Detection and Classification of Collective Beam Behaviour in the LHC

Collective instabilities can lead to a severe deterioration of beam quality, in terms of reduced beam intensity and increased beam emittance, and consequently a reduction of the collider’s luminosity. It is therefore crucial for the operation of the CERN’s Large Hadron Collider to understand the conditions in which they appear in order to find appropriate mitigation measures. Using bunch-by-bunch and turn-by-turn beam amplitude data, courtesy of the transverse damper’s observation box (ObsBox), a novel machine learning based approach is developed to both detect and classify these instabilities. By training an autoencoder neural network on the ObsBox amplitude data and using the model’s reconstruction error, instabilities and other phenomena are separated from nominal beam behaviour. Additionally, the latent space encoding of this autoencoder offers a unique image like representation of the beam amplitude signal. Leveraging this latent space representation allows us to cluster the various types of anomalous signals.