Mawdsley, Benjamin 
ORCID: 0000-0002-2607-4716, Dorkings, Sam, Maskell, Simon ORCID: 0000-0003-1917-2913, Pickup, David, Chong, Samantha ORCID: 0000-0002-3095-875X, Farrell, Thomas and Phillips, Alexander ORCID: 0000-0002-1637-4803
(2025)
Probabilistic modelling to account for uncertainty in portable energy dispersive X-ray diffraction measurements.
Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment, 1075.
p. 170430.
ISSN 0168-9002, 1872-9576
|
Text
NIMA_Version_PostDSTL.pdf - Author Accepted Manuscript Available under License Creative Commons Attribution. Download (5MB) | Preview |
Abstract
Security systems often rely on being able to identify unknown materials in order to categorise them as harmful or safe. The process must be non-destructive, to preserve the material, and be applicable at a distance to reduce the hazard posed when performing the analysis. These restrictions have led to the use of X-ray diffraction imaging, with the challenge then becoming how best to analyse the data to provide well calibrated risk estimates. In this work, we explore probabilistic methods for quantifying our uncertainty when classifying energy dispersive X-ray diffraction data. Changes in the observational environment, or in the condition of the material itself, will introduce significant variation in the data collected. Through the use of a novel detector system, we collect data which aims to represent some of this uncertainty whilst capturing both angular and energy information. We apply Bayesian modelling techniques to the resultant images to enable inference of material properties whilst being robust to changes in experimental geometry. Such a model is more robust to operating in differing environments, and therefore well suited to practical applications. This model additionally aims to provide well quantified probabilities, in contrast to other approaches which do not assign a confidence level, allowing for more informed decision making. Our probabilistic approach is particularly well suited to a small data domain, where collecting data is a costly exercise and incorporating prior knowledge is desirable. We find that our model manages to extract key, material-specific information such that it is able to classify them with approximately 70% accuracy. Finally, we begin to develop an explicit Bayesian model of the experimental design, opening the way for future work to increasingly parameterise the data collection process and more accurately quantify uncertainty, and to do so across more varied environments.
| Item Type: | Article |
|---|---|
| Uncontrolled Keywords: | 5106 Nuclear and Plasma Physics, 51 Physical Sciences, Generic health relevance |
| Divisions: | Faculty of Science and Engineering Faculty of Science and Engineering > School of Electrical Engineering, Electronics and Computer Science |
| Depositing User: | Symplectic Admin |
| Date Deposited: | 17 Mar 2025 10:46 |
| Last Modified: | 09 Jun 2025 14:43 |
| DOI: | 10.1016/j.nima.2025.170430 |
| Related Websites: | |
| URI: | https://livrepository.liverpool.ac.uk/id/eprint/3190844 |
Altmetric
Altmetric