Data-Driven Indication of Flooding in an Industrial Debutanizer Column
The profitability and sustainability of process industries are affected by the performance of each unit involved. A key measure of a unit’s performance is based on whether it operates in a desired production window or whether it trips into an abnormal condition. In this contribution, we study flooding of industrial distillation columns. We aim to improve the performance of an industrial debutanizer column by designing a data-driven flooding indicator. The design of the indicator consists of three steps; (a) the data treatment, (b) a priori labeling, and (c) indicator design. The prior knowledge about flooding within the column is used to design a reference indicator. This knowledge is either unused or fully exploited during the design of the indicator. We compare various design methods and show the potential of data-driven approaches for flooding indication.