Artificial intelligence technology using smart phones for train passenger ride comfort

Huang, Junhui (2024). Artificial intelligence technology using smart phones for train passenger ride comfort. University of Birmingham. Ph.D.

Huang2024PhD.pdf
Text - Accepted Version
Available under License All rights reserved.
Download (6MB)

Abstract

This thesis presents a novel approach to evaluating Ride Comfort (RC) in railway systems by integrating Machine Learning (ML) algorithms with data collected from smartphone sensors. Traditional methods for assessing RC, such as questionnaires and vibration-based measurements, have significant limitations, including reliance on subjective feedback, and complex logistics. These challenges are addressed by developing a scalable, cost-effective, and accurate framework that leverages modern smartphones' widespread availability and advanced capabilities.
The research begins by exploring the potential of ML algorithms in processing large datasets generated from both synthesised data using D-track simulations and real-world data collected via smartphones. The study adheres to the International Organisation for Standardisation (ISO) 2631 standard and the International Union of Railways (UIC) 513 for vibration measurement, ensuring the reliability and accuracy of the data. The ML models developed, including Convolutional Neural Network (CNN), K-means clustering, and ensemble techniques, are employed to quantify dynamic track stiffness, classify various train motions (such as roll, yaw, pitch, and bounce), and evaluate RC at individual points within the train.
A key innovation of this work is using crowd-sourced data from multiple smartphones with Graph Attention Network (GAT) to perform a comprehensive assessment of RC at the train level rather than merely at the individual passenger level. This approach facilitates the transmission of real-time data to a processing centre for subsequent analysis, generating valuable insights to improve passenger experience and operational efficiency, all while ensuring driver focus and maintaining safety standards. The study also demonstrates that this methodology significantly reduces the financial and logistical burdens typically associated with traditional methods while offering a more holistic and nuanced RC evaluation.
The findings of this research have broad implications for the railway industry, particularly in improving infrastructure maintenance, optimising train operations, and enhancing overall RC. By providing a robust framework for RC assessment, this thesis contributes to advancing intelligent transportation systems and supports the future development of more comfortable and efficient railway services. The use of advanced ML models, such as CNN and GAT, in conjunction with crowd-sourced data from multiple smartphones, enables a comprehensive and dynamic assessment of RC at both individual and train-wide levels. This approach allows for more accurate quantification of crucial factors such as track stiffness and train motion, which are critical to understanding and enhancing RC. The cost-effectiveness of the methodology, driven by the use of widely available smartphones, significantly reduces the financial and logistical burdens associated with traditional methods, making it accessible to a wide range of railway services. The research also aligns with international standards, ensuring its applicability across diverse operational contexts. Beyond the railway industry, the findings have broader implications for other modes of transport, such as buses and trams, where similar methodologies could be applied to enhance comfort and operational performance. The adoption of this ML-driven framework not only promises to improve passenger satisfaction and safety but also supports the development of sustainable, passenger-centred transportation systems.

Type of Work:

Thesis (Doctorates > Ph.D.)

Award Type:

Doctorates > Ph.D.

Supervisor(s):