CAD seminar series: Mahbub Islam; Generative AI and Machine Learning for Innovation in Electrolytes
When:
April 16, 2025
2:30 p.m. to 3:30 p.m.
2:30 p.m. to 3:30 p.m.
Where:
Event category:
Seminar
Hybrid
Speaker: Mahbub Islam, Mechanical Engineering, WSU
Time: Wednesday, April 16, from 2:30 pm to 3:30 pm
Location for in-person participants: 1146 FAB
Zoom link for online audience: https://wayne-edu.zoom.us/j/92845590121?pwd=CpRA5Wa5gzSMn2xiVkR2abD83O5nrH.1
Title: Generative AI and Machine Learning for Innovation in Electrolytes and Electrocatalysts for Next-gen Batteries
Abstract: The convergence of artificial intelligence (AI), quantum chemistry, and materials informatics is accelerating the discovery of novel molecules for next-generation energy storage technologies. In this talk, we present an AI-driven framework that integrates generative models, graph-based machine learning, and high-throughput quantum mechanical simulations to design advanced electrolytes and unravel electrocatalyst design principles for next-gen batteries, with a particular emphasis on Li/Na and nonaqueous Mg–CO₂ systems. We developed a generative adversarial network (GAN) with graph neural network (GNN)-based generator and discriminator architectures, trained on ~1 million molecules from the GDB-11 database, to explore the chemical space of potential electrolyte molecules. This approach generated over 30,000 novel, chemically valid molecules. To evaluate their thermodynamic and electronic properties, we applied a message-passing neural network (MPNN) trained on the QM9 dataset (~133,000 molecules) to predict critical quantum properties such as formation enthalpy, HOMO/LUMO energies, and electrochemical stability. Promising candidates were then screened via density functional theory (DFT) calculations to refine predictions of thermodynamic and electrochemical properties, including redox potentials and stability windows. The workflow led to the identification of several high-performance electrolyte molecules with wide electrochemical stability, making them strong contenders for use in Li/Na-metal batteries. In parallel, we explored the design of efficient electrocatalysts for Mg–CO₂ batteries. Leveraging a DFT-derived dataset, we trained ML models using gradient boosting regression (GBR) and artificial neural networks (ANNs) to predict the performance of porphyrin-supported single-atom catalysts (SACs) featuring 3d and 4d transition metals. The models demonstrated high accuracy through K-fold cross-validation. To gain interpretability, we employed SHapley Additive exPlanations (SHAP), permutation importance, and mean decrease impurity (MDI) techniques, identifying key physicochemical features governing catalytic activity. Our results underscore the transformative potential of generative AI and ML in both materials discovery and mechanistic understanding. These approaches offer data-driven alternatives to traditional trial-and-error methods, accelerating the development of safe, stable, and high-performing materials for future electrochemical energy storage systems.
Bio: Mahbub Islam is an Assistant Professor of Mechanical Engineering at Wayne State University. He earned his Ph.D. in Mechanical Engineering from The Pennsylvania State University, followed by a postdoctoral appointment in the School of Materials Engineering at Purdue University. Dr. Islam's research focuses on the application of artificial intelligence (AI), machine learning (ML), and first-principles simulations to understand interfacial chemistries in rechargeable batteries. His work spans metal–sulfur, metal–CO₂, and alkali-ion batteries, as well as electrocatalytic systems such as water splitting and electrochemical synthesis of ammonia. He specializes in density functional theory (DFT)-based modeling to predict functional catalysts and rationally design electrolyte additives for enhanced performance. In addition, Dr. Islam is an expert in reactive molecular dynamics simulations using the ReaxFF force field. He is the developer of eReaxFF, a novel computational technique that introduces explicit electrons into classical MD frameworks to model redox reactions. Dr. Islam has authored over 65 peer-reviewed journal articles and a book chapter. He is a recipient of the Ralph E. Powe Junior Faculty Enhancement Award, and his research is supported by the National Science Foundation (NSF) and the American Chemical Society- Petroleum Research Fund.