Neutron scattering experiments are a critical tool for the investigation of molecular structure in compounds. The HB-2A neutron powder diffractometer at the High Flux Isotope Reactor at ORNL conducts magnetic studies of samples by illuminating them with different energy neutron beams and recording the scattered neutrons. Proper identification and alignment of samples during an experiment is key to ensuring high quality data is collected. At present, this process is performed manually by beamline scientists. RadiaSoft, in collaboration with the beamline scientists and engineers at ORNL, has developed a machine learning-based software automating sample identification. We utilize a fully connected convolutional neural network configured in a U-Net architecture to identify the sample and its center of mass. We then move the sample using a custom Python-based EPICS IOC interfaced with the motors. In this poster, we provide an overview of our machine learning tools and show our results identifying samples at ORNL.
Chen et al. (Thu,) studied this question.