A rapid in-situ hardness detection method for steel rails based on LIBS and machine learning

The railway, as a public transportation system, has played a significant role in global economic development. However, the pursuit of high speed and heavy load in trains has brought significant bending and shear stresses on the rails, making the performance of steel rails a notable focal point. The...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Spectrochimica acta. Part B: Atomic spectroscopy 2024-05, Vol.215, p.106908, Article 106908
Hauptverfasser: Xia, Langyu, Yang, Zefeng, Wei, Wenfu, Wu, Guangning
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:The railway, as a public transportation system, has played a significant role in global economic development. However, the pursuit of high speed and heavy load in trains has brought significant bending and shear stresses on the rails, making the performance of steel rails a notable focal point. The properties of steel rails are crucial factors determining the operational safety of high-speed trains, while the surface hardness is regarded as a key mechanical characteristic. This study is to develop a new rapid in-situ method to measure the hardness of steel rails, by employing Laser-Induced Breakdown Spectroscopy (LIBS) and specified analysis technology. Three distinct methods, including spectral line intensity ratios, plasma excitation temperature and machine learning, have been compared and analysed. Particularly, a multivariate model is established and optimized using the machine learning methods for U71Mn steel rail hardness analysis. In the machine learning algorithms, variance normalization is utilized, resulted in a significant improvement for the information retention during the data dimensionality reduction process. Subsequently, twelve algorithm combinations were explored, revealing that the Particle Swarm Optimization employed in Support Vector Regression (PSO-SVR) yielded the lowest mean squared error (MSE). Further refinement of the PSO-SVR was achieved through the incorporation of adaptive stochastic weights, resulting in an elevated coefficient of determination (R2) to 0.9876. Finally, the performance of the model on the new five samples was validated with an R2 of 0.9864. Otherwise, its potential applicability may be extended to broader domains, providing robust support for enhancing the precision and reliability of LIBS technology in surface hardness quantitative analysis. We proposed a new strategy based on Laser-Induced Breakdown Spectroscopy (LIBS) and machine learning, to realize a rapid in-situ hardness detection for the steel rails. [Display omitted] •This study pioneers the application of LIBS with machine learning for surface hardness analysis of steel rails.•A novel multivariate machine learning approach is introduced to address the complexity of rail matrices.•We have compared and identified the most suitable machine learning algorithm for analyzing the hardness of steel rails
ISSN:0584-8547
1873-3565
DOI:10.1016/j.sab.2024.106908