Back to search

University of Minnesota

Machine Learning in Conjunction with Molecular Simulations for Nanoporous Materials Discovery

Abstract

dc:description.abstract

Chemical storage and separation are critical in solving various energy and environmental problems in the real world, and nanoporous materials are believed to be promising candidates for these applications. The synergy among experiments, computer simulations, and machine learning opens up possibilities to revolutionize nanoporous materials discovery, where new materials can be designed dramatically faster with a fraction of the cost. This dissertation research develops a series of machine learning methods based on high-throughput molecular simulation data for nanoporous materials discovery. First, a brief background on molecular simulations, machine learning, and their application in nanoporous materials discovery is introduced. Then, Chapter 2 develops a machine learning model, named SorbNet, on the optimization of the desorptive drying process. SorbNet is trained using loss function based on statistical mechanics principles of Monte Carlo simulations. The prediction given by SorbNet is used to perform the design of the drying process of alkanediols in zeolites. The capability of transfer learning of a pre-trained SorbNet model is also demonstrated. Chapter 3 extends the research in Chapter 2 to much more complex separation processes and develops SorbNetX, a physics-informed model for mixture adsorption. SorbNetX originates from the ideal-mixture adsorption model in statistical mechanics, and by incorporating the principles of adsorption equilibria, SorbNetX accurately predicts the adsorption of a 8-component mixture using only unary, binary, and ternary data. Next, Chapter 4 employs meta-learning to the joint optimization in material space and state space for vehicular hydrogen storage. The meta-learning model, SorbMetaML, gives higher accuracy and improved generalization compared to fitting a model separately to each material and identifies the optimal hydrogen storage conditions for a variety of nanoporous materials including zeolites, metal--organic frameworks, and hyper-cross-linked polymers. Finally, Chapter 5 analyzes and predicts the adsorption in nanoporous materials at a crystal structure level. This chapter introduces the concept of spatially-resolved loading surface of nanoporous materials and develops the SorbIIT model as the first data-driven method to generate a continuous spatially-resolved loading surface.

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Sun, Yangzesheng

Rights

Language dc:language.iso
en

Identifiers

dc:identifier.*
Dc Identifier Citation
Sun, Yangzesheng. (2022). Machine Learning in Conjunction with Molecular Simulations for Nanoporous Materials Discovery. Retrieved from the University Digital Conservancy, https://hdl.handle.net/11299/269237.

Chain of custody

source
Harvested from
University of Minnesota
Base URL
conservancy.umn.edu/server/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
related terms
citation

Sun, Yangzesheng. Machine Learning in Conjunction with Molecular Simulations for Nanoporous Materials Discovery. 2022. https://hdl.handle.net/11299/269237