Exploring Machine-Learning-Enabled Libs Towards Forensic Trace Attributive Analysis of Fission Products in Surrogate High-Level Nuclear Waste

We investigated the utility of machine-learning-enabled LIBS for direct rapid analysis of selected fission products (FPs), namely, Y, Sr, Rb, and Zr in surrogate high-level nuclear waste mimicking three hypothetical but realistic scenarios: post-detonation glass debris, post-detonation powders, and microliter liquid drops from a radiological crime scene (RCS). Artificial neural network calibration strategies for trace quantitative analysis of the FPs in these materials were developed and achieved >95% prediction for all sample types. Owing to a lack of appropriate certified reference materials synthetic reference standards materials were used to perform method validation to accuracies >91%. Based on the spectral responses of the FPs, principal component analysis successfully differentiated nuclear from non-nuclear waste, demonstrating the method's potential for RCS nuclear forensic and attributive analysis.