Interpretation-enhanced chemical mapping of polymer alloys using low-loss scanning transmission electron microscopy-electron energy loss spectroscopy cluster analysis

This study proposes an advanced analytical method for evaluating the morphology and chemical states of polymer alloys using low-loss spectral imaging in scanning transmission electron microscopy. This approach involves extracting spectral features through Gaussian fitting, as established in our previous study, followed by statistical cluster analysis of 12 extracted spectral features. This technique enables the assessment of component distribution and the detection of reaction layers within polymer alloys. Furthermore, by comparing these feature plots with those of the standard samples, significant spectral shifts were observed, indicating chemical modifications such as conjugated double bond formation and crosslinking. In particular, thermoplastic polyurethane elastomer and Styrene-ethylene butylene-styrene block copolymer exhibited increases in the relative peak fractions of conjugated double bonds (NG2) and unsaturated C=C bonds (NG3), along with a red shift in the former peak position (μG2), consistent with known thermal degradation mechanisms. In contrast, low-density polyethylene showed a blue shift of μG2 with minimal change in NG2, suggesting limited conjugated bond formation. A supervised classification model using these features achieved over 99 % accuracy using support vector machine and random forest algorithms. Permutation importance analysis identified μG2, NG5 (G5: bulk plasmon), and NG2 as the most influential features for classification, underscoring the role of π–π∗ transitions and bulk plasmon features in polymer identification. This method addresses the common limitations of conventional spectral decomposition techniques concerning ambiguity in determining the number of components. This approach offers insights into the chemical state changes induced by alloying, serving as a powerful tool for nanoscale analysis.