Design of An Expert System for a Crude Oil Distillation Column with Bayesian Belief Network

An expert system was developed for atmospheric crude distillation column using the Bayesian Belief Network (BBN). The experimental data was gathered from atmospheric crude tower as the major processing unit in a refinery in Iran, at the normal condition (steady state). In this method nodes of a BBN model present the parameters and state variables of the process under study and edges of BBN present their relationships. In the present work, the process variables are the CDU properties which are feed flow rate, quality and temperature at the tower entrance (furnace product), tower pressure, steam flow rate (replacement for reboiler), and flow rate of products; blend naphtha, kerosene and gasoil. In this work, the two main capabilities of the developed model in the refinery processes, prediction and decision making, were illustrated. Comparing network results and the practical data (history matching) shows that more than 94% of predictions and 91% of suggested decisions are reliable and also using the network saves us twenty three minutes, which indicates high capability of this system in decision making and prediction.