(20c) Dynamic Modeling of Solid Bowl Centrifuges As a Tool for Better Process Design: Advantages of Adaptive Hybrid Models

Solid bowl centrifuges are widely used in many areas of the process industry. Examples are the chemical and pharmaceutical industry, mineral processing, biotechnology and food industry, but also waste water treatment [1]. The choice of the type of centrifuge and the design of the respective apparatus is based on the knowledge of the manufacturers and a large number of pilot tests and simplified analytical equations such as the Sigma theory [2]. This leads to high costs in the design and necessitates considerable product input. For nanoparticles or expensive products such as pharmaceutical product, the necessary quantities are often not available.

The difficulties mentioned above are avoided by dynamic models, which allow the integration of material functions into the calculation of solid bowl centrifuges and thus show a closer approximation to the real process. [3]. Material functions are based on the separation properties that take place in the apparatus and can be determined on a laboratory scale with very small sample volumes. This allows the estimation of the use of solid bowl centrifuges at an early stage of process development and avoids expensive pilot tests [4].

At the same time, however, dynamic models also have limitations, which arise due to the product-dependent characterization of the material functions. If product changes occur during process development, the material functions must be determined again. At the same time, product fluctuations during particle synthesis, which usually takes place upstream, lead to changes in the particle properties. However, knowledge of the particle properties is essential for dynamic modeling of solid bowl centrifuges.

This talk presents the evolution of dynamic modeling for solid bowl centrifuges to a hybrid model that has a physics-based and data-based component. In addition to the interconnection of the hybrid models, which can have a parallel or serial structure, the choice of the data-based model is also crucial [5]. In this context, neural networks show promising results with respect to the accuracy of the training. In addition, this presentation shows the methodical framework to directly integrate the process and material properties recorded during operation by online learning into the hybrid model. This allows the direct integration of the framework into the industrial application in the form of a digital twin. The use and accuracy of the hybrid model is demonstrated and extensively discussed by comparison with experimental data for pilot-scale centrifuge.

Acknowledgement:
The authors would like to thank the German Research Foundation (DFG) for funding in the priority programe SPP1697 (Grant no: NI 414/21-3) and Federal Ministry for Economic Affairs and Climate Action (Grant no: 21638 N).

References:

[1] H. Anlauf, Recent developments in centrifuge technology, Sep. Purif. Technol. 58 (2007) 242–246.

[2] C.M. Ambler, The theory of scaling up laboratory data for the sedimentation type centrifuge, J. Microb. Biochem. Technol. 1 (1959) 185–205.

[3] M. Gleiss, S. Hammerich, M. Kespe, H. Nirschl, Application of the dynamic flow sheet simulation concept to the solid-liquid separation: Separation of stabilized slurries in continuous centrifuges, Chem. Eng. Sci. (2017).

[4] P. Menesklou, T. Sinn, H. Nirschl, M. Gleiss, Scale-up of decanter centrifuges for the particle separation and mechanical dewatering in the minerals processing industry by means of a numerical process model, Minerals. 11 (2021) 1–18.

[5] P. Menesklou, T. Sinn, H. Nirschl, M. Gleiss, Grey box modelling of decanter centrifuges by coupling a numerical process model with a neural network, Minerals. 11 (2021).