Machine learning based longitudinal virtual diagnostics at SwissFEL.

Area of Science:

• Accelerator Physics
• Particle Beam Diagnostics
• Machine Learning Applications

Background:

• Electron bunch length is critical for optimizing Free Electron Laser (FEL) performance.
• Traditional bunch length measurements using Transverse Deflecting Structures (TDS) are invasive and time-consuming.
• Existing non-invasive methods using synchrotron radiation have systematic discrepancies compared to TDS measurements.

Purpose of the Study:

• To implement and optimize a machine learning (ML) approach for predicting electron bunch length.
• To overcome discrepancies between synchrotron radiation monitor (SRM) and TDS-based bunch length measurements.
• To provide a non-invasive and efficient method for characterizing bunch length at SwissFEL.

Main Methods:

• Utilized a machine learning approach to analyze synchrotron radiation emitted by electron bunches passing through a magnetic chicane.
• Leveraged known energy chirp from upstream radio frequency structures.
• Trained ML models to predict bunch length (10 fs to 2 ps) and longitudinal profile.

Main Results:

• Successfully implemented and optimized an ML-based method for bunch length prediction.
• Overcame systematic discrepancies between SRM and TDS measurements.
• Enabled accurate prediction of bunch length and longitudinal profile at SwissFEL compression stages.

Conclusions:

• The developed machine learning approach offers a reliable, non-invasive alternative for electron bunch length characterization.
• This method enhances the optimization of accelerator performance by providing accurate and frequent measurements.
• The approach successfully addresses limitations of previous non-invasive techniques.