E N A B L I N G T E C H N O L O G I E S

1 7 2 H I G H L I G H T S 2 0 2 3 I

Machine-learning-based online data analysis enables autonomous closed-loop experiments

Real-time data analysis based on machine-learning (ML) presents an important opportunity to establish closed-loop feedback systems, enables live- monitoring physical parameters beyond observables and allows for real-time decision-making during synchrotron experiments. Here, an artificial neural network, capable of considering prior knowledge, was used to extract physical thin film parameters during an X-ray reflectometry (XRR) experiment.

X-ray user facilities rank amongst the largest scientific data producers in the world, and recent advances in accelerator development and detector technology are resulting in an increasing volume of data generated in experiments. This is driving a surge in interest regarding the application of machine-learning (ML) techniques to automate data analysis. In order to prepare beamlines for ML-driven experiments, specific solutions to manage acquisition, analysis and storage have been developed at research facilities or in data-driven national and international collaborations such as DAPHNE4NFDI [1], PaNOSC and ExPaNDS.

Using an X-ray reflectivity (XRR) experiment as a case study, this work presents the seamless integration of user-developed ML code with beamline control infrastructure, enabling real-time data analysis and integrated archiving of the analysed results with respect to FAIR (findable, accessible, interoperable, reusable) principles. It also demonstrates the accuracy and robustness of ML methods when applied to the analysis of XRR curves and Bragg reflections of thin film structures [2] through the ability to autonomously control a vacuum deposition setup.

User-developed ML code can be integrated into beamline control and data acquisition software such as BLISS [3] through the underlying TANGO layer [4] that is commonly used in beamline environments. This approach ensures high portability of the user-developed code between multiple synchrotron sources and demonstrates the interoperability of ML codes and TANGO to access entire ML models. Unlike beamline control processes, ML data analysis can run on compute resources in central computing facilities. Using VISA [5] a solution for remote access to IT infrastructure for data processing users can prepare and use IT infrastructure exclusively available to the experimental team shortly before and during specific experiments that can be customised to their needs.

For the case-study, a combined one-dimensional convolutional neural network (CNN) with subsequent multilayer perceptron was trained to extract physical thin- film parameters (thickness, density, roughness). The ML- model was used to reconstruct scattering length density (SLD) profiles in an XRR experiment on beamline ID10.

It is important to note that for a given SLD profile, the corresponding theoretical XRR curve can be swiftly calculated. However, reversing this operation presents a challenge because of the inherent ambiguity that often

Fig. 136: a) The machine learning pipeline with special emphasis on the

injection of priors at inference time. b) Sketch of the autonomous

acquisition and feedback scheme.