Development of Chemistry-Informed Machine Learning Tools for the Analysis of Powder X-Ray Diffraction Data

The collection of powder X-ray diffraction (PXRD) data is a commonly employed technique in the analysis of new materials, especially those formed as crystalline powders. Developments of high-throughput and automated techniques have the potential to accelerate the process of data acquisition of PXRD data, which will increase the data available in structural databases. The emergence of big data in this area merits the development of methods which can be used to perform analysis on, or gain insights into the structural data of materials in an automated fashion. In this work, a series of tools known as the Python Powder X-Ray Diffraction Toolkit (PyPDT) were developed to prepare for the inevitable “big data” generation in the area of PXRD. This work can be broken down into 3 tools; a physics-informed synthetic PXRD data generator, clustering and dimensionality reduction tools for representing and classifying PXRD data, and a data visualisation tool to view the results of the PXRD classification tool, alongside the underlying PXRD data. These tools hold the potential to identify novel structures or experimental anomalies present in unlabelled experimental data by leveraging existing structural data from databases, whilst allowing for dynamic visualisation and analysis of the machine learnt representation alongside the PXRD data. This introduces a level of transparency and interpretability to the process, providing users with increased confidence and validity of the results. The process of developing, and examples of using these tools on real-world examples of experimental significance, are presented in this work and potential areas where further developments can be made are highlighted.