(493e) Machine Learning Aided Tools for Online Monitoring of Separation Processes

Separation systems are crucial in the process industry and find many applications in the separation of various chemical and biological streams. However, the inherent complexity of these processes, coupled with the lack of Process Analytical Technologies (PATs), poses a significant challenge in developing robust analytical technologies for real-time monitoring and measurement (Barringer, 2006). As such, mathematical models of the systems can be used to aid in process design, optimisation, and control.

Chromatographic separation systems are typically described by complex Partial Differential and Algebraic Equations (PDAEs), due to the physicochemical phenomena describing the systems (Kumar et al., 2021). This can often complicate further online applications, such as optimisation and control, due to the resulting high computational cost (Narayanan et al., 2021). To overcome such limitations, different machine learning aided tools can be implemented to ease the computational burden (Rathore et al., 2023).

In this work we are examining various strategies of data-driven and hybrid modelling for chromatographic processes, using Artificial Neural Networks (ANNs). We focus on the twin-column Multicolumn Countercurrent Solvent Gradient Purification (MCSGP) process, used for the purification of monoclonal antibodies (Aumann & Morbidelli 2007). The process model is developed with lumped kinetics and uses PDAEs to describe the solid and liquid phase concentration of each of the separation species (Müller-Späth et al., 2008). Using 50 collocation points, the resulting model comprises 3309 variables and 4119 equations, 805 of which are ordinary differential equations (Papathanasiou et al., 2016).

ANNs are developed as surrogates to describe the chromatographic process and trained to directly predict the liquid phase concentrations of the species at the column outlet at Cyclic Steady State (CSS). The developed machine learning aided models are assessed based on their accuracy and computational cost, as well as their online performance in further applications, such as optimisation and control. The results indicate that data-driven and hybrid models of separation processes can accurately predict the system performance and be used for process design, simulation, and online monitoring.

Acknowledgements

Funding from the UK Engineering & Physical Sciences Research Council (EPSRC) for the i-PREDICT: Integrated adaPtive pRocEss DesIgn and ConTrol (Grant reference: EP/W035006/1) is gratefully acknowledged.

References

Aumann, L., & Morbidelli, M. (2007). A continuous multicolumn countercurrent solvent gradient purification (MCSGP) process. Biotechnology and Bioengineering, 98(5), 1043–1055.

Barringer, G. (2006). Downstream Process Optimization Opportunities Using On-Line and At-Line PAT Instrumentation. BioPharm International, 2006 Supplement(3).

Kumar, V., Leweke, S., Heymann, W., von Lieres, E., Schlegel, F., Westerberg, K., & Lenhoff, A. M. (2021). Robust mechanistic modeling of protein ion-exchange chromatography. Journal of Chromatography A, 1660, 462669.

Narayanan, H., Luna, M., Sokolov, M., Arosio, P., Butté, A., & Morbidelli, M. (2021). Hybrid Models Based on Machine Learning and an Increasing Degree of Process Knowledge: Application to Capture Chromatographic Step. Industrial and Engineering Chemistry Research, 60(29), 10466–10478.

Papathanasiou, M. M., Avraamidou, S., Oberdieck, R., Mantalaris, A., Steinebach, F., Morbidelli, M., Mueller-Spaeth, T., & Pistikopoulos, E. N. (2016). Advanced control strategies for the multicolumn countercurrent solvent gradient purification process. AIChE Journal, 62(7), 2341–2357.

Rathore, A. S., Nikita, S., Thakur, G., & Mishra, S. (2023). Artificial intelligence and machine learning applications in biopharmaceutical manufacturing. Trends in Biotechnology, 41(4), 497–510.