镧元素激光诱导击穿光谱定量分析准确度提高的研究

Improving Accuracy of Quantitative Analysis of La in Graphite Using Laser-In⁃duced Breakdown Spectroscopy

基于激光诱导击穿光谱(LIBS)技术对石墨基底中镧(La)元素进行了定量分析和检测准确度提高的研究。受La含量过高的影响,谱线强度与含量不存在明显的线性关系,基本定标法无法构建有效的校正模型,进而分别采用内标法、偏最小二乘法(PLS)及两者相结合的方法进行了校正模型的构建,并验证了回归分析的可行性。结果表明:以CI247.85 nm线为内标线,所有分析线的积分强度与其比值均与La含量具有较好的线性关系,线性相关系数均在0.95以上,其中,LaⅢ237.93 nm分析线具有最佳线性相关性,以其校正模型获取的LIBS预测含量值与真实值的拟合曲线拟合度(R^(2))为0.9881;采用多条谱线的积分强度为自变量,以La含量为因变量,构建PLS-area模型开展多元分析,预测值与真实值的拟合曲线拟合度为0.9705;当采用多条谱线的积分强度比值作为自变量代入PLS算法,即内标法与PLS相结合(PLS-ratio)后,拟合曲线的拟合度最佳,其R^(2)为0.9983;对比3种模型的预测性能可知,PLS-ratio模型的反演预测精度最高,特别是对高含量样品3#~5#,预测相对误差均在4%以内;此外,3种模型均方根误差(RMSEC)分别为0.5407%,0.8497%和0.2029%,PLS-ratio模型的RMSEC最小,显示该方法具有最佳的预测性能。研究结果表明,采用内标法与PLS相结合的多变量线性回归模型能有效减小LIBS分析误差,提高预测能力,为开展材料中La元素的快速分析提供了可靠基础。

Lanthanum(La)is one of the light rare earth elements and has been playing a very important role in military industry,food hygiene,electronic science and technology for its special physical and chemical properties.Hence,it is necessary to perform accurate and sensitive analysis of La concentration in materials.Meanwhile,it is significant to realize real-time online detection and analysis dur⁃ing the production process.The traditional chemical analysis methods such as inductively coupled plasma-optical emission spectrometry(ICP-OES),inductively coupled plasma mass spectrometry(ICP-MS)and X-ray fluorescence spectrometry(XRF)are time-consum⁃ing,complex in sampling preparation,expensive in cost and experimental device,which are hard to keep up with the modern analytical and testing technology development.Laser-induced breakdown spectroscopy(LIBS),a form of atomic emission spectroscopy,has been widely deployed as a new elemental analysis technique based on photon emission due to the de-excitation of excited atoms and/or ions from intense laser-induced plasma(LIP)created on target.The technique has the characteristics of rapid analysis,simultaneous analy⁃sis of multiple elements,no or minus sample preparation and real-time online analysis,etc.Thus,it has a tremendous growth and been widely applied in a variety of areas,such as industrial process analysis,environmental monitoring,mineral exploitation,phar⁃maceutical preparation,agriculture and food.With the increasing acceptance of LIBS as a quantitative spectral method for element measurement,there is a need for advanced statistical data analysis methods.Conventionally,the peak intensity or peak area of the emission line of interest is calculated in a LIBS spectrum to construct calibration curves from the known concentration of a set of cali⁃bration samples,the so-called standard calibration curve method is the simplest and most widespread.However,the accurate quantita⁃tive analysis of LIBS is still a challenge due to physical-chemical matrix effects,the overlapping emission spectra and especially fluctu⁃ations of experimental parameters.Several chemometric data analysis methods such as partial least square(PLS)regression,artificial neural networks(ANN)and support vector machine(SVM)have been proposed to overcome these difficulties with the goal of enhanc⁃ing analytical performance of LIBS,these advanced methods extract valuable information effectively from a complex LIBS spectrum.Actually,LIBS is nowadays more and more combined with these new methods in order to improve its analytical performances.Com⁃pared to the standard calibration curve method that utilizes only a single emission line,PLS is a pattern recognition technique capable of analyzing a multitude of emission lines.The procedures of PLS modeling cover various techniques such as the principal component analysis(PCA),canonical correlation analysis and multiple linear regression analysis,and the choice depending on which source of variation is deemed most significant.The primary goal of the present work was to achieve rapid analysis of rare earth lanthanum(La)element in materials,PLS method was mainly performed to reduce matrix effects and to improve prediction accuracy.Several mixture samples with La concentrations that varied from 0~17.06%were prepared by the standard addition method,a nanosecond LIBS system was used to quantitatively analyze La in a highly pure graphite matrix.Due to the high concentration of La in the matrix,there was no ob⁃vious linearity between the peak area and the concentration.Therefore,the calibration models were constructed by internal standard method,PLS and the combination of the two methods respectively,and the feasibility of regression analysis was verified.The results showed that when using C I 247.85 nm line as the internal reference line,there existed a strong linear relationship between the ratio of each analysis line and La concentration,and all the linear correlation coefficients of the calibration curves were above 0.95.In view of the fact that the analysis line of La III 237.93 nm had the best linear correlation,according to the calibration model,the regression coef⁃ficients(R^(2))of the fitting curve between the predicted and the real concentration was 0.9881.Then,PLS-area model was constructed by using the peak area of several lines as the independent variable and La content as the dependent variable,the value of R^(2) of the model was 0.9705,which was slightly lower than that of the internal standard method.When the integral intensity ratio of multiple spectral lines was substituted into PLS algorithm as the independent variable,the data for PLS-ratio model showed the best linear correlation be⁃tween the predicted concentration and real concentration,and the value of R^(2) was 0.9983.By comparing the prediction performance of the three models,it could be concluded that the prediction accuracy of PLS-ratio model exhibited the highest accuracy,particularly for high content Samples 3#~5#,with the relative errors of less than 4%.Additionally,the root mean square error(RMSEC)values of the three correction models were 0.5407%,0.8497%and 0.2029%,respectively,and RMSEC value of PLS-ratio model was the minimum,indicating that the model had the best prediction performance.The results showed that the multivariable linear regression model which combined with internal standard method and PLS could effectively reduce the error of LIBS analysis and improve the prediction ability.