|Towards the understanding of railway flange squeal : phenomenon, mechanism, and hybrid model-enabled global sensitivity analysis
|Ni, Yi-qing (CEE)
Zhou, Lu (CEE)
|Railroad tracks -- Noise
Railroad trains -- Noise
Railroads -- Management
Hong Kong Polytechnic University -- Dissertations
|Department of Civil and Environmental Engineering
|xxvi, 238 pages : color illustrations
|Wheel-rail squeal noise during train curving is a tricky problem influencing passengers and nearby residents. The tonal curve squeal that mainly occurs at the inner rail has been extensively studied, while the broadband flange squeal that takes place at the outer rail that can be equally annoying is seldom mentioned and hardly investigated. This first part of this thesis presents a study aiming at understanding the mechanism of wheel-rail flange squeal from phenomenon to the underlying source. A systematic demonstration is presented to establish connection between the observed wheel-rail flange squeal noise and contact force, with measurements taken from a series of well-controlled in-situ experiments on an operating metro line, and numerical simulations. An integrated transient model consisting of three submodels on train dynamics, rail dynamics, and wheel-rail contact incorporating genuine 3D surface irregularities is developed specifically for flange squeal analysis. The developed model can reproduce the characteristics of flange squeal and reasonably correlates wheel-rail contact force with the flange squeal noise. Features of flange squeal under four-speed levels corresponding to in-situ experiments are reappeared from global dynamics to local contact status. An explanation of the generation mechanism of flange squeal is proposed by measurement observation and confirmed through the proposed transient flange squeal analysis model with extensive parameter discussions on multiple influencing key factors, including train speed, contact position, shape of contact patch, contact creepages, and wheel and rail irregularities.
Wheel-rail flange squeal is influenced by not only independent system parameters but also their mutual interactions. The intense level of flange squeal is hard to be thoroughly investigated with a deterministic flange squeal analysis model. Global sensitivity analysis (GSA) is therefore adopted and devised to carry out uncertainty quantification on input parameters to the output wheel-rail flange squeal level, which serves as the second part of this thesis. A more comprehensive investigation, with special attention paid to those adjustable parameters, is carried out. Considering the high computational cost of the proposed flange squeal analysis model, a surrogate model based on polynomial chaos expansion (PCE) and sequential sampling strategy are utilized to reduce the sampling requirement during training. Sensitivity indices are obtained with the trained PCE-based surrogate model, and the importance of parameters on the wheel-rail flange squeal is quantified. The quantified results are finally used for carrying out optimization design, demonstrating the effectiveness and reliability of the proposed PCE-based surrogate model and the results of GSA.
Although GSA has successfully been utilized for capturing the predominant factors and the optimization design has also shown effectiveness in mitigating wheel-rail flange squeal, compromises were still made in the current investigation. Only the front outer wheel was taken as a potential squealing source because of the computational cost of wheel-rail rolling contact analysis. A more efficient flange squeal analysis model with considerations of multiple potential squealing wheels and multiple train vehicles as well as the dynamic interactions among vehicles are expected. At the current stage, one of the largest obstructions is the computationally demanding wheel-rail rolling analysis component integrated in the flange squeal analysis model, which although is an exact solution strategy, requires iterative large size stress-deformation matrices manipulation for any one of the contact solutions. As a result, this dilemma enlights the author to apply a neural network model to surrogate the computationally demanding wheel-rail rolling contact simulation component. An accurate, computationally efficient, and manageable surrogate model that is capable of capturing critical characteristics of wheel-rail rolling contact problems is developed and validated in the last part of this thesis. Three sub-network, including the stress magnitude net with MAE loss function, stress distribution net with modified MSE loss function, and stick-slip classification net with cross-entropy loss function, are adopted to formulate the neural network model and named as CONTACTNet. Comprehensive comparisons are conducted to reveal the necessity and effectiveness of the physical-based data preprocessing inputs, specially designed sub-network, and loss functions. It is demonstrated that the CONTACTNet, even with an entry-level GPU, could provide up to 20 times computational gain over the CONTACT, especially for practical simulation circumstances that take into account an in-service 8-vehicle train or 16-vehicle train with a large number of wheelsets. Great potential and benefits of implementing machine learning techniques to establish a surrogate model for wheel-rail rolling contact simulation are presented.
The present study is expected to establish a footstone for the mechanism comprehension of flange squeal and inspire further research in this area. New technologies in other research fields, such as statistical analysis, machine learning, etc., have been proven to be effective in the flange squeal investigations, which provide a new research vision for the relevant conventional structural dynamic analysis.
|All rights reserved
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: