1.School of Science, Changchun University of Science and Technology, Changchun 130022, China 2.School of Chemical and Environmental Engineering, Changchun University of Science and Technology, Changchun 130022, China 3.School of Physics and Electronic Sciences, Shandong Normal University, Jinan 250358, China
Fund Project:Project supported by the National Natural Science Foundation of China (Grant No. 61575030) and the Natural Science Foundation of Jilin Province, China (Grant Nos. 20180101283JC, 20200301042RQ)
Received Date:25 April 2021
Accepted Date:18 June 2021
Available Online:05 October 2021
Published Online:20 October 2021
Abstract:The concentration of Fe in aluminum alloy can affect the plasticity, heat resistance, strength and stress corrosion resistance of the alloy. The quantitative analysis of aluminum alloy composition is an important part of the online detection of alloy composition. To improve the quantitative analysis accuracy of Fe in aluminum alloy, the spatial confinement nanosecond laser-induced breakdown spectroscopy is combined with the gradient-descent method. By collecting laser-induced aluminum alloy plasma emission spectra, it is found that the plasma radiation intensity under the confinement of the plate space is significantly enhanced. The enhancement factor of the plasma emission spectrum with a plate spacing of 10 mm is 2.3. The internal standard method and the gradient descent method are used to establish the calibration models respectively, and the values of fitting coefficient (R2), root mean square error (RMSE) and average relative error (ARE) of the two models are compared. Without plate spatial confinement, the R2, RMSEC, RMSEP and ARE of the Fe element calculated by the internal standard method are 90.66%, 0.1903%, 0.1910% and 9.2220%, respectively. The R2, RMSEC, RMSEP and ARE of Fe element obtained by the gradient descent method are 97.12%, 0.1467% (weight concentration), 0.1124% (weight concentration) and 7.1373%, respectively. With the plate spatial confinement, the R2, RMSEC, RMSEP and ARE of Fe element calculated by the internal standard method are 95.22%, 0.1409% (weight concentration), 0.1401% (weight concentration), and 6.8893%, respectively. The R2, RMSEC, RMSEP and ARE of Fe element obtained by the gradient descent method are 99.22%, 0.0731% (weight concentration), 0.0756% (weight concentration) and 3.5521%, respectively. Comparing with the internal calibration model, the accuracy and stability of the gradient descent calibration model are improved. The spatial confinement LIBS combined with the gradient descent method can effectively reduce the influence of the alloy matrix effect and the self-absorption effect on the quantitative analysis. Keywords:laser-induced breakdown spectroscopy/ gradient descent/ spatial confinement/ aluminum alloy
优化平板空间约束光谱增强效果最好实验条件(激光能量80 mJ、板间距10 mm、光谱探测延时7 μs)下的铝合金LIBS光谱如图3所示, 选取247—277 nm范围内Fe谱线较丰富的LIBS光谱波段作为铝合金中的Fe成分分析对象. 在平板空间约束条件下, 纳秒激光烧蚀诱导铝合金产生的等离子体发射光谱强度存在明显增强, 且谱线Fe I 274.9 nm增强了2.3倍. 这是由于纳秒激光烧蚀铝合金诱导产生的等离子体快速向外膨胀, 同时伴随产生冲击波向外快速膨胀. 当向外膨胀的冲击波遇到平板空间约束障碍物时, 反弹膨胀传输的冲击波压缩等离子体羽, 使等离子体羽体积变小, 增加等离子羽体内的粒子间碰撞几率, 使低激发态原子经碰撞激发跃迁至高激发态的原子数增加, 从而实现了等离子体发射光谱的增强[25-27]. 等离子体光谱稳定性是评价LIBS实验重复性的重要参数, 在优化的平板空间约束实验条件下, LIBS谱线中Fe元素 Fe I 274.9 nm特征光谱及270 nm处的背景光谱的相对标准偏差(relative standard deviation, RSD)值分别为3.48%和2.56%, 而无空间约束下对应的光谱RSD值分别为10.13%和7.32%. 结果表明, 空间约束不仅能够增强LIBS光谱强度, 同时还能降低LIBS光谱的RSD, 提高LIBS光谱稳定性, 从而提高了LIBS实验重复性. 图 3 空间约束纳秒激光诱导击穿光谱 Figure3. Nanosecond laser-induced breakdown spectroscopy with and without parallel plates confinement.
24.2.定量分析 -->
4.2.定量分析
内标法是LIBS分析中应用较广泛的定量分析方法, 通过内标元素的元素峰高或归一化光谱强度来补偿LIBS检测中的基体效应[28]. 选择Al II 288.2 nm作为内标参考线, 以Fe元素的Fe I 274.9 nm谱线作为定量分析谱线. 在相同的实验条件下, 将每个浓度的铝合金样品采集5组LIBS光谱数据, 其中3组数据作为训练集, 2组作为预测集, 得到有/无空间约束条件下铝合金中Fe元素的LIBS定量分析结果如图4所示. 可以看到, 无空间约束条件下铝合金中Fe元素LIBS定量分析参数R2, 训练集均方根误差(root mean square error calibration, RMSEC), 测试集均方根误差 (root mean square error prediction, RMSEP)和ARE值分别为90.66%, 0.1903% (质量分数), 0.1910% (质量分数)和9.2220%, 板间距为10 mm的平板空间约束条件下的铝合金中Fe元素LIBS定量分析参数R2, RMSEC, RMSEP和ARE值分别为95.22%, 0.1409% (质量分数), 0.1401% (质量分数)和6.8893%. 结果表明, 空间约束提高了铝合金中Fe元素LIBS定量分析精度和准确度. 但是, LIBS内标法获得的Fe元素的标准浓度与预测浓度的共线性不高, 定标曲线的R2值有待提高. 图 4 Fe元素的LIBS内标法定标曲线, 图中的含量均为质量分数 Figure4. Calibration curve of internal standard method, concentrations in the figure are all the weight concentration.
选取分析谱线Fe I 274.9 nm, Fe I 275.6 nm, Fe I 273.9 nm和Fe I 259.9 nm作为输入变量建立Fe的梯度下降定标模型. 设定初始参数θ0—θ4为0, 设定学习率取值范围为0—10 (学习率的常用变化范围不超过10), 最大迭代次数10000次. 无空间约束时, 通过运算得到在学习率α = 0.1时, Fe的定标模型代价函数局部最优, 此时迭代次数为2000次. 当α > 0.1时, 代价函数呈发散状, 无法进行收敛. 有空间约束时, 通过运算得到在学习率α = 0.3时, Fe的定标模型代价函数局部最优, 此时迭代次数为1000次. 当α > 0.3时, 代价函数呈发散状, 无法进行收敛. 收敛曲线如图5所示. 图 5 代价函数随迭代次数的变化图 (a) 无约束; (b) 约束 Figure5. Graph of the change of the cost function with the number of iterations: (a) Without spatial confinement; (b) with spatial confinement.
对应的有/无平板空间约束条件下结合梯度下降法的铝合金中Fe元素的LIBS定标曲线如图6所示. 可以看出, 无空间约束条件下, 采用梯度下降法获得的Fe元素的定量分析参数R2, RMSEC, RMSEP和ARE分别为97.12%, 0.1467% (质量分数), 0.1124% (质量分数)和7.1373%, 而在板间距为10 mm的平板空间约束条件下, 采用梯度下降法获得的Fe元素的定量分析参数R2, RMSEC, RMSEP和ARE分别为99.22%, 0.0731% (质量分数), 0.0756% (质量分数)和3.5521%. 图 6 Fe元素梯度下降法LIBS定标曲线, 图中的含量为质量分数 Figure6. Calibration curve of gradient descent, concentrations in the figure are all the weight concentration.