Output estimation using multiple secondary measurements: High‐purity distillation

Abstract

AbstractMeasurements of temperatures (secondary outputs) and flows (inputs) are used to estimate product compositions (outputs) in a distillation column. The problem is characterized by strong collinearity (correlation) between temperature measurements and an ill‐conditioned model from inputs to outputs. In a linear study, three estimator methods, the Kalman‐Bucy filter, Brosilow's inferential estimator, and principal component regression (PCR), are tested for performance with μ‐analysis. One can achieve remarkably good control performance with the static PCR estimator, which is almost as good as the dynamic Kalman filter. The quality of the estimate for these two estimators is improved by additional temperature measurements, although the improvement is only minor for more than about five measurements. On the other hand, the performance of the Brosilow inferential estimator may not improve by adding measurements due to sensitivity to modeling errors. For all estimators, the use of flow (input) measurements does not improve the estimator performance and does in fact damage the performance if a static estimator is used.