Rapid and accurate identification of steel alloys by femtosecond laser-ablation spark-induced breakdown spectroscopy and machine learning

This work investigates the application of femtosecond laser-ablation spark-induced breakdown spectroscopy (fsLA-SIBS) combined with machine learning algorithms for the rapid and accurate identification of steel alloys. Three algorithms, namely random forest (RF), support vector machine (SVM), and partial least squares identification analysis (PLS-DA), were compared and evaluated. The results indicate that, in 100 independent classifications, the RF model demonstrated an average accuracy of 0.9337, significantly surpassing the accuracies of the SVM model at 0.8281 and the PLS-DA model at 0.8646. In addition, in the evaluation of 5-fold crossvalidation and the prediction set, the RF model achieved a near-perfect micro-average area under curve (AUC) of 0.9996, surpassing the AUCs of the SVM model at 0.9761 and the PLS-DA model at 0.9847. The PCA results provided valuable insights into the spectral features that most significantly contributed to the classification accuracy, further confirming the RF model's robustness and effectiveness. This integrated approach offers a powerful tool for the rapid classification and accurate identification of steel alloys in industrial applications.