Railway tracks are prone to damage after long-term exposure to harsh environments, posing risks to transportation safety. Traditional manual inspection is inefficient and error-prone, while existing deep learning–based detection models, such as YOLO series, often struggle with small targets, complex backgrounds, and multi-scale damage. To address these challenges, this study proposes an improved track damage detection framework, termed SNBF-YOLO, which integrates the Star Net and BiFPN modules. Star Net adaptively enlarges the receptive field to enhance feature representation, while BiFPN optimizes bidirectional multi-scale feature fusion. Experimental results on a real-world dataset demonstrate that SNBF-YOLO improves precision, recall, and mAP by 19.4, 13.2, and 14.3%, respectively, compared to the baseline YOLOv10n model. The enhanced model achieves higher robustness and computational efficiency, enabling accurate real-time detection of track defects, including fine cracks and missing fasteners. Despite these advances, the study is limited by the scale and diversity of the dataset, which may affect generalization in more diverse conditions (e.g., rain, snow, or severe corrosion). Future work will focus on enlarging the dataset and exploring lightweight deployment strategies to further improve real-world adaptability.

Despite the significant impact of network hyperparameters on deep learning car-following models, there has been relatively little research on network hyperparameters of deep learning car-following models. Therefore, this study proposes a car-following model that combines particle swarm optimization (PSO) and gated recurrent unit (GRU) networks. The PSO-GRU car-following model is trained and tested using data from the natural driving database. The results demonstrate that compared to the intelligent driver model (IDM) and the GRU car-following model, the PSO-GRU car-following model reduces the mean squared error (MSE) for the speed simulation of following vehicles by 88.36% and 72.92%, respectively, and reduces the mean absolute percentage error (MAPE) by 64.81% and 50.14%, respectively, indicating a higher prediction accuracy. Dataset 3 from the drone video trajectory database of Southeast University and NGSIM’s I-80 dataset are used to study the car-following model’s cross-dataset adaptability, that is, to verify its transferability. Compared to the GRU car-following model, the PSO-GRU car-following model reduces the standard deviation of the test results by 60.64% and 32.89%, highlighting its more robust prediction stability and better transferability. Verifying the ability of the car-following model to produce the stop-and-go phenomenon can evaluate its transferability more comprehensively. The PSO-GRU car-following model outperforms the GRU car-following model in creating stop-and-go sensations through platoon simulation tests, demonstrating its superior transferability. Therefore, the proposed PSO-GRU car-following model has higher prediction accuracy and cross-dataset adaptability compared to other car-following models.

Accurate and rapid anomaly detection of train control systems is an inevitable requirement for ensuring the safe and efficient operation of high-speed railways. Currently, the manual offline fault diagnosis has issues such as ineffectiveness in fault locating and a relatively large scope of fault impact. In response, an anomaly detection model based on multimodal learning with the attention mechanism is proposed. According to the interrelated relationship between text logs and visual images representing equipment working status, a language-vision fusion two-stream multimodal neural network learning architecture is designed. The entire network structure, centred on the attention mechanism, learns the mapping relationship between inputs and outputs, simultaneously processes log generation of multiple sub-equipment, and separately focuses on the context of the text and changes in indicator light display at specific positions on the visual images. At the final decision-making layer, the learning results of language and vision are organically fused through logical operations, producing a unified output indicating the anomalous state of the system. Experimental results on real train operation datasets demonstrate the model’s superior performance in terms of precision and recall compared to other methods, validating its effectiveness.

Electronic equipment is essential to railway transport, as its performance directly impacts the safety of train operations. However, current operation and maintenance practices face several problems, including unrecognized degradation of equipment performance, difficulty in diagnosing faults after they occur, and coarse-grained maintenance strategies, These problems may lead to an excessively broad scope of fault impact, the continued operation of defective equipment, and other serious consequences. In this context, the paper presents an intermittent fault (IF)-based health management approach for railway electronic equipment. Building on fault mechanism analysis for electronic equipment, a mathematical parameterized characterization of intermittent faults is achieved by integrating the fault frequency and intensity vectors into the state transition probability matrix of a hidden Semi-Markov model (HSMM). To solve the non-convex parameter estimation problem, a global optimization algorithm based on swarm intelligence is constructed to maximize the log-likelihood objective function, so as to evaluate the current operational state of the equipment. The paper then uses the equipment observation sequences predicted by the long short-term memory network (LSTM) to obtain the remaining useful life (RUL) values at any timepoint. The estimation serves as the basis for proactively adjusting maintenance decisions. In experiments based on real-world power module faults, compared with the conventional Baum–Welch trained HSMM method, the proposed method in this paper improves state recognition accuracy by 2.15% and reduces the RMSE of RUL prediction by 16.67%, verifying its effectiveness.