Speaker
Description
In the context of the operational monitoring of the ARRONAX C70XP cyclotron, our previous work addressed the limitations of the Isolation Forest (IF) algorithm in detecting local anomalies, particularly those occurring near the mean of normal data, due to its reliance on axis-parallel splits. To overcome this issue, we developed and validated a hybrid model combining an autoencoder and IF, using time series data from the proton beam intensity on target. This approach significantly improved the detection of both global and local anomalies, with no false alarms observed during evaluation. Building on these results, the present study investigates the use of transfer learning to generalize the hybrid model to other process variables originating from different subsystems, including the source, injector, and cyclotron core. Results suggest that the model can effectively label large volumes of multivariate operational data, supporting the development of a more scalable and integrated anomaly detection framework for the C70XP.
Funding Agency
The Arronax cyclotron is supported by the CNRS, Inserm, INCa, Nantes Université, the Regional Council of Pays de la Loire, local authorities, the French government, and the European Union.
| In which format do you inted to submit your paper? | Word |
|---|---|
| Preprint marking on your proceeding paper | I do not wish my paper to be marked as preprint. |