(455i) Data-Driven Acceleration of Materials Development: Leveraging AI/ML to Scalably Bridge the Gap from Models to Materials

Materials play a vital role across diverse industries, impacting areas from healthcare to consumer goods to automotive technologies and beyond. The process of researching, developing, and scaling up materials involves a multifaceted approach encompassing fundamental understanding of chemical and physical properties, chemical and/or materials science engineering, modeling ranging from molecular to macroscale, experimental workflows, and ultimately product development processes. Materials development requires a blend of fundamental-level insights and empirical knowledge around process development and optimization in order to analyze macroscopic properties and lead to the eventual development of a viable product that can be brought to market.

This presentation will explore methodologies for integrating information from molecular to macroscopic scales while incorporating both fundamental principles and applied domain expertise to address complex multiobjective challenges centered around materials development. Using an AI/ML approach, we will discuss how to leverage both fundamental and applied domain knowledge alongside data-driven insights to accelerate experimental processes with a focus on generating viable material products using small, information-rich datasets (<100 data points).

A case study from the materials industry will be used to illustrate the application of this approach in solving business-driven problems by iteratively using the relevant corpus of data to suggest statistically ranked candidate experimental procedures, experimentally testing those AI-generated protocols, reincorporating the resulting new data to improve models, and repeating this sequential learning process to arrive at novel materials that achieve complex performance targets. This work demonstrates the value of utilizing the entirety of relevant data on hand that an expert experimentalist will understand to have impact upon eventual material performance while not requiring explicit mathematical relationships between material properties or processing conditions when those relationships may not exist.