Authors:

Myron Milad Tadros

Date Created:

Not specified

Course Title:

Professor:

Not specified

About Paper:

Gamma-ray spectroscopy is used by nuclear physicists to better understand the structure of the nucleus in excited states. By observing the decays from an excited nucleus, we can create level-schemes that map its allowed energy states and transitions. With this knowledge, we can identify specific energy states that are useful in application to batteries known as isomers. Isomers are states with a long half-life, allowing for the storage of energy with little decay at a smaller size than their chemical counterparts. However, few convenient isomer states are currently known, and more level-schemes are required to identify more isomer states. The current method to create level-schemes demands months of expert labor, making isomer identification slow and hard to reproduce. We recast the task as an image-to-graph translation problem and propose a machine-learning pipeline that turns an entire coincidence matrix directly into a level-scheme. We explore a solution for this problem by using convolutional neural networks as well as conditioning an autoregressive graph generator on the synthetic gamma-gamma matrix input. To build intuition before tackling sparse experimental data (<1% signal), we train and validate on Monte-Carlo-derived gamma- gamma matrices and level-schemes, progressively increasing spectral complexity to mimic realistic noise. Utilizing the machine learning models mentioned, the level-schemes generated will serve as a benchmark for the accuracy of the model when used on real gamma-gamma data. By staging development in these synthetic environments, we aim to deliver a framework capable of real-time, reproducible spectroscopy once migrated to experimental data. Keywords: Gamma Ray Spectroscopy; Machine Learning; Nuclear Physics; Monte Carlo; Level Scheme

Source:

Purdue University / 2025

Topics:

No topics listed

Co-authors:

Myron Milad Tadros