Remote LIBS based on transformer-CNN method for quantitative analysis of trace elements in steel

Laser-induced breakdown spectroscopy (LIBS) has emerged as a key technique for standoff quantitative analysis in industrial intelligent detection, enabling component measurement under hazardous conditions. However, remote LIBS applications, particularly for trace elements in materials like steel, face significant challenges due to signal attenuation and low signal-to-noise ratio at distances of several meters. This study addresses these issues by developing a remote LIBS system with a 10 m measurement distance and proposing a novel transformer-CNN hybrid method for quantitative analysis of trace elements. Using 40 standard steel samples, a dataset of 2400 spectral data points was constructed. Spectral preprocessing was optimized through wavelet transform, with Coif5 identified as the most effective function for enhancing signal characteristics. The TransCNN-LIBS model integrates transformer architectures to capture global spectral dependencies and CNN layers for local feature extraction, achieving high accuracy quantitative analysis for four trace elements in steel, which are phosphorus (P), sulfur (S), copper (Cu), and molybdenum (Mo). Comparative evaluations demonstrate superiority over conventional methods, including PLS, BPNN, and standalone CNN or transformer models. The root mean square error and mean relative error of the proposed method outperform those in the experiments. This study provides a reliable framework for accurate, non-contact elemental analysis, advancing the remote LIBS into industrial automation processes.