School of Civil and Environmental Engineering

Ph.D. Thesis Defense Announcement

MACHINE LEARNING AIDED INVERSE DESIGN FOR POLYMERIC MEMBRANES: CASE STUDIES IN NANOFILTRATION AND GAS SEPARATION

By Raghav Dangayach

Advisor:

Dr. Yongsheng Chen

Committee Members: Dr. Xing Xie (CEE), Dr. Ameet Pinto (CEE), Dr. Shane Snyder (CEE), Dr. Sankar Nair (ChBE)

Date and Time: April 02, 2026. 9 AM EST

Location: Price Gilbert Memorial Library 4222 (Dissertation Defense Room)

Online Link: https://gatech.zoom.us/j/7448923349

ABSTRACT
Polymeric membranes have been widely used for liquid and gas separation in various
industrial applications over the past few decades because of their exceptional versatility
and high tunability. Traditional trial-and-error methods for material synthesis are
inadequate to meet the growing demands for high-performance membranes. In that context, Machine learning (ML) has demonstrated huge potential to accelerate the
design and discovery of polymeric membranes.
ML models are highly complex and behave as “black boxes”, making it hard for
researchers to interpret the predictions made by them. Using Explainable Artificial
Intelligence tools such as Shapley additive explanations and Partial Dependence Plots
can help in visualizing the impact of features on ML models, giving insight into their
decision-making behavior. These tools were used to develop synthesis-propertyperformance
relationships to understand principles guiding the separation process of
Lithium and Magnesium using polyamide-nanofiltration (NF) membranes. The goal is
to ensure that the predictions made by ML models are consistent with domain
knowledge.
Using insights obtained from SHAP analysis, a novel, strategic framework was
proposed for rational screening of monomers and polymers for NF and gas separation
applications. A metric called "SHAP-score" was devised, which quantifies the impact
that a feature (in this case, a property of a polymer) has on an ML model. This
quantification is a product of the features' importance and their correlation to the
performance metric of choice, i.e., permeability or selectivity. The properties of
potential materials are scaled on basis of this impact, allowing us to score materials and
find potential candidates for membrane testing. This method is an attempt towards the
development of a general-purpose methodology for polymeric material discovery that
has huge applications not only limited to membrane science, but also various other
subfields.