激光诱导击穿光谱的高铁车轮钢晶粒度等级研究  

Grade Evaluation of Grain Size in High-Speed Railway Wheel Steel Based on Laser-Induced Breakdown Spectroscopy

作　　者：欧阳爱国[1] 余斌[1] 胡军[1] 林同征 刘燕德[1] OUYANG Ai-guo;YU Bin;HU Jun;LIN Tong-zheng;LIU Yan-de(Intelligent Electromechanical Equipment Innovation Research Institute,East China Jiaotong University,Nanchang 330013,China)

机构地区：[1]华东交通大学智能机电装备创新研究院,江西南昌330013

出　　处：《光谱学与光谱分析》2022年第11期3428-3434,共7页Spectroscopy and Spectral Analysis

基　　金：国家自然科学基金项目(31760344);水果光电检测项目技术能力提升项目(S2016-90);江西省教育厅科学技术研究项目(GJJ60516);江西省优势科技创新团队建设计划项目(20153BCB24002);南方山地果园智能化管理技术与装备协同创新中心(赣教高字[2014]60号)资助。

摘　　要：我国高速铁路运行距离长,服役环境多变,对车轮钢的性能要求较高。车轮钢的晶粒尺寸直接影响着车轮钢的力学性能,且晶粒的特征和测量对材料科学有着重要的作用,因此为了保证高速列车的安全运行,对高铁车轮的晶粒度等级进行检测是十分必要的。利用激光诱导击穿光谱(LIBS)实验平台对5个不同晶粒度等级的ER8高速列车车轮钢样品(经过不同热处理得到不同晶粒度等级)进行击穿获取光谱信息,比较了基体元素Fe和合金元素(Cr, Mo, Co)的谱线强度与5个不同晶粒度等级的样品之间的相关性,发现均与样品晶粒度等级存在不同程度的相关性。利用此关系建立以谱线强度为变量的偏最小二乘判别分析(PLS-DA)模型,在建立模型前分别采用标准正态变量变换(SNV)、多元散射校正(MSC)和Savitzky-Golay卷积平滑方法进行预处理。通过比较各种预处理方法,得出采用SNV预处理后建立的模型效果最佳,建模集误判个数为4个,准确率为95.7%,预测集误判个数为3个,准确率为90%。在SNV预处理方法的基础上,分别选择竞争性自适应重加权算法(CARS)、连续投影算法(SPA)和CARS-SPA三种波长筛选方法进行波长筛选,比较基于不同特征波长筛选的模型效果,结果表明,使用CARS进行波段筛选后建立的模型效果最佳,建模集误判个数为2个,准确率为97.9%,预测集的误判个数为1个,准确率为96.7%,模型的准确率均高于90%,可以将不同晶粒度等级的样品进行分类。综合分析以上判别分析模型结果,发现结合SNV预处理和CARS波段筛选后的PLS-DA模型的准确率最高。研究表明,采用激光诱导击穿光谱技术结合偏最小二乘判别分析高铁车轮钢晶粒度等级具有一定可行性,可将其用于评估车轮钢表面晶粒度等级,同时也为LIBS技术应用于不同晶粒度等级的高铁车轮钢研究提供了一定的基础依据。China’s high-speed railroads run long distances and have variable service environments, requiring high wheels performance. The grain size of the wheel directly affects the mechanical properties of the wheel, and the characteristics and measurement of the grain have an important role in materials science, so in order to ensure the safe operation of high-speed trains, it is necessary to test the grain size level of high-speed railway wheels. The laser-induced breakdown spectroscopy(LIBS) experimental platform was used to obtain the spectral information of five ER8 high-speed train wheel samples with different grain size grades(different grain size grades obtained after different heat treatments) by breakdown and the correlation between the intensity of the base element Fe and alloying elements(Cr, Mo, Co) and the five samples with different grain size grades was compared. Partial least squares-discriminant analysis(PLS-DA) models with the spectral line intensity as the variable were developed using this relationship, and standard normal variate transformation(SNV), multiplicative scatter correction(MSC). Savitzky-Golay convolutional smoothing methods were used to pre-process the models, respectively. The models were preprocessed by Standard normal variate transformation(SNV), Multiplicative scatter correction(MSC), Savitzky-Golay convolutional smoothing methods. By comparing various preprocessing methods, it is concluded that the effect of the model established after SNV preprocessing is the best. The number of misjudgments in the modeling set is 4, and the accuracy rate is 95.7%. The number of misjudgments in the prediction set is 3, and the accuracy rate is 90%.Based on the SNV preprocessing method, three-wavelength screening methods, competitive adaptive reweighted sampling(CARS), continuous projections algorithm(SPA), and CARS-SPA are selected for wavelength screening, and the model effects based on different characteristic wavelengths are compared. The results show that the model established after band scree

关 键 词：LIBS 车轮钢晶粒度等级 谱线强度 PLS-DA 

分 类 号：O657.3[理学—分析化学]