激光诱导击穿光谱定量分析锂矿石中锂元素

锂元素具有优良的物理和化学性能，因而在军事、电池、特种合金、受控热核反应等领域具有重要作用。现有的锂矿石分析主要是基于酸分解的原子吸收光谱、电感耦合等离子体质谱或原子发射光谱等离线方法。激光诱导击穿光谱(LIBS)是一种无需样品制备、适用于低原子序数元素(包括锂)的原子发射光谱方法。采用LIBS技术，实验采集了11种锂矿石成分分析标准物质的等离子体发射光谱，分别在610.35和670.78 nm附近观测到了Li的特征峰，但由于谱线的重叠，无法应用单变量线性回归进行建模。在全谱积分强度标准化基础上，分别使用偏最小二乘回归(PLSR)和基于主成份分析的支持向量回归(PCA+SVR)对锂矿石标准物质中的锂含量进行建模。校准模型的相关参数通过留一组交叉验证均方根误差(RMSECV)来确定。结果表明，相较于PCA+SVR校准模型，PLSR的决定系数(R2)更大，校准均方根误差(RMSEC)更小，但预测均方根误差(RMSEP)远大于RMSEC，存在过拟合现象。另一方面，PCA+SVR计算得到的RMSEP和平均相对误差(MRE)相对于PLSR更小，因此认为PCA+SVR模型拥有更好适应度。从而证明，LIBS技术可以实现锂矿石中Li含量的分析，有望应用于位于传送带上锂矿石的原位在线定量分析。

Quantitative Analysis of Li in Lithium Ores Based on Laser-Induced Breakdown Spectroscopy

Lithium has excellent physical and chemical properties, so it plays an important role in military, battery, special alloy, controlled thermonuclear reactions and other fields. The existing lithium ore analysis methods are mainly off-line methods such as atomic absorption spectrophotometry, inductively coupled plasma mass spectrometry or atomic emission spectrometry based on acid decomposition. Laser-induced breakdown spectroscopy (LIBS) is an atomic emission spectroscopy method without sample preparation and is suitable for low atomic number elements (including lithium). The plasma emission spectra of 11 lithium ore composition analysis reference materials were collected experimentally using LIBS technology. The characteristic peaks of Li were observed near 610.35 and 670.78nm respectively. However, due to the overlap of spectral lines, univariate linear regression cannot be used for modeling. Based on the standardization of full-spectrum integral intensity, partial least squares regression (PLSR) and support vector regression based on principal component analysis (PCA + SVR) are used to model the lithium content in lithium ore reference materials. The relevant parameters of the calibration model are determined by the root to mean square error of the cross-validator (RMSECV). The results show that compared with PCA + SVR calibration model, the determination coefficient (R2) of PLSR is larger, and the calibration root mean square error (RMSEC) is small, but the prediction root mean square error (RMSEP) is much larger than RMSEC, and there is an overfitting phenomenon. On the other hand, the RMSEP and mean relative error (MRE) calculated by PCA + SVR are smaller than PLSR, so we think that PCA + SVR model has good adaptability. This work proves that LIBS technology can analyse Li content in lithium ore and is expected to be applied to the in-situ online quantitative analysis of lithium ore on a conveyor belt.