Fig. 1. System architecture design of the CSNS intelligent operation attendant

Fig. 2. Distribution of RCS dipole magnet power amplitude variation and its set values (the upper curve represents the measured values, while the lower curve represents the set values)

Fig. 3. Illustration of different data types in accelerator systems (a) Distribution of RFQ reflected power, (b) Beam profile captured by the multi-wire target near the window

Fig. 4. Distribution comparison of normal and abnormal vectors (a) Normal vector distribution, (b) Abnormal vector distribution

Fig. 5. Example of anomalies successfully detected only by the clustering algorithm (not detected by autoencoder or isolation forest)

Fig. 6. Example anomaly detection results for other cooling water temperature data

Fig. 7. Detection results of the clustering model on dc power supply fault data

Fig. 8. Distribution of power supply PV and targeting power during abnormal DC power supply conditions (top curve: current error of LRWBPS01 power supply, bottom curve: beam targeting power)

Fig. 9. Detection results of the clustering model on resonant power supply fault data

Fig. 10. Model deployment functional architecture

Fig. 11. Detection results of the clustering model on cooling temperature fault data

Fig. 12. Example of the web monitoring interface