Machine learning using beam loss monitors for DIAMOND-II

The slow losses measured by Beam Loss Monitors (BLMs) at synchrotron light source facilities offer useful but indirect insight into the state of the beam. Patterns arise across the set of BLMs depending on the movement of insertion devices, beam current, temperature, humidity, and other contributors. A variety of neural network models were designed and evaluated to model this behaviour under user beam operation to enable anomaly detection and aid fault investigations.