1. Education
  2. Thesis Finder
  3. Research groups
  4. SPT
  5. Solvent selection for LLX using Machine Learning Techniques

Solvent selection for LLX using Machine Learning Techniques

Solvent selection for LLX using Machine Learning Techniques

Description:

Solvent screening for liquid-liquid extraction has evolved over time. Originally, you would start in the lab with doing equilibrium measurements on a series of solvents, exploring the distribution coefficient of the solute(s) you would like to separate from the mixture and/or from each other. This is, however, a time- and capital intensive method of screening. To expand on the scope of solvents that you can include in the screening, upon availability of predictive tools like UNIFAC and software packages such as COSMO-RS, researchers have shifted to in silicio pre-screening. However, UNIFAC and COSMO-RS have limited accuracy for systems that are increasingly non-ideal (such as alcohols, acids and amines), in addition quantum chemistry based methods such as COMSO-RS can be computationally expensive. For liquid-liquid extraction, in-silico screening is more important than for example for distillation, because in liquid-liquid extraction there are two liquid phases where the accuracy is limited.

The latest development in predictive modeling is machine learning, and Dr. Vermeire (Assistant Prof at KU Leuven) has experience in the use of graph neural networks for the prediction of solubilities. With a technique called transfer learning models can learn trends in smaller datasets by leveraging other larger datasets of related but less accurate data to learn.

With this project, we will explore the use of this approach for liquid-liquid extraction solvent screening. As part of the assignment, you will visit Dr. Vermeire in Leuven to discuss with her the use of tools such as ChemProp and SolProp for property prediction.

We will select several systems related to metals recycling, and after first making a shortlist of solvent (mixture) systems based on machine learning, we will then study these in the lab to validate the accuracy of the machine learning approach on solvent screening.

Supervisors:

Nick Kumar (tutor) - n.kumar-3@utwente.nl
Dr. Florence Vermeire (external member: KU Leuven, Belgium)
Prof. dr. ir. Boelo Schuur (supervisor) - b.schuur@utwente.nl
​​​​​​​Prof. dr. ing. Meik Franke (other member) - m.b.franke@utwente.nl


Assignment

In this project, you will

  1. Learn about machine learning approaches for solvent screening
  2. Make specific screenings together with Dr. Vermeire (KU Leuven) on extraction of three metals of choice.
  3. Validate the shortlisted solvent compositions in the lab in Twente