Studying complex microstructures with advanced transmission electron microscopy and machine learning

This project aims to develop novel microstructure characterization through the application of novel unsupervised machine learning methods (in particular advanced clustering approaches such as HDBSCAN and probabilistic fuzzy clustering) to correlated structure and composition data. The UoM has a unique transmission electron microscope capable of recording simultaneous high quality diffraction and EDX measurements and the student will optimize the experimental parameters needed for this data acquisition. They will also investigate the data science needed to work with the experimental measurements. Particularly previous studies have identified preprocessing, data merging and manifold transformation method of the multi-dimensional data as essential steps to make it amenable to machine learning. This will allow subtleties of microstructure to be identified with greater reliability than was previously possible. The methods will be developed as part of ongoing research into the structure-composition-property analyses of novel lithium battery cathode materials, engineering alloys and functional ceramics.