Combinatorial Materials Science represents a potentially powerful approach to identifying new and unexpected materials. This involves the rapid, high-throughput synthesis, measurement, and analysis of a large number of different materials. Understanding the functional behavior of the materials requires a characterization of the structure-property relations. Crystalline structure information can be obtained through X-ray diffraction studies. An unsolved challenge is to develop automated techniques for identification of unique diffraction patterns and to cluster the resulting patterns into contiguous phase fields corresponding to regions with different material composite structures.
Intellectual Merit: This exploratory project is aimed at establishing the feasibility of a unique interdisciplinary approach, involving a team of materials scientists and computer scientists, to address the challenge of structure (crystalline phase) identification of the composite materials. Specifically, the PIs propose to extract the key diffraction pattern features from the raw experimental data as a first step towards the development of computational methods for the identification of crystalline phases.
Broader Impacts: The project, if successful, will establish the feasibility of a key first step in an overall methodology to significantly speed the materials scientific discovery process in general, and in the search for new materials for the next generation fuel-cell technology in particular. The project brings together faculty and students, providing training in materials science, engineering, and computer science.
The main goal of the project was to develop techniques that demonstrate the feasibility of automated analysis of X-ray diffraction datasets to infer crystalline phase ranges in ternary and higher composition spaces, using real-world datasets that introduce counting noise, interfering background signals, and other experimental imperfections. Our work focused on three main areas. First, we made improvements to the experimental methodology used to collect X-ray diffraction data. We adopted a new deposition technique to generate composition spreads with small sections of substrate removed and covered with an ultrathin amorphous membrane, eliminating a large source of background noise. We also moved our experiments to an updated beam line with higher photon flux and a detector configuration allowing for multiple images per sample. This allowed us to better characterize measurement noise as well as to detect and remove some artifacts that can appear in the data. Secondly, we sought to improve the signal/noise discrimination in the real-world raw data in order to provide more reliable input for automated algorithms used to determine material structure. In addition to developing pre-processing techniques to exploit opportunities provided by the new experimental methods, we also made significant progress in the automated detection of diffraction peaks in crystalline mixtures. Finally, we produced and published a series of example synthetic and real datasets, and related tools, for the phase map identification problem, making this problem accessible to a wider range of computational researchers.
