Study On Precise Quantification Analysis Methods Of Elements In Complex Matrix Samples Using LIBS Technology

As a promising analytical technique,laser-induced breakdown spectroscopy(LIBS)has rapidly progressed in recent years.Compared with other analytical and detection technologies,LIBS has many inherent advantages,such as simple operation,no complex sample preprocessing,simultaneous detection of multiple elements,fast analysis speed,and remote analysis,thus it has been applied in many fields,including industry,agriculture,medicine,public safety,and space exploration.However,the complex matrix of natural materials such as soils and rocks affect the spectral response and deviates it from ideal rules,which limits the accuracy of LIBS analysis results and hinders the further development and application of LIBS technology and instruments.Due to the complex and unclear mechanism of matrix effects on spectral response,it is very challenging to use these spectra affected by matrix effects for accurate quantitative analysis.Based on the previous achievements of our research team in the development of LIBS instruments and supported by the Sichuan Province Key Research and Development Program(No.2022YFG0235),this paper focused on the precise quantitative analysis of multiple elements in rocks and lithology identification using the LIBS technique combined with chemometrics.The main research contents and innovations are as follows:1.The specialized analysis and testing software for the desktop LIBS instrument was designed and developed.Based on the results of the requirements analysis,the software’s overall framework structure was designed using the ModelView-Controller(MVC)framework,which has the advantages of logical separation,hierarchical clarity,and easy-to-expand.The development of the software’s instrument control module,data acquisition module,data storage module,exception handling module,and other modules was completed,achieving multiple objectives such as instrument initialization,spectral data acquisition,data storage,and so on.The testing results show that the software can meet the needs of the LIBS instrument’s use.2.A partial area normalization preprocessing method and a peak intensity calibration method based on principal component regression(PCR)were proposed,and a quantitative analysis model of Si,Al,Ca,Mg,and K elements was established.This model was integrated into the LIBS instrument,which could effectively detect the target elements in drilling debris samples.The determination coefficients between the detection results of LIBS and the X-ray fluorescence(XRF)instrument’s results were 0.9719,0.9604,0.9479,0.9652,0.9245,respectively.Consistent with the overall trend of the results of the two detection methods,effective lithology recognition and stratigraphic changes analysis could be achieved based on the LIBS technology’s detection results,demonstrating the instrument’s application potential in rock debris logging field.3.A dual-task analysis model based on two-dimensional Convolutional Neural Networks(CNN)was proposed to achieve effective rock identification and quantitative analysis of seven elements including Si,Al,Ca,etc.This study constructed a two-dimensional multi-output CNN model for rock recognition and multi-element quantitative analysis in parallel.The model’s accuracy of rock recognition reached 99.26%,surpassing k NN,SVM,and PLS-DA models.The determination coefficients between the predicted values and the standard values of Si,Al,Ca,and Mg elements were all higher than 0.99,mean absolute error,and relative error were both better than BP-ANN and PLSR models.4.An innovative multi-channel,miniaturized,and low-cost hardware device for LIBS was designed and produced to obtain the dual-dimensional feature information of plasma spectral intensity and spatial distribution.Combined with deep learning algorithms,precise rock identification and accurate multi-element quantitative analysis could be achieved.Firstly,the miniaturized light path and an optical fiber array as the receiver of plasma emission light were combined with a smartphone camera to create a smartphone spectrometer.The device replaced the commercial spectrometer used in traditional LIBS systems,and a series of plasma spectral images of rock samples were captured to explore the spatial distribution characteristics of plasma of different rock types.The CNN model for rock identification was constructed using multi-channel spectral images,and the model’s accuracy of spectral classification for the test set data reached 100%.A CNNCBAM-LSTM hybrid model was constructed by introducing the attention mechanism,and the determination coefficients between the predicted values and standard values of Ca,Mg,Na,and Ba elements on the test set were 0.9953,0.9820,0.9881,and 0.9712,respectively.Compared to other models,RMSE values were significantly reduced,which were 1.101,0.472,0.186,and 0.181,respectively.Lastly,the visualization of the quantitative analysis features used in the hybrid model was performed,intuitively demonstrating the relationship between the element spectral features and their spatial distribution characteristics on the image and the quantitative results,verifying the chemical significance of the deep learning model constructed based on the image processing method.The analysis methods of LIBS technology combined with machine learning strategies studied in this paper have reference value for the accurate analysis of complex matrix samples using LIBS technology and instruments.