In railway condition monitoring, feature classification is a very critical step, and the extracted features are used to classify the types and levels of the faults. To achieve better accuracy and efficiency in the classification, the extracted features must be properly selected. In this paper, maximal information coefficient is employed in two different stages to establish a new feature selection method. By using this proposed two-stage feature selection method, strong features with low redundancy are reserved as the optimal feature subset, which results in the classification process having a more moderate computational cost and good overall performance. To evaluate this proposed two-stage selection method and prove its advantages over others, a case study focusing on the rolling bearing is carried out. The result shows that the proposed selection method can achieve a satisfactory overall classification performance with low-computational cost.

Rail surface defect inspection is of particular importance in modern railways. Accurate and efficient surface defect detection approaches support optimized maintenance. This enables safe operation of the railway network. However, the scale and harsh working environments of the railway still pose challenges to existing manual and vision-based inspection methods. Inspired by recent advances in laser measurement and deep learning in computer vision, this paper proposes a laser-based 3D pixel-level rail surface defect detection method that combines high-precision laser measurement data with the concept of deep semantic segmentation. In the proposed method, the rail surface is first measured in 3D using a low-cost 2D laser triangulation sensor. Then, a new deep semantic segmentation network is introduced. The network is composed of a fully convolutional segmentation module and two symmetric mapping modules, which can take 3D laser measurement data as input and output 3D pixel-level defect detection results in an end-to-end manner. The modular design of the network allows the use of various segmentation modules for different applications or scenarios. Experiments on a 3D rail dataset demonstrate the feasibility of the proposed method with a pixel-level detection accuracy measured by mean Intersection over Union (mIoU) of up to 87.9%. The 3D output provides not only location and boundary information but also the 3D characterization of defects, giving an essential reference for further defect management and repair tasks.

Rail profile measurement is one of the most critical tasks for track quality inspection to ensure the safe operation of track systems. In modern railway systems, laser triangulation sensors have been widely adopted in onboard measuring units to carry out 2-D rail profile measurement due to their robustness and truly noncontact properties. However, existing solutions limit the degrees of freedom of the laser sensor and thus cannot provide full coverage of the rail profile due to the “shadowing effect” of triangulation sensors. Incomplete profiles limit the performance of wear assessment hence more detailed inspections still rely on contact-based tools operated by humans. These processes are time-consuming and incompatible with the ever-shortening maintenance windows available in modern railway systems. Benefiting from the miniaturization of sensing technology and improving processors, multisensing systems combine the strengths of different sensors. This article presents a new solution for a laser-based multisensing system for noncontact rail profile measurement and wear inspection. The addition of an inertial measurement unit (IMU) and a camera module allows portable rail profile measurement without “blind spots.” Optimized iterative closest point (ICP) registration is then applied to generate a complete representation of the rail profile. Experimental results demonstrate that the proposed system can provide accurate and efficient rail profile measurement, and could potentially replace conventional contact-based inspection tools.

An improved method with feature extraction and neural network classification is proposed and applied to address the fault diagnosis problem in railway switch systems. Features are extracted by analysing the non-stationary signal time-frequency characteristics of the electro-mechanical machine under different fault severities. To enhance the fault diagnosis capability, an Energy-based Thresholding Wavelets (EBTW) approach is proposed by reconstructing a low-dimensional feature vector with selected parameters. Different from the conventional discrete wavelet transform and soft-thresholding wavelet transform, the proposed method localizes and redistributes the signal energy to achieve an efficient dimension reduction. The first step of the proposed method is to perform a wavelet transform upon the original signals, whose wavelet coefficients are then thresholded and an extraction of features is made, on the basis of the energy conservation property of wavelet transforms. Once the features are extracted and selected, several state-of-the-art neural-network-based classifiers are applied for fault diagnosis by identifying the signal features from a machine normal operation and faulty operations with multiple severities. The effectiveness of the proposed method is verified using real-world operational railway switch data and comparing with conventional feature extraction methods under different classifiers.

Pattern recognition and classification have been recently applied in the area of condition monitoring to solve the problems of fault diagnosis and prognosis. Usually, the original condition monitoring data will be transformed into representative features that are of lower dimensionality, followed by the application of a classifier. However, most time-domain and frequency-domain features cannot provide good representations of non-stationary signals that are very common in machine faults. This research solves this problem by applying a Discrete Wavelet Transform (DWT) to extract the time-frequency features. Subsequently, a feed-forward artificial neural network is used as a method for fault diagnosis and classification. A railway switch system is used as a specific application in this research to validate the method using data from a field switch system. Features of the scale coefficients are then extracted using DWT for the 'Haar' wavelet type at decomposition level 9. Subsequently, the feature data are split into training, validation and testing sets, with each set containing one input feature and one corresponding desired output fault severity. After training and testing using a two-layer neural network, it is shown that in this case the switch overdriving fault with different severities could be diagnosed with 100% accuracy.

Recent developments in the area of condition monitoring research have been targeted towards predicting machinery health condition for the purpose of preventative maintenance. Typically, published research uses data collected from rotating components (bearings, cutting tools, etc.) working in an idealized lab environment as the case study for prognosis algorithm validations. However, the operational implementation in industry is still very sporadic, mainly owing to the lack of proper data allowing sufficiently mature development of comprehensive methodologies. The prognosis methodology presented herein bridges the gap between academic research and industrial implementations by employing a novel time period for prognosis and implementing random coefficients regression models. The definition of the remaining maintenance-free operating period (RMFOP) is proposed first, which helps to transform the usefulness of the degradation data that is readily available from data short of failure. Degradation patterns are subsequently extracted from the original degradation data, before fitting into either of two regression models (linear or exponential). The system residual life distributions are then computed and updated by estimating the parameter statistics within the model. This RMFOP-based methodology is validated using real-world degradation data collected from multiple operational railway switch systems across Great Britain. The results indicate that both the linear model and the exponential model can produce residual life distributions with a sufficient prediction accuracy for this specific application. The exponential model gives better predictions, the accuracy of which also improves as more of system life percentage has elapsed. By using the RMFOP methodology, switch system health condition affected by an incipient overdriving fault is recognized and predicted.