Optimization of filament-induced breakdown spectroscopy of metal-containing water with deep reinforcement learning

Abstract

Rapid and real-time monitoring of the concentrations of metal elements in water is essential for water quality evaluation and freshwater production through water desalination. Here we show the ability of the deep reinforcement learning (DRL) in assisting the filament-induced breakdown spectroscopy (FIBS) technique for high-sensitivity and standoff detection of trace-level metal elements in water. The DRL agent is trained to determine two important intricately-coupled parameters, the pulse duration and the distance between the filament starting point and the water surface, achieving the optimal control of the FIBS intensity at the air–water interface. The limits of detection of DRL-assisted FIBS for Al, Cu and Pb elements in water reach to 230, 850 and 1120 ppb, respectively. With this method, we further perform high-sensitivity analysis of the diffusion properties of multi-salt species during the freezing desalination, and find that the captured possibility of metal ions into the ice body decreases with the increasing freezing time, which exhibits a strong dependence on the metal species. This work opens up possibilities in controlling the nonlinear optical emissions by the high-intensity filament excitation assisted by the cutting-edge artificial intelligence technologies.