engleski

Batch Rectification Process Modeling For A Systems With Unknown Vapor-Liquid Equilibria Using Hybrid Neural Network

In pharmaceutical and fine chemicals industry, process of batch rectification is often used for solvent recovery, waste water purification and similar applications. Very often exact vapor - liquid equilibria data are not available, either because of the organic or inorganic impurities that exists in the mixture, or mixture itself consists of components for witch VLE data can not be found in literature. In this work, a mathematical model is developed for a simulation of the batch rectification process using hybrid neural network. Method is based on the experimental data obtained trough the experiments carried out on the rectification column with known number of theoretical stages. Each experiment was carried out on different predetermined reflux ratio. During the experiment, in regular time intervals, bottom temperature was recorded and samples of the intermediate distillate composition are analyzed. In that way, sets of discrete data values that connects current time, reflux ratio, bottom temperature and distillate composition are formed. In developed hybrid neural model process dynamics is described by differential equations for a material balance and neural net is used for unknown vapor - liquid equilibria data. Multilayered feedforward net is used and trained using backpropagation method applying the conjugate gradient algorithm as a iearning method. Both simulation and experimental test has shown a good agreement between data obtained with neural network model and data obtainedţtrough simulation or real experiment. Method is proved to be robust and easily applicable.

batch rectification; neural network; modeling; vapor-liquid equilibria

nije evidentirano