A novel approach for the quantitative analysis of multiple elements in steel based on laser-induced breakdown spectroscopy (LIBS) and random forest regression (RFR)

Abstract

A novel method based on laser induced breakdown spectroscopy(LIBS) combined with random forest regression(RFR) was proposed to quantitative analysis of multielement of fourteen steel samples. Normalized LIBS spectrum of steel in which characteristic line(Si, Mn, Cr, Ni and Cu) identified by NIST database was used as analysis spectrum. Then, two parameters of RFR were optimized by out-of-bag (OOB) error estimation. The performance of calibration model was investigated by different input variables(the whole spectral bands(220-800nm) and spectra feature bands(220-400nm), respectively). In order to validate the predictive ability of multielement calibration model in steels, we compared RFR with partial least-squares(PLS) and support vector machines(SVM) to predict the concentrations of multielement in steels. And, the three quantitative techniques are evaluated in terms of prediction accuracy and root mean square error(RMSE).Random forest is shown to correctly model nonlinear effects dues to self-absorption in the plasma and to provide the best results. It confirms that LIBS technique coupled with RFR has a good potential for the in situ rapid determination of multielement in steels and even metallurgy field.A novel method based on laser induced breakdown spectroscopy(LIBS) and random forest regression(RFR) was proposed to quantitative analyze of multi-elements in fourteen steel samples. Normalized LIBS spectra of steel in which characteristic line(Si, Mn, Cr, Ni and Cu) identified by NIST 10 database were used as analysis spectra. Then, two parameters of RFR were optimized by out-of-bag (OOB) error estimation. The performance of calibration model was investigated by different input variables(the whole spectral bands(220-800nm) and spectra feature bands(220-400nm), respectively). In order to validate the predictive ability of multi-elements calibration RFR model in steels, we compared RFR with partial least-squares(PLS) and support vector machines(SVM) by means of prediction accuracy 15 and root mean square error(RMSE). Thus, RFR model can eliminate the influence of nonlinear factors dues to self-absorption in the plasma and provide a better predictive result. It confirms that LIBS technique coupled with RFR has a good potential for the in situ rapid determination of multi-elements in steels and even metallurgy field.

Analysis of time‐resolved laser‐induced breakdown spectra by mean field‐independent components analysis (MFICA) and multivariate curve resolution–alternating least squares (MCR‐ALS)

“…The characteristics of LIBS, which include remote and rapid analysis, no sample preparation, applicability to any type of sample, and potential for field portability, make this quasi nondestructive analytical technique a very attractive method . Laser‐induced breakdown spectroscopy has developed in many fields, such as the nuclear industry, aerosols analysis, cultural heritage, biology, polymers, and metallurgy, and the number of applications is still growing. The LIBS signal stems from highly nonlinear, coupled phenomena driving the sample laser ablation and the laser‐plasma interaction.…”

Quantitative analysis of nonmetal elements in steel using laser-induced breakdown spectroscopy combined with random forest

“…Therefore, the traditional univariate analysis fails to compensate the influence of these interference factors. Hence, multivariable analysis methods such as partial least square regression (PLSR), 24 random forests (RF), 26,27 artificial neural network (ANN) [28][29][30][31] and support vector machine (SVM) 32-34 are adopted to solve matrix effect and improve the accuracy of quantitative analysis for steel sample. It has been proved that the running speed of RF is very fast and the precision of RF algorithm is incomparable, and even can prevent over-fitting, as well as has a good tolerance for the noise.…”

A laser induced breakdown spectroscopy quantitative analysis method based on the robust least squares support vector machine regression model

“…The regression model is built or trained to describe the relationship between the elemental concentrations and the analytical spectral intensities of emission spectra collected from the plasma. The multivariate techniques of principal component regression (PCR) [18], partial least squares regression (PLS) [19], random forest regression (RFR) [20] have been used in LIBS [20] reported that RFR had a better performance potential in LIBS quantitative analysis compared with the methods of SVM and PLS when a feature spectral band was regarded as the input variables and a preprocessing of smoothing and de-noising of the spectra was made before modeling.…”

“…However, the predictive ability of RFR method highly depends on the two parameters, the number of the trees in the forest and the number of the peaks randomly selected as the candidates for splitting at each node [20]. A desired regression model not only should be robust since the measured spectral intensities may fluctuate even under the same condition, but also should have nonlinear fitting ability [21][22] because the relationship between the elemental concentrations and the analytical spectral intensities usually exhibits strong nonlinearity due to the matrix effect, the absorption effect and the influence of background radiation.…”