Abstract:

Machine-learning augmented cross section evaluation

Cross sections are fundamental input data to modeling and simulation tools that researchers use to design reactors, dimension radiation shielding, plan medical treatments and more. Cross section libraries are the result of a convoluted process that converts experimental measurements into tabulated values. The raw data is sometimes affected by large discrepancies between different experimental campaigns, often lack information on uncertainty, and typically features large gaps in the energy range of interest. The evaluators have the challenging task to sort through these data and combine them with physics model, exposing the outcome to the risk of human bias. Furthermore, the evaluated data are tested in computational models of critical experiments with the risk that a compensation of errors could hide issues in the data. In this talk, we suggest that the cross section evaluation process can be improved using machine learning (ML) methods. We will illustrate an evaluation pipeline augmented by ML, present results for selected cross sections, and discuss the benefits and limitations of this approach.

Bio:

Massimiliano Fratoni

Massimiliano Fratoni is Xenel Distinguished Associate Professor in the Department of Nuclear Engineering at the University of California, Berkeley (UCB). He received a Laurea in Nuclear Engineering from Università di Roma “La Sapienza” (Italy), and a MSc and a PhD from the University of California, Berkeley. Prior to joining the Nuclear Engineering Department at UCB, he held a Research Scientist position at the Lawrence Livermore National Laboratory and a faculty position at The Pennsylvania State University. Prof. Fratoni’s main research interests are in sustainable nuclear energy through advanced reactors and advanced fuel cycles that maximize natural resource utilization and minimize nuclear waste.