Research

Research within the Center for Data-Driven Design of Multifunctional Material Systems is organized within four technical thrusts. Within each thrust, a team of physical and computational scientists collaborates with partners at the Air Force Research Laboratory (AFRL) to address materials challenges using AI solutions.

Thrust one: Characterizing and predicting rare events in materials systems

A fundamental tenet of materials science is that processing generates the microstructure that mediates material properties. This thrust is built around discovering processing-structure-processing (PSP) relationships for fatigue failure in polycrystalline aerospace alloys, primarily Ti-based systems. In this thrust, we capitalize on computer vision (CV) methods applied to characterize and analyze the multimodal information contained in microstructural images coupled with machine learning (ML) to discover PSP connections.

CMU: Rollett (MSE), Poczos (ML)

AFRL:  Shah, Simmons, Payton, Criner, Musinski, Shade, John

Process flow for machine learning enabled microstructure characterization and failure prediction.

Thrust two: Multimodal materials data fusion

During the past two decades, the field of 3D materials characterization has matured to the point where high quality, large scale, multi-Tbyte data sets can be acquired. To a large extent, these experimental techniques involve scanning every point in a volume without considering whether or not that point will actually provide additional information about the microstructure. In our second research thrust, machine learning and signal processing is being applied to reveal and take advantage of structures in the measured data to dramatically speed up the acquisition process in scanning electron microscopy (SEM), high angle annular dark field (HAADF) tomography in the transmission electron microscope (TEM), and high energy diffraction microscopy (HEDM) at a synchrotron beam line.

CMU: De Graef (MSE), Rohrer (MSE), Rollett (MSE), Chi (ECE)

AFRL: Uchic, Donegan, Shah

Machine learning enabled data analysis to enhance the effectiveness of material characterization using multimodal characterization data.

Thrust three: Adaptive experimental design

Design of experiments for materials discovery is currently primarily done by humans in an ad-hoc manner and is prone to human errors, biases, and limitations in navigating complex high-dimensional spaces. Further, since collecting and/or labeling data is often costly, experimental datasets are often small with incomplete data representations. In our third thrust, we develop AI methods that can integrate feedback from prior data, simulations, and expert knowledge to guide subsequent experiments to efficiently search for materials with desired properties, with a focus on oxide materials. The approach integrates with the ARES software package developed at AFRL.

CMU: Jayan (ME), Rohrer (MSE), Singh (ML)

AFRL: Maruyama, Hooper

Autonomous synthesis of gold nanoparticles with in-line validation. Distinct colors of solutions indicate precise control of particle size in the nanometer range.

Thrust four: Adaptive materials design in complex environments

Active materials that are capable of organizing "on-command" into superstructures with desired properties promise breakthrough advances across a range of defense and energy related material technologies. A pervasive challenge in the realization of such materials is the vast and interconnected parameter space to be explored. The goal of this thrust is to establish toolsets and protocols for the efficient navigation of large parameter spaces that can be applied to the optimization of complex material systems and their adaptation to evolving constraints. The material testbed is constituted of self-assembled brush particle-based materials that are to be optimized for multiple property combinations.

CMU: Bockstaller (MSE), Kara (ME), Yao (HCII)

AFRL: Harper, Drummy, Buskohl, Gellman, Baldwin

Illustration of a particle brush and the relevant material characteristics that control material properties. Also shown is the approach for machine-learning enabled prediction of structure-property relations in brush particle hybrid materials.

SEED One

The goal is to optimize catalysts for the hydrosilylation to achieve unique silane-based materials which are used across a range of critical Air Force applications. The approach is to optimize the inexpensive reagents first and then to transfer the knowledge/algorithm iteratively to increasingly more expensive reagents.
CMU:

AFRL: 

SEED Two

This project focuses on establishing a data-driven design framework for tailoring the material properties of elastomer composites with MXenes and liquid metal (LM) inclusions. The goal is to develop machine learning classifiers and regression algorithms for establishing a mapping from processing parameters to the desired material nanostructure. 

CMU:

AFRL: