# Forensic classification of gunpowder and fireworks powders by ATR-FT-IR spectroscopy and chemometric modelling

## TL;DR

This paper presents a method using infrared spectroscopy and chemometric models to quickly and accurately distinguish gunpowder from fireworks powders for forensic analysis.

## Contribution

The study introduces a validated ATR-FT-IR and chemometric workflow for forensic classification of energetic materials with high accuracy.

## Key findings

- PCA identified key spectral regions linked to nitrocellulose and nitrate-perchlorate compounds for differentiation.
- LDA and SVM models achieved 97.1% classification accuracy, with SVM capturing non-linear patterns.
- Misclassifications were attributed to compositional overlap, not analytical error, highlighting the method's reliability.

## Abstract

Rapid and reliable discrimination of ammunition propellants from consumer fireworks powders is critical in forensic explosives analysis but remains challenging due to overlapping chemical signatures and variability in formulations.

In this study, attenuated total reflectance Fourier transform infrared (ATR-FT-IR) spectroscopy was combined with multivariate chemometric models to classify sixty-nine real-world gunpowder samples, including forty-three ammunition propellants and twenty-six fireworks powders. Several spectral preprocessing strategies, baseline correction, normalization, standard normal variate (SNV), and multiplicative scatter correction (MSC), were systematically evaluated to determine their effects on spectral variance and classification performance.

Principal component analysis (PCA) revealed that the main discriminant spectral regions correspond to nitrocellulose and nitroglycerin bands characteristic of propellants, and nitrate-perchlorate features typical of fireworks powders, confirming that the observed separation reflects genuine chemical differences. Linear discriminant analysis (LDA) achieved a classification accuracy of 97.1%, while support vector machine (SVM) models captured additional non-linear variance in the dataset. Regression-based approaches, including principal component regression (PCR), partial least squares regression (PLS-R), and support vector regression (SVR), indicated that apparent misclassifications were chemically plausible and largely attributable to compositional overlap rather than analytical error.

The results demonstrate that both the selection and sequence of spectral preprocessing steps significantly influence model performance. The proposed ATR-FT-IR chemometric workflow provides a rapid, non-destructive, and interpretable screening approach for forensic laboratories and establishes a benchmark methodology for differentiating complex energetic materials.