Machine Learning–Enhanced Multiscale Modeling of Spatially Tailored Materials

University Of Iowa, Iowa City IA

Investigators

Abstract

Spatially tailored materials are metal-ceramic composites consisting of two or more different materials. The volume fraction of each material continuously changes in space. Compared to traditional composites, these composites offer advantages over traditional ones for they provide more opportunities to designers to fashion their performance for the operating conditions and environment. Since these composites have features spanning multiple length scales, it is usually necessary to develop predictive models capable of describing physical phenomena at different levels of resolution, e.g., at the nanoscale as well as the microscale. This award supports the development of an innovative multiscale method, enhanced by machine learning techniques and supported by experimental data, to investigate the mechanics of spatially tailored materials under mechanical and thermal loading. The proposed method will accelerate the design of the next generation of metal-ceramic composites for use in the automotive, aerospace, and biomedical industries. In addition, the award will leverage university programs to support: (1) undergraduate teaching and learning in data science and engineering; (2) recruitment of female, underrepresented minority, and LGBTQ students; and 3) outreach to K-12 students. Current state-of-the-art practices for numerical modeling of spatially tailored materials use the principles of micromechanics to bridge the gap from the lower scales under consideration to the macroscale. However, most methods cannot account for the molecular interfacial interactions and the microstructure uncertainties, both of which must be considered to accurately predict material responses at the higher length scales. And even when these considerations are addressed, it results in the difficulty of explicitly deriving effective material properties and constitutive or failure relationships. This project utilizes data science techniques and machine learning to overcome these challenges in the multiscale material modeling of a spatially tailored titanium alloy-titanium diboride metal-ceramic material system. The project will lead to the following general outcomes: (1) development of a data-enabled approach in hierarchical multiscale modeling to link and interface information (including uncertainties) across multiple length and time scales; (2) development of an efficient way to generate homogenized microscale models of heterogeneous composites that account for microstructure uncertainties; (3) development of an adaptive machine learning framework that updates with data accumulation; and (4) design of multiscale experiments under a variety of thermomechanical loading conditions. 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.