Characterisation and probability of detection analysis of rolling contact fatigue cracks in rails using eddy current pulsed thermography

Abstract

With transportation volumes continuously increasing, railway networks are now facing problems of greater axle loads and increasing vehicle speeds. The most direct consequence is the initiation of rolling contact fatigue (RCF) defects in rails, which have become safety issues for all types of railway systems and received more attention due to lack of timely examination and management. Among different RCF defects, the RCF crack probably presents the biggest hazard in rails. Detection and characterisation of RCF cracks aim to provide detailed guidelines for safety management and preventative grinding. Unfortunately, current nondestructive testing and evaluation techniques are still facing several challenges and research gaps. One outstanding challenge is the characterisation of RCF cracks under their complex geometries and clustered distributions. One major research gap is how to evaluate the probabilistic performance in crack characterisation via a proper framework. By combining the advantages of eddy current pulse excitation and infrared thermography, this thesis proposes the use of eddy current pulsed thermography (ECPT) technique to address the detection and characterisation of RCF cracks in rails. To quantitatively investigate the ECPT’s performance in crack characterisation, a performance evaluation framework based on probability of detection (POD) analysis is proposed. The major contributions of the thesis are summarised as follows: (1) implementations of three-dimensional FEM models and a lab-based ECPT system for investigating the characterisation of RCF cracks under clustered distributions and geometric influences; (2) temporal/spatial-thermal-feature-based ECPT for angular slots and RCF cracks detection and characterisation; (3) investigations into the capability and the performance of ECPT for characterising angular slots and natural RCF cracks via a POD analysis framework. The thesis concludes that the proposed feature-based ECPT system can characterise RCF cracks in both light and moderate stages. Based on feature comparison and POD evaluation, tempo-spatial-based patterns are better fits for pocket length characterisation. Temporal domain-based features show better performances for inclination angle characterisation. A spatial domain-based feature, SST, can characterise vertical depths with reasonable POD values. One tempo-spatial-based pattern at the early heating stage, IET-PCA, gives the best performance for characterising surface lengths. Still, several issues need to be further investigated in future work, such as feature selection for crack characterisation, three-dimensional reconstruction of RCF cracks, model-assisted POD frameworks for improving the effectiveness of POD analysis with a limited number of physical specimens.