Flooding Prevention in Distillation and Extraction Columns with Aid of Machine Learning Approaches

Abstract

Flooding of separation columns is a severe limitation in the operation of distillation and liquid‐liquid extraction columns. To observe operation conditions, machine learning algorithms are implemented to recognize the flooding behavior of separation columns on laboratory scale. Besides this, the investigated columns already provided the modular automation interface Module Type Package (MTP), which is used for data access of necessary sensor data. Hence, artificial intelligence (AI) tools with deep learning offer high potential for the process industry and allow to capture operating states that are otherwise difficult to detect or model. However, the advanced methods are only hesitantly applied in practice due to complex combination of operational sensing, data analysis, and active control of the equipment. This article provides an overview on how AI‐based algorithms can be implemented in existing laboratory plants. Process sensor data as well as image data are used to model the flooding behavior of distillation and extraction columns for stable and robust operational conditions.

Design of Module Type Package Services for Modular Downstream Units and Process Analytic Technology

“…To merge such ML methods within the service concept of the MTP, further investigations with demonstrator plants have to take place. First approaches have already been described by Oeing et al [30] with the services and experimental setup presented here.…”

Implementation of a Control Strategy for Hydrodynamics of a Stirred Liquid–Liquid Extraction Column Based on Convolutional Neural Networks

“…Modular automation concept for the DN32 extraction column using a convolutional neural network (CNN)-based optical sensor, adapted with permission under a creative commons CC-BY license from ref .…”

“…Since the shortcut connection is learning only the residual, the whole module is called residual module. The shortcut connection’s skipping of certain layers speeds up the training process of the neural net. , To train a CNN means to improve and thus adjust the weights and the threshold of the model (depicted in Figure ) to improve its performance on the given task. This task here being the classification of states and droplet sizes.…”

“…CAM is a simple technique for generating visual explanations of the predictions of CNNs . CAM uses the global average pooling layer in a CNN to generate a map that highlights which parts of an image the network deems most important with respect to a particular class label. , …”

“…22 CAM uses the global average pooling layer in a CNN to generate a map that highlights which parts of an image the network deems most important with respect to a particular class label. 14,23 In Figure 6, it is observable that the red region of the image section is in the area where a continuous layer of the disperse fluid is formed, that is, at the left and right stirrer blade region. In this region, flooding occurs, thus it can be deduced that the classification decision of the neural net is in a reasonable image section.…”