(513c) Performance Monitoring and Optimization of Extraction Columns By a Hybrid Modeling Approach

1. Background

In most industrial applications, liquid-liquid extraction columns are equipped with only a basic control system and minimal instrumentation, limiting the possibility of effective performance monitoring. Thus, extraction columns are usually operated at a constant solvent flow with a significant safety factor. This safety factor accounts for fluctuations in the feed stream composition and quantity, ensuring a separation performance of the column within the specifications. However, this robust approach leads to increased energy consumption for solvent recovery, which is usually realized in energy-intensive unit operations. Future application of extraction in biotechnological and recycling-based processes might even further increase safety factors due to higher uncertainties regarding feedstock composition compared to current processes. In order to make extraction processes efficient, we propose a model-based and data-driven performance monitoring and optimization for extraction columns. Therefore, we introduce a concept consisting of four steps, including (i) monitoring and (ii) modeling of extraction columns, and an approach for (iii) an online model parametrization and (iv) optimization of the column’s separation performance.

2. Methods

The main objective of the monitoring system is the detection of critical operating states, such as flooding, at an early stage. As early indicators for flooding, key fluid dynamics properties, e.g., drop size distribution and disperse phase hold-up, are monitored by telecentric optics and a differential pressure sensor, respectively. The evaluation of the optical data is pursued online by a Mask Recurrent Convolutional Neural Network (Mask R-CNN) introduced recently [1]. The evaluation was extended to allow quantification of the drop sedimentation velocity between several consecutive images. Apart from monitoring, the operating data is currently used to initialize the column simulation and will serve as the basis for online model parametrization in further developments.

The column model is based on a reduced droplet population balance which tracks the evolution of the drop mean diameter and the hold-up along the column’s axial domain [2]. Both properties, particularly the drop mean diameter, are dependent on drop breakage and coalescence phenomena. For both phenomena, physical-empirical models [3, 4] were enhanced by data-driven parameter estimators [5] establishing an overall hybrid modeling approach. The hybrid column model was validated on a dataset consisting of 292 column experiments retrieved from the literature.

3. Results and Outlook

In this contribution, we will present and discuss the results of the monitoring and modeling of extractions columns. Experimental investigations indicate that the monitoring of fluid dynamics provides the necessary data to assess column operation. Specifically, detailed monitoring of the hold-up and the drop sedimentation velocity allows to detect first signs of flooding, providing a possibility to adjust the column operation in advance. Regarding the column model, the hybrid PBM-based column model combines the extrapolation advantages of a physically motivated model basis with the high precision of a data-driven approach. The hybrid column model predicts the drop mean diameter in the 292 column experiments with significantly higher accuracy than a comparable physical-empirical approach. Additionally, the validity range, e.g., convex hull [6], of the hybrid model is not limited to the data it was trained on in contrast to a data-driven approach. Based on the correct prediction of the drop mean diameter, we deduce that drop breakage and coalescence are correctly depicted by our hybrid approach ensuring a wide applicability of our column model.

Currently, we are developing a holistic parametrization approach for the column model to account for feed fluctuations in biotechnological and recycling-based processes. This parameterization will ensure the simulations’ validity and thus will allow an optimization of the column performance by minimizing the solvent stream tailored to a variable feed composition. The first results of the model parametrization and performance optimization will be presented.

4. References

[1] S. Sibirtsev, S. Zhai, M. Neufang, J. Seiler, A. Jupke, Proceedings of International Solvent Extraction Conference 2022.

[2] M. Attarakih, M. Abu-Khader, H.-J. Bart, Chemical Engineering Science 2013, 91 (4), 180 – 196. DOI: 10.1016/j.ces.2013.01.032.

[3] D. Garthe, Fluiddynamics and Mass Transfer of Single Particles and Swarms of Particles in Extraction Columns, Dissertation, TU München 2006.

[4] M. Henschke, Auslegung pulsierter Siebboden-Extraktionskolonnen, Habilitation, RWTH Aachen University 2003.

[5] J. Brockkötter, M. Cielanga, B. Weber, A. Jupke, Ind. Eng. Chem. Res. 2020, 59 (44), 19726 – 19735. DOI: 10.1021/acs.iecr.0c03282.

[6] O. Kahrs, W. Marquardt, Chemical Engineering and Processing: Process Intensification 2007, 46 (11), 1054 – 1066. DOI: 10.1016/j.cep.2007.02.031.