Collaborative Research: Efficient Learning of Process-Structure-Property Models in Value-Driven Materials Design

Georgia Tech Research Corporation, Atlanta GA

Investigators

Abstract

This award supports research that will contribute knowledge towards the efficient discovery of new material systems. New materials are expected to have a significant impact in a broad range of application domains, ranging from biomedical systems to infrastructure and energy systems, improving the efficiencies of these systems and creating new capabilities that have so far been technologically out of reach. In current practice, however, the development of new materials is very costly and time-consuming because it relies mostly on physical testing and experimentation. Rather than focusing on the development of a specific new material, this award aims to develop modeling approaches and learning algorithms that allow for more efficient and effective exploration and discovery of new materials in general. The results of the investigation will help material scientists and engineers understand when to rely on mathematical analysis models or when to use physical experiments so that new information about so far unexplored materials can be gathered efficiently, and so that the materials design effort can be efficiently guided towards materials with desired and valuable properties. The research is expected to lead to a dramatic acceleration of the materials design process with significant competitive advantages to US industry. Through a university-industry consortium these innovations will be transferred into industrial practice. All new models and algorithms will be shared open-source, and the research findings, methods and tools will be incorporated in on-campus and on-line courses, with the potential to reach a large number of students, researchers and practitioners. The main research objective of this project is to critically evaluate the relative merits of different modeling formalisms and approaches for capturing and utilizing materials domain knowledge in a way that is most valuable to the designer. In the design process, multiple information sources will be combined, including bulk material tests, low-cost experimental assays, and physics-based multiscale Process-Structure-Property models. The hypothesis is that combining information from a portfolio of information sources with synergistic cost-accuracy trade-offs leads to a more efficient and effective design process. A second focus is on combining the information from these sources into integrative reduced-order Process-Structure-Property linkages. These linkages support learning through Bayesian updating as new information is acquired, and they are computationally inexpensive and therefore well-suited for searching the design space. The overall design framework will be applied and validated in the context of dual phase steels. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.