LIBS Info: Element Analysis

Title Authors Material Detector Spectrometer Software
Machine Learning Allows Calibration Models to Predict Trace Element Concentration in Soils with Generalized LIBS Spectra Jin Yu, Nicole Delepine-Gilon, Zengqi Yue, Yuqing Zhang, Hua Li, Tianlong Zhang, Yishuai Niu, Liang Gao, Ye Tian, Chen Sun Soil ICCD Mechelle 5000 SciKit-Learn
Laser: Nd:YAG
Gate Delay: 1.000us
Gate Width: 2.000us
In this paper, the authors use machine learning to improve results for analysis of Ag in Soils - the base material are NIST reference soils, which are then spiked with Ag standard reference solutions before compressing into pellets for analysis. The paper undertakes univariate analysis using the Ag I 328.1nm peak, forming models specific to samples based on a Reference soil and a generalised models across all soil types. The univariate results are compared with output of a Back-propogation Neural Network (BPNN). Input data to the BPNN consists of 150 pixels most correlated to the analyte concentrations. The fully trained BPNN shows a significant improvement in the LOD for the samples.
Element Detection Limit (ppm) Wavelength (nm) Other Wavelengths (nm) Calibration Method Calibration Range (ppm) Notes
Ag 4.9600 (None) -10.0000 N/A BPNN 20.0000-840.0000 Generalised Back Propagation Neural Network model for generalised soil.
Ag 1.4200 (None) -10.0000 N/A BPNN 20.0000-840.0000 Soil specific Back Propagation Neural Network analysis of soil pellets
Ag 23.8300 (Calibration Curve Slope) 328.1000 N/A Univariate 20.0000-840.0000 Generalised Univariate calibration over 4 different soil types.
Ag 18.4700 (Calibration Curve Slope) 328.1000 N/A Univariate 20.0000-840.0000 Soil type specific calibration using NIST Soil as background, spiked with Ag solutions.

Element RMSE (ppm) Wavelength (nm) Calibration Method Notes