Quantitative elemental analysis of ultra-white glass via a hybrid LIBS-RBF neural network approach

This study presents a novel, high-precision method for the quantitative elemental analysis of ultra-white glass. By integrating laser-induced breakdown spectroscopy (LIBS) with a radial basis function (RBF) neural network, this study addresses critical limitations of traditional techniques related to spectral interference and matrix effects. A custom LIBS system was used to collect 940 spectral datasets from 10 ultra-white glass samples, and 925 pre-processed samples (characterized by 15 features) were used to train and evaluate the RBF model. The proposed LIBS-RBF framework achieved exceptional accuracy, with determination coefficients (R2) of 0.964 for Ca II and 0.936 for Si I. Compared to conventional methods (e.g., R2 = 0.85–0.91 for inductively coupled plasma-optical emission spectroscopy and 0.82–0.88 for x-ray fluorescence), the new approach more effectively resolves spectral overlap while reducing the analysis time from hours to seconds. Key innovations include dynamic hyperparameter optimization (spread parameters: 65 for Si I and 46 for Ca II) and the tuning of laser energy (60 mJ) and delay time (500 ns) to enhance signal-to-noise ratios. The micro-destructive nature, rapid processing (<10 s per sample), and robustness (relative standard deviation <2.5%) demonstrate the transformative potential of this method for real-time quality control in photovoltaic glass manufacturing, smart coatings, and other high-performance optical applications where precise composition control is critical.