Research on Quantitative Analysis Method for Aluminum Alloy Using Backpropagation Artificial Neural Network Algorithm Based on LIBS Technology

Rapid online analysis of the elemental distributions in aluminum alloys is critical for classification and quality control. This study focused on the 5052 Al-Mg alloy, utilizing laser-induced breakdown spectroscopy (LIBS) combined with a backpropagation artificial neural network (BP-ANN) to develop a quantitative model for analyzing elemental contents. Spectral data were collected from five 5052 Al-Mg alloy samples, with 188 spectra from each sample, resulting in 940 data sets. These were divided into training and test sets, with 700 data sets used for training and 225 for testing. The model predicted concentrations of Al (396.15 nm) and Mg (279.54 nm) had coefficients of determination of 0.9862 and 0.9646, root-mean-square errors of 0.6609 and 0.75005, mean absolute errors of 1.0986 and 0.5504, and mean bias errors of 0.3565 and 0.0231, respectively. These results demonstrate that the BP-ANN model, in combination with LIBS technology, provides an accurate and stable quantitative analysis method for aluminum alloys.