SIA OpenIR  > 工业控制网络与系统研究室
面向便携LIBS系统的合金材料定量分析与牌号鉴别
Alternative TitleQuantitative analysis and brand identification of alloy materials for portable LIBS system
王国栋1,2
Department工业控制网络与系统研究室
Thesis Advisor孙兰香
Keyword激光诱导击穿光谱 牌号鉴别 岭回归 特征选择 模型迁移
Pages62页
Degree Discipline计算机应用技术
Degree Name硕士
2020-05-26
Degree Grantor中国科学院沈阳自动化研究所
Place of Conferral沈阳
Abstract激光诱导击穿光谱(Laser-induced breakdown spectroscopy,LIBS)是一种基于等离子体发射光谱的元素分析技术,利用LIBS技术可以实现在线、原位、快速检测。但是受限于LIBS分析系统中激光器和光谱仪等组件的体积影响,便携式LIBS系统中很难采用较大能量激光器和高精度的光谱仪,受基体效应和其他实验条件影响,其精度比传统实验室LIBS系统差。在便携式LIBS系统实验条件下,如何对合金材料的牌号进行鉴别、为合金材料元素提供准确的定量分析是便携式LIBS系统在合金材料鉴别上应用的关键。 论文主要内容如下:(1)根据牌号库元素信息生成浓度数据训练分类模型,利用定性-定量-牌号鉴别三步组合方式实现大规模牌号鉴别。在实际应用场景中便携设备需要对多种基体的合金材料进行鉴别,为满足大规模牌号样品鉴别的需求,将整个牌号鉴别分成基体分类、定量分析、牌号鉴别模块。先利用光谱信息对合金材料进行基体分类,根据基体分类结果将光谱数据输入到不同基体的定量分析模块进行浓度定量分析。利用牌号库元素浓度信息生成对应牌号数据训练牌号分类模型。最终将定量分析模型给出的浓度数据输入到牌号分类模型中进行牌号鉴别。利用上述牌号鉴别方法对铝合金中Al-Si系下的5种牌号样品4800张光谱数据进行鉴别,在牌号库30个铝合金牌号样品中的鉴别正确率达到95%以上。(2)利用岭回归模型系数结合递归特征消除实现光谱特征选择,提高定量分析模块的分析性能。光谱数据中不仅包含丰富的特征谱线信息,也包含噪声和干扰信息。为了提高定量分析性能,需要对光谱数据进行特征选择。利用岭回归(ridge)系数绝对值对光谱特征进行筛选,通过递归特征消除(RFE)方式剔除光谱特征中最小系数绝对值对应光谱特征,以筛选后的特征子集作为偏最小二乘回归(PLS)模型输入,以PLS模型拟合结果评判整个特征子集优劣。结果表明:对铝合金Fe、Si、Mg、Cu、Zn、Mn 6种元素进行浓度标定,利用特征选择之后的光谱特征作为PLS模型输入进行曲线拟合,经特征筛选之后拟合结果相比全谱输入有显著提高。(3)对传统分段直接标准方式进行优化,利用优化后的方法实现了两台设备光谱数据全谱标准化,为多台设备之间光谱一致性提供了解决方案。针对两台便携设备之间的模型转移进行了相关研究,利用分段直接标准化方法对从机光谱数据进行标准化,并分析了影响光谱标准化精度的主要原因。进一步探究了标准化中最少传递样本个数和参与拟合的光谱特征个数对标准化精度的影响,证明了少量的特征和一定的传递样本数量能够提高分段直接标准化的拟合精度。
Other AbstractLaser-induced breakdown spectroscopy is an element analysis technique based on the emission spectrum of plasma. However, due to the influence of the volume of the components such as laser and spectrometer in the LIBS analysis system, it is difficult to adopt the high-energy laser and high-precision spectrometer in the portable LIBS system. Therefore, the portable LIBS system is affected by the matrix effect and other experimental conditions, and its accuracy is worse than that of the traditional laboratory LIBS system. Under the condition of portable LIBS system, how to identify the brands of alloy materials and provide accurate quantitative analysis of alloy material elements is the key to the application of portable LIBS system in the identification of alloy materials. The main contents of the paper are as follows: (1) Based on the information of brand library, the training classification model of concentration data is generated, and the three-step combination method of qualitative, quantitative and brand identification is used to realize the large-scale brand identification. In practical application scenarios, portable devices need to identify alloy materials of a variety of substrates. In order to meet the requirements of large-scale brand sample identification, the whole brand identification is divided into matrix classification, quantitative analysis and brand identification modules. The spectral information was used to classify the matrix of the alloy material, and the spectral data were input into the quantitative analysis module of different matrix for concentration quantitative analysis. Using the element concentration information of brand library to generate the corresponding brand data to train the brand classification model. Finally, the concentration data from the quantitative analysis model was input into the brand classification model for brand identification. The method of brand identification was used to classify 4800 pieces of spectral data of 5 kinds of al-si series samples in aluminum alloy, and the classification accuracy reached more than 95% in 30 brand samples. (2) In order to improve the performance of quantitative analysis module, spectral feature selection is realized by ridge regression model coefficient combined with recursive feature elimination. Spectral data contain not only abundant information of characteristic spectral lines, but also information of noise and interference. In order to improve the performance of quantitative analysis, it is necessary to select the characteristics of spectral data, it is necessary to select the characteristics of spectral data. The absolute value of ridge coefficient was used to screen the spectral features, and the spectral features corresponding to the absolute value of the minimum coefficient were eliminated by recursive feature elimination (RFE). The filtered feature subset was used as partial least squares regression (PLS) model input, and the PLS model fitting results were used to judge the advantages and disadvantages of the whole feature subset. The results show that: the concentration of six elements of aluminum alloy Fe, Si, Mg, Cu, Zn, Mn were calibrated, and the spectral characteristics after spectrum selection were used as the input of PLS model for curve fitting. The fitting result after feature selection was significantly improved compared with the full-spectrum input. (3) This paper optimizes the traditional segmented direct standard method, and realizes the full spectrum standardization of the spectral data of two devices by using the optimized method, which provides a solution for the spectral consistency between multiple devices. In this paper, the model migration between two portable devices is studied, and the spectral data of the slave is standardized by means of direct segmentation standardization. On the premise of not affecting the accuracy of the model, the influence of the number of transferred samples and the number of spectral features of the slave machine on the accuracy of the fitting is proved that a small number of features and a certain number of transferred samples can improve the accuracy of piecework direct standardization.
Language中文
Contribution Rank1
Document Type学位论文
Identifierhttp://ir.sia.cn/handle/173321/27117
Collection工业控制网络与系统研究室
Affiliation1.中国科学院沈阳自动化研究所;
2.中国科学院大学
Recommended Citation
GB/T 7714
王国栋. 面向便携LIBS系统的合金材料定量分析与牌号鉴别[D]. 沈阳. 中国科学院沈阳自动化研究所,2020.
Files in This Item:
File Name/Size DocType Version Access License
面向便携LIBS系统的合金材料定量分析与(3494KB)学位论文 开放获取CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[王国栋]'s Articles
Baidu academic
Similar articles in Baidu academic
[王国栋]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[王国栋]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.