|Alternative Title||Research on Aluminum Alloy Grade Identification Method Based on Fiber-LIBS|
|Keyword||激光诱导击穿光谱 牌号鉴定 光纤激光器 支持向量机 偏最小二乘|
|Place of Conferral||沈阳|
针对目前传统的牌号检测技术检测周期长、元素分析种类受限、有辐射源等缺陷和局限性，本文采用激光诱导击穿光谱(Laser-induced breakdown spectroscopy，LIBS）技术作为铝合金牌号的检测技术。激光诱导击穿光谱技术是近年来新兴的一种原子发射光谱分析技术，具有分析速度快、无需样品制备、易开发便携式设备以及可以分析几乎所有元素等优点，因此，该技术在越来越多的领域内展现出突出的应用价值。论文的主要研究内容如下：（1）基于光纤激光器的LIBS实验系统设计：针对实验室中使用的传统LIBS系统的缺点以及工业现场面向便携式、在线分析的要求，设计了基于光纤激光器（光纤激光器具有体积小、成本低、功耗低等优点）的LIBS实验系统。另外，针对光纤激光器频率高、单脉冲能量低等特点设计了实验系统的控制时序和实验方案。在考虑到光谱强度、光谱信背比以及光谱稳定性的基础上对激光器功率、能量、脉宽、频率等参数进行合理设置，成功收集到了铝合金样品的特征光谱。（2）基于LIBS光谱分类的牌号鉴定方法研究：采用支持向量机算法对铝合金光谱数据进行多分类建模，用铝合金牌号作为模型的类别标签，然后用分类模型预测待测样品牌号，实现铝合金按牌号分类。通过对原始光谱数据进行筛选、归一化等方式逐步提高模型预测精度。对于光谱数据维度过高，分类模型训练时间较长的问题，采用主成分分析算法对光谱数据进行降维，结果表明，采用筛选+归一化+PCA数据集作为SVM的输入，可以使模型平均预测准确率达到99.83%，平均建模时间缩短至0.14s，实现了铝合金样品牌号的快速准确鉴定。（3）基于元素浓度定量分析的牌号鉴定方法研究：针对基于LIBS光谱分类的牌号鉴定方法普适性差、建模需要大量样本等缺点，研究了基于元素浓度定量分析的牌号鉴定方法。在定量分析方面，介绍了LIBS单变量分析方法和基于偏最小二乘法的多变量分析方法，说明了单变量分析方法的缺点，建立了偏最小二乘分析模型，实现了对铝合金的定量分析。然后在牌号匹配方面，介绍了常用的基于模糊隶属度函数的牌号匹配算法和基于最小卡方值的牌号匹配算法，分析了其缺陷。最后提出了基于虚拟样本生成的牌号匹配方法，并通过逐步优化模型，实现了对单张光谱牌号鉴定86.7%的准确率，对平均光谱牌号鉴定96.94%的准确率。表明了基于定量分析的牌号鉴定方法的可行性。
In view of the defects and limitations of the traditional testing technology, such as long detection cycle, limited element analysis and radiation sources, this thesis uses laser induced breakdown spectroscopy as the testing technology for the grade of aluminum alloy. Laser induced breakdown spectroscopy (laser-induced breakdown spectroscopy) is a new technology of atomic emission spectroscopy in recent years. It has the advantages of fast analysis, no sample preparation, easy development of portable equipment and the analysis of almost all elements. Therefore, this technology has shown outstanding application value in more and more fields. The main contents of this thesis are as follows: (1) Design of LIBS experimental system based on fiber laser. In view of the shortcomings of the traditional LIBS system used in the laboratory and the requirements of the portable and on-line analysis of the industrial site, a LIBS experimental system based on fiber lasers (fiber lasers with small size, low cost, low power consumption, etc.) was designed. In addition, the control timing and experimental scheme of the experimental system are designed for the high frequency of the fiber laser and low single pulse energy. Spectral intensity, spectral signal-to-background ratio and spectral stability were taken into account to set the parameters such as laser power, energy, pulse width and frequency, and the characteristic spectra of aluminum alloy samples were successfully collected. (2) Research on brand identification method based on LIBS spectral classification. A multi-classification model of aluminum alloy spectral data was modeled using a support vector machine (SVM) algorithm. Aluminum alloy grades were used as category labels for the model, and then the classification model was used to predict the grades of the samples to be tested. Finally, aluminum alloys were classified by grade. Through the screening and normalization of the original spectral data, the prediction accuracy of the model is gradually improved. For the problem that the spectral data is too high dimensional and the training time of the classification model is long, the principal component analysis algorithm is used to reduce the dimension of the spectral data. The results show that the use of screening + normalization + PCA data set as the input of SVM can make the average prediction accuracy of the model reach to 99.83%, the average modeling time is shortened to 0.14s, and the aluminum alloy sample grades can be quickly and accurately identified. (3) Research on brand identification method based on quantitative analysis. Due to the poor applicability and the large number of samples required for modeling of the grades identification method based on LIBS spectral classification, the grades identification method based on elemental concentration analysis was studied.In terms of quantitative analysis, LIBS univariate analysis method and multivariate analysis method based on partial least squares method are introduced. The disadvantages of univariate analysis methods are described. The partial least squares analysis model is established and quantitative analysis of aluminum alloy is realized. . Then in the aspect of grades matching, the commonly used grades matching algorithm based on fuzzy membership function and the grades matching algorithm based on the minimum chi-square value are introduced, and the defects are analyzed. Finally, a grades matching method based on virtual sample generation was proposed. Through the gradual optimization of the model, an accuracy of 86.7% for the identification of a single spectrum and an accuracy of 96.94% for the identification of the average spectrum were achieved. It shows the feasibility of the grades identification method based on quantitative analysis.
|周中寒. 基于Fiber-LIBS的铝合金牌号鉴定方法研究[D]. 沈阳. 中国科学院沈阳自动化研究所,2018.|
|Files in This Item:|
|基于Fiber-LIBS的铝合金牌号鉴定（2556KB）||学位论文||开放获取||CC BY-NC-SA||Application Full Text|
|Recommend this item|
|Export to Endnote|
|Similar articles in Google Scholar|
|Similar articles in Baidu academic|
|Similar articles in Bing Scholar|
Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.