A parameter estimation method for chromatographic separation process based on physics-informed neural network.

Area of Science:

• Chemical Engineering
• Computational Science

Background:

• Chromatographic separation processes are typically modeled using partial differential equations (PDEs) to capture complex adsorption equilibria and kinetics.
• Parameter identification in these PDE models is computationally intensive and time-consuming.

Purpose of the Study:

• To develop and validate a novel Physics-informed Neural Network (PINN) model for parameter estimation in chromatographic separation.
• To assess the accuracy, efficiency, and robustness of the PINN approach compared to conventional methods.

Main Methods:

• A Physics-informed Neural Network (PINN) model was developed for a binary component chromatographic system.
• The numerical accuracy of the PINN model was verified against the finite element method (FEM).
• Model parameters were estimated using the PINN from column outlet data, including noisy experimental data.

Main Results:

• The PINN model demonstrated high numerical accuracy, comparable to FEM simulations.
• Parameter fitting error was reduced by up to 35.0% compared to conventional methods.
• Computational time was reduced by up to 95% using the PINN approach.

Conclusions:

• The developed PINN model offers a computationally efficient and accurate alternative for parameter estimation in chromatographic separations.
• The PINN model shows robustness in handling noisy experimental data.
• This approach has the potential to accelerate process modeling and optimization in chromatography.