How can data science help researchers detect where radioactive material is? Ken Dayman, nonproliferation data scientist, talks data collection and analytics.
We collected data around ORNL’s high performance research reactor and collocated radiochemical processing facility with networks of detectors. We used multiple measurement modalities to generate a rich dataset relevant to a range of nuclear nonproliferation applications. A diverse team of researchers collaborated to develop and evaluate methods to extract interpretable information from rare events embedded in these large heterogenous data. Under these efforts, our team focused on the selective detection, characterization, and tracking of radiological material around the testbed facilities. Over the life of the project, we developed and demonstrated analytics to ingest streamed acoustic data to detect large vehicles; the analytics differentiated the vehicles from other sources of noise. We then fused the results with a physics-informed gamma-ray analysis to track the movements of shielded radiological material around the facilities. As part of this research, we developed an iterative self-supervision method that may enable long-term deployment of sensors and associated data analysis.
I fell into a career at ORNL somewhat by chance. My PhD advisor was on the review panel for an independent review on an ORNL-led project, and he let me know the project was starting work that was closely related to work I had done in graduate school. This led to an introduction and a post-doc job offer to support that project. Once I was at ORNL, I was lucky to expand my horizons and learn about the diverse work going on at the lab and amazing collection of facilities and capabilities.
My move into data science for nuclear nonproliferation was a very organic process. In school and during my post-doc at ORNL, I focused on data analysis for various experiments and simulation efforts such as neutron activation and gamma-ray spectrometry and post-detonation fallout debris generation and transport. During these projects, I started incorporating data analytics with more traditional statistical analyses and numerical optimization techniques. Over time, data analytic methods became a larger aspect of my work, and today, we have a group of nonproliferation data science specialists working to develop novel data analytic methods and solutions for the unique combination of problems pervasive in the national security space and fairly unique relative to the greater data science landscape.
