A method and system for detecting the degree of concrete damage

Abstract

This invention provides a method and system for detecting the degree of concrete damage. The invention relates to the field of concrete testing technology. The method includes acquiring multiple fiber optic strain monitoring data during the operation of the concrete lining of an immersed tunnel; constructing a strain monitoring map; processing the strain monitoring map based on a graph neural network to determine multiple first abnormal damage regions of the concrete lining of the immersed tunnel; determining multiple second abnormal damage regions of the concrete lining of the immersed tunnel based on underwater laser scanning video of the multiple first abnormal damage regions; determining a target damage degree detection route based on the multiple second abnormal damage regions; and performing damage degree detection of the concrete lining of the immersed tunnel based on the target damage degree detection route. This method can efficiently and accurately plan the optimal damage degree detection route for the concrete lining of an immersed tunnel.

Description

Technical Field

[0001] This invention relates to the field of concrete testing, and specifically to a method and system for detecting the degree of concrete damage. Background Technology

[0002] As a core underwater infrastructure in cross-sea transportation networks, the safety and stability of the concrete lining structure of subsea immersed tunnels directly affects the long-term operational life and traffic safety of the tunnels. Long-term exposure to the high-salt, high-pressure environment of the deep sea, coupled with the continuous effects of temperature cycles and dynamic traffic loads, makes the concrete lining prone to damage such as micro-cracks and localized spalling. Furthermore, the initial damage is often concealed, and if not detected in time, it can gradually expand, threatening the overall structural safety of the tunnel. Traditional methods for detecting the extent of damage to the concrete lining face significant limitations. The large span and linear distribution of subsea immersed tunnels mean that traditional methods of large-area, comprehensive inspections are significantly limited. This not only requires substantial investment of manpower, resources, and time but also suffers from low efficiency, making it difficult to achieve precise focusing and efficient detection of damaged areas.

[0003] Therefore, how to efficiently and accurately plan the optimal damage detection route for the concrete lining of the immersed tunnel is an urgent problem to be solved. Summary of the Invention

[0004] The main technical problem solved by this invention is to efficiently and accurately plan the optimal damage detection route for the concrete lining of the immersed tunnel.

[0005] According to a first aspect, the present invention provides a method for detecting the degree of concrete damage, comprising: acquiring multiple fiber optic strain monitoring data during the operation of the concrete lining of an immersed tunnel; constructing a strain monitoring map, wherein the strain monitoring map includes multiple fiber optic monitoring nodes and multiple edges between the multiple fiber optic monitoring nodes, the node features of the fiber optic monitoring nodes include fiber optic strain monitoring data and three-dimensional coordinates of the fiber optic sensor deployment, and the edges between the fiber optic monitoring nodes represent the similarity of the fiber optic strain monitoring data of the monitoring points; processing the strain monitoring map based on a graph neural network to determine multiple first abnormal damage regions of the concrete lining of the immersed tunnel; acquiring underwater laser scanning videos of the multiple first abnormal damage regions of the concrete lining of the immersed tunnel; determining multiple second abnormal damage regions of the concrete lining of the immersed tunnel based on the underwater laser scanning videos of the multiple first abnormal damage regions of the concrete lining of the immersed tunnel; determining a target damage degree detection route based on the multiple second abnormal damage regions of the concrete lining of the immersed tunnel; and performing damage degree detection of the concrete lining of the immersed tunnel based on the target damage degree detection route.

[0006] In one possible implementation, determining the target damage level detection route based on multiple second abnormal damage areas of the concrete lining of the subsea immersed tunnel includes: determining multiple risk damage point information based on underwater laser scanning videos of multiple second abnormal damage areas of the concrete lining of the subsea immersed tunnel; clustering the multiple risk damage point information to obtain K clusters; determining severe risk damage point information, moderate risk damage point information, and mild risk damage point information based on the K clusters; generating multiple preliminary damage level detection route information based on the severe risk damage point information, the moderate risk damage point information, and the mild risk damage point information; and determining the target damage level detection route based on the multiple preliminary damage level detection route information.

[0007] In one possible implementation, the input to the graph neural network is the strain monitoring map, and the output of the graph neural network is multiple first abnormal damage areas of the concrete lining of the submerged tunnel.

[0008] In one possible implementation, the risk damage point information includes the geometric morphology data of the risk damage point and the lining loss data.

[0009] According to a second aspect, the present invention provides a concrete damage detection system, comprising:

[0010] The first acquisition module is used to acquire multiple fiber optic strain monitoring data during the operation of the concrete lining of the submerged tunnel.

[0011] A construction module is used to construct a strain monitoring map, which includes multiple fiber optic monitoring nodes and multiple edges between the multiple fiber optic monitoring nodes. The node features of the fiber optic monitoring nodes include fiber optic strain monitoring data and three-dimensional coordinates of the fiber optic sensor layout. The edges between the fiber optic monitoring nodes represent the similarity of the fiber optic strain monitoring data of the monitoring points.

[0012] The first abnormal region determination module is used to process the strain monitoring spectrum based on a graph neural network to determine multiple first abnormal damage regions of the concrete lining of the submerged tunnel.

[0013] The second acquisition module is used to acquire underwater laser scanning videos of multiple first abnormal damage areas of the concrete lining of the submerged tunnel.

[0014] The second abnormal area determination module is used to determine multiple second abnormal damage areas of the concrete lining of the subsea immersed tunnel based on underwater laser scanning video of multiple first abnormal damage areas of the concrete lining of the subsea immersed tunnel.

[0015] The detection route determination module is used to determine the target damage level detection route based on multiple second abnormal damage areas of the concrete lining of the submerged tunnel.

[0016] The damage detection module is used to detect the degree of damage to the concrete lining of the submerged tunnel based on the target damage detection route.

[0017] In one possible implementation, the detection route determination module is further configured to: determine multiple risk damage point information based on underwater laser scanning videos of multiple second abnormal damage areas of the concrete lining of the submerged tunnel; cluster the multiple risk damage point information to obtain K clusters; determine severe risk damage point information, moderate risk damage point information, and mild risk damage point information based on the K clusters; generate multiple preliminary damage degree detection route information based on the severe risk damage point information, the moderate risk damage point information, and the mild risk damage point information; and determine the target damage degree detection route based on the multiple preliminary damage degree detection route information.

[0018] In one possible implementation, the input to the graph neural network is the strain monitoring map, and the output of the graph neural network is multiple first abnormal damage areas of the concrete lining of the submerged tunnel.

[0019] In one possible implementation, the risk damage point information includes the geometric morphology data of the risk damage point and the lining loss data.

[0020] According to a third aspect, embodiments of the present invention provide an electronic device, including: a processor; a memory; and a computer program; wherein the computer program is stored in the memory and configured to be executed by the processor to implement the method described above, the method comprising: acquiring multiple fiber optic strain monitoring data during the operation of the concrete lining of an immersed tunnel; constructing a strain monitoring map, the strain monitoring map including multiple fiber optic monitoring nodes and multiple edges between the multiple fiber optic monitoring nodes, the node features of the fiber optic monitoring nodes including fiber optic strain monitoring data and three-dimensional coordinates of the fiber optic sensor deployment, and the edges between the fiber optic monitoring nodes representing the similarity of the fiber optic strain monitoring data of the monitoring points; processing the strain monitoring map based on a graph neural network to determine multiple first abnormal damage regions of the concrete lining of the immersed tunnel; acquiring underwater laser scanning videos of the multiple first abnormal damage regions of the concrete lining of the immersed tunnel; determining multiple second abnormal damage regions of the concrete lining of the immersed tunnel based on the underwater laser scanning videos of the multiple first abnormal damage regions of the concrete lining of the immersed tunnel; determining a target damage degree detection route based on the multiple second abnormal damage regions of the concrete lining of the immersed tunnel; and performing damage degree detection of the concrete lining of the immersed tunnel based on the target damage degree detection route.

[0021] According to the fourth aspect, this embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned concrete damage degree detection method. The method includes: acquiring multiple fiber optic strain monitoring data during the operation of the concrete lining of an immersed tunnel; constructing a strain monitoring map, the strain monitoring map including multiple fiber optic monitoring nodes and multiple edges between the multiple fiber optic monitoring nodes, the node features of the fiber optic monitoring nodes including fiber optic strain monitoring data and three-dimensional coordinates of the fiber optic sensor layout, and the edges between the fiber optic monitoring nodes representing the similarity of the fiber optic strain monitoring data of the monitoring points; processing the strain monitoring map based on a graph neural network to determine multiple first abnormal damage regions of the concrete lining of the immersed tunnel; acquiring underwater laser scanning videos of the multiple first abnormal damage regions of the concrete lining of the immersed tunnel; determining multiple second abnormal damage regions of the concrete lining of the immersed tunnel based on the underwater laser scanning videos of the multiple first abnormal damage regions of the concrete lining of the immersed tunnel; determining a target damage degree detection route based on the multiple second abnormal damage regions of the concrete lining of the immersed tunnel; and performing damage degree detection of the concrete lining of the immersed tunnel based on the target damage degree detection route.

[0022] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0023] This invention provides a method and system for detecting the degree of concrete damage. The method includes acquiring multiple fiber optic strain monitoring data during the operation of the concrete lining of an immersed tunnel; constructing a strain monitoring map, which includes multiple fiber optic monitoring nodes and multiple edges between the nodes. The node features of the fiber optic monitoring nodes include fiber optic strain monitoring data and three-dimensional coordinates of the fiber optic sensor deployment. The edges between the fiber optic monitoring nodes represent the similarity of the fiber optic strain monitoring data of the monitoring points. The strain monitoring map is processed using a graph neural network to determine multiple first abnormal damage regions of the concrete lining of the immersed tunnel; underwater laser scanning videos of the multiple first abnormal damage regions of the concrete lining are acquired; multiple second abnormal damage regions of the concrete lining are determined based on the underwater laser scanning videos of the multiple first abnormal damage regions; a target damage degree detection route is determined based on the multiple second abnormal damage regions; and damage degree detection of the concrete lining of the immersed tunnel is performed based on the target damage degree detection route. This method can efficiently and accurately plan the optimal damage degree detection route for the concrete lining of an immersed tunnel. Attached Figure Description

[0024] Figure 1 A schematic flowchart of a method for detecting the degree of concrete damage provided in an embodiment of the present invention;

[0025] Figure 2 This is a schematic diagram of a partial structure of an immersed tunnel provided in an embodiment of the present invention;

[0026] Figure 3 is a schematic diagram of a partial structure of an immersed tunnel with concrete lining provided in an embodiment of the present invention;

[0027] Figure 4 is a schematic diagram of an underwater laser scanner provided in an embodiment of the present invention;

[0028] Figure 5 A flowchart illustrating a detection route for determining the degree of damage to a target, provided by an embodiment of the present invention;

[0029] Figure 6 This is a schematic diagram of a concrete damage detection system provided in an embodiment of the present invention. Detailed Implementation

[0030] The present invention will now be described in further detail with reference to specific embodiments and accompanying drawings. Similar elements in different embodiments are referred to by associated similar element reference numerals. In the following embodiments, many details are described to facilitate a better understanding of the invention. However, those skilled in the art will readily recognize that some features may be omitted in different situations, or may be replaced by other elements, materials, or methods. In some cases, certain operations related to the present invention are not shown or described in the specification. This is to avoid obscuring the core parts of the invention with excessive description. For those skilled in the art, detailed description of these related operations is not necessary; they can fully understand the related operations based on the description in the specification and general technical knowledge in the art.

[0031] In this embodiment of the invention, the following are provided: Figure 1 The method for detecting the degree of concrete damage shown includes steps S1 to S7:

[0032] Step S1: Obtain multiple fiber optic strain monitoring data during the operation of the concrete lining of the submerged tunnel.

[0033] Multiple fiber optic strain monitoring data during the operation of the concrete lining of the subsea immersed tunnel are strain measurements periodically collected by fiber optic sensors deployed on the concrete lining of the subsea immersed tunnel during the tunnel's operation. Figure 2 Figure 3 is a schematic diagram of a partial structure of an immersed tunnel provided in an embodiment of the present invention. Figure 3 is a schematic diagram of a partial structure of an immersed tunnel including a concrete lining provided in an embodiment of the present invention.

[0034] Fiber optic strain monitoring data records the deformation information of concrete lining under various factors such as water pressure, temperature changes, and loads, including strain amplitude, strain rate of change, and strain time-series curves.

[0035] Step S2: Construct a strain monitoring map. The strain monitoring map includes multiple fiber optic monitoring nodes and multiple edges between the multiple fiber optic monitoring nodes. The node characteristics of the fiber optic monitoring nodes include fiber optic strain monitoring data and three-dimensional coordinates of the fiber optic sensor layout. The edges between the fiber optic monitoring nodes represent the similarity of the fiber optic strain monitoring data of the monitoring points.

[0036] A strain monitoring map is a structured data representation of strain monitoring information for the concrete lining of an immersed tunnel. The strain monitoring map consists of multiple fiber optic monitoring nodes and multiple edges connecting these nodes.

[0037] Each fiber optic monitoring node represents a physical strain monitoring location. The node characteristics of each fiber optic monitoring node include the fiber optic strain monitoring data acquired at that location and the three-dimensional coordinates of the fiber optic sensor deployment.

[0038] Multiple edges between fiber optic monitoring nodes can quantify the similarity of fiber optic strain monitoring data between different monitoring points, and the edges can characterize the correlation and difference of strain at different fiber optic monitoring nodes.

[0039] In some embodiments, a deep neural network model can be used to determine the similarity of fiber strain monitoring data between monitoring points of fiber optic monitoring nodes.

[0040] Deep neural network models include Deep Neural Networks (DNNs). A deep neural network is an artificial neural network containing multiple hidden layers. It can automatically learn complex feature representations and abstract patterns from input data through multiple layers of nonlinear transformations. Deep neural networks possess powerful learning and generalization capabilities and can handle high-dimensional, nonlinear data. By adjusting network weights using backpropagation algorithms and gradient descent optimizers, deep neural networks can achieve accurate classification or regression analysis for complex tasks.

[0041] Step S3: Based on the graph neural network, the strain monitoring spectrum is processed to determine multiple first abnormal damage areas of the concrete lining of the submerged tunnel.

[0042] Graph Neural Networks (GNNs) are deep learning models capable of processing graph data. GNNs learn feature representations on a graph by aggregating and transforming information about nodes and their neighbors. Through message passing mechanisms, GNNs enable node features to propagate and iteratively update across the graph, thereby capturing dependencies between nodes and global topological information. GNNs can effectively process non-Euclidean spatial data. By learning the latent patterns of fiber optic monitoring node features and edges in a strain monitoring map, GNNs can identify signs of local anomalies or structural instability. GNNs can be used to detect damage, anomalies, and potential risks in complex structures. The input to the GNN is the strain monitoring map, and the output is multiple first-order abnormal damage areas in the concrete lining of the submerged tunnel.

[0043] The multiple first abnormal damage areas of the concrete lining of the subsea immersed tunnel were determined by a graph neural network based on the strain monitoring spectrum analysis results, which were sets of areas with abnormal strain data and structural behavior deviating from the normal pattern.

[0044] The strain monitoring data in the first abnormal damage area are significantly different from those in the surrounding normal area, and there may be structural damage such as cracks or breaks.

[0045] Strain monitoring maps are a structured knowledge representation that integrates scattered strain monitoring information and their spatial relationships. The strain data and 3D coordinate features of each fiber optic monitoring node accurately describe the structural stress state at its corresponding location, and the similarity edges between nodes quantify the synergy of strain changes in different regions. For example, high strain similarity between physically adjacent monitoring points indicates uniform stress in the structure, while low similarity suggests potential local stress concentration or damage. Constructing such strain monitoring maps transforms abstract strain data into a computable map structure, preserving the microscopic features of individual monitoring points while capturing the macroscopic correlations of structural strain through edge relationships. This provides graph neural networks with clearly defined and interconnected input data. Furthermore, constructing strain monitoring maps reduces the redundancy of raw strain monitoring data and highlights the dependencies between key features. This allows graph neural networks to efficiently aggregate information from nodes and their neighbors, accurately identify the superposition effects of local and globally correlated anomalies, and improve the accuracy and efficiency of identifying the first abnormal damage region.

[0046] Graph neural networks (Graph Neural Networks) can iteratively process strain monitoring maps through multi-layer graph convolution operations. In each layer, the Graph Neural Network aggregates information from neighboring nodes of each fiber optic monitoring node and fuses the features of these neighboring nodes with the features of the fiber optic monitoring node itself. During processing, the Graph Neural Network can identify nodes or groups of nodes in the strain monitoring map that significantly deviate from the normal strain pattern. This deviation manifests as a sharp increase or persistently high value in the fiber optic strain monitoring data of the node features, while the similarity of the fiber optic strain monitoring data represented by the edges between nodes decreases sharply. The Graph Neural Network can establish a strain pattern standard based on historical normal operation data, then calculate the deviation between the features of each fiber optic monitoring node in the current strain monitoring map and the strain pattern standard, using the deviation value as the node's anomaly score. Simultaneously, the connectivity analysis mechanism of the Graph Neural Network can focus on detecting fractures or weakening in the strain field, i.e., monitoring points that are physically close. If the similarity of the fiber optic strain monitoring data of the monitoring points represented by the edges between fiber optic monitoring nodes is significantly lower than a threshold, it indicates the possible presence of cracks or localized damage. The final graph neural network can mark fiber optic monitoring nodes with high strain anomaly scores and low similarity connection features, along with their surrounding areas, as multiple first-order abnormal damage areas in the concrete lining of the submerged tunnel.

[0047] Step S4: Obtain underwater laser scanning videos of multiple first abnormal damage areas of the concrete lining of the submerged tunnel.

[0048] The underwater laser scanning video of multiple first abnormal damage areas is obtained by scanning and capturing video data of multiple first abnormal damage areas of the concrete lining of the submerged tunnel using an underwater laser scanner. Figure 4 is a schematic diagram of an underwater laser scanner provided in an embodiment of the present invention.

[0049] Step S5: Based on the underwater laser scanning video of multiple first abnormal damage areas of the concrete lining of the subsea immersed tunnel, determine multiple second abnormal damage areas of the concrete lining of the subsea immersed tunnel.

[0050] In some embodiments, an abnormal damage area localization model can be used to determine multiple second abnormal damage areas of the concrete lining of the subsea immersed tunnel. The abnormal damage area localization model is a long short-term neural network model. The input to the abnormal damage area localization model is underwater laser scanning video of multiple first abnormal damage areas of the concrete lining of the subsea immersed tunnel, and the output of the abnormal damage area localization model is the multiple second abnormal damage areas of the concrete lining of the subsea immersed tunnel.

[0051] Long Short-Term Memory (LSTM) neural network models are a type of deep learning model that excels at processing long sequences of data. LSTMs address the vanishing or exploding gradient problems that commonly occur in long data processing through unique gating mechanisms and cell state designs. The core of an LSTM consists of three gates: an input gate, a forget gate, and an output gate. The input gate filters and incorporates new, valid information; the forget gate selectively discards redundant or irrelevant data; and the output gate controls the output and transmission of information. LSTMs can effectively capture temporal dependencies in long sequences of data and extract dynamic features and key information from consecutive video frames.

[0052] The multiple secondary abnormal damage areas of the concrete lining of the subsea immersed tunnel are a set of damage areas with a more precise range determined by the abnormal damage area localization model.

[0053] Underwater laser scanning videos of multiple first-abnormal damage areas in the concrete lining of the submerged tunnel fully recorded continuous visual information for each first-abnormal damage area. This visual information includes dynamic features such as morphological changes and surface details of the suspected damage area, while also capturing environmental correlation traces and temporal morphological differences related to the damage. This video data can supplement the limitations of strain monitoring data, verifying the existence of actual damage in the first-abnormal damage areas and its specific manifestations through intuitive visual evidence. This provides a comprehensive and targeted visual data foundation for the model to screen for real damage. The video data contains clear geometric features, motion characteristics, and continuity of real damage. The model can filter out real damage with clear features and time-tested characteristics from a broad range of first-abnormal damage areas, achieving precise narrowing and localization of the damage range, and ultimately identifying more accurate second-abnormal damage areas.

[0054] Long Short-Term Neural Networks (LSTNs) can decompose underwater laser scanning videos of multiple first-abnormal damage areas of the concrete lining of an immersed tunnel into frames to extract visual features from each frame, including grayscale values, edge contours, and texture structure. These features are then arranged into a feature sequence in temporal order. The LTN can focus on key damage-related features, such as crack edges and surface texture changes, through the input gate, and control the transmission of these key features through the output gate. The model can utilize cell states to store the patterns of damage feature changes over long time sequences, such as the morphological consistency of damaged areas in consecutive frames and the stability of damage boundaries. Through feature analysis and temporal dependency modeling of the entire video sequence, the LTN can determine whether there is actual damage in each first-abnormal damage area and identify areas with actual damage as second-abnormal damage areas of the concrete lining of the immersed tunnel.

[0055] In some embodiments, determining multiple second abnormal damage areas of the concrete lining of the subsea immersed tunnel based on underwater laser scanning videos of multiple first abnormal damage areas includes steps S21 to S23:

[0056] Step S21: Based on the underwater laser scanning video of multiple first abnormal damage areas of the concrete lining of the submerged tunnel, determine multiple highly suspicious damage feature division areas, the basic features of each highly suspicious damage feature division area, and the dynamic morphological change sequence of each highly suspicious damage feature division area.

[0057] In some embodiments, a long short-term neural network can be used to determine multiple highly suspicious damage feature division regions, the basic features of each highly suspicious damage feature division region, and the morphological dynamic change sequence of each highly suspicious damage feature division region.

[0058] The high-suspicious-damage-feature segmentation region is the sub-region with prominent damage features output by the long short-term neural network.

[0059] The basic features of each highly suspicious damage feature area include quantitatively described morphological features such as crack length and direction, depression depth and area, and pore diameter and extent.

[0060] The morphological dynamic change sequence of each highly suspicious damage feature segmentation region is a record of the morphological changes of each highly suspicious damage feature segmentation region in the video time dimension, output by a long short-term neural network, such as a continuous change sequence of contour integrity and a continuous change sequence of detail clarity.

[0061] Long Short-Term Neural Networks (LSTNs) possess advantages in time-series data processing and feature extraction, and can adapt to the dynamic characteristics of underwater laser scanning videos. Through gating mechanisms, LSTNs can effectively capture the temporal dependencies between video frames, thereby accurately identifying key visual information within the first abnormal damage area. Simultaneously, the model can deeply mine damage-related local and global features in each frame, and then, based on indicators such as contour recognition and feature saliency, delineate highly suspicious damage feature regions. Through continuous analysis of multi-frame data, LSTNs can extract the morphological evolution patterns of each delineated region over time, forming a dynamic morphological change sequence. Furthermore, through feature quantization learning, LSTNs can accurately identify and output basic features such as crack length and direction, and depression depth and area.

[0062] Step S22: Based on the multiple highly suspicious damage feature division regions, the basic features of each highly suspicious damage feature division region, and the morphological dynamic change sequence of each highly suspicious damage feature division region, determine the suspicion score, dynamic feature fluctuation degree, and feature consistency index of each highly suspicious damage feature division region.

[0063] In some embodiments, a deep neural network can be used to determine the suspicion score, dynamic feature fluctuation degree, and feature consistency index of each highly suspicious damage feature segmentation region.

[0064] The suspicion score for each highly suspicious damage feature region is a quantitative assessment of the degree to which each highly suspicious damage feature region conforms to the true damage features, obtained through the output of a deep neural network.

[0065] The degree of dynamic feature fluctuation is a numerical value representing the discreteness of the dynamic changes in the morphology of regions divided by each highly suspicious damage feature output by a deep neural network.

[0066] The feature consistency index of two adjacent highly suspicious damage feature regions is a numerical value that is a quantitative assessment of the degree of fit between adjacent highly suspicious regions in terms of basic feature type and morphological change trend, through the output of a deep neural network.

[0067] Deep neural networks, through hidden layers, can progressively deepen the abstract representation of the basic features of each highly suspicious damage feature region. By combining the temporal patterns in the dynamic change sequence of the corresponding region's morphology, the model can quantitatively evaluate the degree of fit between each highly suspicious damage feature region and the actual damage features, and output a suspicion score for each region. Simultaneously, the model can capture the discrete distribution characteristics of the dynamic change sequence of each highly suspicious damage feature region's morphology. The model calculates the degree of dynamic feature fluctuation for each region by analyzing the fluctuation amplitude of contour integrity and detail clarity. Furthermore, deep neural networks can effectively learn the feature association patterns between adjacent highly suspicious damage feature regions and compare the matching degree of basic feature types and the consistency of morphological change trends between adjacent regions. Through similarity calculation, the model can calculate the feature consistency index between two adjacent highly suspicious damage feature regions.

[0068] Step S23: Based on the multiple highly suspicious damage feature division regions, the suspicion score of each highly suspicious damage feature division region, the dynamic feature fluctuation degree, and the feature consistency index of two adjacent highly suspicious damage feature division regions, multiple second abnormal damage regions of the concrete lining of the submerged tunnel are determined.

[0069] In some embodiments, deep neural networks can be used to identify multiple second abnormal damage areas in the concrete lining of an immersed tunnel.

[0070] Deep neural networks, leveraging their multi-parameter integrated decision-making and logical reasoning capabilities, can accurately identify the second abnormal damage area in the concrete lining of submerged tunnels. The deep neural network uses a suspicion score threshold as the core screening criterion, retaining regions with high-suspicious damage features that meet the score. Then, through dynamic feature fluctuation analysis, unstable regions with drastic morphological fluctuations caused by environmental interference are eliminated, while reliable candidate regions with stable morphology are identified. Simultaneously, the model compares the feature consistency index of adjacent high-suspicious damage feature regions, merging adjacent regions with the index meeting the standard as the same damage extension, while regions with the index not meeting the standard remain independent. The final output is a second abnormal damage area with clear boundaries and accurate range.

[0071] Step S6: Determine the target damage level detection route based on multiple second abnormal damage areas of the concrete lining of the submerged tunnel.

[0072] In some embodiments, Figure 5 This is a flowchart illustrating a detection route for determining the degree of target damage according to an embodiment of the present invention. The route for determining the degree of target damage includes steps S31 to S35:

[0073] Step S31: Determine information on multiple risk damage points based on underwater laser scanning videos of multiple second abnormal damage areas of the concrete lining of the submerged tunnel.

[0074] The underwater laser scanning videos of multiple second abnormal damage areas of the concrete lining of the subsea immersed tunnel are a collection of video segments extracted from the underwater laser scanning videos of multiple first abnormal damage areas of the concrete lining of the subsea immersed tunnel, which precisely correspond to multiple second abnormal damage areas.

[0075] In some embodiments, a risk damage point determination model can be used to determine multiple risk damage point information. The risk damage point determination model is a long short-term neural network model. The input to the risk damage point determination model is underwater laser scanning video of multiple second abnormal damage areas of the concrete lining of the submerged tunnel, and the output of the risk damage point determination model is multiple risk damage point information.

[0076] The multiple risk damage point information is detailed data on the specific damage points on the surface of the concrete lining of the subsea immersed tunnel, determined by underwater laser scanning video analysis of multiple secondary abnormal damage areas of the concrete lining using a risk damage point determination model. The risk damage point information includes geometric data of the risk damage points and lining loss data.

[0077] Geometric morphological data of risk damage points is used to describe the physical shape characteristics of risk damage points. Geometric morphological data includes the shape, size, distribution range, and edge contour of the damage points.

[0078] Lining loss data records the degree of damage to the concrete lining of the submerged tunnel at risk points. Lining loss data includes the amount of lining material falling off, the structural thickness loss value, and the damaged area.

[0079] Underwater laser scanning videos of multiple secondary abnormal damage areas in the concrete lining of the submerged tunnel recorded the continuous visual dynamics of these areas, including key information such as morphological changes, surface details, and spatial relationships. This video data can provide intuitive and comprehensive visualization for the identification of risk damage points, enabling the model to accurately extract specific feature data for each damage point.

[0080] Long Short-Term Neural Networks (LSTNs) can utilize time-series processing capabilities to perform frame-by-frame scanning and feature integration on underwater laser scanning video data of the input second-abnormal damage area. The model receives the laser scanning depth information of the current frame through an input gate and, combined with the hidden state of the previous time step, identifies whether there is a depth abrupt change relative to the normal lining surface at the current scanning location. Due to the complexity of the underwater environment, anomalies in a single frame may be affected by noise. LTNs can utilize the memory function of cell states to compare the features of the current frame with those of historical frames. Only when deep indentations or texture breaks matching damage characteristics appear in multiple consecutive frames will the model classify it as a valid damage signal. After confirming the existence of damage, the model can store the changing patterns of damage features over a long time series, such as the morphological consistency of the damage area in consecutive frames and the stability of the damage boundary. Furthermore, by analyzing these temporal and spatial features, the model can accurately identify each risk damage point. For each identified risk damage point, the model can further extract its geometric features such as shape, size, and distribution range, transforming them into geometric morphological data of the risk damage point. Simultaneously, it can calculate indicators such as the amount of lining material detachment and thickness loss at the damage point, forming lining loss data.

[0081] Step S32: Cluster K clusters are obtained based on information from multiple risk damage points.

[0082] The clustering algorithm mentioned is the K-means clustering algorithm, which is an iterative clustering analysis algorithm. The main purpose of the K-means clustering algorithm is to divide the dataset into K independent clusters, so that the data points within the same cluster exhibit high similarity in the feature space, while the data points between different clusters are significantly different. The K value can be obtained by setting it in advance.

[0083] Each of the K clusters represents a set of risk damage points that are spatially adjacent and highly similar in damage characteristics. These similarities are reflected in the three-dimensional distribution, geometric dimensions, and degree of material loss of the damage points. For example, clustering information on multiple risk damage points within a section of an immersed tunnel yields three clusters. Cluster 1 contains a set of severely damaged points concentrated at the tunnel top, with significant damage depth and obvious concrete spalling. Cluster 2 contains a set of lightly damaged points distributed in the middle section of the tunnel sidewall, mainly consisting of fine network cracks with minimal lining loss. Cluster 3 contains a set of structurally damaged points located at the joints of tunnel sections, exhibiting a linear distribution and moderate depth.

[0084] The process of clustering multiple risk damage points in the concrete lining of an immersed tunnel using the K-means clustering algorithm is as follows: First, the algorithm randomly selects K risk damage points from the set containing all risk damage point information as initial cluster centers. Next, for each risk damage point in the set, based on its feature vectors (coordinates, geometric data, and lining loss data), the Euclidean distance formula is used to calculate its distance to the K initial cluster centers, and the risk damage point is assigned to the corresponding cluster according to the principle of closest proximity. After all risk damage points have been initially divided, the average value of each feature of all risk damage points in each cluster is recalculated, and this value is used as the updated cluster center. This process of assigning points and updating cluster centers is repeated until the new cluster center positions show minimal change from the previous iteration. At this point, the clustering process is considered converged, thus completing the K-means clustering.

[0085] Clustering can effectively integrate the scattered and complex information on risk and damage points in the concrete lining of immersed tunnels. Because the number of originally identified risk and damage points is large and scattered across the long tunnel space, analyzing each discrete risk and damage point directly results in high computational redundancy. Clustering, however, can group damage points with similar characteristics into independent operational units, greatly simplifying the data structure. By dividing multiple risk and damage point information into K clusters, the degree of damage concentration and the distribution pattern of damage in different areas inside the tunnel can be visually displayed.

[0086] Step S33: Determine the information of severely risky damage points, moderately risky damage points, and lightly risky damage points based on the K clusters.

[0087] In some embodiments, a risk analysis model can be used to determine information on severe risk damage points, moderate risk damage points, and mild risk damage points. The risk analysis model is a deep neural network model. The input to the risk analysis model is the K clusters, and the output of the risk analysis model is the information on severe risk damage points, moderate risk damage points, and mild risk damage points.

[0088] The information on severe risk damage points is a set of detailed information on risk damage points that pose a significant threat to the safety of the concrete lining structure of the submerged tunnel, as determined by a risk analysis model.

[0089] The information on moderate-risk damage points is a collection of detailed information on risk damage points that have a certain impact on the concrete lining structure of the submerged tunnel but have not yet reached an emergency state, as determined by the risk analysis model.

[0090] Information on minor risk damage points is a set of detailed information on risk damage points that are identified by a risk analysis model and only manifest as minor surface defects or early-stage damage.

[0091] K clusters categorize multiple risky damage points based on feature similarity, with each cluster containing damage points exhibiting similar geometric shapes and lining loss characteristics. The data from these K clusters not only includes the spatial distribution of each damage point but also the statistical characteristics of damage density within the region. This categorization method reduces data redundancy and highlights the group characteristics of different types of damage, enabling the risk analysis model to perform concentrated risk assessments on damage points within the same cluster, thereby improving the accuracy and efficiency of the assessment.

[0092] Deep neural networks can organize the feature data of K clusters. Each cluster's features include statistical values ​​of the geometric morphology of all risk damage points within the cluster, statistical values ​​of lining loss data, and spatial distribution characteristics of damage points within the cluster. After receiving this cluster feature data through the input layer, the model undergoes nonlinear transformations through multiple hidden layers to gradually extract deeper risk-related features, such as the correlation between lining loss data and structural bearing capacity, and the damage propagation potential reflected by the geometric morphology data. During training, the model can continuously adjust network weights based on historical damage data and corresponding risk level labels using a backpropagation algorithm to optimize its ability to identify risk features. The model can compare the extracted cluster features with preset risk assessment standards and, combined with factors such as the geometric complexity of damage points within the cluster, the severity of lining loss, and the spatial distribution density of damage points, calculate a risk score for each cluster. Based on the risk score, the deep neural network can classify the risk damage points within a cluster into three levels: severe, moderate, and mild. The damage points corresponding to clusters with risk scores higher than a preset high threshold are classified as severe risk damage points, the damage points corresponding to clusters with risk scores between the preset medium and high thresholds are classified as moderate risk damage points, and the damage points corresponding to clusters with risk scores lower than the preset medium threshold are classified as mild risk damage points.

[0093] Step S34: Generate multiple preliminary damage level detection route information based on the information of severe risk damage points, the information of moderate risk damage points, and the information of mild risk damage points.

[0094] In some embodiments, a route simulation model can be used to generate multiple preliminary damage level detection route information. The route simulation model is a generative adversarial network (GAN). The input to the route simulation model is the information on high-risk damage points, medium-risk damage points, and low-risk damage points; the output of the route simulation model is the information on multiple preliminary damage level detection routes.

[0095] Generative Adversarial Networks (GANs) are deep learning models consisting of two core components: a generator and a discriminator. The generator is responsible for producing seemingly realistic data samples from random noise or specific conditions, while the discriminator distinguishes between fake samples and real data samples generated by the generator. During training, the two components compete and challenge each other. The generator continuously optimizes its generation strategy, while the discriminator continuously improves its discrimination ability, ultimately enabling the generator to produce high-quality target data that conforms to complex distribution patterns.

[0096] Multiple preliminary damage detection route information is generated by analyzing and simulating information on severely risky, moderately risky, and lightly risky damage points using a route simulation model. These are multiple preliminary path schemes used to detect the degree of damage.

[0097] Each preliminary damage assessment route passes through all the high-risk damage points, as well as some medium-risk and low-risk damage points. The information for each preliminary damage assessment route clearly specifies the starting point, ending point, the order of various risk damage points along the route, and the route direction.

[0098] Generative Adversarial Networks (GANs) possess powerful feature mining and generation capabilities. The generator in a GAN can deeply analyze the spatial correlation patterns of severely risky, moderately risky, and lightly risky damage points. The discriminator can accurately verify path feasibility and provide feedback on optimization directions, thereby efficiently generating multiple preliminary damage detection routes that meet the detection requirements. The generator first sets all severely risky damage points as essential nodes in the path planning. It then attempts to construct various topological sequences connecting these essential nodes to form the basic framework of the route, ensuring that each generated route covers all areas posing the greatest threat to structural safety without omission. Subsequently, the generator calculates the distance offset between these secondary nodes and the basic framework path based on the principle of spatial proximity for moderately and lightly risky damage points. If a moderately or lightly risky point is within the effective scanning radius of the basic framework path or has a small offset along the path, the generator includes it in the route sequence; otherwise, if the distance is too far and causes a severely circuitous path, it is discarded. The discriminator network rigorously verifies and scores the routes output by the generator based on pre-defined constraint logic. The discriminator evaluates the overall effectiveness of the routes, rewarding those that, while ensuring the total distance or time remains controllable, also connect more damage detection points of moderate and mild risk. After multiple rounds of adversarial training, the generator-generated detection routes achieve a high level of realism, rationality, and effectiveness, ultimately outputting multiple preliminary damage detection route information that meets the requirements.

[0099] Step S35: Determine the target damage detection route based on the information of the multiple initially selected damage detection routes.

[0100] In some embodiments, a target route determination model can be used to determine the target damage level detection route. The target route determination model is a deep neural network model. The input to the target route determination model is the information of the multiple initially selected damage level detection routes, and the output of the target route determination model is the target damage level detection route.

[0101] The target damage level detection route is the optimal detection path selected by the target route determination model after comprehensively evaluating information from multiple initially selected damage level detection routes.

[0102] The target damage detection route comprehensively considers multiple factors such as detection efficiency, coverage integrity, and path rationality, and can complete the detection of damage to the concrete lining of the submerged tunnel in the optimal way to ensure accurate detection results and the lowest detection cost.

[0103] Multiple preliminary damage detection route information provides a diverse range of detection path samples. Each route differs in detection order, coverage, path length, and other aspects. These differences provide a wealth of comparative analysis objects for the target route determination model. The model can select the detection route with the best overall performance by evaluating the advantages and disadvantages of different routes.

[0104] Deep neural networks can map the performance metrics of each initially selected damage detection route to an input vector. The hidden layers of the deep neural network can construct an evaluation standard that integrates multiple evaluation factors such as detection efficiency, energy consumption, equipment safety, and detection quality. The model can calculate the score of each route on each evaluation dimension through forward propagation and use internally learned weight parameters to handle the trade-offs between different metrics. For example, the model can identify a route that, although the total distance is shortest, may lead to increased wear and tear on the detection equipment or unstable data acquisition due to frequent sharp turns, thus penalizing it in the equipment safety dimension. Conversely, another route, although slightly longer, can smoothly scan all high-risk areas in one go, and the model will reward it in the detection quality dimension. Deep neural networks can capture the complex nonlinear constraints between these metrics, thereby calculating the comprehensive performance score of each route. Finally, the model can compare the comprehensive scores of all initially selected routes and determine the initially selected damage detection route with the highest score as the target damage detection route, which achieves the optimal balance between efficiency, quality, and cost.

[0105] Step S7: Detect the damage level of the concrete lining of the submerged tunnel based on the target damage level detection route.

[0106] Once the target damage detection route is determined, the damage level of the concrete lining of the submerged tunnel is detected based on the target damage detection route.

[0107] Based on the same inventive concept Figure 6 This is a schematic diagram of a concrete damage detection system provided in an embodiment of the present invention. The concrete damage detection system includes:

[0108] The first acquisition module 41 is used to acquire multiple fiber optic strain monitoring data during the operation of the concrete lining of the submerged tunnel.

[0109] Module 42 is used to construct a strain monitoring map, which includes multiple fiber optic monitoring nodes and multiple edges between the multiple fiber optic monitoring nodes. The node features of the fiber optic monitoring nodes include fiber optic strain monitoring data and three-dimensional coordinates of the fiber optic sensor layout. The edges between the fiber optic monitoring nodes represent the similarity of the fiber optic strain monitoring data of the monitoring points.

[0110] The first abnormal region determination module 43 is used to process the strain monitoring spectrum based on a graph neural network to determine multiple first abnormal damage regions of the concrete lining of the submerged tunnel.

[0111] The second acquisition module 44 is used to acquire underwater laser scanning videos of multiple first abnormal damage areas of the concrete lining of the submerged tunnel.

[0112] The second abnormal area determination module 45 is used to determine multiple second abnormal damage areas of the concrete lining of the subsea immersed tunnel based on underwater laser scanning video of multiple first abnormal damage areas of the concrete lining of the subsea immersed tunnel.

[0113] The detection route determination module 46 is used to determine the target damage degree detection route based on multiple second abnormal damage areas of the concrete lining of the underwater immersed tunnel.

[0114] The damage detection module 47 is used to detect the damage level of the concrete lining of the submerged tunnel based on the target damage level detection route.

[0115] It should be noted that, in order to simplify the description disclosed herein and thus aid in the understanding of one or more embodiments of the invention, the foregoing description of embodiments of this specification sometimes combines multiple features into a single embodiment, drawing, or description thereof. However, this method of disclosure does not imply that the subject matter of this specification requires more features than those mentioned in the claims. In fact, the embodiments contain fewer features than all the features of a single embodiment disclosed above.

[0116] Finally, it should be understood that the embodiments described in this specification are merely illustrative of the principles of the embodiments described herein. Other variations may also fall within the scope of this specification. Therefore, alternative configurations of the embodiments described herein are intended to be illustrative rather than limiting, and should be considered consistent with the teachings of this specification. Accordingly, the embodiments described herein are not limited to those explicitly introduced and described herein.

Claims

1. A method for detecting a degree of damage of concrete, characterized by, include: Acquire multiple fiber optic strain monitoring data during the operation of the concrete lining of the submerged tunnel; A strain monitoring map is constructed, which includes multiple fiber optic monitoring nodes and multiple edges between the multiple fiber optic monitoring nodes. The node characteristics of the fiber optic monitoring nodes include fiber optic strain monitoring data and three-dimensional coordinates of the fiber optic sensor layout. The edges between the fiber optic monitoring nodes represent the similarity of the fiber optic strain monitoring data of the monitoring points. The strain monitoring spectrum is processed based on a graph neural network to identify multiple first abnormal damage areas in the concrete lining of the submerged tunnel. Underwater laser scanning videos of multiple first-abnormal damage areas in the concrete lining of the subsea immersed tunnel were obtained; Based on underwater laser scanning videos of multiple first abnormal damage areas of the concrete lining of the subsea immersed tunnel, multiple second abnormal damage areas of the concrete lining of the subsea immersed tunnel were determined. Based on multiple second abnormal damage areas of the concrete lining of the submerged tunnel, a target damage level detection route is determined. Damage to the concrete lining of an underwater immersed tunnel is detected based on the target damage detection route.

2. The method for detecting the degree of concrete damage as described in claim 1, characterized in that, The target damage level detection route based on multiple second abnormal damage areas of the concrete lining of the submerged tunnel includes: Underwater laser scanning video of multiple secondary abnormal damage areas of the concrete lining of the subsea immersed tunnel was used to identify information on multiple risk damage points. K clusters were obtained by clustering based on information from multiple risk damage points; Based on the K clusters, information on severe risk damage points, moderate risk damage points, and mild risk damage points are determined. Based on the information on severe risk damage points, moderate risk damage points, and mild risk damage points, multiple preliminary damage level detection routes are generated. The target damage detection route is determined based on the information from the multiple preliminary damage detection routes.

3. The method of claim 1, wherein the step of determining the damage degree of the concrete structure is performed by using a damage degree determination table. The input to the graph neural network is the strain monitoring map, and the output of the graph neural network is multiple first abnormal damage areas of the concrete lining of the submerged tunnel.

4. The method for detecting the degree of concrete damage as described in claim 2, characterized in that, The risk damage point information includes the geometric morphology data of the risk damage point and the lining loss data.

5. A system for detecting the degree of damage of concrete, characterized by include: The first acquisition module is used to acquire multiple fiber optic strain monitoring data during the operation of the concrete lining of the submerged tunnel. A construction module is used to construct a strain monitoring map, which includes multiple fiber optic monitoring nodes and multiple edges between the multiple fiber optic monitoring nodes. The node features of the fiber optic monitoring nodes include fiber optic strain monitoring data and three-dimensional coordinates of the fiber optic sensor layout. The edges between the fiber optic monitoring nodes represent the similarity of the fiber optic strain monitoring data of the monitoring points. The first abnormal region determination module is used to process the strain monitoring spectrum based on a graph neural network to determine multiple first abnormal damage regions of the concrete lining of the submerged tunnel. The second acquisition module is used to acquire underwater laser scanning videos of multiple first abnormal damage areas of the concrete lining of the submerged tunnel. The second abnormal area determination module is used to determine multiple second abnormal damage areas of the concrete lining of the subsea immersed tunnel based on underwater laser scanning video of multiple first abnormal damage areas of the concrete lining of the subsea immersed tunnel. The detection route determination module is used to determine the target damage level detection route based on multiple second abnormal damage areas of the concrete lining of the underwater immersed tunnel. The damage detection module is used to detect the degree of damage to the concrete lining of the submerged tunnel based on the target damage detection route.

6. The concrete damage degree detection system of claim 5, wherein, The detection route determination module is also used for: Underwater laser scanning video of multiple secondary abnormal damage areas of the concrete lining of the subsea immersed tunnel was used to identify information on multiple risk damage points. K clusters were obtained by clustering based on information from multiple risk damage points; Based on the K clusters, information on severe risk damage points, moderate risk damage points, and mild risk damage points are determined. Based on the information on severe risk damage points, moderate risk damage points, and mild risk damage points, multiple preliminary damage level detection routes are generated. The target damage detection route is determined based on the information from the multiple preliminary damage detection routes.

7. The concrete damage degree detection system of claim 5, wherein, The input to the graph neural network is the strain monitoring map, and the output of the graph neural network is multiple first abnormal damage areas of the concrete lining of the submerged tunnel.

8. The concrete damage degree detection system of claim 6, wherein, The risk damage point information includes the geometric morphology data of the risk damage point and the lining loss data.

9. An electronic device, comprising: include: processor; Memory; And a computer program; wherein the computer program is stored in the memory and configured to be executed by the processor to implement the concrete damage detection method as claimed in any one of claims 1 to 4.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, When executed by the processor, the program implements the concrete damage detection method as described in any one of claims 1 to 4.

Images