Accurate prediction of mega-electron-volt electron beam properties from UED using machine learning

Area of Science:

• * Ultrafast electron diffraction (UED) and microscopy (UEM)
• * Accelerator physics
• * Machine learning applications

Background:

• * Accurate characterization of electron beam properties (emittance, energy spread, pointing jitter, energy fluctuation) is essential for UED/UEM.
• * Existing diagnostics are often invasive and cannot operate at high repetition rates, limiting real-time analysis.
• * Inherent fluctuations in UED/UEM instruments necessitate per-bunch beam property knowledge.

Purpose of the Study:

• * To develop a real-time, noninvasive method for characterizing electron beam properties in UED/UEM.
• * To enable accurate prediction of beam properties for each electron bunch during experiments.
• * To overcome limitations of current invasive and low-repetition-rate diagnostics.

Main Methods:

• * Implementation of a machine learning (ML) strategy.
• * Training an ML model on a dataset of fully diagnosed electron bunches.
• * Utilizing easily recorded detector parameters at high repetition rates for prediction.

Main Results:

• * Accurate prediction of electron beam properties for every shot using ML.
• * The developed technique is noninvasive and operates at high repetition rates.
• * Successful application of ML for real-time beam characterization.

Conclusions:

• * ML-based diagnostics offer a powerful solution for real-time electron beam characterization in UED/UEM.
• * This approach enables online optimization of beam stability and quality.
• * Facilitates advanced data analysis, including event filtering and online corrections, unlocking new experimental capabilities.