Integrating Machine Learning Process and Thermodynamic Modeling for Separation Processes

To reduce the environmental impact of the chemical industry, energy-efficient processes are required. Separation processes contribute a large portion of the total energy demand of the production of chemicals. To replace conventional separation processes, adsorption-based separation with solid materials has been demonstrated to be a promising alternative. Many materials have been reported, but it is challenging to estimate the efficiency and separation performance of each material due to the number of possible materials and the computational cost of process modeling. First principle adsorption process models need adsorption properties of materials in many conditions. To accelerate the performance evaluation, ML models have been trained to reproduce adsorption processes calculations at a lower computational cost. However, it remains unclear if the ML models can transfer to different separation from the ones they have been trained on. In addition, currently reported machine learning models require simplified thermodynamic models for the adsorption properties as inputs. At the Energy and Process Systems Engineering (EPSE) group, we evaluate these adsorption properties of materials with classical Density Functional Theory (cDFT), which allows predictions at molecular simulation accuracy. This project aims to combine the ML-based process models with cDFT adsorption calculations and explore the transferability of ML models to new separation processes.

To reduce the environmental impact of the chemical industry, energy-efficient processes are required. Separation processes contribute a large portion of the total energy demand of the production of chemicals. To replace conventional separation processes, adsorption-based separation with solid materials has been demonstrated to be a promising alternative. Many materials have been reported, but it is challenging to estimate the efficiency and separation performance of each material due to the number of possible materials and the computational cost of process modeling.

First principle adsorption process models need adsorption properties of materials in many conditions. To accelerate the performance evaluation, ML models have been trained to reproduce adsorption processes calculations at a lower computational cost. However, it remains unclear if the ML models can transfer to different separation from the ones they have been trained on. In addition, currently reported machine learning models require simplified thermodynamic models for the adsorption properties as inputs. At the Energy and Process Systems Engineering (EPSE) group, we evaluate these adsorption properties of materials with classical Density Functional Theory (cDFT), which allows predictions at molecular simulation accuracy.

This project aims to combine the ML-based process models with cDFT adsorption calculations and explore the transferability of ML models to new separation processes.

This project will consist of several steps: • The first step will be to familiarize yourself with the given machine learning process model and the required input parameters of the models • In the second step, you will use classical Density Functional Theory to simulate the adsorption properties that are required as input to the ML model. You will develop a pipeline to couple cDFT calculations with ML process model evaluation. • Finally, you will compare the results of the ML model with already implemented first-principle models to evaluate the performance of the ML model on a separation different from the one it was trained on.

This project will consist of several steps:
• The first step will be to familiarize yourself with the given machine learning process model and the required input parameters of the models
• In the second step, you will use classical Density Functional Theory to simulate the adsorption properties that are required as input to the ML model. You will develop a pipeline to couple cDFT calculations with ML process model evaluation.
• Finally, you will compare the results of the ML model with already implemented first-principle models to evaluate the performance of the ML model on a separation different from the one it was trained on.

If you are interested in the project or want to learn more about it, please contact Marcel Granderath (mgranderath@ethz.ch). We look forward to hearing from you.

If you are interested in the project or want to learn more about it, please contact Marcel Granderath (mgranderath@ethz.ch). We look forward to hearing from you.