LIBS-Based Classification of Thermal Aging Levels in High-Voltage Wire Harnesses of New Energy Vehicles Using MAD-RF

High-voltage wiring harnesses in new energy vehicles (NEVs) are susceptible to thermal aging in high-temperature environments, whereas conventional assessment methods are difficult to deploy rapidly. This study combines laser-induced breakdown spectroscopy (LIBS) with random forest (RF) classification to assess thermal aging levels in cross-linked polyethylene (XLPE) insulation. Thirteen laboratory-aged XLPE wiring harness samples were prepared, and 100 single-shot spectra were acquired at fresh positions for each aging level. The specific methodological contribution is the use of class-wise median absolute deviation (MAD) at each wavelength as a variable-selection criterion before RF training. Three models were compared: RF, principal component analysis (PCA) combined with RF (PCA–RF), and MAD combined with RF (MAD–RF). RF achieved 100% internal hold-out accuracy for the non-aged versus 60-day comparison and 84.23% across all 13 aging levels. PCA–RF increased the multiclass accuracy to 87.31%, whereas MAD–RF reached 95.00% under the original exploratory hold-out workflow. These results indicate that wavelength-wise robust dispersion can provide a compact, discriminative representation of the present LIBS dataset.