Bayesian optimization of the conditions for highly sensitive detection of surface contamination by laser-induced breakdown spectroscopy

Abstract

Bayesian optimization based on Gaussian process regression was applied to optimize the conditions for the highly sensitive detection of surface contamination by laser-induced breakdown spectroscopy (LIBS). Three experimental parameters for laser ablation–pulse energy, stage height, and detection height, with 2160 possible combinations were employed for simultaneous optimization. The reciprocal value of the limit of detection (LOD) was defined as a target variable. This value was obtained from ten LIBS measurements of a pair of samples: a clean substrate and a sample with adhered silicone oil with a specified surface concentration. In total, 173 experiments under 91 conditions were completed in 16 optimization rounds. Six candidate conditions for LOD evaluation were specified based on the optimization results. The LOD evaluation by LIBS employing a detection system using an intensified charge-coupled device provided the lowest LOD value of 0.084 ± 0.008 μg cm⁻².