Speaker
Description
In order to improve the day-to-day particle accelerators operations and maximize the delivery of the science, new analytical techniques are being explored for anomaly detection, classification, and prognostications. We describe the application of an uncertainty aware machine learning (ML) method using Siamese neural network model to predict upcoming errant beam pulses as Spallation Neutron Source (SNS) accelerator. Predicting errant beam pulses reduces downtime and can prevent potential damage to the accelerator. The uncertainty aware machine learning model was developed to be able to detect upcoming errant beam pulses not seen before. Additionally, we developed a gradient class activation mapping for our model to identify relevant regions within a pulse that makes it anomalous and we use these regions for clustering fault types. We describe the accelerator operation, related ML research, the prediction performance required to abort beam while maintaining operations, the monitoring device and its data, the uncertainty aware Siamese method with its results, and fault type clustering results.
