Multivariable stochastic control of a pilot plant packed bed tubular reactor

Multivariable regulatory computer control of the concentrations of three independent species at the exit of a pilot plant packed bed tubular reactor for the hydrogenolysis of butane is successfully carried out using only temperature measurements along the central axis of the reactor. A low order linearized state space model for the reactor system is developed from basic material and energy balances, and the unknown parameters of the system are estimated using dynamic data collected directly from the reactor. A Kalman Filter is used to estimate the states (central axis temperatures) of the system, and the multivariable feedback control algorithm is designed to optimize a quadratic function of the predicted exit concentrations. Hydrogen and butane feedrates are used as manipulated variables. The control algorithm is implemented on the pilot plant reactor in a series of DDC control studies. It is shown to perform very well and considerably better than a single loop PI controller with feedback on the hotspot temperature. Performance is evaluated under both the normal stochastic disturbances in the system and under a severe deterministic load disturbance. Exit concentrations predicted by the model and used in the performance index are shown to compare very favourably with actual concentrations obtained from a process gas chromatograph.