Thor Mejdell

ESTIMATORS FOR ILL-CONDITIONED PLANTS: HIGH-PURITY DISTILLATION

Temperatures and flows are often used as secondary measurements to estimate the product compositions (outputs) in distillation columns. The problem is characterized by strong collinearity (correlation) between the temperature measurements, and often between the effects of the inputs on the outputs. In a linear study three different estimator methods, the Kalman-Bucy Filter, Brosilows inferential estimator, and Principal Component Regression (PCR) are tested for performance with mu-analysis. It is found that use of input flow measurement has a damaging effect on the estimator performance for this ill-conditioned plant (with high RGA-elements). This is the main reason why the Brosilow inferential estimator is found to perform poorly. It is found that the static PCR-estimator performs remarkable well compared with the dynamic Kalman filter. Contrary to some claims in the literature, it is found that the performance of the estimate generally is improved by adding temperature measurements.