Towards machine-learning-based on-the-fly analysis of neutron reflectometry

Reflectometry experiments can benefit substantially from recent advances in machine learning, enabling real-time data analysis, informed decision making during measurements, optimized experimental conditions and ultimately closed-loop experimental workflows. While most prior automation efforts have focused on X-ray reflectometry, neutron reflectometry can also benefit to the same extent. We report the first machine-learning-based pipeline for real-time neutron reflectometry deployed at the Institut Laue-Langevin (ILL), Grenoble, France. The system integrates the reflectorch package into the data acquisition workflow using the IT infrastructure of the facility. This enables analysis that is two orders of magnitude faster than conventional tools, allowing real-time estimation of physical parameters with associated uncertainties and feedback through a graphical user interface. The pipeline has been successfully tested at the ILL, and can be adapted to work at other neutron facilities seeking to enable real-time feedback-driven reflectometry analysis.