双脉冲激光诱导光谱结合多变量GA-BP-ANN检测合金钢中C元素

在合金钢众多成分中碳(C)属于微量非金属元素，其含量决定了合金钢的主要力学性能，准确、实时掌握C元素的含量，对合金钢的生产及分类起到关键作用。双脉冲激光诱导击穿光谱(DP-LIBS)是一种可用于在线快速分析合金钢中元素的有效手段，不仅具有实时、样品预处理简单等优点，还能够增强物质的烧蚀度和信号强度，从而提高LIBS技术的检测灵敏度。为了减小基体效应影响，进一步提高LIBS技术对合金钢中微量C元素定量分析的精确性，采用多元素多谱线的修正方法，通过DP-LIBS结合反向传播人工神经网络(BP-ANN)，建立多变量GA-BP-ANN定标法。首先在氩气环境对合金钢样品进行DP-LIBS采集，目标C元素选择了谱线强度变化能够体现其含量变化的C 193.09 nm处的原子谱线，同时选取共存元素Fe，Cr，Mn和Si对应的特征谱线，以提供更多的光谱信息，提高C元素定量分析的准确度，共选择15条特征分析谱线，其中Fe元素含量丰富且相对稳定，作为内标元素引入以减小谱线波动；之后通过遗传算法(GA)寻优，对C/Fe，Cr/Fe，Mn/Fe和Si/Fe的谱线强度比进行优化选择；最后将GA选择的多谱线强度比作为BP-ANN网络的输入，输出为目标C元素浓度值，建立多变量GA-BP-ANN定标方法。为比较该方法预测结果的精确性，同时建立传统定标曲线法与以C/Fe为输入的单变量BP-ANN定标方法。利用标准合金钢样品，通过留一法交叉预测C元素含量值，与内标法和单变量BP-ANN定标方法相比，预测样品的平均相对误差分别由14.78%和14.75%减小到8.29%，预测值与真实值之间的决定系数R2分别由0.967 4和0.974 4提升至0.989 3。结果说明了多变量GA-BP-ANN定标法预测的C元素含量更接近于真实含量，证明了该方法用于LIBS检测合金钢中C元素含量的可行性。

Detection of C Element in Alloy Steel by Double Pulse Laser Induced Breakdown Spectroscopy With a Multivariable GA-BP-ANN

Carbon(C)is a trace nonmetallic element in many components of alloy steel.Its content determines the main mechanical properties,grasp the content of element C accurately and timely plays a vital step in the process and sort of alloy steel.Double Pulse Laser-Induced Breakdown Spectroscopy(DP-LIBS)is an effective method for on-line rapid analysis of elements in alloy steel.It not only can deal sample real-time and simple sample pretreatment but also can enhance the intensity of signal and extent of ablation.In order to diminish the matrix effect and raise the precision in quantitative detection of trace element C in LIBS,a modified method of multi-element and multi spectral lines was used with an artificial neural network of back-propagation(BP-ANN).Thus a way of multivariable GA-BP-ANN was created.Firstly,the spectrum of alloy steel samples collected by DP-LIBS in an argon atmosphere,the atomic spectrum at C 193.09 nm was selected as the spectral analysis line of element C.Its intensity can correlate with the content of element C.In order to offer more spectral information and raise the accuracy of quantitative analysis,fifteen characteristic analysis spectral lines of the coexisting elements Fe,Cr,Mn and Si were selected simultaneously.The content of element Fe is rich and relatively stable in samples,which can be used as a standard internal element to reduce the fluctuation of spectral lines;then through the genetic algorithm(GA)searched,the ratios of C/Fe,Cr/Fe,Mn/Fe and Si/Fe were optimized;finally,the input of the three-layer BP-ANN was the intensity ratios of the multi-spectral line pairs selected by GA,and the output was the content of element C,the multivariable GA-BP-ANN calibration was established.In order to contrast the results of predicted,the traditional calibration curve and the univariate BP-ANN calibration methods with C/Fe as input were established.Predicted element C content in alloy steel with leave one sample,compared with the conventional calibration curve and the univariate BP-ANN methods,the average relative error of the predicted samples decreased from 14.78%and 14.75%to 8.29%,the coefficient of fitting determination between the predicted content and the certified content of element C increased from 0.9674 and 0.9744 to 0.9893,respectively.The results showed that the predicted content of element C by the multivariable GA-BP-ANN calibration method was closer to the real content,which proved the feasibility of this method for the LIBS quantitative analysis of element C in alloy steel.