"Machine Learning and Radiological Characterization"

Thursday 17 Apr 2025, 10:00 → 11:15 Europe/Zurich

24/1-016 - HSE Unit Head Meeting Room (CERN)

Matteo Magistris (CERN)

10:00 → 11:15 Machine Learning and Radiological Characterization

Machine learning (ML) refers to a set of computational techniques that can identify patterns in data and make predictions. Over the past decade, ML has advanced significantly, driven in part by the increasing availability of large datasets in fields such as social media, finance, and healthcare.
In radiological characterization, however, acquiring large datasets is more challenging due to the high costs of sampling, radiochemical analysis, and gamma spectrometry. To address this, the software ActiWiz can compute the complete radionuclide inventory of a radioactive item within seconds. At CERN, we can generate extensive datasets by introducing probability distributions for key irradiation parameters originating from TREC, such as position and waiting time, and simulating thousands of activation scenarios using ActiWiz.
The combination of ActiWiz and machine learning has already led to promising applications. These include developing decision criteria for clearance of electric cables, replacing gamma spectrometry with dose-rate measurements to characterize medium-level radioactive waste, and detecting unexpected materials in waste packages through gamma spectrometry. This approach opens exciting new possibilities in radiological characterization and in radiation protection in general.

Speakers: Andrea Gomes (CERN), Elisso Stamati (University of Ioannina (GR)), Jean-Baptiste Potoine, Matteo Magistris (CERN), Raphael Neugebauer (University of Vienna (AT))

Machine Learning and Radiological Characterization

 

Machine learning (ML) refers to a set of computational techniques that can identify patterns in data and make predictions. Over the past decade, ML has advanced significantly, driven in part by the increasing availability of large datasets in fields such as social media, finance, and healthcare.

In radiological characterization, however, acquiring large datasets is more challenging due to the high costs of sampling, radiochemical analysis, and gamma spectrometry. To address this, the software ActiWiz can compute the complete radionuclide inventory of a radioactive item within seconds. At CERN, we can generate extensive datasets by introducing probability distributions for key irradiation parameters originating from TREC, such as position and waiting time, and simulating thousands of activation scenarios using ActiWiz.

The combination of ActiWiz and machine learning has already led to promising applications. These include developing decision criteria for clearance of electric cables, replacing gamma spectrometry with dose-rate measurements to characterize medium-level radioactive waste, and detecting unexpected materials in waste packages through gamma spectrometry. This approach opens exciting new possibilities in radiological characterization and in radiation protection in general.