Classification of gamma-irradiated aluminum-epoxy nanocomposites utilizing laser-induced breakdown spectroscopy and machine learning techniques

Abstract This study investigates the degradation characteristics of gamma-irradiated aluminum-filled epoxy nanocomposites through the integration of Laser-Induced Breakdown Spectroscopy (LIBS) and supervised machine learning techniques. To simulate prolonged aging in high-voltage insulation settings, epoxy-based composites containing 0 wt% and 5 wt% aluminum nanofillers were subjected to gamma radiation. The LIBS spectra exhibited distinct alterations in the elements of gamma-aged samples, including shifts in the emission lines of carbon and oxygen. This indicates that radiation induced chain scission and oxidation. Principal Component Analysis was employed to reduce spectral dimensionality and reveal underlying trends across four unique material states: unaged and aged, both with and without nanofillers. Principal Component Analysis (PCA) successfully reduced the spectral dimensionality, with the first three principal components (capturing 85.43% of variance) clearly separating the four material states of unaged/aged, with/without nanofillers. Seven machine learning classifiers were evaluated on the PCA-transformed data using a rigorous 5-fold cross-validation protocol. Linear Discriminant Analysis (LDA) and Support Vector Machine (SVM) achieved perfect classification accuracy, while K-Nearest Neighbors (KNN) also performed exceptionally (98.3% ± 3.7%). The results underscore the profound effectiveness of combining LIBS with ML for the non-destructive evaluation of insulation degradation and provide quantitative evidence for the stabilizing role of core-shell structured Al nanofillers. This methodological pipeline demonstrates significant potential for real-time, ML-enhanced condition monitoring of polymeric insulating materials in radiation-prone environments.