Utilizing Distributional Measurements of Material Characteristics from SEM Images for Inverse Prediction

• Activity Number: 619 - Topics in Defense and National Security
• Type: Contributed
• Date/Time: Thursday, August 1, 2019 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Defense and National Security
• Author(s): Daniel Ries* and John Lewis and Adah Zhang and Christine M Anderson-Cook and Marianne Wilkerson and Gregory L Wagner and Julie Gravelle and Jacquelyn Dorhout
• Companies: Sandia National Laboratories and Sandia National Laboratories and Sandia National Laboratories and Los Alamos National Laboratory and Los Alamos National Laboratory and Los Alamos National Laboratory and Los Alamos National Laboratory and Los Alamos National Laboratory
• Keywords: inverse prediction; functional data analysis; forensics

Abstract

The U.S. Government has been conducting actinide material processing experiments with the goal of identifying processing signatures of nuclear forensic value. Signatures have the ability to credibly predict the source characteristics of material. A large effort is invested in gathering scanning electron microscope (SEM) images of processed material with a subsequent analysis of particles using image analysis software, such as Morphological Analysis of MAterials (MAMA). Based on many measured particles, the software calculates many distributional characteristics of particles, including the perimeter, vector area, etc. Often, each distribution is summarized as a mean and standard deviation for use in the prediction of source characteristics. However, distributional measurements contain a wealth of information such as shape and skewness, which can provide meaningful information in discriminating source characteristics. Leveraging statistical functional regression approaches for entire distributions improves the prediction of source characteristics over the traditional approach of using simple summaries. The methodology is demonstrated with data from a bench-scale uranium study.