Identification of a Pulsed Liquid-liquid Extraction Column

Abstract

In this paper, the identification of a pulsed liquid-liquid extraction column is dealt with. This chemical operation consists in creating an intimate contact of the mixture containing the component to be extracted with a solvent (extractor) and separate the obtained immiscible phases. The behaviour of the columns is governed by hydrodynamic and mass transfer phenomenons. The conductivity measured under the distributor and the pulse frequency have been selected as the control variables. For control purposes, the complex dynamics of the considered process have been modeled by a linear discrete-time model. The recursive least squares identification methods using U-D factorization have been implemented on a PDP-11 computer, which is used for on-line estimation of the model parameters. Several problems associated with on-line identification such as model parametrization, plant operation and supervision and signals conditioning are examined. Model validation is carried out using several criterions, such as Akaike's criteria for order selection, residuals analysis and loss functions evaluation. Some techniques that have been used to estimated the time delay are also described. The outcome of the experiments is a model which highlights the behaviour of this very sensitive plant.