16–21 Aug 2026
Daejeon Convention Center
Asia/Seoul timezone

Automated conditioning utilizing machine learning: network optimization and reinforcement learning

TUPO037
18 Aug 2026, 16:00
2h
1F Exhibition Hall (Daejeon Convention Center)

1F Exhibition Hall

Daejeon Convention Center

Poster Presentation MC4.A07: Room temperature RF TUPO - Poster Session

Speaker

Stephan Wagner-Rossel (Goethe University Frankfurt)

Description

RF-conditioning is an essential pre-processing step of normal conducting cavities. This time-intensive work can pose great risks to the equipment and cavity if conditioning-effects such as multipacting, discharges or degassing aren’t taken seriously.
To reduce the workload for human personnel, it was proposed to develop a deep-learning based algorithm to conduct conditionings on its own. This algorithm is trained on experimental data recorded from various conditionings performed by several experimenters. During initial training, the algorithm is tasked to predict the experimenters’ action based on the power-levels, pressure and frequencies recorded over the last seconds. So far developed networks have been able to perform this task with average errors of few hundred Hertz and few tenths of dBm respectively.
To teach the network to not only reproduce human behaviors, but identify the optimal course during conditioning, pre-trained networks are planned to be combined with reinforcement learning to enable continued learning during experiments. The results of experiments using optimized, pre-trained networks in two different modus opperandi are being presented.

I have read and accept the Privacy Policy Statement Yes

Author

Stephan Wagner-Rossel (Goethe University Frankfurt)

Co-authors

Holger Podlech (Goethe University Frankfurt) Dr Klaus Kuempel (Goethe University Frankfurt)

Presentation materials

There are no materials yet.