19–21 Oct 2026
Frascati
UTC timezone
The deadline for the abstracts submission has been delayed to July 20, 2026

Improving the time resolution of bolometric tomography in TCV with modelling of AXUV-diodes

20 Oct 2026, 16:00
10m
Sala Bruno Brunelli (Frascati)

Sala Bruno Brunelli

Frascati

Building F23 Via Enrico Fermi 45, 00044 Frascati, Rome
Short Oral AI in support of diagnostics AI in support of diagnostics

Speaker

Valentina D Agostino (Enea/Tor Vergata)

Description

The measurement of the total radiated power emitted by a tokamak plasma is a key diagnostic, as radiation represents a significant fraction of the power exhaust. Beyond its role in the global energy balance, the spatial distribution of radiation is essential for assessing impurity behaviour and ensuring the safe operation of plasma facing components. Bolometers provide accurate absolute measurements but are limited in temporal resolution, restricting the observation of fast events. AXUV diodes, part of the TCV RADCAM system, offer high temporal resolution but suffer from hardware related limitations that prevent stable absolute calibration. This work aims to reconstruct fast radiative events in TCV plasmas, such as ELMs and MARFEs, with the absolute sensitivity of bolometric diagnostics while exploiting the higher temporal resolution of AXUV measurements.
A first approach uses the emissivity reconstructed from bolometric tomography to compute the AXUV signals expected from the plasma. The ratio between measured and expected signals provides a time‑dependent recalibration factor applied to rescale the AXUV data, enabling synthetic bolometric signals with significantly improved temporal resolution. Preliminary results show that this factor not static, but vary between discharges and across campaigns, consistent with the known progressive degradation of diode sensitivity and eventually saturation. Ongoing statistical analysis across multiple shots is used to quantify this evolution and to assess the reliability of the method across a wide range of plasma conditions. As a second approach, a machine learning-based method is proposed. Neural networks are trained to learn the mapping between diode measurements and the corresponding bolometric outputs. Compared to the tomographic method, this data driven model exploits the complementary nature of the two diagnostic systems, capturing the underlying correlations despite their different spectral sensitivities, and can implicitly account for the long term sensitivity drift of the diodes. Once trained, the model provides calibrated AXUV signals with bolometer equivalent absolute sensitivity and high temporal resolution, allowing the study of radiative behaviour during fast transient, otherwise inaccessible with conventional bolometry. The robustness of the method is further enhanced by constraining the model with additional correlated measurements, such as electron temperature and electron density.

Author

Valentina D Agostino (Enea/Tor Vergata)

Co-authors

Andrea Murari (Consorzio RFX) Dr Gabriele Partesotti (EPFL-SPC) Gerarda Apruzzese (ENEA) Ivan Wyss (Università degli studi di Roma Tor Vergata) Michela Gelfusa (University of Rome Tor Vergata) Riccardo Rossi (Università di Roma 2 Tor Vergata) Mr Simone Kaldas (University of Rome Tor Vergata, Industrial Engineering Department and Enea Frascati Research Centre, Nuclear Department) Dr Umar Sheikh (EPFL-SPC)

Presentation materials

There are no materials yet.