Railways speedily transport many people and goods nationwide, so railway accidents can pose immense damage. However, the infrastructure of railways is so complex that its maintenance is challenging and expensive. Therefore, using artificial intelligence for railway safety has attracted many researchers. This paper examines artificial intelligence applications for railway safety, mainly focusing on deep learning approaches. This paper first introduces deep learning methods widely used for railway safety. Then, we investigated and classified earlier studies into four representative application areas: (1) railway infrastructure (catenary, surface, components, and geometry), (2) train body and bogie (door, wheel, suspension, bearing, etc.), (3) operation (railway detection, railroad trespassing, wind risk, train running safety, etc.), and (4) station (air quality control, accident prevention, etc.). We present fundamental problems and popular approaches for each application area. Finally, based on the literature reviews, we discuss the opportunities and challenges of artificial intelligence for railway safety.

We propose a method for detecting earthquakes for high-speed trains based on unsupervised anomaly-detection techniques. In particular, we utilized autoencoder-based deep learning models for unsupervised learning using only normal training vibration data. Datasets were generated from South Korean high-speed train data, and seismic data were measured using seismometers nationwide. The proposed method is compared with the conventional Short Time Average over Long Time Average (STA/LTA) model, considering earthquake detection capabilities, focusing on a Peak Ground Acceleration (PGA) threshold of 0.07, a criterion for track derailment. The results show that the proposed model exhibit improved earthquake detection capabilities than STA/LTA for PGA of 0.07 or higher. Furthermore, the proposed model reduced false earthquake detections under normal operating conditions and accurately identified normal states. In contrast, the STA/LTA method demonstrated a high rate of false earthquake detection under normal operating conditions, underscoring its propensity for inaccurate detection in many instances. The proposed approach shows promising performance even in situations with limited seismic data and offers a viable solution for earthquake detection in regions with relatively few seismic events.

The dynamic behavior of structures in liquefiable sand exhibits more complicated characteristics, due to the development of excess pore pressure caused by cyclic loading, than that in dry sand. Therefore, it is crucial to accurately predict the soil–pile structure behavior during liquefaction to prevent damage to the structures. In this study, three-dimensional numerical modeling was performed to predict the dynamic soil–pile behavior during liquefaction. To directly simulate pore pressure generation due to soil shear deformation, the Finn liquefaction model was applied and coupled with the Mohr-Coulomb elasto-plastic model. Soil nonlinearity was considered by applying hysteretic damping, and the interface model was applied to simulate various dynamic phenomena between the soil and pile. Simplified continuum modeling was introduced to prevent reflection wave generation and increase analysis efficiency. The applicability of the proposed numerical model was validated using the experimental results. Thereafter, a parametric study was conducted to provide a better understanding of the dynamic behavior of pile foundation during liquefaction. From a series of parametric studies, several important factors that can affect the dynamic pile responses in liquefiable sand were identified. Also, the characteristics of the dynamic soil–pile structure interactive behavior, which are significantly different from each other in liquefied and dry sand, were analyzed qualitatively and quantitatively.

In this study, centrifuge model tests were used to examine the lateral behavior of a monopile embedded in dry sand through cyclic lateral loading tests. The soil specimens used in the tests were dry Jumunjin sand with a relative density of 80% and a friction angle of 38°. A static loading test was performed once, and cyclic loading tests were performed four times using four magnitudes of cyclic load (30%, 50%, 80%, and 120% of static lateral capacity). The experimental cyclic p-y curve was obtained through the tests, and the maximum soil resistance points that were found for each load were used to find the cyclic p-y backbone curve for each depth. The two variables which are needed to define the cyclic p-y backbone curve, i.e., the initial modulus of subgrade reaction (k_ini) and ultimate soil resistance (p_u), were suggested as functions of the soil’s physical properties and the pile. The cyclic p-y curve of the first cycle and the 100th cycle were formulated to present the upper limit and lower limit. The suggested cyclic p-y curve had an overestimated soil resistance compared with the existing API (1987) method, but the initial modulus of subgrade reaction was underestimated.

When an off-road tracked vehicle travels, shearing action and ground sinkage occur on the soil–track interface, severely affecting the tractive performance of the vehicle. Notably, ground sinkage, which is induced by the vehicle’s weight (static sinkage) and longitudinal forces in the direction of travel producing slip (slip sinkage), develops motion resistance, directly restricting the tracked vehicle’s performance. Thus, it is critical to consider both static sinkage and slip sinkage to assess the tractive performance of a tracked vehicle. In this research, model track experiments were conducted to investigate slip sinkage. The experimental results showed that the slip sinkage increased as the slip ratio increased, but the rate of increase decreased. The slip sinkage was found to increase as the density of the ground decreased and imposed vertical load increased. The experimental results were used to calculate normalized slip sinkage, and an empirical equation for slip sinkage in terms of slip ratio was developed. This equation will allow vehicle operators to predict the slip sinkage and associated motion resistance for given soil and vehicle conditions.

Railway infrastructure maintenance faces a challenge posed by the laborious task of monitoring widespread deformation, which is critical for ensuring safety. This study utilized the Persistent Scatterer Interferometric Synthetic Aperture Radar (PS-InSAR) technique with high-resolution TerraSAR-X satellite images to measure and analyze deformation at the operating track. Tailored through parameter optimization to fit regional topography and validated against field measurements, PS-InSAR was applied across a 30km by 50km area over two years. The analysis demonstrated that PS-InSAR could efficiently generate high-accuracy time-series data, capable of detecting both the uplift and subsidence processes. It highlighted the importance of augmenting single-point subsidence values with comprehensive time-series analysis for a complete deformation assessment. The study concluded that PS-InSAR is an accurate and cost-effective tool for large-scale linear infrastructure monitoring, despite technological constraints such as radar imaging frequency and the lack of high-resolution sources. Considering these constraints, future research will prioritize developing an enhanced algorithm capable of analyzing both urban and suburban areas, accommodating varying numbers of point scatterers.