A DANN-Based Qualitative Spectral Analysis Method for Varying Motion Parameters in High Pulse Repetition Frequency LIBS

The classification and recycling of alloys hold strategic significance in promoting sustainable development.Traditional LIBS technology has been widely used in factories for online analytical detection.However, most miniaturized handheld LIBS instruments, which are more suitable for factory applications, employ high pulse repetition frequency lasers.Nevertheless, due to its high sensitivity to operational parameters, this poses significant challenges to industrial production.This research tackles three critical challenges in high-speed LIBS analysis: speed-dependent intensity variations, overlapping alloy spectra, and rigid algorithmic processing.Our solution combines adaptive neural networks with spectral focusing mechanisms, tested on 21 industrial alloys (Cu, Al, Fe).Key findings show that faster scanning speeds (4 vs. 1 mm/s) intensify spectral signals through reduced laser defocusing, yet degrade traditional neural network accuracy to 78.00%.The developed dual-focus network maintains 93.95% classification accuracy at high speeds by learning speed-independent patterns and prioritizing critical emission lines, Ssuch as Mg I 285.2 nm and Fe 275.6 nm.Compared to conventional methods, this approach reduces prediction inconsistencies in high PRF LIBS spectra acquired at handheld-device movement speeds by 42.3%, while maintaining over 90% accuracy within the operational speed range of handheld devices used in practical industrial applications.This effectively addresses the challenges of identifying copper, aluminum, and iron alloys in dynamic LIBS environments.