This session will feature research on railroad motive power and rolling stock.
Fault Diagnosis for Rolling Bearings of a Freight Train Under Limited Fault Data: A Few-shot Learning Method
Chenzhong Li, Southwest Jiaotong UniversityShow Abstract
Ping Wang, Southwest Jiaotong University
Jiebo Li, China Academy of Railway Sciences
Haichuan Tang, CRRC
Kanghua Yang, Southwest Jiaotong University
Qing He (firstname.lastname@example.org), Southwest Jiaotong University
This paper develops a novel few-shot learning framework for the fault diagnosis of freight train rolling bearings. The proposed method has the capability to transfer the learning outcome from one bearing fault diagnosis model to another different but related task for which very limited training data are available. To this end, first, the acceleration signals of different types of bearing faults are collected by a single wheelset experimental platform. Second, we preprocess the data through data segmentation and frequency domain transformation and divide the data into training and test sets according to a certain ratio. Third, we establish two models, a fine-tuned Convolutional Neural Network (CNN) and a Conditional Wasserstein Generative Adversarial Network (C-WGAN), which are used to classify the fault types of rolling bearings with different volumes of training data. Afterward, a case study is provided to demonstrate the classification performance of the proposed models. The results show that the diagnosis capability of the traditional CNN in the frequency domain is significantly superior to that in the time domain. However, when the amount of data is small, the traditional CNN model does not work. In contrast, the few-shot learning of bearing faults works well for both the fine-tuned CNN and C-WGAN models. Further, the classification performance of the C-WGAN is better than that of the fine-tuned CNN when the training data are extremely limited.
An Online-learning Framework Using LSTMs and CDF for Traction Motor Temperature Prediction in High-speed Trains
Hao Ma, Beijing Jiaotong UniversityShow Abstract
Honghui Dong, Beijing Jiaotong University
Limin Jia, Beijing Jiaotong University
Yong Qin, Beijing Jiaotong University
Traction motor is an important part of the traction system of high-speed trains. In the current alarm system, traction motor temperature is regarded as an important indicator to evaluate the health of traction motors and monitored in real time. Based on the existing high-speed train failure prediction architecture, an online-learning framework is proposed to predict the temperature of traction motors in real time. In the proposed framework, LSTMs are used to receive various data streams and predict temperature online. To adapt to the dynamic changes of data stream, we also propose a complementary resampling process using cumulative distribution function (CDF), which can improve model stability and accuracy in online prediction. Finally, experiments are conducted based on the data of China Railway High-speed trains. The results show that models in proposed framework can accurately predict traction motor temperature online. The addition of the resampling process significantly improves the online prediction performance of the model, and this improvement becomes more obvious as the prediction length increases which will contribute to fault warning. After adaptaion and expansion, the framework can also be easily applied in other scenarios.
Study on the Maintenance Plan Optimization model of High-speed Railway ATP Equipment with Improved GA Solution
Lijuan Shi, Tongji UniversityShow Abstract
Xuan Liang, Tongji University
Huize Sun, Tongji University
The purpose of this study is to explore the appropriate maintenance intervals for ATP (Automatic Train Protection) on-board equipment of CTCS-3(Chinese Train Control System) in considering of economic loss factor due to train delay for ATP equipment faults. Both reliability-centered maintenance theory and the multi-objective optimization method are applied in this study. The field data of train delay caused by ATP equipment malfunction were collected and applied to model parameter construction. Several contributions are achieved in the following. Firstly, three categories of maintenance behavior (mechanical maintenance-(1a), adjustment or partial replacement-(1b) and overall replacement-(2p)) were put forward in reflecting the actual maintenance behavior. Secondly, a nonlinear optimization model was established with the objective function of minimizing the economic cost in sum of maintenance cost and train delay loss, while considering the constraints of availability and reliability. Finally, an improved genetic algorithm based on penalty function is proposed to determine the reasonable preventive maintenance cycles for ATP devices. The results show that the optimized maintenance intervals are variable in considering of dynamic accumulated train delay loss caused by ATP equipment malfunction, which are different from other previous studies. This investigation may help us to make more appropriate maintenance decisions in view of practice and economy. Keywords: Maintenance Plan Optimization Model, ATP Vehicle Equipment, Reliability-centered maintenance, Improved Genetic Algorithm, Delay Caused Economic Loss
Prediction Model of Train Fault Probability on Urban Rail Transit Main Line
zhenbo Wang, Tongji UniversityShow Abstract
Ke Pan, Tongji University
Yucheng Li, Tongji University
Xiafei Ye, Tongji University
To predict the train fault probability on urban rail transit main line reasonably, train formation, cumulative running kilometer and un-wheeling repair or overhaul experience are determined as the main influencing factors of train fault probability through qualitative analysis. Subsequently, discrete dataset of fault occurrence number for single train in a certain 120,000 running kilometer period is generated based on the actual data. According to the distribution characteristics of data, Poisson distribution and zero-inflated Poisson distribution are used to build three alternative models with possible function forms. After comparison and selection, a prediction model of train fault probability on urban rail transit main line based on Poisson distribution is finally proposed. The results indicate that train fault probability will increase when train formation increases, and will decrease first and then rise when cumulative running kilometer increases. When a train has been put into use, the minimum train fault probability will appear in the 4th 120,000 running kilometer period, and the initial value will be exceeded in the 7th 120,000 running kilometer period.
Total Cost of Ownership and Hydrogen Demand for Fuel Cell-Powered Railroad Locomotives
Theodore Krause (email@example.com), Argonne National LaboratoryShow Abstract
Dionissios Papadias, Argonne National Laboratory
Rajesh Ahluwalia, Argonne National Laboratory
Jui-Kun Peng, Argonne National Laboratory
Greg Moreland, Oak Ridge National Laboratory
The goal of this U.S. Department of Energy (DOE) Hydrogen and Fuel Cell Technologies Office (HFTO)-funded project is to compare the total cost of ownership (TCO) for a diesel locomotive with the TCO for a conceptual hydrogen-fueled, fuel cell-powered locomotive for three railroad applications: line haul freight, regional passenger, and switching. TCO for the diesel locomotive is based on the purchase price of the locomotive and industry average values for locomotive utilization and maintenance and fuel costs over a 30-year lifetime. TCO for the fuel cell locomotive considers three different scenarios based on the current, interim, and ultimate cost and performance targets for the fuel cell system, hydrogen storage, and hydrogen fuel as these technologies advance and manufacturing volume increase which lowers cost. Our analysis shows that the regional passenger and switcher fuel cell locomotives become cost competitive with their diesel counterparts in the ultimate scenario if performance and cost targets for the fuel cell, hydrogen storage, and hydrogen fueling can be met. However, even under our ultimate scenario, the line haul freight fuel cell locomotive is still more expensive than its diesel counterpart by 22%. Opportunities for lowering the cost and improving the performance of fuel cell and hydrogen technologies were identified. HFTO will use this information to develop technical targets for its R&D portfolio to enable fuel cell-powered locomotives to be cost-competitive with incumbent diesel locomotives in all three applications.
A Fuzzy Programming Approach for the Electric Multiple Unit Circulation Planning Problem Using Simulated Annealing
Boliang Lin, Beijing Jiaotong UniversityShow Abstract
Yinan Zhao, Beijing Jiaotong University
Jian Li, Beijing Jiaotong University
Ruixi Lin, National University of Singapore
The Electric Multiple Unit (EMU) circulation plan is regarded as one of the key operation issues for a high-speed railway transportation system. The EMU circulation plan consists of determining the connections of trains while accomplishing the passengers’ demands. It is notable that the EMUs need regular maintenance every certain kilometers or minutes for safety reason. Consequently, the circulation plan must ensure that EMU trains can reach the maintenance depots if they require maintenance in the forthcoming days. In this paper, we propose a 0-1 integer programming model for the EMU circulation plan with the aim of minimizing the total costs of the mileage losses, which is incurred by revising the EMUs before the corresponding travel mileage reaches to the cycle. Given the accumulated travel mileage of EMUs is allowed to be 10% above the standard mileage cycle in practical , an ingenious fuzzy maintenance constraint is thereby presented to tackle the mileage cycle constraint with a certain elasticity. The exterior penalty function is employed to deal with the fuzzy constraint and a simulated annealing heuristic is employed to solve the model. The modeling and solving approach are applied on a practical instance in the context of China High-speed railway system. Compared with the average travel mileage of EMU trains of the manual solution, that of the SA solution is 293 km increased. It can be then concluded that the optimization method presented in this paper can improve the quality of EMU circulation plan effectively.
A Novel Axle Temperature Forecasting Method based on Graph Convolutional Network and Gated Recurrent Unit
JIE MAN, Beijing Jiaotong UniversityShow Abstract
Honghui Dong, Beijing Jiaotong University
Xiaoming Yang, Beijing Jiaotong University
Limin Jia, Beijing Jiaotong University
Yong Qin, Beijing Jiaotong University
At present, the scale of China's high-speed railways and the number of high-speed train are increasing, which makes research on train equipment fault diagnosis and health management more and more meaningful. Bearings are one of the most prone to failure equipment on high-speed trains. The temperature of the faulty bearing will increase suddenly during the working process, so the axle temperature prediction has become a key research direction. This paper proposes a new form of axle temperature data organization, which connects axle temperature measurement point based on their locations so as to form a graph. Then, based on the Graph Convolutional Network (GCN) and Gated Recurrent Units (GRU) models, this paper proposes the GCN and GRU framework (GCG) to extract features and predict axle temperature. Finally, this paper conducts experiments based on actual data. The experimental results show that the prediction accuracy and tracking sensitivity are the best compared with current-advanced methods.
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.