Speaker
Description
We present an unsupervised pipeline that learns a compact representation of beam-on machine state at the ALS, detects anomalies preceding beam-loss events, and highlights the responsible channels for operator diagnosis. Archiver data are resampled to a uniform time grid, filtered to beam-on intervals using stored current, and pruned by variability and principal-component analysis. A variational autoencoder with residual encoder-decoder stacks is trained on the standardised PV vectors; the global anomaly score and the per-PV attribution are both derived from per-PV reconstruction z-scores, so the score is an exact decomposition of the channel ranking.
We apply the pipeline to 34 beam-loss events from the 2025 ALS user run; in several cases it surfaces early-stage anomalies in the PV subsystems that subsequently led to the beam dump, indicating a framework can act as an early-warning aid for operators.
Funding Agency
This work was supported by the Director of the Office of Science of the U.S.Department of Energy under Contract No. DEAC02-05CH11231.
| In which format do you inted to submit your paper? | LaTeX |
|---|---|
| Preprint marking on your proceeding paper | I wish my paper to be marked as preprint. |