Machine-learning-aided accelerated discovery of energy materials with enhanced properties

Abstract

The discovery of novel materials plays a crucial role in advancing cutting-edge technologies, enabling innovative performance improvements in fields such as energy storage and electronics. As existing materials reach their limits, there is a growing demand for novel materials with superior performance to surpass current ones. However, the discovery of novel materials remains a challenging task, with traditional trial-and-error methods being time-consuming and resource-intensive. This dissertation explores the application of mahine learning (ML) to accelerate the discovery of high-performance materials, with a focus on two key areas: metal-organic frameworks (MOFs) for ammonia (NH3) adsorption and cathode active materials for magnesium (Mg) ion batteries. The first study leverages an ML model to efficiently screen over 12,000 MOFs, identifying eight MOFs with high NH3 working capacity through the integration of ML predictions and theoretical validation through GCMC and molecular dynamics (MD) simulations. The second study investigates Mg-based garnet structures for Mg-ion batteries, employing ML to accelerate the discovery process and first-principle calculations for theoretical validation. Together, these studies demonstrate the transformative potential of ML in materials science, offering a pathway to significantly reduce the time and resources required for materials discovery, while ensuring the reliability of predictions through computational validation.