Automated Crystal Structure Identification from X-ray Diffraction Patterns

X-ray diffraction is a commonly used experimental technique for identifying the structure of crystalline materials[3]. A crystalline material has its atoms arranged in periodic lattices with long-range symmetry. As a result, it scatters a beam of X-rays in well-defined patterns. Analysis of an X-ray diffraction pattern produced by a material yields extensive information about the structure of the material, in particular, about the kind of periodic lattice its atoms are arranged in. Currently, basic analysis of X-ray diffraction patterns is often performed manually, often requiring human input and some guesswork at what the structure of an unknown material is likely to be. This makes it hard to use X-ray diffraction data in automated analysis of large datasets. Automating the process of structure determination from X-ray patterns would be a step toward high-throughput computational searches for materials with desirable properties a goal that the materials science community has been moving to in recent years as available computational power has increased. This project aims to automatically classify the crystal structure of an unknown material. There are seven possible basic crystal systems (also known as Bravais lattices) that a crystalline material can adopt, and the measured XRD pattern of the material contains sufficient information to say which crystal system the material belongs to. The input to our program is an XRD pattern that consists of scattered X-ray intensity as a function of a variable called the reciprocal lattice vector (denoted q). We discretize this pattern by slicing up the q-space into a variable number of pieces and use the intensity at each q as a feature. The output is a classification of the given XRD pattern into one of the seven basic crystal systems. We implement a Naive Bayes classifier and a neural network to perform the classification and compare their performance.