Elsevier

Measurement

Volume 182, September 2021, 109672
Measurement

A feature engineering framework for online fault diagnosis of freight train air brakes

https://doi.org/10.1016/j.measurement.2021.109672Get rights and content

Highlights

  • This study diagnoses automatic train air brake systems with data collected from a three-car in-lab experiment platform.

  • A divided-and-integrated framework is proposed for car-level and component-level fault location.

  • A feature selection algorithm based on modified reinforcement learning is developed.

  • The outcome of this study will benefit existing automatic air brake system fault diagnosis.

Abstract

Automatic air brake systems are widely used in freight train braking to ensure railway operation safety. Various types faults pose an enormous threat to freight operations. Existing algorithms lack a unified framework for generating key features. In this research, we propose a novel feature engineering framework for the fault diagnosis of freight train air brakes. First, experimental data are collected through a three-car in-lab experimental platform. Second, a peak detection method combined with first-order difference function to partition and classify the air pressure time series into the braking phase and releasing phase. Third, a divided-and-integrated framework is designed for feature engineering. Feature selection is carried out via a modified reinforcement learning method. Finally, multiple machine learning algorithms are explored and the results indicate that random forest method shows the best performance. The proposed model achieves about 99% accuracy for car-level fault detection and over 94% accuracy for component-level fault diagnosis.

Introduction

As a vital guarantee of railway trains' safe operation, working state recognition of the braking system is critically important [1]. Due to the novel internal configuration, any potential safety hazard or fault-related deterioration may influence the train braking system's integrity and durability and even cause significant losses. To prevent the interference of brake force transmission faults, it is necessary to diagnose and locate the faults in a timely and accurate manner. However, the current fault diagnosis method relies heavily on human experience and is often laborious, cumbersome, and time-consuming [2]. Additionally, the accuracy of manual recognition is relatively low. Thus, the development of an automated diagnosis system that includes braking phase identification and car-level and component-level fault diagnosis while not interfering with train operations is an urgent need but remains a challenge.

Train braking systems have been studied extensively since 1867 [3]. Braking systems can be divided into pure pneumatic brake systems and electro pneumatic brake systems, depending on how the brake command is performed. At present, pure pneumatic brake systems are widely used in freight trains, while electric brake systems are mainly adopted in electric multiple units (EMUs). George Westinghouse invented the first conventional pure pneumatic brake system, which has since been further developed and changed many times [4]. The pure pneumatic brake used for freight trains refers to the automatic air brake system [5]. Questions have been raised about the safety of their prolonged use. In particular, pure pneumatic brake system failure has become one of the main obstacles to achieving high speed and heavy haul capabilities for freight trains, which has received considerable attention. The failure of any of the components in a pure pneumatic brake device decreases overall performance and increases the need for system revision. For example, the leakage of pipelines in a pure pneumatic brake system drastically affects braking performance by increasing the time required to reach the steady state, leading to longer stopping distances. In modern freight railways, the brake system not only affects operational safety but also restricts the further improvement of train speed and traction quality. However, so far, there has been little discussion about the method involved in pure pneumatic brake fault diagnosis.

Fig. 1 presents the relationships and connections between the parts of a freight train's automatic air brake system. As indicated in the figure, the automatic air brake system utilizes pressurized air provided by an external air compressor and piped to the air reservoir as the operating medium. The pipeline transmits the pressurized air along the entire length of the train. The brake plays a foundation role in the system, which uses the air pressure variations of the pipeline to generate a control signal and force for each car [6]. A typical brake usually contains a three-way valve, brake cylinder, auxiliary air reservoir, and foundation braking equipment to carry out its main functions. In practice, when the air pressure in the pipeline decreases, the three-way valve allows pressurized air from the auxiliary air reservoir into the brake cylinder. Then, the brake cylinder acts on the brake riggings, resulting in clamping shoes on the wheels. Conversely, releasing actions are commanded by increasing the air pressure in the pipeline such that a pneumatic connection is established between the brake cylinder and atmosphere via the three-way valve, leading to the release of the braking force on the riggings.

The performance of the brake is related to the stable operation of the entire railway system. Fault diagnosis can increase the stability and reliability of the railway train brake system [7]. Automatic air brake systems have been developed into very complex structures. This poses the main difficulty in fault diagnosis. Empirical observations have suggested that approximately 70–90% of maintenance time is related to fault diagnosis in the process of unexpected failure handling, and 10–30% is allocated to fault repair by a skilled worker [8]. Automatic brake fault diagnosis has become an emergent task for researchers and railway managers.

Overall, there still exist two major challenges in fault diagnosis of automatic air brake systems. First, numerous components have been introduced into automatic air brake systems, which are tightly coupled. In this case, the fault types for these components appear in various forms [9]. In addition, different fault types for the same component may produce different data phenotypes. Second, a freight train often consists of dozens or even hundreds of cars. The automatic air brake system relies only on a single pipeline to transmit pressurized air, which is initiated from the locomotive and has to be sent to all cars along the train. Due to each car's successive connection, a series of failures is simultaneously observed of the entire brake system. Such complexity and dependence create the ambiguity of fault location, causing difficulties in automatic fault diagnosis. To mitigate such phenomena, it is very important to design a fault diagnosis system with a straightforward strategy that preferably does not rely on the human operators’ experience to reduce cost and increase efficiency.

Many studies on brake fault diagnosis have been conducted [10], [11], [12], [13]. Air pressure readings, which are directly used for brake control, could act as a monitoring signal for automatic air brake systems. This would reveal a variety of phenomena under different conditions. For several years, many methods for air pressure feature extraction have been proposed [14], [15], [16]. Nevertheless, the features generated by past studies may only be effective for a certain defect at a certain stage and do not reflect the full picture of the dynamics. Most of the fault diagnosis algorithms studied in previous works have suffered from the lack of an interpretable and systematic feature engineering framework. The framework should consider the automatic air brake system's operating principles to select the most representative features from the air pressure signal. Designing a framework has become one of the key tasks in fault diagnosis of freight train air brakes. Motivated by the above discussions, we propose a novel framework in which the air pressure at key positions in an automatic air brake system is monitored. With the price reduction and miniaturization trends of electronic devices, the prospect of deploying sensor networks on-board freight trains is promising.

To address the above-noted gaps in the research to date, our primary objective is to develop an appropriate feature engineering framework for air brake fault diagnosis. Under this objective, we study both the detection of abnormal cars along a freight train and the diagnosis of faulty components in an abnormal car, referred to hereafter as car-level fault and component-level fault diagnosis. First, experimental data are collected through a three-car data acquisition platform equipped with a sensor network. Second, air pressure data are partitioned and classified into two phases: braking and releasing. Third, considering the characteristics of strong dependence on air pressure between adjacent freight cars, this study designs a divided-and-integrated framework to identify the corresponding features for fault diagnosis. At the same time, a modified reinforcement learning-based feature selection is used to select the optimum combinations to solve the dependence of features. Fourth, the proposed car-level and component-level fault diagnosis framework is verified with various machine algorithms according to the diversity of fault types.

The main contributions of this paper are as follows:

  • A three-car in-lab experimental platform was developed with a real-world automatic air brake system. Such a platform can simulate braking and releasing faults that are close to the real-world situation. To comprehensively reflect each component's air pressure variation, we established a car-level sensor network consisting of five air pressure sensors distributed to crucial locations in each car. Thus, the method improves the fault diagnosis system capacity through simulation of numerous types faults and conditions.

  • Considering the automatic air brake system's characteristics, a divided-and-integrated framework is proposed for car-level and component-level fault diagnosis. The divided features can identify abnormal cars along a train with relatively higher accuracy. Integrated features are extracted afterward to classify faulty component types in abnormal cars based on the detected states and connections between cars.

  • A feature selection algorithm based on modified reinforcement learning is developed to reduce the feature subset size in which irrelevant, noisy, and redundant features are discarded.

The rest of this paper is organized as follows: Section 2 presents a brief literature review of various studies and reports findings for a freight train braking system's fault diagnosis. Section 3 proposes a feature engineering-based framework for fault diagnosis. Section 4 presents tests conducted on an experimental platform and discusses the results. Finally, Section 5 gives some conclusions and future research recommendations.

Section snippets

Literature review

Previous fault diagnosis methods rely heavily on manual inspection. Minor faults that are easily ignored will cause damage and may even result in failure of the brake system over a long service time. With sensor technology development, many improved fault diagnosis methods have been proposed and have achieved better performance in on-board condition monitoring for freight railway cars [17]. Overall, the brake system diagnosis methods can be divided into model-based, knowledge-based, and

Methodology

Fig. 3 shows the proposed architecture of the automatic fault diagnosis model. The model conducted in this study can be divided into four major stages: (a) In the data acquisition stage, after establishing a three-car in-lab experiment platform, we deploy the air pressure sensor network at the critical points of each car. The sensed air pressure is collected as training datasets for fault diagnosis. (b) The next stage is phase partitioning and classification, in which the phase is divided into

Experiments and results

The air brake system sensor readings were acquired from experiment cars with the help of a data collection system. The data collection system comprises several piezoelectric transducers to measure air pressure value across critical locations in the automatic air brake system.

Conclusions and future work

In this study, we have developed a feature-engineering framework for fault diagnosis in automatic air brake systems. It is demonstrated that the divided-and-integrated framework we proposed can improve the fault diagnosis accuracy. With modified reinforcement learning-based feature selection, the model focuses on whether the feature should be part of our model. We studied its performance concerning other traditional state-of-the-art methods and verified using the three-car in-lab experimental

CRediT authorship contribution statement

Qihang Wang: Writing - original draft. Tianci Gao: Writing - original draft. Haichuan Tang: Writing - original draft. Yifeng Wang: Writing - original draft. Zhengxing Chen: Writing - original draft. Jianhui Wang: Visualization, Investigation. Ping Wang: Writing - review & editing. Qing He: Conceptualization, Methodology, Writing - review & editing.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgment

This work was supported by National Science Foundation of China (U1934214, 51878576), Sichuan Science and Technology Project NO. 2020YFG0049, and CCRC Research Institute Co., Ltd..

References (51)

  • S.W. Usselman

    Air brakes for freight trains: technological innovation in the American railroad industry, 1869–1900

    Business History Review

    (1984)
  • C. Cruceanu and X. Perpinya, “Train braking,” Reliability and safety in railway, InTech, pp. 29–74,...
  • L. Jun-Hong et al.

    Development of braking system on freight cars and the present condition of operation in China

    Railway Locomotive & Car

    (2006)
  • S. rong Yan, Research and Application on Key Technologies of Fault Diagnosis Expert System for Train Brake, Master’s...
  • H. Sheng et al.

    Analysis of the root reasons affecting safety of 120-type valve through statistics of freight Wagon’s brake faults

    Railway Locomotive & Car

    (2013)
  • Z.D. Hua, J. H. Quan, H. E. Xiao, T. University, Fault diagnosis techniques for the information control system of...
  • A. Manghai et al.

    Brake fault diagnosis through machine learning approaches – a review

    Struct. Durability Health Monit.

    (2017)
  • H. Chen et al.

    A review of fault detection and diagnosis for the traction system in high-speed trains

    IEEE Trans. Intell. Transp. Syst.

    (2019)
  • M. Zhang, Fault Diagnosis on Train Brake System Based on Multi-dimensional Feature Fusion and GBDT Enhance...
  • D. Pei, Research on Fault Diagnosis of Railway Wagon Air Brake System Based on Bayesian Network, Master’s Thesis,...
  • W. guang Liang, Research on Fault Diagnosis of Railway Freight Car Braking System Based on PHM Technology, Master’s...
  • E. Bernal et al.

    Onboard condition monitoring sensors, systems and techniques for freight railway vehicles: a review

    IEEE Sens. J.

    (2018)
  • X.T. Zhuan and others, Optimal handling and fault-tolerant speed regulation of heavy haul trains, PhD Thesis,...
  • Q. Wu et al.

    Railway air brake model and parallel computing scheme

    J. Comput. Nonlinear Dyn.

    (2017)
  • S. Ramarathnam et al.

    Development of a model for an air brake system with leaks

    Am. Control Conf.

    (2009)
  • Cited by (17)

    • Method to enhance time series rolling fault prediction by deep fast Fourier convolution

      2024, Measurement: Journal of the International Measurement Confederation
    View all citing articles on Scopus
    View full text