Answers are grounded in this paper and cite their sources.
The braking system directly determines the stopping capability, adhesion utilization, and operational safety of high-speed trains. However, braking performance is simultaneously affected by electric–pneumatic braking-force distribution, actuator delay, wheelset dynamics, and wheel-rail adhesion that varies with speed and contact conditions. This paper first systematizes the braking structure at three levels: vehicle, carriage, and wheel, while summarizing the longitudinal dynamics, slip-ratio formulation, and wheel-rail adhesion characteristics to define the feasible braking domain. On this basis, artificial-intelligence approaches are reviewed according to three levels of development: machine learning (ML) for estimation, deep learning (DL) for representation learning, and deep reinforcement learning (DRL) for sequential braking decision-making. The analysis shows that ML and DL improve the estimation of adhesion, stopping response, and wheel-slip conditions, whereas DRL extends toward direct traction/braking command generation; however, the physical feasibility of the resulting actions is not guaranteed when adhesion and actuator states vary outside the training domain. To address this gap, the paper proposes a conceptual CTD4–Digital Twin framework in which the twin updates physical states and parameters, CTD4 learns the return distribution under uncertainty, and a safety projection constrains the action before it is transmitted to the braking system.
No takes yet. Share an insight, caveat, or question.
Nguyen et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: