Condition monitoring of a heterogeneous azeotropic distillation column using a self-organizing map

Abstract

A method is proposed to monitor the conditions of a chemical plant by considering the relationships among many types of process variables. The proposed method was evaluated in case studies involving a simulator on a heterogeneous azeotropic distillation column. First, a set of good-condition data and ill-condition data of process variables was stored. In this process, a classifier algorithm was applied to select a set of principal process variables representing the process condition. With the selected variables, the input vector for a self-organizing map algorithm was defined. Second, a set of typical patterns of good-condition data was generated by using selforganizing map with the stored good-condition data. Third, the condition of the column was monitored by calculating the dissimilarity value between the test data and the typical patterns of good-condition data. Consequently, it was clarified that the proposed method provides effective information on plant conditions in the early stages.