| The development goal of intelligent manufacturing demand higher request to the fast component detection of materials,products and intermediate products.The common detection techniques,such as spark direct reading spectrometry,inductively coupled plasma emission spectrometry,have high demand for sample preparation,complicated operation and poor real-time capability.The research and development of a rapid analysis technique for industrial scenario is important for improving metallurgic process automation level,reducing energy consumption and optimizing smelting process.Laser-induced breakdown spectroscopy(LIBS)is a new atomic emission spectrometry with the advantages of low requirement of sample preparation,fast analysis speed,capability of real-time measurement and suitability for multi-elemental analysis.LIBS is appealing for its good prospects of the in-situ,quick and remote applications.In view of the component detection request of mining and metallurgy filed and the advantages of LIBS technique,an LIBS quantitative analysis method for quick detection of multi-elemental components is proposed.Because of the complex components,less standard samples at high temperature and the nonlinear relationship between spectral intensities and elemental concentrations,a series of researches on the optimization of multi-experimental parameters and nonlinear modeling of spectral information are carried out,in order to obtain high quality spectra,improve quantitative accuracy and quicken the applications in mining and metallurgy scenario.The specific contents can be summarized as follows:(1)For the optimization of multi-experimental parameters in LIBS,an evaluation function of spectral quality based on intensities and SBRs of multi-lines and a multivariate optimization model of multi-parameters are proposed.Taking alloy steels and vanadium slags as analysis targets,an LIBS experimental system was developed and the effect of laser pulse energy,integral time,delay time and lens-sample distance on the quality of spectral signal was studied.An evaluation function named multiple line-intensity-SBR(MLIS)was proposed to judge the spectral quality of different analytical lines.Furthermore,a multivariate model was built by means of quadratic regression orthogonal design.The integral time and delay time of LIBS were optimized by the multivariate model.The experimental results showed that for alloy steels,the optimized integral time and delay time were 5.7?s and 2.4?s,respectively;for vanadium slags,the optimized integral time and delay time were 9.8?s and 2.3?s,respectively(2)To address the problem of LIBS spectral interference effect in multi-elemental measurement of vanadium slags and pellets,a method based on selective ensemble learning is proposed for LIBS quantitative analysis.Taking vanadium slags and pellets as analysis targets,different combinations of the candidate interference lines of the analytical element were added into calibration process and a series of single regression learners were obtained.Then a subset of the single learners with larger diversity and higher accuracy was picked out by K-means clustering and selective ensemble learning.The selected single learners were synthesized to construct a strong regression model,which was used to predict various oxides in vanadium slags and pellets.The proposed method does not depend on experience to select analytical lines and interference lines,and the predicted results are decided by a number of single learners.The experimental results showed that compared with the regression model built without interference lines,the mean relative errors of CaO,MgO,SiO2,TiO2 and V2O5 in three testing vanadium slags were reduced from 18.430%,12.721%,8.777%,7.475%and 12.483%to 1.403%,4.390%,4.532%,2.438%and 2.086%,respectively.The mean relative errors of CaO,MgO,SiO2 and TiO2 in three testing pellets were reduced from 14.586%,14.544%,12.972%and 12.583%to 3.728%,3.642%,2.813%and 3.768%,respectively.(3)For the problem of less number of standard samples at high temperature,a method based on the improved TrAdaboost transfer learning is proposed for high-temperature LIBS quantitative analysis.Taking alloy steels as analysis targets,the LIBS spectral characteristics of samples at room temperature were transferred to different but similar spectral information of samples at high temperature.The sample weights of room-temperature and high-temperature training samples were adjusted by different strategies.All single learners generated in the modeling process were synthesized to obtain an ensem ble regression model to predict the testing alloy steel samples at high temperature.The proposed method makes full use of transferred spectral information to reduce the effect of less standard samples on LIBS quantitative accuracy.The experimental results showed that compared with the regression model built by traditional machine learning with less standard samples,the mean relative errors of Cr,Ni,Mn and Fe in three high-temperature testing samples were reduced from 15.182%,52.678%,22.279%and 9.437%to 4.619%,5.804%,7.032%and 2.966%,respectively.(4)For the problem of difficult to obtain standard samples,a calibration-free LIBS method based on improved self-absorption correction and optimized estimation of plasma temperature is proposed.Taking alloy steels as analysis targets,the in-ternal reference line of each analytical species was automatically selected by a programmable procedure through transition probability and low energy level.Then the self-absorption effect of the selected internal reference line and analytical lines was corrected and the plasma temperature was estimated by the particle swarm optimization algorithm.Finally,a calibration-free LIBS quantitative model with good performance is obtained.The proposed method can be realized with less easily accessible parameters and further expand the application field of LIBS.The experimental results showed that compared with the traditional calibration-free LIBS quantitative method,the mean relative errors of Cr,Ni,Mn and Fe in twenty-two alloy steels were reduced from 26.756%,30.652%,26.644%and 9.764%to 4.733%,7.135%,7.207%and 3.926%,respectively.The related methods and ideas of this paper for the optimization of experimental parameters,multi-elemental interference,high-temperature modeling with less standard samples and calibration-free LIBS quantitative analysis,can be extended to other spectrometry analysis(e.g.X-ray fluorescence spectroscopy and Raman spectrometry)and have wide application prospects in the fields of electricity,chemical engineering,biology and environmental protection. |
PDF Full Text Request