An application of digital estimation and adaptive control techniques to a real chemical process.

A continuous stirred tank reactor operating under direct digital control has been constructed, commissioned and used for the practical evaluation of estimation and control techniques of interest to chemical engineers. Several increasingly complex algorithms, based on the Kalman filter, were applied to estimate process states and/or unmeasurable parameters under a variety of simulated and practical conditions. A number of multivariable suboptimal control algorithms were also investigated and, by combination with the Kalman-type estimators, on-line adaptive control was realised. The simulation studies showed that for single input- single output systems, a most useful technique for process identification was the summed sine wave method; however, for multivariable systems, it was necessary to use estimators based on state-space models. Increasing estimator complexity progressively enhanced estimates and, in the case of the most sophisticated, eliminated biases: unfortunately, the unavoidable increase in computational burden sometimes precluded on-line use. A reduced, order filter, which operated on exact measurements, gave satisfactory (but slightly biased) parameter estimates. Simulation studies confirmed that all the suboptimal control schemes considered were feasible and, when coupled with an estimator, capable of controlling the process adaptively in stochastic environment. Practical results verified the simulation work; for example, close agreement was obtained between experimental estimates of a highly nonlinear parameter and the value quoted in the literature. Real process sensors were designed to be essentially noise-free and under these conditions proportional plus integral control gave good results, illustrating the power of feedback. Multivariable control, operating under more adverse conditions, gave less satisfactory results because of modelling errors, which produced state biases. However, the more sophisticated adaptive algorithms were effective in correcting parameter errors, (and therefore state biases), and gave enhanced control