Self-absorption correction in LIBS-based lithium isotope analysis with a modified 1D U-Net

Abstract

Lithium isotopes, particularly ⁶Li, play a crucial role as tritium breeding materials in nuclear fusion research and are essential components of fusion fuel. Laser-Induced Breakdown Spectroscopy (LIBS) offers a rapid and preprocessing-free method for isotope analysis. However, strong self-absorption effects cause spectral distortion, complicating the precise determination of isotope ratios. This study proposes a deep learning-based modified 1D U-Net model to address self-absorption effects. Our model was trained using simulation data, which was validated against two types of experimental data: measured spectral data minimizing self-absorption effects and self-reversal spectrum data. This proposed model effectively corrected self-absorption effects resulting in an accurate restoring the central wavelengths of peaks critical to isotope ratio analysis. This research highlights the potential of deep learning in resolving a challenge of self-absorption for LIBS-based lithium isotope analysis, demonstrating that training solely on simulation data can achieve effective results.