Low-Level Radio Frequency (LLRF) Workshop 2025

Fault Prediction and Diagnosis for SRF Systems–Focus on RF Power Source Failures

Oct 14, 2025, 5:20 PM

20m

Newport News Marriott at City Center

Oral or Poster

Speaker

Jiayi Peng (Institute of Modern Physics)

Description

This research aims to develop an efficient and reliable fault detection method for radio frequency (RF) superconducting cavity system power sources. Superconducting cavities are core components of large-scale scientific facilities such as particle accelerators and synchrotron radiation light sources, and their stable operation is crucial for successful experiments. However, as a key component driving superconducting cavities, potential faults in RF power sources can lead to system performance degradation or even interruption. Traditional fault detection methods often rely on manual experience or simple threshold judgments, which suffer from low detection accuracy, poor real-time performance, and difficulty in handling complex fault modes.

This abstract proposes an intelligent fault detection framework based on machine learning, designed to overcome the limitations of traditional methods. The method first involves real-time data acquisition of critical operating parameters of the RF power source, such as output power and LLRF output. Subsequently, through feature engineering on these multi-dimensional data, effective features that can characterize the system's health status are extracted. During the fault detection model training phase, a combined strategy of supervised learning and unsupervised learning will be employed. For known fault types, classification models such as support vector machines will be constructed for precise identification; for unknown or novel faults, anomaly detection algorithms such as local outlier factor will be utilized for real-time early warning.

Experimental validation on CAFe2 demonstrates that this method can detect power source faults 15 days in advance. The research provides effective technical support for the predictive maintenance of SRF systems, enhancing the operational reliability and efficiency of large-scale scientific facilities.Compared to traditional methods, this intelligent fault detection system exhibits stronger adaptability and robustness, effectively reducing system downtime and ensuring the stable operation of large-scale scientific facilities. This research provides new insights for the predictive maintenance and intelligent management of RF superconducting cavity system power sources.