Sambo, Bello (2019) Incorporating automated rail fatigue damage detection algorithms with crack growth modelling. Doctoral thesis, University of Huddersfield.
Abstract

This thesis examines the feasibility of incorporating Non Destructive Testing (NDT) of rail surface damage by means of combining image processing with damage prediction models. As rail traffic and adherence to safety measures become increasingly strict on the network, the associated maintenance cost of rail infrastructure must be kept at a minimum. Proactive maintenance is crucial to maintaining the competitive advantage of rail transport. A considerable amount of research has been done on improving the practical tediousness associated with popular condition monitoring techniques in rail industry e.g. Ultrasonic, and Eddy current method. This thesis aims to fill the gap of yet to be explored benefit, of combining detection and prediction of RCF damage. This research project will contribute to the rail industry by simplifying maintenance operations and support decision making.

In this thesis, a summary of existing image-based NDT and crack growth models is presented as a foundation on which the novel application is built.It could be said that similar research mainly focuses on quantifying severity of damage without predicting crack behaviour. The simulated results of the proposed image processing algorithm confirm superiority of local illumination invariant enhancement, multi-window segmentation, and cascaded feature extraction. The influential parameters of these methods are consistent within each image data set but differ across all sets. This is observed to be as a result of difference in environmental and reflection properties of acquired images.A sensitivity analysis of the proposed algorithm on data set 2 suggests a non-linear relationship between severity of damage and pixel mean intensity including variance.

Taking to account fracture mechanics aspect of this thesis, the influence of crack geometry on growth rate and path has been established by case study of newly initiated and critically grown cracks. It was further established that larger cracks are observed to grow faster than smaller ones. In addition, the influence of track curve radius and supporting structures on wheel rail contact dynamics is well understood from the structural mechanic’s tests related to contact forces and bending moment. These translate to increase or decrease in contact stresses, strains, and the propagation rate of defects.

Unlike other predictive models, the method developed in this thesis focuses on replicating the actual surface condition of the rail prior to estimating the fracture parameters (using detailed 3D Finite Element model) that dictate residual life of the rail asset. The model makes it possible to combine two separate maintenance activities i.e. detection and prediction without inducing down time of the service.

A direct impact of this novel application is the utilisation of the actual crack boundary for prediction of fracture behaviour. It is insinuated that stress distribution of actual crack boundary differs from elliptical equivalent assumptions.

Further work would include improving detection aspect of the novel application to avoid intersecting boundary coordinates, which are not readily imported into the Linear Elastic Fracture Mechanics (LEFM) prediction model. It is also beneficial to expand the prediction aspect of the research work to include influence of neighbouring cracks and fluid entrapment for more flexible analysis of other environmental and contact conditions. To improve on current work, it will be useful to conduct laboratory investigations on the influence of Image Acquisition System (IAS) light source in relation to illumination inequality within the captured image. Also fracture mechanics experimental validation can be used to assert the accuracy of the method

Information
Library
Documents
[img]
Preview
Sambo THESIS.pdf - Accepted Version
Available under License Creative Commons Attribution Non-commercial No Derivatives.

Download (24MB) | Preview
Statistics

Downloads

Downloads per month over past year

Add to AnyAdd to TwitterAdd to FacebookAdd to LinkedinAdd to PinterestAdd to Email