This study focuses on the acoustic emission wave classification for the sake of more accurate and comprehensive rail crack monitoring in the field typically with complex cracking conditions, high-operational noise, and mass data. There are mainly three types of acoustic emission waves induced by operational noise, impact, and crack propagation, respectively. Synchrosqueezed wavelet transform was introduced to represent intrinsic characteristics of acoustic emission waves more clearly in the time–frequency domain, where acoustic emission waves induced by different mechanisms were found to show various patterns of energy distribution. Then, a multi-branch convolutional neural network model with two branches was developed to automatically classify the three types of acoustic emission waves by taking into account their synchrosqueezed wavelet transform plots in various time–frequency scales. Training, validation, and test data sets were constructed using acoustic emission waves collected through a series of field and laboratory tests with various noise levels and loading conditions. The transfer learning was used to train the model faster, and the Bayesian optimization algorithm was applied to tune the hyperparameters. Finally, the multi-branch convolutional neural network model achieved higher accuracy and robustness than the traditional convolutional neural network model of single branch in identifying different acoustic emission mechanisms. The proposed acoustic emission wave classification method based on synchrosqueezed wavelet transform and multi-branch convolutional neural network is able to detect not only surface rail cracks, where both impact-induced and crack propagation-induced acoustic emission waves would be identified, but also internal rail cracks where only crack propagation-induced acoustic emission waves would be captured.

This article focuses on the rail crack monitoring using acoustic emission technique in the field typically with complex cracking conditions and high operational noise. A novel crack monitoring strategy based on Tsallis synchrosqueezed wavelet entropy was developed, where synchrosqueezed wavelet transform was introduced to explore the time–frequency characteristics of acoustic emission signals and Tsallis entropy was adopted to quantify the local variation of acoustic emission wavelet coefficients more accurately. The mother wavelet of synchrosqueezed wavelet transform and three key parameters of time-Tsallis synchrosqueezed wavelet entropy, including characteristic frequency band, non-extensive parameter, and time window length, were appropriately determined. The performance of the strategy was validated through field tests with an incipient rail crack and trains running at operating speeds. Time-Tsallis synchrosqueezed wavelet entropy efficiently detected and located the crack by extracting the crack-related transients in acoustic emission signals that were easily submerged in the operational noise. Synchrosqueezed wavelet transform further helped to analyze the mechanisms of these crack-related transients, which were distinguished to be either crack propagation or impact. The experimental results demonstrated that the crack monitoring strategy proposed is able to detect both surface and internal rail cracks even in the noisy environment, highlighting its potential for field applications.

A high-strength bolt connection is the key component of large-scale steel structures. Bolt loosening and preload loss during operation can reduce the load-carrying capacity, safety, and durability of the structures. In order to detect loosening damage in multi-bolt connections of large-scale civil engineering structures, we proposed a multi-bolt loosening identification method based on time-frequency diagrams and a convolutional neural network (CNN) using vi-bro-acoustic modulation (VAM) signals. Continuous wavelet transform was employed to obtain the time-frequency diagrams of VAM signals as the features. Afterward, the CNN model was trained to identify the multi-bolt loosening conditions from the raw time-frequency diagrams intelligently. It helps to get rid of the dependence on traditional manual selection of simplex and ineffective damage index and to eliminate the influence of operational noise of structures on the identification accuracy. A laboratory test was carried out on bolted connection specimens with four high-strength bolts of different degrees of loosening. The effects of different excitations, CNN models, and dataset sizes were investigated. We found that the ResNet-50 CNN model taking time-frequency diagrams of the hammer excited VAM signals, as the input had better performance in identifying the loosened bolts with various degrees of loosening at different positions. The results indicate that the proposed multi-bolt loosening identification method based on VAM and ResNet-50 CNN can identify bolt loosening with a reasonable accuracy, computational efficiency, and robustness.

The acoustic emission (AE) technique is a promising approach for detecting and locating fatigue cracks in metallic structures such as rail tracks. However, it is still a challenge to quantify the crack size accurately using this technique. AE waves can be generated by either crack propagation (CP) or crack closure (CC) processes and classification of these two types of AE waves is necessary to obtain more reliable crack sizing results. As the pre-processing step, an index based on wavelet power (WP) of AE signal is initially established in this paper in order to distinguish between the CC-induced AE waves and their CP-induced counterparts. Here, information embedded within the AE signal was used to perform the AE wave classification, which is preferred to the use of real-time load information, typically adopted in other studies. With the proposed approach, it renders the AE technique more amenable to practical implementation. Following the AE wave classification, a novel method to quantify the fatigue crack length was developed by taking advantage of the CC-induced AE waves, the count rate of which was observed to be positively correlated with the crack length. The crack length was subsequently determined using an empirical model derived from the AE data acquired during the fatigue tests of the rail steel specimens. The performance of the proposed method was validated by experimental data and compared with that of the traditional crack sizing method, which is based on CP-induced AE waves. As a significant advantage over other AE crack sizing methods, the proposed novel method is able to estimate the crack length without prior knowledge of the initial crack length, integration of AE data or real-time load amplitude. It is thus applicable to the health monitoring of both new and existing structures.