Estimate of process outputs from multiple secondary measurements

Temperature and flow measurements are used to estimate the product compositions in a distillation column. The problem is characterized by strong colinearity (correlation) between the temperature measurements. Contrary to some claims in the literature, it is found using a Kalman-Bucy Filter that the goodness of the estimate, even when used for feedback control, is improved by adding temperature measurements. This does not apply to Brosilows inferential estimator which in its original form is very sensitive to colinearity in the measurements. It is important to use only those directions in the measurement space which are excited by the independent variable (inputs and disturbances). The Partial Least Square Regression (PLS) method used in statistics adresses this explicitly. In the paper we use the PLS method to gain insight into the directions of the temperature space.