Machine learning-enhanced traceability and particle analysis of pyrotechnic post-explosion residues using SEM/EDS

Abstract

Mordern techniques such as spectroscopic and chromatographic techniques have advanced pyrotechnic precursor analysis, yet the forensic investigation methods of post-detonation residues remains limited, thereby constraining the identification of explosive sources. Pyrotechnic post-explosion residues (PPERs) retain stoichiometric signatures that can be linked to precursor formulations through machine learning-enhanced scanning electron microscopy / energy-dispersive X-ray spectroscopy (SEM/EDS) analysis, offering a foundation for further forensic study. This study presents a systematic approach to PPERs analysis with three key components: (1) the development of a test vessel and particles collection system for PPERs; (2) the employment of an automated SEM/EDS protocol incorporating the Particle X Perception System for high-throughput elemental and morphological characterization (15,000–38,000 particles per sample), and a data pretreatment analytical protocol that included morphometric screening criteria and multivariate statistical methods; (3) the construction of a machine learning framework integrating t-distributed Stochastic Neighbor Embedding (t-SNE) for dimensionality reduction and Random Forest Regression (RFR) for predictive modeling. The hybrid model demonstrated excellent clustering performance (Normalized Mutual Information (NMI) > 0.80) and high predictive accuracy (R² > 0.95, Root Mean Squared Error (RMSE) < 0.07), supporting the potential for pyrotechnical traceability from SEM/EDS data of PPERs.