulated empirically or based upon first-principles, are necessary for the development and implementation of high-performance control strategies.;In this study, an advanced nonlinear control strategy was developed and experimentally applied to a laboratory fixed-bed water-gas shift (WGS) reactor. This was achieved by designing and constructing the reactor with computer facilities capable of acquiring data and implementing control. A network of three dedicated computers employed in an open-architecture, distributed control system (DCS) with multi-drop data acquisition hardware was used.;A control-relevant, first-principles model for the WGS reactor, consisting of three partial differential equations, was developed. The model was discretized spatially using weighted residual techniques on finite elements, and a predictor-corrector method was used to integrate the discretized system in time. Model parameters were determined using information-rich, dynamic experimental data. Model predictions compared well with experimental results.;Two on-line parameter estimation schemes capable of both efficiently and robustly estimating nonlinear model parameters were developed. Both estimators were designed to exploit the WGS model structure and the large number of available measurements. The first required the solution of a relatively large nonlinear programming problem. The second resulted in a normalized MIT-type estimation rule. For the WGS system, the two methods produced similar results.;Good control of the WGS reactor was achieved experimentally over a broad operating region when Nonlinear Model-Predictive Control (NMPC) was implemented. NMPC, a feedback strategy for constrained nonlinear processes, requires optimization of a performance objective over a time horizon at each sampling interval. It was implemented using a sequential optimization and solution technique. Closed-loop behavior was superior to that obtained using PID or adaptive linear control. Simulation experiments were used to demonstrate that controller performance was enhanced by nonlinear parameter adaptation and strategies for feed-forward implementation. NMPC feedback properties for systems experiencing gain sign changes were also investigated.