Intelligent identification of differential subgrade settlement of ballastless track system based on vehicle dynamic responses and 1D-CNN approach

The millimeter-scale deformation control standards of high-speed railways make the monitoring and identification of subgrade settlement a key technology, especially for ballastless track systems. Addressing the uncoordinated deformation between track and subgrade caused by high track stiffness, this paper proposes an intelligent identification method for subgrade settlement based on vehicle vibration signals and deep-learning technology. Firstly, a high-speed vehicle-ballastless track-subgrade coupled dynamics model considering the track-subgrade nonlinear contact state is utilized, to obtain various vibration responses of the vehicle subsystem due to differential subgrade settlement. After data preprocessing, a multi-channel one-dimensional convolutional neural network (1D-CNN) is established to extract features from the vehicle vertical acceleration dataset and output subgrade settlement status automatically. Finally, the robustness of the model is verified using Gaussian white noise. The research indicates that the vertical accelerations of the car body and bogie frame contain dynamic characteristics caused by medium-long wavelength settlements, while the vertical acceleration of the wheelset is more sensitive to short-wavelength settlements. It is feasible to identify the degree of subgrade settlement using car body, bogie frame and wheelset vertical accelerations as parameterized multi-source input of the 1D-CNN network. The three-channel input model significantly improves the identification accuracy and demonstrates the robustness against noise interference compared to the single-input and dual-input models. This method can effectively enhance the identification capability of 'concealed' subgrade settlements in high-speed railways, offering a scientific methodology for status maintenance of ballastless track lines.