Railway damage detection is of great significance in ensuring railway safety. The cracks on the rail surface play a key role in studying the formation and development process of rail damage, predicting the occurrence of rail defects, and then improving the service life of the rail. However, due to the small shape of the cracks, the typical detection method is relatively complicated, and the speed is quite slow. Although traditional magnetic particle inspection technology is fairly accurate at detection, it is costly and inconvenient to carry and install, while also limiting the detection speed and affecting the system’s operation. In this paper, a semantic segmentation detection method is developed by using various collected rail surface crack data and deep learning through a neural network. By comparing the inspection of the same rail surface with magnetic particle inspection technology, only inexpensive cameras are used and the inspection speed is increased while maintaining relatively high accuracy. In addition, the method can achieve fast detection speeds if it is extended to be combined with high-frequency cameras. It is an economical, efficient, and environmentally friendly method for future rail surface detection.

In this paper, an application of computer vision and machine learning algorithms for common crossing frog diagnostics is presented. The rolling surface fatigue of frogs along the crossing lifecycle is analysed. The research is based on information from high-resolution optical images of the frog rolling surface and images from magnetic particle inspection. Image processing methods are used to pre-process the images and to detect the feature set that corresponds to objects similar to surface cracks. Machine learning methods are used for the analysis of crack images from the beginning to the end of the crossing lifecycle. Statistically significant crack features and their combinations that depict the surface fatigue state are found. The research result consists of the early prediction of rail contact fatigue.

Railway ballast tamping is one of the cost-expensive renewal and maintenance works of railway superstructure. The quality of ballast consolidation influences its resistance to residual deformations and long-term deterioration of track geometry. The process of ballast compaction along the sleeper under the vibration loading is complex and depends on many factors. The ballast flow processes under the vibration loading can produce both consolidation and un-consolidation of ballast material. The present study is devoted to the experimental investigation of ballast consolidation inhomogeneity. The method of ballast local consolidation measurement is proposed. The method is based on the velocity of impact wave propagation that is measured with device. The application of modern microcontroller and sensor techniques provided simple and reliable multi-point velocity measurements in a ballast layer. That enables well enough spatial resolution of ballast consolidation inhomogeneity. The measurement analysis has shown more than 4 times higher consolidation under the sleeper center than for unconsolidated ballast.

This paper focuses on the experimental study of an alteration in the railway crossing dynamic response due to the rolling surface degradation during a crossing’s lifecycle. The maximal acceleration measured with the track-side measurement system as well as the impact position monitoring show no significant statistical relation to the rolling surface degradation. The additional spectral features are extracted from the acceleration measurements with a wavelet transform to improve the information usage. The reliable prediction of the railway crossing remaining useful life (RUL) demands the trustworthy indicators of structural health that systematically change during the lifecycle. The popular simple machine learning methods like principal component analysis and partial least square regression are used to retrieve two indicators from the experimental information. The feature ranking and selection are used to remove the redundant information and increase the relation of indicators to the lifetime.

The paper deals with remaining useful life (RUL) prognosis of common crossings based on inertial measurements. Axle-box inertial measurements on operational trains could be a cheap alternative to conventional inspection means. The low correlation between maximal wheel acceleration and useful life of the crossing is considered and the reasons are analysed. A machine learning approach, including feature extraction, selection, fusion and degradation modelling, is then used to cope with the problem. More time domain and spectral features are extracted from measured vertical accelerations to provide a higher utilization of the available information. After removing redundant features, the data is fused using principal component analysis to obtain a condition indicator for common crossings. A data-driven prognostic methodology is proposed based on an iteratively updated exponential degradation model. The assessment of the prognosis quality is carried out depending on the crossing lifetime and the reached value of the condition indicator.