Search

• A spatiotemporal feature fusion-based deep learning (FFDL) framework is developed. • The FFDL framework is utilized to forecast synchronous excavation stability. • It's designed with feature fusion, deep learning, and global sensitivity analysis. • Feature fusion-based indicators are generated for excavation database preparation. • A case study is conducted to verify the applicability of the FFDL framework. Under rapid urbanization, analyzing and perceiving synchronous excavation stability is essential to generate valuable and foresightful information in identifying and mitigating potential risks in deep excavations. Although the existing forecasting methods can achieve high accuracy in the safety state perception of deep excavations, it is prone to lack proper consideration of the monitoring item's fusion and cause error and uncertainty via the single-parameter risk evaluation. To address it, this paper proposes a spatiotemporal feature fusion-based deep learning (FFDL) framework to implement the feature fusion-based data analysis and time-series forecasting on the factor of safety (FS) on behalf of excavation stability, and global sensitivity analysis. The fusion indicators and proposed FFDL framework are validated through a comprehensive case study encompassing 100 typical metro station constructions. The feature fusion-based data analysis results confirm the statistical significance of fusion indicators, contributing to establishing a high-quality spatiotemporal sequential database. As for the real-time prediction, the Bayesian optimization CNN-LSTM model provides a reliable prediction of excavation stability and the relative errors in the training and testing datasets do not exceed 1% and 3% in construction duration. It also enhances the predictive accuracy and outperforms the existing models LSTM, GRU, BiLSTM, and CNN-LSTM, reaching an average improvement of 38%, 32%, and 32% in RMSE, MAE, and MAPE, respectively. Furthermore, the proposed model has an excellent generalization performance in accurately predicting the excavation stability of the other ten excavation cases with an average relative error of 5.37%. In short, this work contributes to not only providing a new perspective in risk management by integrating feature fusion-based indicators and excavation stability but also verifying the applicability of the proposed framework. [ABSTRACT FROM AUTHOR]

• A richly annotated dataset is proposed for the segmentation of tunnel point clouds. • 3D deep learning is used to identify tunnel structural elements. • A hierarchical dataset architecture is designed to enable rigorous evaluation. • 3D deep learning networks show promising segment segmentation results. • A fully automatic scanning-analysis workflow is established for large-scale tunnels. Point clouds provide a novel and effective alternative to understanding the structural behaviours of segmental tunnel linings. 3D deep learning (DL) has emerged as a promising technology capable of automatically deriving point-wise semantic and instance labels from point clouds. The utilisation of 3D DL in segment segmentation of tunnel point clouds has not been explored and the development of tailored 3D DL networks has been hindered by the absence of specialised datasets and benchmarks. To bridge this gap, this paper introduces a richly annotated hierarchical dataset: 'Seg2Tunnel', acquired from five tunnels and including 1,300 tunnel rings. Using the Seg2Tunnel dataset, the feasibility of applying 3D DL to the segment segmentation is demonstrated for the first time. Experiments are conducted to investigate the influences of training set size, data augmentation strategy, input size, and hyperparameter on the performance of trained 3D DL models and to provide benchmarks and insights for future uses of the Seg2Tunnel dataset. The 3D DL models trained by the Seg2Tunnel dataset outperform currently existing image- and voxel-based DL methods. The Seg2Tunnel dataset and benchmarks are fundamental in shaping the design of 3D DL networks tailored for tunnel point clouds. The study provides a novel paradigm for automatically understanding the tunnel structural elements in the point clouds, paving the way for unmanned construction and intelligent evaluation of segmental tunnel linings. [ABSTRACT FROM AUTHOR]

• A hybrid graphical network approach considering causality named AC-GGN is proposed. • It can accurately make spatio-temporal prediction of deep excavation-induced ground settlement. • It is verified in a practical excavation case to reach high accuracy for each monitoring point. • Data augmentation techniques help to strengthen the long-term prediction ability of AC-GGN. • Global sensitivity analysis is conducted to quantitatively explain prediction behavior of AC-GGN. With the increasing demand for deep and large-scale excavation pits, the deformation response during excavation has become exceedingly complex, especially located in building-intensive areas. This paper proposes a hybrid deep learning model named attention-causality-based graphical gated network (AC-GGN) to accurately make the spatio-temporal prediction about the excavation-induced ground settlement at different monitoring points during the foundation pit construction. The novelty of the AC-GGN model lies in its flexible integration of four key components, including the Granger causality (GC) test, graph convolutional network (GCN), the gated recurrent unit (GRU), and attention mechanisms, which work together to effectively capture casual relationships along with spatial and temporal dependence embedded in the observed time-series from each monitoring points and then boost the prediction performance. To validate its applicability and superiority, a case study about a metro station excavation project in the Shanghai Metro Line 14 is conducted. Results indicate that the AC-GGN model outperforms state-of-the-art algorithms, which can make precise predictions for each monitoring point. The proper data augmentation technique facilitates long-term prediction with high accuracy, thereby expanding the scope of AC-GGN application beyond short-term prediction. Moreover, the global sensitivity analysis can be used to reveal which monitoring points have the most significant impact on ground settlement prediction. It can aid in identifying key risk areas for monitoring and control. In summary, the novel architecture of AC-GGN is beneficial to dynamically capture and predict the trend of ground settlement across different areas of the construction site. Practically, the accurate prediction results generated by AC-GGN offer rich evidence to not only perceive the excavation-induced risk development but also formulate corresponding measures in advance for risk mitigation. [ABSTRACT FROM AUTHOR]

[Display omitted] • BIM-centred framework TunGPR is to enhance data-driven maintenance of tunnels. • GPR synthetic dataset is created using domain randomisation and FDTD modelling. • Dual-rotational CNN is customised to enhance accuracy of hyperbola detection. • Digital workflow is to enable interoperability among BIM, simulations, and GPR sensing. Non-destructive Testing (NDT) techniques and data-driven technologies are increasingly applied in underground infrastructure maintenance, which can facilitate predictive monitoring for informed decision-making. Ground Penetrating Radar (GPR) is extensively utilised in rapid condition assessment of tunnel linings, particularly for detecting defects that lead to unforeseen changes of dielectric properties of materials. In this paper, a prototyped framework is proposed, namely TunGPR, for GPR-based tunnel lining assessment by incorporating Building Information Modelling (BIM), synthetic database and deep learning-enabled interpretation. The first module integrates laser-scanned point clouds and GPR Scan-to-BIM of tunnel lining with geological model. Subsequently, interoperability is achieved between the geo-integrated BIM and GPR simulation software. From the dielectric model retrieved from the BIM model, a database is established, considering a variety of condition combinations (i.e., voids, cavities, delamination, and water intrusion) leveraging domain randomisation and Finite-Difference Time-Domain (FDTD) modelling, as well as monitored field data. The dataset is then fed into the diagnostic module underpinned by a dual-rotational Convolutional Neural Network (CNN) that is customised to enhance accuracy and automation of hyperbola detection. Lastly, a preliminary risk assessment matrix is implemented into the BIM model for data management and action prioritisation. These efforts serve as an initial step to validate the feasibility and effectiveness of the GPR-enabled data-driven maintenance for tunnel linings in a BIM-centred framework. [ABSTRACT FROM AUTHOR]

• Novel coupled approach for undrained stability of rectangular tunnels. • Isogeometric analysis and deep learning ensure accuracy and efficiency. • High accuracy with mean squared error of 10−6 in predicting tunnel stability. The problem considered in this paper is for the stability evaluation of a rectangular tunnel in undrained clay during lining process. The approach adopted involved the use of isogeometric analysis (IGA) and the upper bound limit analysis formulation for the stability analysis. For the geometrical representation, B-spline basis functions are used to generate a set of B-spline surfaces that define the boundary of the soil domain, allowing for the exact representation of the tunnel geometry. The upper bound limit analysis is then formulated as a second-order cone program (SOCP), which can be solved by using a numerical optimization algorithm. The accuracy and reliability of the proposed method are validated by comparing the results with those published in previous studies. Furthermore, a large dataset is generated by randomly varying the input parameters, and a deep learning model is trained to learn the dataset. The deep learning model is trained using the mean squared error (MSE) metric, which yields an MSE as small as 10−6, indicating the high accuracy and precision of the proposed approach. Feature analysis and Partial Dependence Plots (PDPs) were used to gain insights into physical behavior by identifying important variables and understanding their individual and collective effects. In conclusion, the coupling of IGA and upper bound limit analysis provides a comprehensive and reliable solution to the stability of rectangular tunnels. The results obtained from this approach are accurate and can be achieved with reduced computational cost, making it an attractive approach for practical engineering applications. This approach can be used as a basis for future research on tunnel and tunnel stability analysis and may be extended to other types of soil structures under complex geometries. [ABSTRACT FROM AUTHOR]

• Forty U-shaped models are proposed to detect tunnel leakages rapidly and accurately. • UNet and UNet++ are coupled with six types of classification CNNs. • Multiple evaluation indices are introduced to obtain the optimal detection models. • Grad-CAM++ is utilized to understand the 'black box' of deep learning models. Tunnel lining leakage is a crucial indicator of metro shield tunnels' safety status. For automatic, rapid and accurate detection of leakages, this paper proposes a deep learning-based approach with enhanced efficiency, accuracy and interpretability. First, a total of forty U-shaped semantic segmentation models are developed by coupling UNet and UNet++ with six types of classification convolutional neural networks. Then, multiple evaluation indices, i.e., accuracy, computational complexity, and model complexity are introduced to determine the optimal leakage detection models which achieve a balance between efficiency and accuracy. Finally, Gradient-weighted Class Activation Mapping Plus Plus (Grad-CAM++) is leveraged to understand the mechanisms behind the 'black-box' of deep learning-based models. The experimental results show that UNet with encoder EfficientNetB6 is the most applicable model amongst UNet-based models, which gets 84.97 % intersection over union (IoU), requires 115.77 billion floating-point operations (FLOPs), and has 42.49 million parameters. UNet++ with encoder EfficientNetB5 achieves 87.70 % IoU, needs 327.26 billion FLOPs, and has only 30.86 million parameters, thereby becoming the best-performing model among UNet++-based models. The quantitative comparison with existing dominant approaches also proves that our proposed models have high accuracy and low time and space complexities. Furthermore, based on the visual explanations, it concludes that deepening and widening the encoder networks can boost models' performance, and the two feature fusion methods, namely simple skip connection and dense skip connections, play a crucial role in achieving precise tunnel leakage segmentation. The developed algorithms of this study provide a pixel-wise segmentation basis for fast and accurate quantitative assessment of metro tunnel health. [ABSTRACT FROM AUTHOR]

[Display omitted] • An automatic method is established for localization of underground pipelines using GPR. • A deep learning algorithm is proposed for automatic detection of underground pipelines. • The YOLOv3 model is trained and tested using a real GPR dataset. • The buried depth of the pipeline is estimated with a high accuracy by migration and binarization. • High recognition and location accuracy is achieved on the real GPR dataset. Ground penetrating radar (GPR) is a popular non-destructive method for detecting and locating underground pipelines. However, manual interpretation of a large number of GPR B-scan images is time-consuming, and the results highly relies on the practitioner's experience and the priori information at hands. An automatic GPR method for recognition and localization of underground pipelines is proposed based on a deep learning model in the paper. Firstly, a dataset containing 3,824 real GPR B-scans of pipelines is established. Secondly, a You Only Look Once version 3 (YOLOv3) model is trained to recognize the regions of the underground pipelines in a GPR image. Thirdly, the hyperbolic response of a pipeline is focused by migration, and transformed into a binary image by an iterative thresholding method. Finally, the apex of the hyperbola is employed to estimate both the horizontal position and the buried depth of the pipeline. Field experiments validated that the absolute errors of the estimated depths are less than 0.04 m and the average relative error is lower than 4 %. It is demonstrated that the proposed method is automatic, high-speed, and reliable for recognition and localization of underground pipelines in urban area. [ABSTRACT FROM AUTHOR]

• A new index is proposed to predict disc cutter wear during shield tunnelling. • Cutter radial wear and working time of shield machine are used to develop the index. • The index can change the wear data into time series data to use AI model conveniently. • An ensemble model is developed based on sequential deep learning techniques. • The proposed method can predict wear of a certain disc cutter in real time. This paper proposed a new index for evaluation of disc cutter life during earth pressure balance (EPB) tunnelling. This new index was defined as the ratio of accumulated cutter radial wear to working time of the shield machine. With this new index, the measured disc cutter wear can be transformed into a time series data. To predict cutter wear with construction process, an ensemble intelligent model integrating one-dimensional convolutional neural network (1D-CNN) and gated recurrent unit (GRU) was developed via incorporating the proposed cutter wear index. A multi-step-forward prediction mode was adopted to train the ensemble model to predict cutter wear in advance. Field data collected from an EPB tunnelling section in Guangzhou-Foshan intercity railway, Guangzhou, China, was used for validation. Results showed that the proposed index and ensemble model can predict wear of a certain cutter with high accuracy. Three other sequential deep networks were employed for comparison to verify the applicability of the proposed index and ensemble model. The proposed index and ensemble model is convenient to be used on site and can predict wear of a certain cutter on cutterhead to help determine which cutter to be replaced during real-time construction. [ABSTRACT FROM AUTHOR]

• A Deep learning training comparison was conducted to specify optimal technique. • A Transfer learning model is used to create a Crack detection CNN for tunnel lining. • A Directional crack classification CNN was created using Transfer Learning. • The sliding window technique is used in conjunction with CNN's to test the Models. • The CNNs were developed using Images from the Tunnel lining at CERN. One early sign of tunnel structure deterioration originates in the form of cracking, and therefore crack detection and resultant classification is integral for tunnel structural inspection and maintenance. Conventionally tunnel cracks are manually recorded and classified by trained professionals, which is costly, time-consuming and inevitably subjective. Recent advances in deep learning space have allowed for automatic cracks detection algorithms to be developed and subsequently utilized in structural health assessment of surface buildings, bridges, roads and other civil infrastructure. Nevertheless, these methods of development underperform when implemented for a tunnel structure in an underground environment due to the disparity of illumination combined with the congested image data caused by pipes, steel mesh, wires, and other tunnel facilities. To overcome these challenges, this paper proposes an innovative image-based crack detection method accompanied with crack classification using Convolution Neural Network (CNN) specifically for underground infrastructure environment. Unlike conventional CNN development from scratch, the proposed CNN incorporates Transfer Learning in the form of the VGG16 model with weights transferred from ImageNet. The transfer model was trained under different scenarios in order to find the optimal model for the novel dataset. The various models are trained using over 10′000 images validated on 2′500 images all of which are 256 × 256 pixels in size, these models are all subsequently tested using 30 images 3072 × 4096 pixels in size all of which are collected from the underground infrastructure at CERN. The Transfer Learning model used outperforms that of the traditional CNN training method of training from scratch. The optimum transfer model accomplished better testing metrics 96.6%,87.3%,92.4%,89.3% for Accuracy, Precision, Recall and F1 score respectively all of which were achieved with a shorter training time. The proposed CNN determines the existence and location of cracks within an image which are then subjected to a secondary classification CNN where the crack is categorized into one of the four crack classes which include the three directional classes of Horizontal, vertical and diagonal with the last crack classes incorporated to represent complex crack regions. The secondary classification CNN attains an Accuracy of 92.3% a Precision of 83.9% a Recall value of 82.3 % and a final F1 score of 81.5%. The performance of the proposed methods shows that a CNN crack detector/classifier can effectively overwhelm the unfavourable tunnel environment and accomplish results to a high standard. [ABSTRACT FROM AUTHOR]

• Instance segmentation model is proposed for complex rock fragment recognition. • Multilevel feature fusion, prior anchors, and self-attention modules are designed. • Dark illumination is solved by homomorphic filtering and histogram equalisation. • Extremely large and small rock fragments can be recognised with good accuracy. • Effectiveness is verified by comparing different models in ablation experiments. Timely recognition of rock fragments and their morphological sizes can allow adjustment of excavation parameters during tunnel boring machine (TBM) tunnelling. Traditional manual inspection strongly relies on the experience and subjective determination of the human operators, and sieving tests cannot be conducted in real time and are energy-consuming. Rock fragments in real-world images are frequently observed against a dark background, distributed with a high size diversity, and blocked by each other. A novel deep learning-based method is proposed in this paper to achieve real-time on-site rock fragment recognition. The proposed instance segmentation model comprises two subnetworks: object detection and semantic segmentation. The object detection subnetwork is designed based on a modified single-shot detector architecture, and multilevel feature fusion, prior anchors, and self-attention modules are utilised to localise rock fragment regions. The semantic segmentation subnetwork is designed based on U-net. Down-sampling stages use the structures of the object detection subnetwork to share the extracted features of the rock fragments, and up-sampling stages employ skip connection and self-attention modules to accomplish binary segmentation in each detected bounding box. A total of 50 original images with a resolution of 4096 × 3072 were collected: 35 for training and 15 for testing. The results showed that 88.8% of the rock fragments can be recognised and that the average recall and average intersection-over-union values reach 0.87 and 0.76, respectively. Small rock fragments inevitably missed in the labelling process and extremely large ones can also be recognised. The predicted size distributions of the rock fragments fit well with the ground truth ones. Ablation experiments were conducted to further demonstrate the effectiveness of the proposed method. This study presents both visual recognition and statistical results of the size distribution of rock fragments during TBM tunnelling, which can assist in the prediction of rock properties and the adjustment of excavation parameters. [ABSTRACT FROM AUTHOR]

• Tunnel depth images are generated from point clouds for segment joints labeling. • Two deep CNNs are designed to label segment joints accurately and completely. • RANSAC algorithm improves accuracy and reliability of dislocation value calculation. This paper presents a longitudinal dislocation detection method using an accurate tunnel segment joint labeling algorithm featured by deep CNNs (Convolutional Neural Networks). This method is proposed to be four steps. First, a mobile scanning system is used to acquire 3D point clouds of metro shield tunnels. Then, we use cylinder projection to generate tunnel surface depth images from 3D point clouds for segment joint labeling. Subsequently, two deep CNNs are designed to accurately label the segment joints on the depth images. The first CNN can roughly locate the segment joint positions, and the second precisely label the segment joints. Based on the labeled segment joints, two point data sets are obtained on both sides of each segment joint. By using the RANSAC algorithm, the two point sets can fit into two planes, the equation of which is then calculated to generate the dislocation value of the tunnel segment. Experiment results show that this method can label segment joints integrally and accurately without being affected by nearby tunnel equipment. Compared with traditional image edge detection algorithms (Canny and Sobel with Hough Transform), the CNNs are more powerful in labeling segment joints. When the distance measuring accuracy of scanner is 1.2 mm + 10 ppm, the internal and external accuracy of our detection method are evaluated to be 0.4 mm and 0.9 mm respectively. Compared with the scanning line method, the external accuracy of our method is higher and more reliable when there is tunnel equipment around segment joints. [ABSTRACT FROM AUTHOR]