Turnout Fault Diagnosis Based on CNNs with Self-Generated Samples

Abstract

China’s rapid development of high-speed railways has imposed increasing requirements for safety and reliability of signal systems, especially the critical part: turnouts. In this paper, we propose an intelligent fault diagnosis approach that can effectively detect turnout faults based on self-generated fault samples. First, the action mechanism of a switch machine is analyzed and we establish a turnout action model to simulate the turnout operation current curves, thus considerable samples for a following diagnosis can be obtained. Second, we develop a turnout fault diagnosis model based on convolutional neural networks (CNNs). The networks can be trained by those simulated samples. Our experiments verify that the turnout action model can accurately simulate turnout fault curves and the diagnosis model can effectively identify faults through various formats of curve pictures.