Zhang, Jingqiong ORCID: https://orcid.org/0000-0003-2980-8145, Farr, Nicholas T. H. ORCID: https://orcid.org/0000-0001-6761-3600, Nohl, James, Lai, Yufeng ORCID: https://orcid.org/0000-0002-9987-0975, Abrams, Kerry J., Black, Kate, Willmott, Jon ORCID: https://orcid.org/0000-0002-4242-1204, Rodenburg, Cornelia ORCID: https://orcid.org/0000-0002-9590-375X and Mihaylova, Lyudmila ORCID: https://orcid.org/0000-0001-5856-2223 (2025) Toward automated chemical analysis of materials using secondary electron hyperspectral imaging and unsupervised learning. IEEE Access, 13. pp. 173976-174000. ISSN 2169-3536 (Online) (doi:10.1109/ACCESS.2025.3615908)

Preview

PDF (Open Access Article) 53618 ZHANG_Toward_Automated_Chemical_Analysis_Of_Materials_Using_Secondary_Electron_(OA)_2025.pdf - Published Version Available under License Creative Commons Attribution. Download (8MB) | Preview

Abstract

Advancements in materials science have significantly transformed materials discovery and advanced manufacturing. This, along with the rapid development of sensing and instrumentation, results in a continuous increase in data volumes. To address the limitations of conventional manual analysis, this paper introduces an AI-driven framework for high-throughput chemical analysis of material surfaces at the micro- and nano-scale. The framework integrates unsupervised machine learning with secondary electron hyperspectral imaging (SEHI). It consists of four stages: hyperspectral image processing via tiling, spectral peak extraction, peak categorisation by probabilistic clustering, and chemical analysis. Tiling enables the capture of local spatial-spectral information and generation of a large number of training samples from a single SEHI image stack. After tile-wise spectral peak extraction, the distribution of the peak positions is accurately represented by probabilistic clustering with a Gaussian mixture model (GMM) or a Dirichlet process Gaussian mixture model (DPGMM). Each peak corresponds to a specific chemical bond or element in a material, reflecting the unique spectral characteristics. The performance of the GMM and GPGMM approaches is validated over a case study for identifying chemical elements or bonds of complex metal alloy and carbon films. The results demonstrate accurate chemical analysis, yielding relative errors within ±15% compared to the theoretical model of the valence band density of states. This work is a step forward towards automated material analysis across different tasks such as identifying chemical elements and bonds, visualizing surface (in)homogeneity in metal alloy films for guiding film printing, and supporting digital twins integration for advanced manufacturing.

Item Type:	Article
Uncontrolled Keywords:	advanced manufacturing, artificial intelligence, Gaussian mixture models, material surface chemistry, microscopy, probabilistic clustering, secondary electron spectroscopy, unsupervised learning
Subjects:	Q Science > Q Science (General) Q Science > QA Mathematics > QA75 Electronic computers. Computer science T Technology > T Technology (General) T Technology > TP Chemical technology
Faculty / School / Research Centre / Research Group:	Faculty of Engineering & Science Faculty of Engineering & Science > School of Engineering (ENG)
Last Modified:	28 May 2026 15:13
URI:	https://gala.gre.ac.uk/id/eprint/53618

Actions (login required)

View Item

Downloads