Enormous amounts of data are available to design smart materials and to optimize their performance. The capability to process big data is one of the major challenges in the U.S. government’s recent Materials Genome Initiative (MGI), which will establish an infrastructure for materials and technology innovation to revolutionize state-of-the-practice in various industries.
In this research, we will investigate fundamental questions that will enable the use of this big data for MGI:
- How to extract useful and relevant feature information from big data
- How to model materials knowledge in such feature information
- How to use this knowledge to design smart materials
- How to optimize the design
Modern imaging techniques for material characterization (such as X-ray tomography) now allow us to capture high-resolution 3D images of materials. Each image, revealing detailed material compositions, is stored as voxel values and have the size on the order of gigabytes. In order to extract useful information and knowledge out of this data, feature identification and reduced-order representation of microstructures are necessary.
Recently, a surfacelet transform approach was developed to identify important boundary features of materials (e.g. grain boundary in crystals, nano filler in composites). Parametric representation of microstructures based on surfacelets was developed (Fig. 1). Computationally, these inverse problems can be formulated as solving under-constrained linear equations. The main challenge for high-volume materials data is the computational complexity. A key task is designing algorithms for inverse problems on high-performance computing platforms (e.g., clusters, GPU and other accelerators).
Additionally, computational tools have been developed to simulate and predict the physical phenomena at multiple scales (quantum, atomistic, and macroscopic). Engineering design of smart materials requires an effective approach to integrate the information obtained from these tools. Robust decision-making based on the large amount of data generated by these tools is also a challenge.
Students may choose to do research on one or both objectives:
- Develop models of high-performance materials using data extracted from X-ray tomography of material samples (3D microstructure images)
- Solve inverse problems using image data to establish relationships between salient material structures and physical properties
- Create and use new software tools for large-scale materials image data analysis as part of an information infrastructure for the Materials Genome Initiative
- Develop robust computer simulation techniques to study functional materials, e.g., batteries for energy storage and phase-change memories for information storage
- Design and run multiscale simulations ranging from quantum to continuum scales
- Create methodologies for evaluating and quantifying uncertainty in models and simulations
Knowledge and Skills Expected to Be Developed for Students
- Materials Design
- Scientific and Numerical Computing
- High-Performance Computing
- Data Analytics for Images and Multiscale Data