基于LIBS光谱与机器学习的钢轨表面状态分类方法

高 云众, 车 长金, 王 增颜, 姚 强, 许 帅旗, 康 晓帆
北华大学

摘要


钢轨表面状态影响轮轨接触与行车安全。本文采用激光诱导击穿光谱(LIBS)结合机器学习,对正常表面、剥落掉块、擦伤、鱼鳞伤损和侧边压溃五类钢轨表面状态进行分类。实验包含10个钢轨试件、50个局部采样点(PointID),每个PointID连续采集10条单脉冲光谱,共500条光谱。经Savitzky-Golay平滑、非对称最小二乘基线校正和标准正态变换预处理后,按PointID实施10折分组交叉验证,比较PCA-LDA、Linear-SVM和随机森林(RF)的性能。PCA-LDA获得最高单条光谱分类准确率(Accuracy=Macro-F1=0.8560)。同一PointID内10条预测结果经多数投票汇总后,50个PointID中正确判别46个(Accuracy=0.9200)。波长贡献分析显示模型敏感区间主要分布在372-374 nm、410-415 nm、520-527 nm和548-551 nm附近。结果表明LIBS光谱可反映本实验样品中不同表面状态的响应差异,但模型泛化能力需通过更多独立试件进一步验证。

关键词


激光诱导击穿光谱;钢轨表面状态;机器学习;分组交叉验证;波长贡献分析

全文:

PDF


参考


[1]Song L, Song D.Rail surface defect detection based on improved YOLOv8n[J].Engineering Research Express, 2026, 8(4):045223.Song L, Song D.Rail surface defect detection based on improved YOLOv8n[J].Engineering Research Express, 2026, 8(4):045223.DOI:10.1088/2631-8695/ae45f0.

[2]Xia L Y, Yang Z F, Wei W F, et al.A rapid in-situ hardness detection method for steel rails based on LIBS and machine learning[J].Spectrochimica Acta Part B:Atomic Spectroscopy, 2024, 215:106908.

[3]Cui M C, Shi G Y, Deng L X, et al.Microstructure classification of steel samples with different heat-treatment processes based on laser-induced breakdown spectroscopy (LIBS)[J].Journal of Analytical Atomic Spectrometry, 2024, 39:1361-1374.

[4]Jeon G, Keum H, Lee H, et al.Effects of feature engineering on the robustness of laser-induced breakdown spectroscopy for industrial steel classification[J].Spectrochimica Acta Part B:Atomic Spectroscopy, 2024, 212:106857.


Refbacks

  • 当前没有refback。