This project will be integrated within the activities of the GEMS at MLZ, a research entity focused on the development and characterization of engineering materials. GEMS operates alone or in cooperation with the TUM a TOF reflectometer, a SANS instrument, a diffractometer, and imaging instruments at MLZ.
Typical data of these instruments consist of 2D detector images to which other dimensions are often added (such as neutron wavelength, some external field affecting the sample, isotopic variation, to name a few). The resulting multidimensional data often display some characteristic patterns, offering the opportunity to use machine learning (ML) methods for their analysis. The experimental data acquired at different facilities can be augmented by simulated experiments to provide the data necessary to train ML models. Applications could include data enhancement (resolution upscaling) or rapid pattern recognition / classification of measurements.
A key characteristic of the experiments performed at neutron and synchrotron facilities is the large amount of data they generate. This is of course an opportunity for ML methods but it raises some technical, organizational and legal problems – for example regarding data privacy and limited possibilities of data transfer. This is also an issue for the involved industry partner, where decentralized, federated learning approaches are being explored.