Learning control of a pulsed liquid-liquid extraction column

Abstract

Liquid-liquid extraction has proved to be a very competitive technique compared to conventional separation methods such as distillation. It allows a higher degree of extraction, separation and concentration of the desired products to be obtained. Generally, a liquid-liquid extraction column is run manually. This, however, is not satisfactory, especially concerning new valuable applications in the field of fine chemistry where processes must reveal themselves flexible and polyvalent. The objective of this study was to improve the quality of control even with poor a priori information on the process and thereby to maintain the column in its optimal behaviour zone. This optimal behaviour has been shown to be reached for flooding conditions which can be controlled by measuring the conductivity of the liquid medium at a place located at the bottom of the column (below the distributor), the pulsing action being the command variable. The column considered is modelled as a random medium and the search tactics for the control problem are formulated as those of automation behaviour in a random medium. The automation is considered to be stochastic with a varying structure. This control approach requires minimum possible a priori information on the process. Usually the developed model of chemical plants is static or very complex and cannot be used to derive a control strategy. The experimental results illustrate the efficiency of learning control.