Design and development of a FOC algorithm based on machine learning for the HEPS
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Abstract This paper presents a Multi-layer Perceptron (MLP) neural network-based Fast Orbit Feedback (FOFB) system for the High Energy Photon Source (HEPS), replacing traditional PID control.The approach eliminates explicit response matrix dependence by implicitly encoding beam dynamics within learned parameters, reducing commissioning overhead. The MLP model maps 576 beam position monitor (BPM) measurements to 384 fast corrector outputs.Trained on HEPS data with systematic hyperparameter optimization and 8-bit quantization, the model was deployed on FPGA using custom pipelined architecture.The implementation achieves 0.732 ms inference time, contributing to 0.857 ms total system latency — critical for high-frequency disturbance rejection.Validation demonstrates mean absolute error of 1.04 mA with excellent hardware-software consistency ( R 2 = 0.99).The low latency enables 140 Hz feedback bandwidth and achieves 49.2% orbit RMS reduction (from 3.364 μm to 1.707 μm), meeting HEPS stability requirements.
Fields of Study & Subjects
Particle accelerators and beam dynamics
Particle Accelerators and Free-Electron Lasers
Magnetic confinement fusion research
Publication details
- Journal / source
- Journal of Instrumentation
- Volume
- 21
- Issue
- 03
- Pages
- T03012–T03012
- Publication date
- 1 March 2026
- Language
- EN