Rail Wear Detection with Wheel-Rail Contact Images: A Deep Learning Approach
Qihang Wang, Southwest Jiaotong UniversityShow Abstract
Tianci Gao, Southwest Jiaotong University
Zhengxing Chen, Southwest Jiaotong University
Xiaoming Wang, Southwest Jiaotong University
Yong Liu, China Railway Chengdu Group Co
Ping Wang, Southwest Jiaotong University
Qing He (email@example.com), Southwest Jiaotong University
Rail wear, which occurs continuously owing to the running of trains, is a fundamental aspect of railway operational safety. Currently, rail wear grade is measured manually, but this practice is unable to satisfy the increasing demand for condition-based maintenance. Therefore, there is a pressing need for rapid and continuous detection of rail wear. This paper proposes a three-step image-based method for detecting rail wear on the running band, which is a collection of the interaction traces between the wheel and the rail and contains a significant amount of useful information on the rail-wheel contact. The following three steps are involved in the proposed method: (i) The running band of a rail is segmented and extracted from images using an improved mask region-based convolutional neural network (Mask R-CNN), which uses the scale and ratio information to perform instance segmentation of the running band images. (ii) A multi-level residual neural network model that classifies the rail running band images is constructed. (iii) Reclassification and focal loss are used for enhancing the detection performance by addressing the imbalanced wear class problem. In order to validate our approach, we collected real-world running band images together with the rail wear data measured using a high-accuracy rail profile measurement tool. The results obtained via the application of our method to the data show that our method can rapidly and accurately detect rail wear under different ambient light conditions. In the future, the proposed model can be integrated with the existing onboard rail-inspection system.
Ballast Inspection Tool Using Seismic Surface Waves
Charles Oden, Earth Science Systems, LLCShow Abstract
Timothy Stark, University of Illinois, Urbana-Champaign
Choon Park, Park Seismic, LLC
Carton Ho, University of Massachusetts, Amherst
A man-portable seismic instrument has been developed for assessing the elastic properties of railroad ballast. The Seismic Ballast Inspection Tool (SeiBIT) is based on the multi-channel analysis of surface waves (MASW) method and is designed to resolve the ballast properties as shallow as 1 foot (30 cm) and to depths greater than 16 feet (5 m). The SeiBIT consists of a seismic source and five seismic receivers. The system uses GPS time synchronization and localization so that each unit can be placed individually on the track ballast surface with no location constraints other than lying on the survey line. The computer controlled seismic source can generate horizontal displacements in the in-line or cross-line directions so that either Rayleigh or Love waves can be preferentially generated (respectively). After field data have been collected, Love and Rayleigh wave dispersion curves (i.e., wave velocity versus frequency) are generated from survey data, which are then inverted for seismic properties (e.g., shear wave velocity and layer thickness) using a one-dimensional model. The result is a vertical section of properties that represent the ballast and subgrade. Software has been created for the tablet computer for wireless data acquisition, MASW analysis, and dispersion curve inversion. Complete results can be obtained in minutes. The system response has been assessed with MASW experiments in both the laboratory and in the field. Preliminary results of surveys on railroads indicate that the Rayleigh wave method produces expected results in clean and fouled ballasts.
APPLICABILITY OF LASER DOPPLER VIBROMETER PLACED ON A MOVING PLATFORM FOR RAIL VIBRATION MEASUREMENTS
Korkut Kaynardag (firstname.lastname@example.org), University of Texas, AustinShow Abstract
Chi Yang, University of Texas, Austin
Salvatore Salamone, University of Texas, Austin
This paper investigates the feasibility of measuring rail’s flexural modes using a laser Doppler vibrometer (LDV) placed on a moving platform (i.e., rail car). For this purpose, two different types of experiments are conducted at Transportation Technology Center Inc. (TTCI) in Pueblo, Colorado. First, flexural modes of a rail span (rail section between two consecutive sleepers) are identified using accelerometers measurements collected under a moving rail car excitation. Afterward, the modes of a rail span are obtained using measurements recorded by an LDV placed on a moving rail car. Both the accelerometer and LDV measurements were carried out at four different moving rail car speeds. Comparison of results demonstrated that LDV placed on a moving platform was capable of identifying the flexural modes of rail spans.
Analytical Estimation of Impact Forces Due to Abrupt and Rapid Changes in Track Profile at Rail Ends and Turnout Crossings
Niyazi Özgür Bezgin, Istanbul Universitesi, CerrahpasaShow Abstract
Mohamed Wehbi, Network Rail
Variations in railway track profiles can increase the vertical forces imposed unto them by train wheels. Rail profile can vary abruptly at rail joints and rapidly within turnouts and generate high dynamic impact forces on the track. Increased wheel forces on the track can damage the rail, the rail bearing elements and the supporting layers of the track along with the wheels and the wheel bearing elements of the trains. Railhead and wheel thread plastification, fracture and rolling contact fatigue, rail seat fracture, ballast pulverization and cumulative track settlement due to increased stresses on ballast and subgrade are some of the damages caused by increased wheel forces due to abrupt and rapid changes in track elevation. Analysis of the wheel and track interaction and quantification of the resulting forces takes considerable time due to high degree of indeterminacy inherent in their interaction. Researchers frequently resort to numerical analyses or data extracted from instrumented tracks to analyze these interactions. However, in the absence of these extensive resources engineers and researchers lack an explicit analytical tool, which they can use to investigate these interactions. Therefore, development of analytical tools that can provide rapid and reliable estimates of dynamic impact forces is imminent. This paper demonstrates the application of an analytical method recently introduced to the railway engineering literature, known as the Bezgin Method to estimate the wheel impact forces generated due to track and wheel roughness and an assessment of the consequences of these forces on ballasted railway tracks through case studies.
Bilevel Optimization of Intercity Railway Alignment
Dongying Yang, Southwest Jiaotong UniversityShow Abstract
Sirong Yi, Southwest Jiaotong University
Qing He, Southwest Jiaotong University
Intercity railway transit is the primary way to transfer passengers and freight over a large distance on land. The design of an intercity railway alignment is usually challenging and complex because of great topographic variations and intersections with many existing geographic objects. This paper presents a bilevel optimization model for intercity railway alignment (BIRAO) following the horizontal-vertical alignment design philosophy. The upper level is the horizontal alignment optimization, while the nested lower level is the vertical alignment optimization. A multistage augmented differential evolution (DE) algorithm is adopted for the solution at both levels. BIRAO could generate corridor alignment, initial alignment, and optimal alignment through different allocations of the number and boundaries of decision variables. A real-world case study of railway design is conducted to verify the effectiveness of the BIRAO model based on a GIS system.
Track Performance in Tunnels and Rail Transition Areas with Under Tie Pads and Under Ballast Mats
Stephen Gonzalez, University of FloridaShow Abstract
Zachary White, University of Florida
Jennifer Bridge, University of Florida
Shanyue Guan, East Carolina University
Justin Davis, University of Florida
Kyle Riding, University of Florida
Under tie pads (UTP) and under ballast mats (UBM) have been increasingly used in rail track construction to reduce maintenance costs by better distributing loads, reducing the track modulus, and increasing ballast contact areas with ties. Locations such as tunnels, bridges, and bridge approaches are especially strong candidates for UTP and UBM use due to the high support stiffness they provide to the ballast. In this study, the Virginia Avenue Tunnel in Washington D.C. was instrumented during construction to monitor track pressure distribution, tie acceleration, and tunnel floor vibration during the first 20 months of use. It was found that UTP and UBM reduced the track modulus. Use of UBM in lieu of an extra 6 in. of ballast was found to result in more than a 10% reduction of the average force on the tie directly under the train axle, providing much better distribution of pressure across ties. Settlement measurements showed that the track settlement occurred over the first 6 months of measurement, after which it stabilized to less than 0.157 in. (4 mm).
Improving Railway Maintenance Schedules by considering Hindrance and Capacity Constraints
Floris Nijland, ProRailShow Abstract
Konstantinos Gkiotsalitis (email@example.com), University of Twente
Eric C. van Berkum, Universiteit Twente
The availability of railway networks is important for society and the economy. To keep the infrastructure in good condition, regular maintenance is needed. Regular maintenance is achieved by devising maintenance schedules that assign safe work zones to crews that need to execute preventive maintenance activities. This study aims to optimize the maintenance schedules for both train operators and maintenance contractors, by considering (a) hindrance for parked passenger trains and planned freight trains, and (b) the workload for track workers. Further, maintenance operations are distinguished into different engineering fields since this influences the amount of hindrance. The method presented for designing maintenance schedules is a novel mixed-integer linear programming (MILP) model that considers these aspects. In our Dutch case study, we assess the new scheduling model on its performance and show that large improvements can be made in terms of mean workload for work crews and total hindrance for train operators.
SETTLEMENT AND STRESS DISTRIBUTION CHARACTERISTICS OF A RAILWAY BALLAST LAYER UNDER A DYNAMIC LOAD
Jianxing Liu, Southwest Jiaotong UniversityShow Abstract
Ganzhong Liu, Southwest Jiaotong University
Tianci Gao, Southwest Jiaotong University
Ping Wang, Southwest Jiaotong University
Jieling Xiao, Southwest Jiaotong University
Chenyang Hu, Southwest Jiaotong University
The settlement and stress distribution characteristics of a ballast layer under low-frequency cyclic load were studied with laboratory tests and the discrete element method (DEM). Then the discrete element model was used to further explore the load transfer path of the ballast layer and the movement state of the ballast particles under a high-speed train load. The relationship between the settlement characteristics of the ballast layer under a dynamic load and the state changes of the particles was clarified. The results showed that although the changes of the stress at the bottom of the ballast layer under a low-frequency cyclic load and a high-speed train load could be divided into five stages, the performances of each stage were significantly different. The high-frequency and high-intensity vibration accelerated and aggravated the speed of the settlement and height of the ballast layer and caused the internal particles to need more time to stabilize. As the cycle times of the train load increased, the movement speed of the particles in the ballast layer gradually decreased, and their average coordination number gradually increased. After the train cyclic load lasted for a period, the stress distribution at the bottom of the ballast layer homogenized, and the stress value increased. Thus, the tamping was needed by the ballasted bed during a long-term train load. Otherwise, the stress at the bottom of the ballasted bed would surge, causing potential safety hazards to the subgrade.
Automatic Rail Surface Defects Inspection Based on Mask R-CNN
Feng Guo, University of South CarolinaShow Abstract
Yu Qian (firstname.lastname@example.org), University of South Carolina
Dimitris Rizos, University of South Carolina
Zhi Suo, Beijing University of Civil Engineering and Architecture
Rail surface defects have negative impacts on the riding comfort and track safety, even could lead to rail breakage and cause serious accidents. Based on the reports of the Federal Railroad Administration (FRA), rail surface defects have been among the main factors causing derailments. During the past decades, there have been many efforts trying to quantify rail surface defects. However, the applications of the earlier methods are limited by the high requirements of specialized equipment and personnel training. Till now, rail surface defects inspection is still a very labor-intensive and time-consuming process, which hardly satisfies the field expectations. Therefore, a cost-effective and user-friendly automatic system that can inspect the rail surface defects with high accuracy is in urgent need. To address this issue, this study proposes to use a computer vision-based instance segmentation framework to inspect rail surface defects. A rail surface database having 200 images has been built and released online. The classic instance segmentation model, Mask R-CNN is trained and evaluated. The influences of different backbones and learning rate on performance are investigated. The results indicate ResNet50 backbone has the best inspection capability. With a learning rate of 0.02, the Mask R-CNN model achieves promising performance on the bounding box and mask predictions. Defect images are used to validate the inspection performance of the developed model. The results show the proposed approach is promising, and the orientation of the rail has little impact on the inspection results. The proposed approach has the potential for future field applications.
Application of Large Datasets for Finding the Correlation between the Rate of Settlement and Changes in Geometry Indices
saeed goodarzi (email@example.com), University of Massachusetts, AmherstShow Abstract
Hamed F. Kashani, HyGround/Loram
Steven Chrismer, Amtrak-retired
Carton Ho, University of Massachusetts, Amherst
Settlement of railway tracks amplifies the dynamic loads of trains, which results in high maintenance costs and poor quality of ride. Direct settlement measurement in the field is costly due to the requirement for installing many gauges along the track. Different datasets from the northeast corridor consisting of more than 300 million elements were employed to propose a new method for measuring the rate of settlement based on the reported space curve. In this method, cross power spectral density (CPSD) was employed for determining the quality of data and choosing reliable observations for calculating the settlement. Correlations between the rate of settlement and changes in different geometry indices were investigated and equations were proposed for predicting rate of settlement by geometry indices. Results indicate that standard deviation index has the highest correlation with the rate of settlement. Track quality index (TQI) is sensitive to high-frequency noise and does not yield very accurate results when there is noise in the dataset.
New Application of Thermoplastic Polyurethane/Waste Rubber Powder Blend for Waterproof Seal Layer
Xin Xiao, Tongji UniversityShow Abstract
Jiayu Wang, Tongji University
Degou Cai, China Academy of Railway Science Corporation Limited
Liangwei Lou, China Academy of Railway Science Corporation Limited
Feipeng Xiao, Tongji University
The moisture induced damage is considered a serious threat to the durability of high-speed railway system. In this paper, an innovative application of thermoplastic polyurethane/waste rubber powder blend (TPU/WRP) with higher elastic modulus was introduced to enhance the water resistance of high-speed railway structure as a waterproof seal layer. The thermoplastic polyurethane has been proved to have excellent mechanics and waterproof properties, and the waste rubber has the advantages in cost and sustainability. The objective of this study was to design the TPU/WRP material and evaluate its moisture resistance and mechanical properties. The compressive test was used to determine the elastic modulus and the permeability test was conducted to verify waterproof performance. Further, the finite element method (FEM) was applied to analyze its mechanical responses under both dynamic and static conditions. The results indicated that TPU/WRP had excellent properties for the field utilization of transferring stress, resistance to tensile fracture, supporting the superstructure, and hence is a suitable functional waterproof material for preventing subbase and subgrade layers from moisture damage in heavy load traffic environment such as high-speed railway.
Optimization of Multi-Period Rail Procurement Plan
Hsin-Cheng Shih, National Taiwan UniversityShow Abstract
Chih-Heng Yeh, National Taiwan University
Yung-Cheng Lai, National Taiwan University
Rail is one of the most expensive assets in railway infrastructure. Hence, a well-prepared rail procurement plan could benefit the asset management. For metro system in Taiwan, the rail procurement problem includes two major uncertainties, namely, currency rates and the global steel price. In this study, we propose a scenario-generation process, deterministic and stochastic optimization models, to minimize the expected cost of the rail procurement plan. Results demonstrate that the proposed model can successfully incorporate the uncertainties and obtain the optimal procurement plan. The sensitivity analysis on budget, storage capacity, and expiration period also provides the metro operator the best strategy to further lower the procurement cost. Adopting these models to rail procurement problems can improve the process and results of rail asset management.
Pilot Study of High Speed Rail Track Monitoring using AI and Machine Learning
Konstantin Popov, University of EdinburghShow Abstract
Hwa-Kian Chai, University of Edinburgh
Robert De Bold, University of Edinburgh
Michael Forde, University of Edinburgh
James Hyslip, Loram Maintenance of Way, Inc.
Paul Long, Loram UK Ltd
SinSin Hsu, Network Rail High Speed
Carton Ho, University of Massachusetts, Amherst
Monitoring of High Speed railway track is the focus of this paper. To achieve a more efficient and targeted maintenance schedule, engineers have identified different methods of evaluating the safety of tracks. Research and testing off-track has been successful in evaluating track components. Testing on-track is key to developing a holistic over-view of a track. This paper will report a literature review followed by a successful Pilot Study of on-track testing using geometry measurements collected from geometry cars / moving measurement trains. The Pilot Study reported herein looks at a larger data set from railway lines and aims to identify similarities in the response of their geometry. Data analysis techniques used include: Statistical Models – deterministic, stochastic, and probabilistic models; Artificial Intelligence (AI) and Machine Learning (ML) techniques - Artificial Neural Networks (ANN), Support Vector Machines and Random Forest. The Pilot Study has successfully demonstrated that these tools, programmed in Python, will address the effect that different factors have on the wearing of the track and the scheduling of targeted cost-effective maintenance intervention.
Investigation of Variation of Track Response to Wheel Forces with Bogie Axle Spacing and Introduction of the Concept of Effective Track Stiffness
Erdem Balcı, Istanbul Universitesi, CerrahpasaShow Abstract
Niyazi Özgür Bezgin (firstname.lastname@example.org), Istanbul Universitesi, Cerrahpasa
Mohamed Wehbi, Network Rail
Track stiffness is an important parameter affecting the track response. Axle spacing influences the response of the track to a wheel force and hence has an effect on track stiffness. Stresses generated within the track due to train wheels within a bogie or between neighboring bogies interfere to varying degrees depending on the mechanical characteristics of the layers composing the track, axle spacing and bogie spacing. This interference affects the force-deflection characteristic of the railway track under a wheel. Dynamic impact forces that occur due to track and wheel roughness relate to track stiffness. Therefore, everything else being the same, two trains with different bogie spacing may generate different dynamic impact forces on the railway track. As a result, the accumulated damage on a railway track in time can relate to not only tonnage but also the axle spacing of the trains operated on the railway track. Through the superposition of the estimated track deflections and looking at it from a new perspective, this paper discovers a set of relations between the variations of track stiffness with bogie axle spacing. The paper introduces a new concept of effective track stiffness and hypothesizes that dynamic impact forces on the railway tracks relate to axle spacing. The paper then presents a numerical study that includes dynamic analyses of wheel and track interaction along stiffness transition zones for different values of axle spacing and shows that bogie axle spacing has an influence on dynamic impact forces on railway tracks.
DISCLAIMER: All information shared in the TRB Annual Meeting Online Program is subject to change without notice. Changes, if necessary, will be updated in the Online Program and this page is the final authority on schedule information.