Adaptive inferential control of packed-bed reactors

Abstract

Three different estimators, i.e. Kalman filter, linear and nonlinear estimators, are presented to estimate the primary, unmeasured output. Control is accomplished using secondary measurements of temperature along the reactor length. Linear model estimators for product composition are found to perform poorly. Estimators based on nonlinear models of the plant are needed to accurately predict the composition. This is accomplished by a least-square curve fitting to match the observed data with the model and hence to estimate the input disturbances. Tests using some deliberately introduced plant/model mismatches indicate that the method is robust. In the second part of this paper, two distinct types of adaptive inferential control strategies for the single-input/single-output (SISO) system of the packed-bed reactor are investigated, i.e. the nonlinear inferential control, NLIC [the state feedback control (SFC) scheme] and the linear inferential control, LIC [the internal model control (IMC) scheme], to deal with the nonlinear and time-varying process control problems. The SFC scheme is developed by a nonlinear controller which is derived from the steady-state reactor model, and the IMC scheme by a linear controller which consists of an inverse of the transfer function of the process model and a filter with adjustable parameters. The basic structure of an adaptive inferential control system is coupled with a state estimator and an on-line parameter estimator. Here, we use a reduced-order, nonlinear process model to design the estimators and controller. Simulation results have shown that the performances of the SFC scheme are much better than those of the IMC scheme.