With the increase in heavy-haul train formations and speeds, braking-induced dynamic issues have emerged as critical constraints affecting operational safety and efficiency. Traditional longitudinal train dynamic models exhibit limitations in deriving dynamic performance indicators, whereas three-dimensional train dynamic models suffer from computational complexity. To bridge this methodological gap, a hybrid one-dimensional-three-dimensional train dynamic model is proposed, integrating braking system and line conditions to achieve optimized computational efficiency through selective refinement at critical segments. In addition, a data-driven surrogate modelling approach is established by integrating numerical simulations with machine learning algorithms, including least squares support vector machine (LSSVM), back propagation neural network (BPNN), and long short-term memory (LSTM) network. This methodology enables rapid and accurate assessment of safety indicators during emergency braking. Results indicate that machine learning-based surrogate models achieve high-precision safety assessment whilst significantly improving computational efficiency. The LSTM model attained the highest accuracy with coefficients of determination consistently exceeding 0.98. The proposed surrogate modelling paradigm effectively balances computational accuracy with efficiency, providing theoretical support for heavy-haul train dynamics evaluation and optimization.
Recent advances in computer vision have promoted digital transformation of railway maintenance, providing a novel solution for online monitoring of wheel-rail interaction. This paper proposes an end-to-end framework to achieve real-time detection of wheel-rail relative motion, using the YOLOv8-Pose-based feature-point identification method. Validated by the field full-scale testing, the trained model attains a relative error of 2.78
The safety of road vehicles can be compromised by crosswinds, particularly during severe thunderstorm conditions. Nevertheless, most studies on vehicle stability typically consider only generic synoptic wind conditions. In this work, a methodology to assess the safety of road vehicles under downburst winds is presented. The workflow includes defining the wind velocity field during the downburst, calculating the aerodynamic loads, and evaluating the dynamic response of the vehicle-driver system. Wind tunnel tests have been performed to obtain the aerodynamic coefficients. Co-Simulation with Simulink and VI-CarRealTime has been used to analyse the behaviour of the vehicle. To include the effect of human driver a driver controller has been defined and validated with the results obtained from the driving simulator. The results demonstrated that a high-sided lorry represents a serious safety risk when exposed to the intense and rapidly varying winds generated during a downburst event.
Dynamic modeling plays a critical role in the stability control of Long Combination Vehicles (LCVs). However, accurately capturing the complex lateral dynamics of LCVs remains a significant challenge due to the presence of multiple articulations between vehicle units. In particular, solving for the articulation forces is often computationally intensive and analytically complex. This paper presents a structured approach to derive the state-space matrices governing the lateral dynamic model of LCVs directly from their design parameters. Two generalized methods for formulating articulation forces are proposed: the Calculating from Front to Rear (CFTR) method and the Calculating from Rear to Front (CRTF) method. By integrating one of these general force expressions into the system’s differential equations, we identify consistent regularities in the resulting state-space matrices. The proposed framework simplifies model construction and enhances its scalability to multi-unit vehicle systems. The findings offer valuable insights for model-based controller design, system optimization, and linear stability analysis of LCVs.
Localization is one of the key technical factors for autonomous driving. Existing methods include GNSS (Global Navigation Satellite System) or methods using visual sensors such as cameras or LiDARs (Light Detection and Ranging). However, there are conditions where their accuracy can degrade, even when multiple methods are utilized. This paper proposes using the environmental magnetic field (EMF) for localization. EMF is a combination of the natural geomagnetic field and the magnetic field induced by man-made objects. The proposed method was validated in simulation and experiment using a test vehicle equipped with magnetic sensors. The results showed that the proposed method can accurately localize the vehicle with errors of less than 0.5 m.
