Railway infrastructure has a vital role in the rapid development of an economy. In the past decade, urban rail transit has been developed rapidly because the volume of railway transport has increased significantly. However, as railway usage grows, rail tracks are subjected to more stressful cycles, with heavy loads causing wear and the appearance of cracks. Consequently, train derailment could occur if such defects have propagated. According to world railway administrations, the rail track surface and subsurface defects are major causes of train accidents. Particularly for subsurface defects that are hidden inside the rail structure, continuously ignoring the growth of such defects may lead to derailment. Many types of non-destructive testing methods (NDTs) are available to detect flaws in rail tracks. They include the eddy current method (EC), ultrasonic testing method (UT) and acoustic emission method (AE). However, these methods have limitations: EC is affected by grinding marks and lift-off variations; AE cannot detect internal defects; and UT is usually restricted to low train speeds of around 20–30 mph. Therefore, there is an urgent need to develop advanced technology that can overcome the shortcomings mentioned above and provide a non-contact, long-range and fast detection of rail defects.

In this regard, laser-generated ultrasonic technology (LUT) is the latest candidate to provide a non-contact, reliable and efficient inspection of train rails. Further, LUT is an efficient method of generating guided waves (GWs), such as Rayleigh and Lamb waves. Rayleigh waves have little attenuation during their propagation along the surface of a specimen, and most of their energy is confined within a wavelength. Therefore, the Rayleigh wave can detect subsurface defects too. Further, Rayleigh waves can also travel along curved surfaces, making them ideal for rail inspection.

This research aimed to investigate the capability of laser-generated narrowband Rayleigh waves to detect surface and subsurface defects that occurred at different parts (head, web and foot) of the inspected rail track. The laser-based thermoelastic generation of narrowband Rayleigh waves were investigated in detail using the two-dimensional (2D) finite element method (FEM) to ensure their ability to detect surface and subsurface defects in railheads. In experiments, a newly designed optical system was used to convert point-source laser excitation to a line-array pattern (LAP) that helped to generate the desired narrowband Rayleigh waves. A three-dimensional (3D) SLDV (Scanning Laser Doppler Vibrometer) was used to capture the generated waves. The experiments were performed on defective rail samples that had artificial surface defects (on the head, web and foot) and circular subsurface defects (on the head and web) in the inspected rail track. The time and frequency analyses of both simulation and experimental results show that the LAP type of laser illumination is promising in generating narrowband Rayleigh waves that have great potential in detecting rail defects. In field measurements, it is usually challenging to find the location of the defect because unwanted wave packets mostly surround defect echoes in a laser-generated ultrasonic signal. In this regard, a criterion named time-of-flight-based flaw detection (TOFFD) was proposed to automatically identify defect echoes surrounded by high noise peaks. TOFFD was successfully applied to laser-generated ultrasonic signals captured at the rail track to find the location of the flaw by identifying the defect echo. Further, several experiments were performed to determine (by using TOFFD) the inspection range of the proposed non-contact system for different types of defects that occurred in the rail track. In addition, a conceptual design of LUT equipped with a six degrees of freedom robotic arm was simulated to visualise the proposed system’s future implementation for field inspection of rails. This is a pioneer application of such laser-generated Rayleigh waves for inspecting surface and subsurface defects occurring in train rails because such detailed research work has not been reported before.

The laser-generated Rayleigh wave signals measured at the railhead are usually contaminated with a high level of noise and unwanted wave packets that complicate the identification of defect-triggered echoes in the captured signals. Further, the laser-generated Rayleigh wave signals are not usually repeatable because the shape of the incident wave and the appearance of noise look different at the same detection point when time-domain signal measurements are taken at different times. To deal with these issues, the transduction system of LUT was combined with an enhanced matching pursuit (EMP) signal processing method to achieve a fully non-contact rail track inspection. Matching pursuit (MP) was enhanced by developing two novel dictionaries, which are composed of an FEM simulation dictionary and an experimental dictionary. The designs of an FEM simulation dictionary and an experimental dictionary are part of the major contribution of this research work. The EMP was used to analyse the experimentally captured laser-generated Rayleigh wave signals. The results show that the EMP is highly effective in detecting defects by suppressing noise. Further, it could overcome the deficiency in the low repeatability of the laser-generated signals. The comparative analysis of EMP by using both FEM simulation and experimental dictionaries shows that the EMP with the FEM simulation dictionary is more efficient in noise removal and defect detection from the experimental signals captured by a laser-generated ultrasonic inspection system.

In industry, rail track inspection data are mainly analysed manually to look for defects in a given signal and then classify them. This is a time-consuming, costly and error-prone activity, making the overall inspection process very tedious. In the literature, rail defect identification and classification have been automated using machine learning models to process rail image data (acquired using cameras). However, such a visual-inspection-based automated method has significant drawbacks. These drawbacks include failure to detect subsurface defects, picture data requiring a high-end GPU with extensive computational time and poor image quality influencing the accuracy of machine learning. The laser-generated Rayleigh waves are a potential candidate for rail inspection to replace video-camera-based visual inspection because they can reveal surface and subsurface defects in a non-contact and remote monitoring manner. In this regard, the completely non-contact transduction system of LUT was combined with a deep learning (DL) approach for intelligent detection and classification of railhead surface and subsurface defects. The LUT was used to actuate and capture laser-generated Rayleigh wave signals on railhead specimens to create a database of A-scan signals. The A-scan signals were captured at the railhead of a healthy rail and three rails that had a surface, subsurface and edge defects. The classification capabilities of a support vector machine (SVM), a fully connected deep neural network (DNN) and a convolutional neural network (CNN) were examined after they were applied to the pre-processed signals without extracting any statistical- or signal-processing-based characteristics. The comparative analysis demonstrates that a CNN is robust in classifying railhead defects.

In summary, the designed non-contact inspection system integrated with TOFFD, EMP and DL could automatically detect and classify rail surface and subsurface flaws. Since the system can perform in a line-to-line and long-distance inspection manner, it is superior to the slow and tedious point-to-point inspection systems that employ conventional NDT methods. The proposed system is a very strong candidate to be a next-generation railway inspection technique since it can indicate railway errors in real time and would minimise train accidents caused by derailment.