Speaker
Description
Beam position monitor health is critical for fast-orbit feedback and orbit interlocks, yet many facilities still rely on fixed Q-noise thresholds and manual inspection. We propose a context-aware anomaly detection framework that uses multi-signal machine learning on BPM Q/U/V/S data and machine-state variables (beam current, time since injection, FOFB state, building temperatures, fill pattern). From periods of nominal operation we learn per-BPM baselines and low-dimensional ring-wide noise patterns using straightforward statistical models (empirical distributions, PCA). For any subsequent data set, archived or live, the system computes an interpretable anomaly score per BPM and time window by combining a context-aware rarity measure with a spatial residual from the ring-pattern model. Scores are then grouped into episodes with severity levels and simple spatial classifications (isolated BPM, local cluster, global effect). We will present initial results on SPEAR3 historical data and first comparisons with online deployment in shadow mode
| In which format do you inted to submit your paper? | Word |
|---|