Study To Improve The Accuracy Of LIBS Measurement And The Application To Coal Analysis

The expense of coal makes up more than60%of the cost for a thermal power plant.Timely and fast coal on-line measurement is very important for improving energyefficiency. Traditional coal off-line measurement in laboratory cannot meet thedemand of industry due to the complicated procedures and long time. Laser-inducedbreakdown spectroscopy (LIBS), as a rising technology for chemical analysis, hasadvantages such as full-element analysis and no need for pre-treatment, so it ispotential for coal on-line measurement.Considering the lack of physical laws and linear nature in the popular chemometricmethod PLS, the paper proposed a dominant-factor model based on the understandingof the evolution of LIBS calibration curve. The model used a dominant factor tomodel self-absorption and inter-element interference, extracting the mainconcentration information. PLS with full spectra input was applied to compensate forthe residues from the fluctuation of plasma parameters. The dominant-factor modelcombined the physical laws with PLS and partially avoided overuse of the noise.Moreover, the dominant factor utilized non-linear transformation to improve theability of modeling the non-linear relationship, compensating for the lineardisadvantage of PLS.The dominant-factor model was applied to calculate Cu concentration using LIBSspectra for brass alloys. The non-linearized multivariate dominant factor model basedon PLS achieved the same value of R2as conventional PLS and largely improved theprediction accuracy. RMSEP was reduced from5.25%to1.97%, and RMSEC&Pdecreased from2.81%to1.05%. The results showed that the dominant-factor modelobtained a better quantitative result than conventional PLS. Furthermore, theconcentrations of the four major elements, C, H, N, S, in coal were determined usingLIBS and the dominant-factor model. The paper analyzed the possible reasons forlow precision of the raw spectra. It was found that segmental spectral areanormalization reduced the spectra fluctuation more effectively than whole spectra area normalization. Using LIBS spectra obtained in the atmosphere, thedominant-factor model achieved better accuracy than conventional PLS for C and Hconcentration calculation. For C concentration determination, R2was0.999, RMSEPand RMSEC&P were3.95%and2.60%, respectively, for the non-linearizedmultivariate dominant factor model. The averaged relative error for all samples wasonly2.66%. For H concentration determination, R2was0.884, RMSEP andRMSEC&P were3.95%and2.60%, respectively. However, the averaged relativeerror for all samples was8.43%, showing that more work was needed to furtherimprove the accuracy.For N and S concentrations determination, inert gas protection was used to avoidthe influence from air. Results showed argon was more favored than helium forplasma stability. For N concentration determination, the non-linearized multivariatedominant factor model improved the calibration quality and maintained the predictionaccuracy compared to conventional PLS. The averaged relative error for all sampleswas2.28%. For S concentration determination, the dominant-factor model was worsethan conventional PLS because it only considered the inter-element interference fromC, H, N, O, neglecting other elemental interference. Even for conventional PLS, theaveraged relative error for all samples was as high as39.0%, which indicated that theexperiment and data processing method had to be further optimized.It should be noted that with the development of the understanding for plasmamechanism, more appropriate non-linear transformation and more elementalcharacteristic line intensities can be extracted to improve the dominant factoraccuracy and thus the final results of the model. Therefore, the dominant-factor modelis very potential to be further improved.