Research on Pantograph Defect Classification Based on Vibration Signals

Abstract

Pantograph-based electrical current transmission systems are used in electric traction vehicles. The contact surface between the pantograph and the catenary wire experiences mechanical and thermal effects during the train’s movement. Typically, this contact surface on the pantograph is covered by a segmented carbon or copper rod, attached to an aluminum base. Railways implement organizational measures for pantograph condition monitoring, based on scheduled inspections. Constitutionally, the option to replace contact elements or individual segments of the pantograph exists if wear is detected. Many scientific publications describe ideas for pantograph visualization and automated condition monitoring. These ideas are based on analyzing mechanical vibrations generated by the pantograph, acoustic vibration signal analysis, 3D geometric data of the pantograph surface captured by laser scanning, and combinations of several methods. However, in these publications, mechanical vibration analysis is limited to signal shape and spectral analysis. The possibility of treating the vibration signal as a random process using statistical methods has not been utilized. This study describes the possibility of evaluating classified mechanical pantograph vibrations using the signal’s autocorrelation transformation. A laboratory experiment confirmed the proposed method for evaluating informative signal classification features. The proposed method can distinguish between signals generated by a defective pantograph surface and identify different types of defects.

Keywords

• train pantograph
• classification features
• Mel spectrograms
• convolutional neural networks