Liquid-liquid extraction column flooding detection

A liquid-liquid extraction column serves the purpose of separation of solvent mixtures with the help of a third solvent immiscible with the carrier solvent. In the investigated case, a value component dissolved in a light phase is contacted with the heavy phase, from which it can be separated more easily in a following process step. Two operating states within an extraction column can be identified. The regular operating state, characterized by a high separation efficiency through a large mass transfer of the solute from light to heavy phase. On the other hand, the separation efficiency decreases significantly as flooding, the second but undesired operating state occurs, and the volumetric throughput breaks down.
As the laboratory extraction column is optically accessible, a different approach based on computer vision is implemented for flooding detection. Here, image or video data is fed into a deep learning algorithm,i.e. a convolutional neural network (CNN). The network’s objective is to distinguish between the desired normal operating mode and flooding of the column.