This article presents a novel transfer learning approach for evaluating structural conditions of rail in a progressive manner, by using acoustic emission monitoring data and knowledge transferred from an acoustic-related database. Specifically, the low-level layers of a model pre-trained on large audio data are leveraged in our model for feature extraction. Compared with conventional transfer learning approaches that transfer knowledge from models pre-trained on normal images, the proposed approach can handle acoustic emission spectrograms more naturally in terms of both data inner properties and the way of data intaking. The training and testing data used for rail condition evaluation contains two months of acoustic emission recordings, which were acquired from an in situ operating rail turnout with an integrated acoustic emission –based monitoring system. Results show that the evolving stages of rail from intact to critically cracked are successfully revealed using the proposed approach with excellent prediction accuracy and high computation efficiency. More importantly, the study quantitatively shows that audio source data are more relevant to the acoustic emission monitoring data than Image data, by introducing a maximum mean discrepancy metric, and the transfer learning model with smaller maximum mean discrepancy does lead to better performance. As a by-product of the study, it has also been found that the features extracted by the proposed transfer learning model (“bottleneck” features) already exhibit a clustering trend corresponding to the evolving stages of rail conditions even in the training process before any label is annotated, indicating the potential unsupervised learning capability of the proposed approach. Through the study, it is suggested that selecting source data more correspondingly and flexibly would maximize the benefit of transfer learning in structural health monitoring area when facing heterogenous monitoring data under varying practical scenarios.

Electro-mechanical impedance (EMI) has been proved as an effective non-destructive evaluation indicator in monitoring the looseness of bolted joints. Yet due to the complex electro-mechanical coupling mechanism, EMI-based methods in most cases are considered as qualitative approaches and are only applicable for single-bolt monitoring. These issues limit practical applications of EMI-based methods in industrial and transportation sectors where real-time and reliable monitoring of multiple bolted joints in a localized area is desired. Previous research efforts have integrated various machine learning (ML) algorithms in EMI-based monitoring to enable quantitative diagnosis, but only one-to-one (single sensor single bolt) case was considered, and the EMI–ML integrations are basically unnatural and ingenious by learning the EMI measurements from isolated sensors. This paper presents a novel EMI-based bolt looseness monitoring method incorporating both physical mechanism (acoustic attenuation) and data-driven analysis, by implementing a lead zirconate titanate (PZT) sensor network and a built-in graph convolutional network (GCN) model. The GCN model is constructed in such a way that the structure of the PZT network is fully represented, with the sensor-bolt distance and sweeping frequency encoded in the propagation function. The proposed method takes into account not only the EMI signature but also the relationship between the sensing nodes and the bolted joints and can quantitatively infer the torque loss of multiple bolts through node-level outputs. A proof-of-concept experiment was conducted on a twin-bolt plate, and results show that the proposed method outperforms other baseline models either without a graph network structure or does not consider sensor-bolt distance. The developed hybrid model provides new thinking in interpreting sensor networks which are widely adopted in structural health monitoring, and the approach is expected to be applicable in practical scenarios such as rail insulated joints and aircraft wings where bolt joints are clustered.

The contact condition between the wheel and the rail is paramount to the lifespan, safety, and smooth operation of any rail network. The wheel/rail contact condition has been estimated, calculated, and simulated successfully for years, but accurate dynamic measurement has still not been achieved. Methods using pressure-sensitive films and controlled air flow have been employed, but both are limited. The work described in this paper has enabled, for the first time, the measurement of a dynamic wheel/rail contact patch using an array of 64 ultrasonic elements mounted in the rail. Previous work has successfully proved the effectiveness of ultrasonic reflectometry for static wheel/rail contact determination. The dynamic real-time measurement is based on previous work, but now each element of an array is individually pulsed in sequence to build up a linear measurement of the interface. These cross-sectional, line measurements are then processed and collated resulting in a two-dimensional contact patch. This approach is able to provide not only a contact patch, but more importantly, a detailed and relatively high-resolution pressure distribution plot of the contact. Predictions using finite element methods (FEM) have also been carried out for validation. Work is now underway to increase the speed of the measurement.

Rail stress levels are vital to the lifespan of rail tracks, and are responsible for the safe operation and ride comfort of train services. In particular, wheel–rail contact stress is a dominating factor affecting wear, cracking, fatigue and failure of both wheel and rail. The wheel–rail interaction problem has long been investigated, yet detailed contact information on real cases remains obscure due to the interface complexity, including the varying wheel and rail profiles and lack of effective stress characterisation methods. Ultrasound image study, as an excellent non-destructive evaluation (NDE) method, is widely used in railway systems for defect detection, stress determination and rail profile checking. Specifically, ultrasonic reflectometry has proved successful in making static machine-element contact measurements. This article introduces a novel measuring method for both short-term and long-term dynamic wheel–rail contact monitoring purposes based on ultrasonic reflectometry. The method is investigated in detail, including the study of ultrasound propagation pathways in the rail, and the optimum placement of ultrasonic elements as well as actuator–receiver combinations. The proposed monitoring technique is expected to characterise and monitor the contact behaviour of operating high-speed rail system in real-time.

The interfacial contact conditions between a railway vehicle wheel and the rail are paramount to the lifespan, safety and smooth operation of any rail network. The wheel–rail interface contact pressure and area conditions have been estimated, calculated and simulated by industry and academia for many years, but a method of accurately measuring dynamic contact conditions has yet to be realised. Methods using pressure-sensitive films and controlled air flow have been employed, but both are limited. Ultrasonic reflectometry is the term given to active ultrasonics in which an ultrasonic transducer is mounted on the outer surface of a component and a sound wave is generated. This ultrasonic wave packet propagates through the host medium and reflects off the contacting interface of interest. The reflected waveform is then detected and contact area and interfacial stiffness information can be extracted from the signal using the quasi-static spring model. Stiffness can be related to contact pressure by performing a simple calibration procedure. Previous contact pressure measurement work has relied on using a focusing transducer and a two-dimensional scanning arrangement which results in a high-resolution image of the wheel–rail contact, but is limited to static loading of a specimen cut from a wheel and rail. The work described in this paper has assessed the feasibility of measuring a dynamic wheel–rail contact patch using an array of 64 ultrasonic elements mounted in the rail. Each element is individually pulsed in sequence to build up a linear cross-sectional pressure profile measurement of the interface. These cross-sectional, line measurements are then processed and collated resulting in a two-dimensional contact pressure profile. Measurements have been taken at different speeds and loads.