Abstract: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.

Deep Learning-Based Suppression of Strong Noise in GPR Data for Railway Subgrade Detection

Lightweight Deep Learning Model for Identifying Tunnel Lining Defects Based on GPR Data

Combined CNN and RNN Neural Networks for GPR Detection of Railway Subgrade Diseases

Differential Settlement of Track Foundations Identification Based on GRU Neural Network.

Construction Environment Noise Suppression of Ground-Penetrating Radar Signals Based on an RG-DMSA Neural Network

Intelligent Detection of Urban Road Underground Targets by Using Ground Penetrating Radar based on Deep Learning

Towards 3D Metric GPR Imaging Based on DNN Noise Removal and Dielectric Estimation

Safety Monitoring of Transportation Infrastructure Foundation: Intelligent Recognition of Subgrade Distresses Based on B-Scan GPR Images

GPR detection localization of underground structures based on deep learning and reverse time migration

Application of Air-Coupled Ground Penetrating Radar Based on F-K Filtering and BP Migration in High-Speed Railway Tunnel Detection

Study on intelligent recognition of urban road subgrade defect based on deep learning

Efficient Underground Target Detection of Urban Roads in Ground-Penetrating Radar Images Based on Neural Networks

Mutual Interference Suppression and Signal Enhancement Method for Ground-Penetrating Radar Based on Deep Learning

GPR image denoising with NSST-UNET and an improved BM3D

GPR-based Subsurface Object Detection and Reconstruction Using Random Motion and DepthNet

DMRF-UNet: A Two-Stage Deep Learning Scheme for GPR Data Inversion under Heterogeneous Soil Conditions

Noise Suppression for CRP Gathers Based on Self2Self with Dropout

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

CycleGAN-Based Data Augmentation for Subgrade Disease Detection in GPR Images with YOLOv5

Deep Learning for Improved Subsurface Imaging: Enhancing GPR Clutter Removal Performance Using Contextual Feature Fusion and Enhanced Spatial Attention

Automatic identification of GPR targets on roads based on CNN and Grad-CAM