(493f) Efficient Simulation of Simulated Moving Bed Chromatography By Physics-Informed Neural Network

Simulated Moving Bed (SMB) chromatography is a continuous separation process, offering an efficient method for the purification of high-value compounds across wide variety of industries, including pharmaceuticals, biotechnology, and food. [1-3] Optimizing the SMB process is crucial to increase productivity, reduce operating costs, and maximize product purity and recovery. However, simulation and optimization of the SMB process presents challenges due to its inherent complexity and dynamics. Mathematical models are usually used to replicate the dynamic behavior of the separation process for the SMB process. These mathematical models include partial differential equations (PDEs) that describe mass transfer, adsorption equilibrium, and column switching in the SMB system. Techniques such as the finite difference, and finite element methods are commonly used to solve the PDEs. However, these traditional numerical methods often face computational challenges due to the complex cyclic dynamics of the system.

Physics-Informed Neural Networks (PINNs) is recently introduced as an efficient approach for solving PDEs.[4] In PINNs, the loss function incorporates PDEs that describe the physical laws governing the system. These PDEs ensure that predictions of the neural network comply with the physical principles. By minimizing the loss function, PINN learns to solve the PDEs. Compared to the traditional numerical methods, this approach not only increases calculation robustness but also significantly improves computational efficiency.

This work aims to develop an application of Physics-Informed Neural Networks to the simulation of the SMB process, marking a significant advance over traditional numerical methods. A comparative analysis finds the superiority of PINNs, particularly in terms of reducing computation time compared to the finite element method. This study demonstrates the effectiveness of PINNs in simulating complex SMB chromatographic systems which allows efficient optimization of operating conditions and design.

[1] Bentley, J., Sloan, C., & Kawajiri, Y. (2013). Simultaneous modeling and optimization of nonlinear simulated moving bed chromatography by the prediction–correction method. Journal of Chromatography A, 1280, 51-63.

[2] He, Q. L., Leweke, S., & von Lieres, E. (2018). Efficient numerical simulation of simulated moving bed chromatography with a single-column solver. Computers & Chemical Engineering, 111, 183-198.

[3] Subraveti, S. G., Li, Z., Prasad, V., & Rajendran, A. (2022). Can a computer “learn” nonlinear chromatography?: Physics-based deep neural networks for simulation and optimization of chromatographic processes. Journal of Chromatography A, 1672, 463037.

[4] Lu, L., Meng, X., Mao, Z., & Karniadakis, G. E. (2021). DeepXDE: A deep learning library for solving differential equations. SIAM review, 63(1), 208-228.