Research On Quantitative Analysis Method Based On Laser-induced Breakdown Spectroscopy Technology

Laser-induced breakdown spectroscopy(LIBS)is a new spectral analysis technology,which can realize fast,in situ,on-line and remote detection,and has outstanding application value in these aspects,in recent years by the majority of researchers keen attention.In Laser-induced breakdown spectroscopy,however,the Quantitative analysis of the spectral element content is the most important,and the traditional Quantitative analysis method is based on a linear relationship between spectral intensity and element content,this results in lower accuracy of the test results.In order to improve the Quantitative analysis effect of LIBS technology,the Laser-induced breakdown spectroscopy Quantitative analysis method is studied in this paper based on the application background of the determination of elements in aluminum alloy.The main research contents of this paper are as follows:(1)This paper introduces the Laser-induced breakdown spectroscopy system for Aluminum Alloy analysis,describes several important components of the system and the main experimental devices,and analyzes its principle and purpose of use.Display the type and element contents of the standard sample of aluminum alloy used in the experiment.This paper introduces the principle and calculation process of the traditional Laser-induced breakdown spectroscopy Quantitative analysis method,and analyzes its limitation.The introduction of machine learning applications in Laser-induced breakdown spectroscopy related algorithms,BP neural network,RBF neural network and BP-boosting Algorithm,describe the metrics used to measure the performance of Quantitative analysis using ML algorithms and explain their meaning and calculation methods.(2)in Laser-induced breakdown spectroscopy based ML Quantitative analysis,there are a number of experimental and environmental factors that partially interfere with the Quantitative analysis results.This paper optimizes these parameters before exploring the performance of the Quantitative analysis model to find the optimal parameters for this Quantitative analysis,and to find the optimal laser pulse energy,delay time and laser focusing position,respectively,at the same time,the optimal number of laser pulse superposition is determined,and based on these,the spectral lines of the Laser-induced breakdown spectroscopy elements are obtained.After the spectral lines are obtained,the spectral lines are pre-processed.Firstly,the intensity of the whole spectrum is normalized,the contrast between the spectral lines is reduced,the stability of the detected signal is improved,and the deviation caused by the fluctuation of the parameters is weakened.Then the normalized spectral lines are denoised and the background is corrected,and the spectral data are denoised using wavelet transform.The background of the Laser-induced breakdown spectroscopy is mainly composed of the ambient blank background and the spectral continuous emission background,in the process of background correction,the filtered data is subtracted from the fitted background signal points to get the corrected spectrum.After processing the original spectral data,it can be used in the modeling of machine learning method.(3)select the characteristic of the pre-processed spectral line data,and use select kbest Algorithm to reduce the dimension of the complex spectrum,which is the input of the machine learning algorithm.The BP neural network,RBF neural network and BP-boosting Algorithm are established respectively,and the best algorithm model is found by comparing the performance index,finally,the most suitable machine learning method for Laser-induced breakdown spectroscopy Quantitative analysis is determined.