Sandia Labs FY22 Laboratory Directed Research & Development Annual Report

MACHINE LEARNING TO GO BEYOND THE DARKEN EQUATIONS IN MULTICOMPONENT MIXTURES.

Prediction of fluid transport properties— particularly in confined environments—can accelerate the materials discovery process for applications including energy storage, separations, catalysis, and drug delivery. Transport properties of fluid mixtures in porous environments are extremely difficult to obtain experimentally, and extensive modeling efforts are required to thoroughly sample the composition space needed for material prediction. To overcome these challenges, the Sandia LDRD team used ML to predict individual diffusion rates quickly and accurately for single and multi-component fluids in bulk and porous environments. By integrating different response functions directly in the model

development process, the underlying relationships describing the physics and chemistry controlling diffusion were identified. The resulting ML derived models provide extremely fast diffusion prediction for a wide range of fluid systems and pore architectures. This work has resulted in nine publications with numerous invited and contributed conference talks. One publication was highlighted on the Kudos Research Showcase website (click the Read Article section to view this article). Another publication was an Editor’s Pick for inclusion in a collection of hottest articles of 2021 in the journal Physical Chemistry Chemical Physics . (PI: Jeffery Greathouse)

The workflow for the development of ML-derived models for diffusion prediction involves iterative refinement of the features and algorithm.

ML model development enables quick and accurate prediction of individual diffusion rates for single and multi-component fluids.

75

LABORATORY DIRECTED RESEARCH & DEVELOPMENT