Fig. 1. Schematics of (a) the JET horizontal and vertical bolometer cameras, (b) low-spatial-resolution views, and (c) the intersection region.

Fig. 2. Main plasma parameters of the analyzed database: number of pulses and percentages of disruptions.

Fig. 3. Radiation cases: core radiation in the top row (#94161), low-field-side radiation in the middle row (#94447), and high-field-side radiation in the bottom row (#94615). Tomographic reconstructions have been obtained by averaging bolometer signals in a time window of 10 ms.

Fig. 4. Observed frequency maps in the space of radiated power vs plasma energy for the regions where radiation anomalies appear more often (core, high-field side, and low-field side). The red lines represent the linear boundary between safe and anomalous regions of the operational space. Analogous results are obtained for the other macro pixels.

Fig. 5. Pulse 96 486: time traces of the plasma current and plasma energy (first row); input power and radiated power (second row); locked-mode amplitude and dimensionless core radiation factor (third row); dimensionless divertor, HFL, and HFT radiation factors (fourth row); dimensionless top, LFR, and LFT radiation factors (fifth row); outer Be II photon flux (used to detect ELMs) and total gas rate (sixth row). The right column shows the same plots with a zoom near the disruption.

Fig. 6. Pulse 96 486: electron temperature profile (colormap) and Λ_n,core as a function of time.

Fig. 7. Pulse 94 650: time traces of plasma current and plasma energy (first row); input power and radiated power (second row); locked-mode amplitude and dimensionless core radiation factor (third row); dimensionless divertor, HFL, and HFT radiation factors (fourth row); dimensionless top, LFR, and LFT radiation factors (fifth row); outer Be II photon flux (used to detect ELMs) and total gas rate (sixth row). The right column shows the same plots with a zoom near the disruption.

Fig. 8. Pulse 94 650: tomographies (at t = 15.5, 15.8, 15.9, 16, and 16.5 s) and visible camera frames (from t = 15.58 s to t = 16.74 s) before the disruption.

Fig. 9. Pulses 94 447 (a) and 94 655 (b): time traces of the plasma current and plasma energy (first row); input power and radiated power (second row); locked-mode amplitude and dimensionless core radiation factor (third row); dimensionless divertor, HFL, and HFT radiation factors (fourth row); dimensionless top, LFR, and LFT radiation factors (fifth row); outer Be II photon flux (used to detect ELMs) and total gas rate (sixth row). The right column shows the same plots with a zoom near the disruption.

Fig. 10. Pulse 94 447: tomography inversions before disruption using the ML tomography algorithm.

Fig. 11. Distribution of anomalies for each macro-pixel: the histograms are for the region of the plasma where the anomaly is detected first by the Λ_n,i indicators described in Sec. IV.

Fig. 12. Distribution of anomalous radiation events detected in discharges that do not disrupt.

Fig. 13. Relative times of the various types of anomalous events detected by the indicators considered.

Fig. 14. Pulse 96 491: Evolution of the main plasma quantities of interest for disruption avoidance.

Fig. 15. Pulse 96 491: evolution of the temperature profile (colormap), showing the period when it becomes hollow from t ∼ 10.1 s to t ∼ 12.5 s. The red line shows the core radiation anomaly signal (Λ_n,core).

Fig. 16. Pulse 94 611: evolution of the main plasma quantities of interest for disruption avoidance (left) and visible camera frames (right).