| Author: | Dang, Dazhi |
| Title: | Machine learning-empowered, ultrasonic guided wave-based testing and evaluation methods for railway tracks |
| Advisors: | Ni, Yi-qing (CEE) Wang, You-Wu (CEE) |
| Degree: | Ph.D. |
| Year: | 2025 |
| Subject: | Railroad tracks -- Inspection Ultrasonic testing Machine learning Hong Kong Polytechnic University -- Dissertations |
| Department: | Department of Civil and Environmental Engineering |
| Pages: | xxxi, 262 pages : color illustrations |
| Language: | English |
| Abstract: | The safe and efficient operation of modern railway systems relies heavily on robust health monitoring and defect inspection techniques to mitigate potential risks posed by track defects. Traditional ultrasonic inspection methods, while effective in identifying various types of railway defects, face significant limitations in terms of inspection efficiency, sensor durability, and adaptability to the expanding scale of railway networks. Thereupon, this thesis addresses these challenges by developing advanced guided wave testing (GWT) techniques for railway track inspections, integrating innovative sensing technologies and state-of-the-art signal processing methods to enhance the accuracy, reliability, and efficiency of rail defect detection and health evaluation. The research first addresses the challenge in ultrasonic sensing efficiency and reliability. To fully adapt to the harsh environment of railway on-site monitoring, a novel hybrid sensing system for railway GWT is first proposed in Chapter 3, combining piezoelectric (PZT) actuators and fiber Bragg grating (FBG) sensors to generate and detect ultrasonic guided waves (UGWs). This system leverages the inherent advantages of FBG sensors, including their electromagnetic interference (EMI) resistance, durability, and cost-effectiveness. A high-speed optical interrogation strategy based on edge filters is designed to optimize the performance of FBG sensors in capturing UGW signals. Numerical simulations using COMSOL Multiphysics and experimental validations are conducted to demonstrate the system’s effectiveness in detecting rail defects, with a particular focus on wave propagation characteristics and defect sensitivity. The results reveal that UGWs generated and received by the proposed hybrid system can propagate on the rail interface and are highly sensitive to rail geometric inconsistencies. Building on this hybrid sensing system, this thesis introduced comprehensive defect detection and identification frameworks to overcome the challenges in features extraction posed by perplexing UGW signals. A defect detection and evaluation framework is introduced in Chapter 4, utilizing nonlinear autoregressive models with exogenous inputs (NARX) and a probabilistic damage-sensitive feature (DSF) derived from the probability density function (PDF) of network prediction errors. Experimental studies are conducted to validate the framework, with three different types of railway defects configured to generate training and testing datasets. The NARX models are optimized through hyperparameter fine-tuning, achieving high detection accuracy. Large-scale testing further validates the framework, demonstrating an overall accuracy of 98.0% and confirming its robustness and effectiveness in real-world applications. For further defect identification, an orthogonal matching pursuit (OMP)-based method is developed in Chapter 5, incorporating a customized interfering reflection components (IRC) dictionary to reconstruct defect-related reflective waves. This method leverages the sparse representation capabilities of the OMP algorithm to accurately identify crack locations. Numerical investigations are first conducted to verify the effectiveness of the proposed approach, followed by experimental validations on a railway track segment. The results show that the proposed method can accurately predict crack locations with fitting errors of less than 6 mm, highlighting its potential for practical engineering applications. Comparative studies further demonstrate the superiority of the IRC dictionary in improving defect identification accuracy. Finally, Chapter 6 proposes a rapid railway track diagnosis approach, using pencil lead break (PLB)-induced ultrasound as a cost-effective and portable excitation source for on-site inspections. This method is combined with adversarial autoencoders (AAEs) to process and analyze the ultrasonic signals, enabling the detection of subtle changes indicative of rail defects. A probabilistic damage indicator based on the Jensen-Shannon Divergence (JSD) is developed to evaluate rail health status. The proposed approach is validated through comparative studies in both laboratory and field tests, achieving 95.5% accuracy for intact rails and 97.3% for defective ones. The results demonstrate the method’s potential for on-site inspections, offering a practical solution for rapid and reliable rail defect detection. The proposed methods in Chapter 4–6, though developed based on machine learning (ML) algorithms, are not label-dependent, making them suitable for prospective engineering applications. In conclusion, this thesis makes significant contributions to the field of railway GWT by addressing key research gaps in sensing instrumentation, signal processing, and defect detection methodologies. The proposed techniques offer more efficient, durable, and reliable inspection methods for railway tracks. The findings have the potential to revolutionize intelligent railway maintenance strategies, enhancing safety and reducing operational costs. Future work is recommended to further validate these methods under real-world conditions, explore their scalability for large-scale railway networks, and investigate their applicability to a wider range of defect types and sizes. |
| Rights: | All rights reserved |
| Access: | open access |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/14176