CLIC RF Breakdown Prediction

The proposed CLIC accelerator at CERN relies on RF cavities operating at a very high gradient. The main limiting factor on gradient of an RF cavity is RF breakdowns. This report documents the work on machine learning methods for predicting RF break- downs in the CLIC accelerating structures. The work was done during an internship at CERN during the period of September 2021 to December 2021. The work consisted of developing a framework for applying machine learning models on data from the XBOX2 test stand, in order to predict breakdowns. It also includes a study on the use of data augmentation for mitigation of the inherent class imbalance of the XBOX2 data-set. It was found that data augmentation yields an increase in prediction performance on follow-up breakdowns when using an FCN neural network.