Enhanced classification of aluminum alloys via time-frequency dual-doma1'in acoustic feature fusion with laser-induced breakdown spectroscopy.

Accurate classification of aluminum alloys with minor compositional differences remains a challenge for laser-induced breakdown spectroscopy (LIBS) due to spectral similarity and matrix effects. Herein, we propose a time-frequency dual-domain enhanced acoustic processing (TFDEAP) model to amplify the subtle physical differences in laser-induced plasma acoustic waves. By fusing TFDEAP-enhanced acoustic features with LIBS spectra, we establish a multimodal classification framework. This method significantly enhances the classification performance of five challenging aluminium alloys, with substantial improvements in classification accuracy. The most notable enhancement was observed with SVM, where accuracy rose from 52.00 % (LIBS-only) to 84.67 %. RF achieved the optimal AUC, increasing from 0.771 to 0.977. Although PLS-DA had already achieved a baseline classification accuracy of 67.33 %, significant improvement was attained following optimization by this method, ultimately yielding a classification accuracy of 82.67 %. The TFDEAP model effectively mitigates spectral confusion by extracting complementary physical information from acoustic signals, providing a robust foundation for high-precision, non-destructive material identification in recycling and manufacturing.