UC San Diego

In the last ten years, over 1,000 train derailments that occurred in the US railroad system were caused by undetected rail internal defects that suddenly and dramatically emerged as track breakages. The total cost of those accidents is quantifiable in a few hundred millions of dollars, without considering the associated tragic losses of life and bodily injuries. While there already exist a few methods for the detection of rail internal flaws, several well-known limitations prevent each of them from detecting all of the critical flaws. As a proposal to address this issue, the object of this dissertation has been the development of a novel rail inspection system based on ultrasonic guided waves propagating through rails. Both the generation and the detection of these waves are achieved in a non-contact manner through the use of piezoelectric air-coupled transducers. One of the advantages of employing such non-contact method is the potential to perform tests as the train or inspector car travels at high speed along the railroad. Nevertheless, the main drawback of using air-coupled transducers on steel rail is represented by the significant energy losses occurring at the interface between air and steel due to the large acoustic impedance mismatch between these media. As a result, the signal to noise ratio available when analyzing the data is severely penalized. In an attempt to overcome this limitation, very effective electrical impedance networks have been designed. In parallel, a statistical analysis method based on multivariate outlier detection has been implemented to enhance the defect- sensitivity of the system. Numerical analyses of the ultrasonic wave propagation and interaction with different types of rail internal defects have been carried out using both a finite difference method, based on the Local Interaction Simulation Approach (LISA), and a commercial finite element method software. The results of these analyses were instrumental in understanding salient aspects of the guided wave propagation phenomenon in rails and throughout the ever ending decision-making process for the definition of the many system operating parameters involved. prototype based on these technologies has been built and tested both in-house at the UCSD Rail Defect Test Facility located at Camp Elliott, and in the field at the Transportation Technology Center located in Colorado, in October 2014. Receiver Operating Characteristics curves were used to characterize the performance of the defect detection based on the trade-off between defect detection rate and false alarm rate. In particular, the results of the field test were found to be quite satisfactory and perfectly in line with the predictions of the numerical analyses. Future work should be aimed at improving the prototype performance, particularly in terms of test speed, based on the lessons learnt from the October 2014 field tests and in the laboratory