Pantograph-catenary Arc Fault Detection Based on Lightweight and Single-stage Deep Detector

Pantograph-catenary system plays a significant role in the transmission of electrical energy from catenary to high-speed electrical multiple unit. Due to the influence of train body vibration, height change of catenary, and insufficient contact pressure, a gas gap is easily allowed to be occurred between carbon skateboard and contact wire, which further generates electrical arcs. This arc poses a serious threat to the normal operation of electrical multiple unit. However, the existing arc detection methods heavily depend on prior knowledge or expert experience for extracting arc-related features. The arc fault recognition capacity of them is limited and the false alarm rate is a little high, especially when confronted with complex and variety of running condition. As a breakthrough in computer vision, deep detector based on convolution neural network holds the potential to overcome the aforementioned issues. Thus, a customized lightweight and single-stage deep detector is proposed. It considers the arc detection as a regression problem with using a single convolution neural network, in order to meet the practical demands of high-efficiency and high accuracy in high-speed operation. The location of arc is detected directly from full images in one forward computation of network. The effectiveness of the proposed method is demonstrated on two real monitoring videos of pantograph-catenary system.