On Quantitative Analysis Method Of Detecting The Content Of Alloy Steel Elements Based On LIBS Technology

Alloy steel is an important raw material for the national economy.In spite of its main contents being iron,alloy steel often contains silicon,manganese,chromium,nickle,copper and other elements.Whether the alloy steel contains these elements or how much it contains them,impacts on its performance and quality directly.Therefore,during the smelting of the alloy steel,the rapid detection of the concentration of some important elements in the alloy steel is of great significance to controlling metallurgical process.The traditional method of detecting the concentration of elements in alloy steel has a lot of disadvantages,such as complex sample preparation,time consuming and the off-line detecting,which increases the production time of metallurgy and leads to a lot of waste of manpower,energy and cost.Laser-induced breakdown spectroscopy(LIBS)technology is based on analyzing characteristics spectrum lines of atomic which are generated from the ablation of sample with high power laser pulse in the transient plasma emission.Hence LIBS technology has the excellent advantages which include no pretreatment,rapid on-line detecting and multi-element detecting at the same time.Therefore,LIBS technology has been gradually applied to the on-line detecting of alloy steel elements in the practical work.We can use the wavelength of the characteristic spectrum collected by LIBS instrument to qualitatively analyze which elements are involved in it,and we can use the intensity of the characteristic spectrum to quantitatively analyze how much these elements are included in it.However,the production of laser-induced plasma is very complex process.This is easily influenced by various factors,which will have an effect on the quantitative analysis results,leading to a low prediction precision.In addition,while the spectral data measured by LIBS technology has been denoised,noise still exists.Therefore,choosing a good algorithm with better stability,lower prediction error and tolerance to noise is the key to the quantitative analysis of alloy steel elements based on LIBS.This paper aims to solve this problem.The details are elaborated as follows:Firstly,this paper considers that the traditional LIBS quantitative analysis method is calibration curve method,which does not take the interference of matrix effect and absorption effect into account and often results in the low prediction accuracy of alloy steel elements and a lack of robustness.Since that partial least squares(PLS)method can reduce the influence of the matrix effect,it can be used for on-line detection of the elements in alloy steel based on LIBS technology.However,due to the wide variation range of training set samples in partial least squares model,the prediction accuracy will be reduced.Therefore,we introduce the clusteringalgorithm,we come up with a quantitative analysis model of cluster analysis-partial least square method.and constructs the partial least squares prediction model in each sub-interval.Experimental results show,compared with the traditional partial least squares model,the cluster analysis-partial least squares method has been validated feasible and has higher prediction accuracy than the model of partial least squares.Secondly,this paper investigates that,to some extent,LIBS spectrum lines exist the various self-absorption influence,which results in the non-linear relationship between the characteristic spectral intensities of various elements in alloy steel and elemental concentrations.On one hand,partial least squares has the disadvantages such as inability of the non-linear correction,over-fitting and poor robustness,and on the other hand the random forest regression can ensure that the diversity of characteristic samples set,is capable of dealing with the non-linear spectral data,having good tolerance to spectral noise and overcoming the over-fitting at the same time.Consequently,we will construct the model of LIBS technology combined with random forest regression applied to quantitative analysis of the elements in alloy steel.Through the experimental analysis,random forest regression has higher prediction accuracy,stronger generalization ability and higher tolerance to spectral noise than the method of cluster analysis-partial least square.And random forest regression is a more suitable method for quantitative analysis in alloy steel.The random forest regression method combined with LIBS technology provides a potential method for the quantitative analysis in the alloy steel elements in the metallurgical industry.