Fast evaluation on the fatigue level of copper contact wire based on laser induced breakdown spectroscopy and supervised machine learning for high speed railway

Abstract

High-strength copper contact wire is of great significance to the electrified railway power supply system, which constantly provides electric power to the trains during operation. However, contact wire is subject to pressure, vibration, and natural forces such as wind, rain, ice, etc. which inevitably result in mechanical fatigue over time. This mechanical fatigue can lead to a decrease in the mechanical strength of the contact wire, and ultimately lead to problems such as wire detachment, fracture, or breakage, posing a serious safety hazard to the electrified railway system. Herein, the authors propose a strategy using nanosecond pulsed laser induced breakdown spectroscopy (LIBS) combined with machine learning technique to realise a fast evaluation on the fatigue level of copper contact line. Three different fatigue levels of copper samples have been made related with the requirement of operational conditions, and a total of 898 LIBS spectra were collected. Twenty-four combinations of spectral pre-processing, feature extraction, and optimisation algorithms were used to compare the recognition results with the accuracy, recall rate, and time cost taken into accounted. Results have shown that the standard normal variable transform–principal component analysis–genetic algorithm improve support vector machine (SNV-PCA-GASVM) model have presented a most satisfactory performance than the others. The cross-validation accuracy of the SNV-PCA-GASVM model was 92.97% while the dimensionality of input variables was reduced by 99.62%. This work is useful for the safety operation of power supply system in high speed railway, and technique development concerning the fast evaluation on materials fatigue in other industrial fields.