Accurate and efficient assessment of structural damage in shield tunnels is essential for ensuring the safety and reliability of transportation systems. Cracks in tunnel linings are common, necessitating regular structural integrity assessments to ensure safety. Traditional modeling of such damage is often complex and time-consuming. Therefore, the objective of this study is to automate the entire process from detecting tunnel damage in images to conducting numerical analyses for shield tunnels, thereby enabling rapid assessment of structural integrity. We propose a segment-based method that updates a finite element (FE) model of shield tunnels to reflect geometric changes due to cracks, utilizing computer vision (CV) techniques and geometric analyses. Firstly, the Segment Anything Model, along with CV techniques, is used to identify the shapes and sizes of tunnel components from full and partial tunnel segment images. Then, a Dual VMamba U-Net (DVMamba-UNet) is proposed to identify cracks and provide detailed crack information, i.e., crack masks. Finally, geometric analysis is employed to develop algorithms that automatically transform coordinates and select elements within FE models, facilitating the update of geometric changes. Residual capability assessments of updated FE models are used to evaluate the structural damage and the tunnel segment condition. Two case studies are conducted to verify the effectiveness of the proposed approach and algorithms. The results show that the proposed method allows for automatic updates to the FE tunnel model based on damage detected in images through CV techniques and geometric analyses. Additionally, updated FE tunnel models representing different damage levels are developed and analyzed using numerical simulations. This approach not only proves effective in evaluating structural damage in shield tunnels but also offers potential as a data processing and model updating modules within future Digital Twin frameworks for tunnel infrastructure.

Wheel flats (WF) are a common defect in railway systems, posing risks to operational safety, passenger comfort, and the longevity of infrastructure. Existing detection methods face significant challenges, including sparse labeled data, high noise interference, and limited adaptability to complex operational conditions. To address these issues, this study introduces a semi-supervised learning workflow integrating multi-sensor data from Distributed Acoustic Sensing (DAS) and accelerometers, with a novel Graph Vector-Quantization Variational AutoEncoder (GVQVAE) as the core component. The model combines time-frequency analysis for feature extraction, a graph-based architecture for data fusion, and a vector quantization mechanism to effectively leverage both labeled and unlabeled data. Experimental results from an operational subway system demonstrate the model’s robustness and high accuracy, with an average detection accuracy of 97.08%. These findings highlight the potential of the proposed DAS-accelerometer fusion and GVQVAE model as an effective, scalable solution for enhancing WF detection in modern railway systems.

Monitoring the health of vehicle-track system using deep learning and distributed fiber optic sensing presents a significant challenge due to the vast volume of real-time data and the difficulty of directly assessing the system’s condition. This often results in a severe imbalance in the distribution of extreme samples within the dataset, as large-scale signal collection typically lacks manual labeling. Consequently, supervised deep learning models face limitations due to insufficient labeled training data, while unsupervised deep learning models struggle with contamination from ambiguous samples whose health status remains unclear, hindering the development of robust and accurate models. To address this challenge, we propose SemAnoDiffusion, a semi-supervised model based on blur diffusion and an enhanced contrastive loss training approach. SemAnoDiffusion leverages a small set of labeled data alongside a large amount of unlabeled samples to accurately differentiate between anomalous data, normal data, and ambiguous samples that fall between these categories. In a case study of a metro system in Singapore, Distributed Acoustic Sensing and accelerometer arrays were used to collect track vibration responses as trains passed, with wheel flats occurring in a small subset of the trains. SemAnoDiffusion achieved 100% accuracy in classifying manually labeled normal and anomalous samples and effectively identified semi-damaged samples with unclear damage levels from the labeled data, successfully detecting all trains with wheel flats.

Neural network-based models have been implemented to predict various health indicators of asphalt pavement using pavement historical detection data. Unfortunately, their accuracy and reliability are not acceptable owing to their shallow architecture. To solve the issue, this study proposed an improved Transformer network to predict asphalt pavement health, called the Transformer with forward and reversed time series (Transformer FRTS). In terms of the input data, Transformer FRTS uses a new data form, so-called the random difference time series, to reduce the time dependency of the network prediction. In terms of the network architecture, the proposed network uses its encoder and decoder to obtain the data association from the forward and reverse time series. In addition, Transformer FRTS uses a post-processing decision criterion to improve the accuracy and reliability of prediction. The numerical experiment using the detection data from RIOHTrack full-scale track demonstrates that the proposed network has state-of-the-practice performance in asphalt pavement health prediction.

Image-based intelligent detection of road cracks with high accuracy and efficiency is vital to the overall condition assessment of the pavement. However, significant problems of continuous cracks interruption and background discrete noise misidentification are frequently observed in current semantic segmentation of pavement cracks, which mainly caused by traditional segmentation convolutional neural networks. This paper proposes a skip-level round-trip sampling block structure with the implementation of convolutional neural networks, thereby constructed a novel pixel level semantic segmentation network called CrackW-Net. After that, two datasets, including the widely recognized Crack500 dataset and a self-built dataset, were used to train two versions CrackW-Net, FCN, U-Net and ResU-Net. Meanwhile, comparative experiments are conducted among all these network models for crack detection. Results show that CrackW-Net without residual block performs the best in the task of pavement crack segmentation.