Trigger-free LIBS using kHz and a few mJ laser in combination with random forest regression for the quantitative analysis of steel elements

Abstract

In this work, the use of a low pulse energy and high repetition rate diode pumped solid-state laser was investigated for trigger-free laser induced breakdown spectroscopy (LIBS) based quantitative analysis of commonly used steel grades. Owing to the low pulse energy of 2 mJ, the signal was accumulated over a large number of pulses to obtain a good signal quality and was acquired in a free-running mode, i.e. without any triggering. Although the trigger-free signal acquisition lacked a continuum suppression capability, the weak plasma events caused by the low pulse energy did not produce a significant continuum and hence continuum suppression was not an issue. Further, the 2 kHz repetition rate of the laser required the sample to be continuously scanned to avoid signal decay as a result of the formation of excessively deep craters. As scanning could introduce additional signal fluctuations, a combination of sample scanning velocity and pulse energy leading to the least signal fluctuations was used to acquire LIBS spectra for the steel samples. Of the total 19 samples, 15 steel samples were used to train random forest regression models for the quantitative analysis of various elements, including C, Si, Mn, Cr, Ni, Mo, Cu, and Co. The remaining 4 samples were used for testing/validating the models. It was observed that compared to the full spectral range of ∼185–545 nm, a smaller and information-rich spectral range of ∼185–310 nm produced better results. With the selected spectral range, the relative mean error (RME) ranged from 1.4% to 15.77%, with the lower values belonging to the major elements and the higher values to the minor elements. The root mean square error (RMSE) values were found to range from 0.014 wt% for C to 0.26 wt% for Ni. The R² values, which indicated the correlation between the predicted and reference concentrations, were found to be >96% for all the elements except Si and Cu. These results suggested that trigger-free LIBS based on a low-pulse-energy, high-repetition-rate laser and a single channel spectrometer covering a narrow spectral range, in combination with machine learning techniques, is suitable for the quantitative analysis of steel elements.