Data-Driven Multiscale 1D-CNN-Based Fault Diagnosis of High-Speed Railway Turnout Welded Joints

High-speed railway turnouts, critical for train switching and direction changing, are pivotal components in the railway transportation system. The welded joints are important parts of high-speed railway turnouts and hence their health conditions are paramount for ensuring the safe operation of high-speed railway lines. Traditionally, there are two methods for the health detection of railway turnout welded joints, that is, measurement of wheel-rail forces by putting strain gauges on the welded joints and by straightedge by human. The former measurement method relies on strain gauge method, where strain gauges are attached to the steel rail at welded joints, and the wheel-rail forces are recorded when the train passes through. The strain gauge method is limited to selective inspections and not applicable for a large-scale use. The latter measurement method is labor-intensive and the accuracy depends on the skills of inspectors. The turnout welded joints detection based on vehicle dynamic response has the potential to overcome the shortcomings aforementioned. However, vehicle dynamic response-based health condition detection of turnout welded joints currently has the problem of the sample shortage, which hinders the health condition of turnout welded joints based on big data. Therefore, a vehicle-turnout coupling model is constructed using multibody dynamics to tackle the data shortage problem. The dynamic model can simulate the response of vehicles passing over turnout welded joints with various health conditions. A defect diagnosis method based on a multiscale 1-D convolutional neural network (MS 1D-CNN) is, then, proposed, achieving the welded joints intelligent diagnosis of high-speed turnouts.