Study On The Quantitative Analysis With Spectral Lines Of Multi Element In DP-LIBS Based On Neural Network

Laser induced breakdown spectroscopy(LIBS)is a kind of detection technology to determine the composition and content of material elements by analyzing the plasma emission spectrum.It has great potential in metallurgy,geology,environment,scientific research and other fields.However,the accuracy and stability of LIBS analysis method still need to be further improved.In order to promote the practical application of LIBS technology,the double-pulse laser induced breakdown spectroscopy(DP-LIBS)is used to analyze the alloy steel samples.The process of spectral pretreatment and automatic identification of spectral lines are explored,and the importance is to research the method of quantitative analysis trace elements in alloy steel.The main research contents are as follows:(1)The commonly used quantitative analysis methods were summarized.(2)In order to reduce the noise and complexity in LIBS spectral data of alloy steel.Firstly,the background baseline was removed based on the local minimum,and an improved wavelet threshold function was designed to denoise the spectral data.Then,the second derivation combined with threshold setting was used to find the characteristic spectrum peaks.According to the principle of spectrum generation combined with the database of NIST,the automatic identification of the characteristic spectrum was realized through the program,which convenient for further research.(3)In order to analyze the non-metal C element,the input multi-element spectral data were reduced by genetic algorithm(GA),and based on artificial neural network(ANN),a multi-element correction was designed called GA-BP-ANN.Compared with the internal standard method and the univariate BP method,the average relative error of the prediction results was reduced from 14.78%and 14.75%to 8.29%,and the R~2 between the predicted value and the true value was increased from 0.9674 and 0.9744 to 0.9893,respectively.(4)In order to analyze the metal Mn element,used the three characteristic spectral lines within the broadening as input,designed a Principal Component Analysis-Generalized Regression Neural Network(PCA-GRNN)analysis method.Compared with the multiple linear regression method,the average relative error of the prediction results was reduced from 12.68%to 11.50%,and the R~2 between the predicted value and the true value was increased from 0.9870 to 0.9931.In summary,this article studies the detection of alloy steel samples with DP-LIBS technology from the aspects of spectral data preprocessing,automatic identification of characteristic spectral lines and quantitative analysis of trace elements.The wavelet threshold function is improved to denoise the spectrum data,the signal-to-noise ratio of the spectrum is improved,the automatic identification of characteristic spectral lines is realized through the program,and analysis spectral line selection is improved effectively.The quantitative analysis methods of metal element Mn and non-metallic element C have been improved respectively,which reduce the influence of matrix effect and improve the accuracy and stability of the analysis results.