Analysis of particle size effect of stainless steel powder by laser-induced breakdown spectroscopy based on multivariate quantitative model

This study aims to establish a cross-particle-size quantitative LIBS analytical approach for stainless steel powder analysis. Six grades of stainless-steel powder were analyzed across four particle-size fractions (<25 μm, 25–53 μm, 53–105 μm, and 105–150 μm), generating a dataset that captures both compositional and physical variability. Local spectral-interval feature selection was coupled with multivariate regression to incorporate complete emission-line profiles into model construction. Specifically, local intervals containing the full profiles of Fe, Cr, and Ni emission lines were used as inputs, rather than full-spectrum variables or isolated peak intensities. Spectral analysis showed that larger particles reduced laser-powder coupling efficiency, decreased plasma emission intensity, and degraded inter-pulse stability, thereby causing deviations in conventional univariate internal-standard calibration. To address this effect, partial least squares regression (PLSR) and support vector regression (SVR) models were trained with mixed-particle-size data. The SVR model with a radial basis function (RBF) kernel showed the best predictive performance, with root mean square error of prediction (RMSEP) values of 0.350 wt.% for Cr and 0.284 wt.% for Ni, and coefficients of determination (R2) exceeding 0.99. Variable importance in projection (VIP) and permutation feature importance (PFI) analyses were further used to identify the spectral regions contributing to model performance. These analyses indicated that target-element emission lines made the dominant contribution, while matrix-element spectral features enhanced prediction stability. The proposed strategy provides a practical and adaptable approach for improving LIBS quantification in complex powder systems and supports metal-powder quality monitoring in additive manufacturing (AM).