A method and system for tunnel defect detection

Abstract

The application provides a tunnel defect detection method and system, relates to the technical field of tunnel defect detection, and comprises the following steps: acquiring first information, second information, third information and fourth information; performing fusion modeling processing on the first information to obtain a tunnel structure response matrix; performing dynamic coupling analysis based on the structure response matrix and the third information to generate a tunnel dynamic response model; performing defect detection processing on the second information to obtain a defect distribution atlas; forming a risk intensity distribution map; performing environmental correction analysis on the risk intensity distribution map based on the fourth information to determine tunnel defect information. The application constructs a tunnel defect detection method for multi-source information collaborative analysis by fusing tunnel structure data, image data, train operation data and environmental data. The method improves the identification accuracy and response efficiency of defects, can effectively realize dynamic evaluation and prediction of defect risks, and has strong environmental adaptability and practical application value.

Description

Technical Field

[0001] This invention relates to the field of tunnel defect detection technology, and more specifically, to a tunnel defect detection method and system. Background Technology

[0002] With the rapid development of rail transit infrastructure, tunnel structures operate in a high-frequency environment for extended periods, and are susceptible to structural defects such as cracks, leaks, and spalling due to multiple factors including train loads, environmental changes, and structural aging. Traditional tunnel defect detection methods largely rely on manual inspections or single image recognition techniques, which suffer from problems such as response lag, isolated information, and high false positive rates, making it difficult to meet the real-time and systematic requirements of modern rail systems for defect identification. Therefore, there is an urgent need for an intelligent detection method that can integrate multi-source heterogeneous information and achieve accurate defect identification and risk prediction to improve the level of tunnel structure safety monitoring and maintenance efficiency. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for detecting tunnel defects, thereby improving the aforementioned problems. To achieve this objective, the technical solution adopted by this invention is as follows:

[0004] Firstly, this application provides a method for detecting tunnel defects, including:

[0005] Acquire first information, second information, third information and fourth information, wherein the first information is tunnel structure sensor data, the second information is tunnel image data, the third information is train operation data, and the fourth information is tunnel operating environment data;

[0006] The first information is fused and modeled to obtain the tunnel structure response matrix;

[0007] Dynamic coupling analysis based on structural response matrix and third information is used to generate tunnel dynamic response model;

[0008] The second piece of information is processed for defect detection to obtain a defect distribution map;

[0009] By jointly analyzing the tunnel dynamic response model and the defect distribution map, a risk intensity distribution map is generated;

[0010] Based on the fourth information, an environmental correction analysis was performed on the risk intensity distribution map to determine the tunnel defect information.

[0011] Secondly, this application also provides a tunnel defect detection system, comprising:

[0012] The acquisition unit is used to acquire first information, second information, third information and fourth information, wherein the first information is tunnel structure sensor data, the second information is tunnel image data, the third information is train operation data and the fourth information is tunnel operating environment data;

[0013] A fusion modeling unit is used to perform fusion modeling processing on the first information to obtain the tunnel structure response matrix;

[0014] The first analysis unit is used to perform dynamic coupling analysis based on the structural response matrix and third information to generate a tunnel dynamic response model.

[0015] The detection unit is used to perform defect detection processing on the second information to obtain a defect distribution map.

[0016] The second analysis unit is used to jointly analyze the tunnel dynamic response model and the defect distribution map to form a risk intensity distribution map;

[0017] The correction unit is used to perform environmental correction analysis on the risk intensity distribution map based on the fourth information to determine the tunnel defect information.

[0018] The beneficial effects of this invention are as follows:

[0019] This invention constructs a tunnel defect detection method based on multi-source information collaborative analysis by integrating tunnel structure data, image data, train operation data, and environmental data. This method improves the accuracy and response efficiency of defect identification, effectively enables dynamic assessment and prediction of defect risks, possesses strong environmental adaptability and practical application value, and helps improve the intelligent level of tunnel operation and maintenance and operational safety assurance capabilities.

[0020] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing embodiments of the invention. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a schematic diagram of the tunnel defect detection method described in an embodiment of the present invention;

[0023] Figure 2 This is a schematic diagram of the tunnel defect detection system described in an embodiment of the present invention.

[0024] The diagram is labeled as follows: 10, acquisition unit; 20, fusion modeling unit; 30, first analysis unit; 40, detection unit; 50, second analysis unit; 60, correction unit. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0026] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0027] Example 1:

[0028] This embodiment provides a method for detecting tunnel defects.

[0029] See Figure 1 The figure shows that the method includes steps S10, S20, S30, S40, S50 and S60.

[0030] Step S10. Obtain first information, second information, third information and fourth information. The first information is tunnel structure sensor data, the second information is tunnel image data, the third information is train operation data, and the fourth information is tunnel operating environment data.

[0031] Specifically, the first piece of information consists of data collected by structural sensors deployed at key locations within the tunnel structure, including but not limited to basic physical quantities such as strain, acceleration, vibration, and temperature. This information directly reflects the real-time stress and response characteristics of the tunnel structure during train passage. The second piece of information comprises images of the tunnel's inner wall surface acquired through a tunnel image monitoring system, which can be used to identify defects such as cracks, spalling, and leakage. The third piece of information covers the train's speed, acceleration, load distribution, and vehicle type information during its operation within the tunnel, used to analyze the impact of train operation on the tunnel structure's response. The fourth piece of information consists of tunnel operating environment parameters, including external environmental data such as temperature, humidity, wind speed, and water stains, which plays a crucial role in correcting the generation, expansion, and accuracy of defect identification. The systematic acquisition and organization of this multi-dimensional data provides a comprehensive, dynamic, and correlated data foundation for subsequent response analysis and defect identification.

[0032] Step S20. Perform fusion modeling processing on the first information to obtain the tunnel structure response matrix;

[0033] Specifically, a response model of the tunnel structure under actual operating conditions is constructed using primary information. This involves collecting multi-dimensional sensor data (vibration, acceleration, stress, etc.) from different measuring points, and employing data cleaning, time alignment, and feature reconstruction methods to construct a spatiotemporal response feature vector reflecting the tunnel structure under load conditions. Based on this, a high-dimensional tunnel structure response matrix is ​​generated using a multi-dimensional data fusion algorithm. This response matrix accurately depicts the stress and deformation states of the structure under different sections, at different times, and under different train loads, serving as a crucial data foundation for subsequent dynamic response modeling and coupled analysis.

[0034] Step S30. Perform dynamic coupling analysis based on the structural response matrix and third information to generate a tunnel dynamic response model;

[0035] Specifically, by coupling the tunnel structure response matrix with third-party information, a dynamic mapping relationship between structural response and train operating status is established. Specifically, based on time synchronization mechanisms and trajectory reconstruction technology, structural response data is spatiotemporally matched with train operating parameters (such as speed, acceleration, load distribution, wheelbase, etc.) to construct a dynamic interaction model between the train and the tunnel. A machine learning regression model is used to generate a dynamic response model for the tunnel. This model can dynamically predict the response trend of the tunnel structure under specific train operating conditions, providing a precise foundation for modeling the causal relationship between defects and responses.

[0036] Step S40. Perform defect detection processing on the second information to obtain a defect distribution map;

[0037] Specifically, image defect recognition and localization processing is performed on the second piece of information to identify structural defects such as cracks, leaks, spalling, and deformation inside the tunnel.

[0038] Step S50. Perform joint analysis of the tunnel dynamic response model and the defect distribution map to form a risk intensity distribution map;

[0039] Specifically, the tunnel dynamic response model is deeply integrated with the defect distribution map to conduct defect risk level analysis. By analyzing the response intensity, load sensitivity, and structural degradation characteristics corresponding to defect areas, the potential hazards of various defects under dynamic loads are assessed. A risk intensity distribution map is constructed by combining coupling factor matrices, spatial convolution models, or graph network-based propagation mechanisms to achieve a quantitative assessment of potential safety hazards caused by structural defects under dynamic conditions. This map not only reflects the physical location and extent of defects but also incorporates the structural response background, possessing high timeliness and early warning value.

[0040] Step S60. Based on the fourth information, perform environmental correction analysis on the risk intensity distribution map to determine the tunnel defect information;

[0041] Specifically, the existing risk intensity distribution map is re-analyzed and revised based on fourth information to improve the accuracy and reliability of defect identification. Under the influence of environmental factors such as humidity, temperature difference, and water seepage, the defect characteristics of tunnel structures will change dynamically, such as crack propagation, material fatigue, or coating peeling.

[0042] Furthermore, step S20 includes steps S21 to S24:

[0043] Step S21. Based on the sensor deployment locations corresponding to the first information, the tunnel is divided into multiple regional units, and the sensor data in each regional unit is clustered and integrated to obtain the structural state response vector of each regional unit;

[0044] Specifically, based on the deployment locations of each sensor in the first information, the tunnel is divided into multiple regional units, each containing one or more sensors. For the sensor data in each regional unit, clustering and integration methods are applied, and structural state response vectors reflecting the overall structural state of the region are extracted through weighted averaging, principal component analysis, and other means.

[0045] Step S22. Map the structural state response vector of each region unit to three-dimensional spatial coordinates to generate a structural response mapping matrix;

[0046] Specifically, the structural state response vector of each regional unit is... Based on its corresponding three-dimensional spatial coordinates ), which is mapped onto the tunnel coordinate system to form a structural response mapping matrix.

[0047] The mapping matrix is Mapping matrix Each row in the table corresponds to a triplet of a region unit:

[0048]

[0049] in, It is a mapping matrix; For the first The three-dimensional spatial coordinates of each region unit; For the first Structural state response vector of each regional unit; The number of regional units, .

[0050] Step S23. Analyze and process the structural response mapping matrix based on the tunnel structure information to obtain the structural response feature matrix;

[0051] Specifically, based on the structural response mapping matrix, tunnel design information, such as lining thickness, cross-sectional dimensions, and support methods, is further incorporated for analysis and processing to conduct in-depth analysis of the response characteristics of each regional unit. Through feature extraction, characteristic information that can characterize the local structural performance state of the tunnel is mined, such as circumferential response differences, longitudinal gradient changes, and local anomaly indicators, ultimately generating a structural response feature matrix for each regional unit.

[0052] Step S24. Integrate the structural response feature matrices of all regional units according to the tunnel structure information to obtain the tunnel structure response matrix;

[0053] Specifically, firstly, based on multiple pre-divided regional units, the structural response feature matrix of each regional unit is collected. Then, based on the tunnel structural layout information provided in the first set of information, the spatial order and relative positional relationship of each regional unit within the overall tunnel structure are determined.

[0054] Specifically, the arrangement of each regional unit in the overall matrix is ​​determined based on the longitudinal mileage coordinates and circumferential zone numbers of the tunnel. During the integration process, the following rules are used for data organization: regional units of the same or adjacent sections are arranged in circumferential order; different sections are stacked sequentially along the longitudinal direction; for missing or abnormal data units, neighborhood interpolation or smoothing is used to ensure the continuity and integrity of the overall matrix; the dimensions and scale of the feature matrices of each region are unified to ensure the comparability and superposition of data from each unit during integration.

[0055] Through the above steps, the structural response feature matrices of all regional units are systematically integrated to generate a two-dimensional or three-dimensional tunnel structural response matrix. This response matrix, indexed by the tunnel's longitudinal mileage, circumferential angle, and feature type, can intuitively represent the overall response state of the tunnel at different locations and in different feature dimensions, providing fundamental data support for subsequent train load analysis and tunnel health assessment.

[0056] Furthermore, step S30 includes steps S31 to S305:

[0057] Step S31. Analyze the third information and construct the train operation feature vector;

[0058] Specifically, the third information includes, but is not limited to, dynamic operational data such as train speed, acceleration, axle load distribution, train formation, transit time, and train type. The parameters in the third information are normalized to unify the data dimensions of each feature dimension, facilitating subsequent neural network processing.

[0059] The normalized parameters are combined and arranged in a predetermined order to form a train operation feature vector. The train operation feature vector defined in this application is as follows:

[0060]

[0061] in, For the first The operational feature vector of a train; For the first The speed of the train; For the first The acceleration of the train; For the first The first train Axle load of each axle; For the first Train type code for each train; For the first The time when the trains pass through the tunnel; For the number of train axles, .

[0062] Step S32. Input the train operation feature vector and the tunnel structure response matrix into a preset temporal convolutional neural network model to generate a coupling weight matrix;

[0063] Specifically, the operational feature vector of each train and the tunnel structural response matrix at the corresponding time point are input into a temporal convolutional neural network model. The temporal convolutional neural network extracts the coupling relationship features between train operation and tunnel response in the time dimension through a series of convolution operations. During network training, supervised learning is employed, using actually observed response data as labels to optimize model parameters. After convolution, pooling, and normalization processing, the model outputs a coupling weight matrix between the train operation state and the tunnel response state. This matrix characterizes the correlation strength between train operation features and the responses of different areas of the tunnel.

[0064] Step S33. Based on the coupling weight matrix, construct a chain graph model representing the relationship between the train's operating state and the tunnel's structural response;

[0065] Specifically, the tunnel's various regional units are used as nodes in a graph model, with each node carrying the response characteristics of its corresponding regional unit. Directed edges are established between nodes based on the weight values ​​in the coupling weight matrix, with the edge weight determined by the magnitude of the correlation between the nodes. This chain-like structure effectively captures the correlation characteristics and transmission relationships of the tunnel's regional unit responses under train operation dynamics.

[0066] The resulting chain diagram model can be represented as:

[0067]

[0068] in, It is a chain diagram model; It is a set of nodes, i.e., a set of region units; Let it be the set of edges; It is an adjacency matrix, whose elements are determined by the values ​​in the coupling weight matrix.

[0069] Step S34. Calculate the degree of mutual influence between different regional units under each operating state based on the chain diagram model, and generate the regional coupling strength matrix;

[0070] Specifically, graph traversal and influence measurement methods are used to calculate the interaction strength of all node pairs (c, z) in the chain graph. The degree of influence can be calculated comprehensively through indicators such as path length, cumulative weight value, or information propagation capability, and is defined as:

[0071]

[0072] in, For nodes and nodes The intensity of the influence between them; For nodes To the node Path weights between them; For nodes To the node The Path; For nodes To the node The number of feasible paths between them .

[0073] Step S35. Construct a tunnel dynamic response model based on the joint modeling of the tunnel structure response matrix, train operation feature vector, and regional coupling strength matrix;

[0074] Specifically, the tunnel structure response matrix, train operation feature vector, and regional coupling strength matrix are jointly input into the joint modeling framework, which can be modeled using a multi-input fusion neural network.

[0075] The modeling process is optimized using the following objective function:

[0076]

[0077] in, The objective function is... For a genuine response; Output the predicted response for the model; for; and These are the weighting coefficients; The fitting loss between the actual response and the predicted response; This represents the loss in preserving the coupling relationships between regional units; This is a regularization term.

[0078] By continuously optimizing the above model parameters, the final tunnel dynamic response model can comprehensively reflect the influence of train dynamic operation on the structural state of different areas of the tunnel, and achieve accurate assessment of tunnel stress changes, potential anomalies and health status.

[0079] Furthermore, step S40 includes steps S41 to S45:

[0080] Step S41. Preprocess the second information to obtain the target image;

[0081] Specifically, the acquired tunnel image data, i.e., the second information, undergoes preprocessing operations, including but not limited to: image denoising (such as mean filtering and Gaussian filtering), image enhancement (such as contrast enhancement and histogram equalization), and size normalization (such as adjusting to a uniform resolution and size), thereby obtaining a clear and standardized target image.

[0082] Step S42. Divide the target image into structural regions based on tunnel structure information to obtain a structural region mapping matrix;

[0083] Specifically, based on pre-defined tunnel structure areas, such as the arch, sidewalls, and foundation, and combined with tunnel geometric parameters, the target image is... It is divided into several structural regions according to structural characteristics.

[0084] Let the total be divided into If there are multiple structural regions, then a structural region mapping matrix is ​​generated. ,in:

[0085]

[0086] Each pixel corresponds to a structural region number. In this application, the structural region numbers are:

[0087] , indicating the vaulted area; , indicating the area of ​​the left wall;

[0088] , indicating the area of ​​the right wall; , indicating the base area.

[0089] Step S43. Extract and identify the defect image features of each structural region based on the preset convolutional neural network model to obtain the defect detection results. The defect detection results shall at least include the defect detection coordinates.

[0090] Specifically, for each structural region, a local image is obtained through cropping. ,in Indicates the tunnel number Each structural region is input into a pre-trained defect detection convolutional neural network model. In this process, image features are extracted and defects are identified.

[0091] The defect detection process is as follows:

[0092]

[0093] in, For the first Detection results for each structural region; It is a convolutional neural network model; For the first Local images of structural regions.

[0094] The defect detection results should include at least a defect category label (such as crack, peeling, deformation, etc.) and the coordinates of the defect detection box, represented as follows:

[0095]

[0096] in, For the first Detection results for each structural region; For the first Defect category labels for each structural region; For the first The first structural region The x-coordinate of the center of the detection frame for each defect; For the first The first structural region The vertical coordinate of the center of the detection frame for each defect; For the first The first structural region The width of the detection frame for each defect; For the first The first structural region The length of the detection frame for each defect; For the first The number of defects detected in each structural region. .

[0097] Step S44. Convert the defect detection coordinates to the defect location in the tunnel structure coordinate system, and construct a defect location mapping model;

[0098] Specifically, using the center point of the detection frame as a reference, and combining camera calibration parameters or tunnel unfolding rules, the two-dimensional image coordinates are accurately mapped to the three-dimensional space of the tunnel physical structure, forming a one-to-one correspondence between the defect detection results and the structural coordinates, thereby constructing a defect location mapping relationship model.

[0099] Step S45. Based on the defect location mapping relationship model, perform spatial fusion processing on the defect detection results in each structural region to construct a defect distribution map;

[0100] Specifically, for each structural region The converted physical space defect location information is then fused, including: duplication elimination (if the distance between two defect locations is less than a set threshold, they are considered the same defect and merged); clustering and merging (closely adjacent defect points are clustered to form defect groups); and defect type encoding (spatial visualization encoding is performed based on defect categories, with different types marked with different colors / shapes). This generates a complete defect distribution map.

[0101] Specifically, step S50 includes steps S51 to S55:

[0102] Step S51. Based on the defect distribution map and the tunnel dynamic response model, construct the coupling correspondence matrix between defect location and structural response;

[0103] Specifically, the defect distribution map provides the location and type of each defect in the tunnel, while the dynamic response model reflects the dynamic response of the tunnel structure under different operating conditions. By coupling these two pieces of information, a matrix can be established to represent the response characteristics of the tunnel structure at a given defect location. This coupling matrix can reveal how different defects affect the structure's response and the dynamic behavior caused by the defects on the structure.

[0104] Step S52. Perform feature vector fusion on the structural response data corresponding to each defect location in the tunnel dynamic response model to obtain the local response feature set at each defect location;

[0105] Specifically, for each defect location, relevant structural response data, such as stress, displacement, and vibration, are extracted from the tunnel's dynamic response model. This response data varies depending on the defect's location. Therefore, feature vector fusion is required to obtain a local response feature set that comprehensively reflects the local impact of the defect on the tunnel structure.

[0106] Step S53. Assign a corresponding risk weight to each defect location based on the defect distribution map;

[0107] Specifically, the system assigns a risk weight to each defect location based on the defect distribution map. The risk weight reflects the potential risk that a defect may pose to the tunnel structure at different locations. These risk weights may be related to the type (such as cracks, corrosion, etc.), size, and location (such as near support points or curves) of the defect.

[0108] Risk weights are typically assigned based on historical data, expert experience, or calculated using models. For example, defects near support points may have higher risk weights, while defects far from support points may have lower risk weights.

[0109] Step S54. Perform risk scoring based on the local response feature set and risk weights to obtain a preliminary risk level score vector;

[0110] Specifically, the risk score quantifies the risk level of each defect by combining the response feature vector with the corresponding risk weight. The risk score can be calculated using a weighted average, as shown in the following formula:

[0111]

[0112] in, For the first Risk score for each defect; For the first The first defect and the first The coupling coefficient of a local response feature; For the first Risk weight of each defect; This represents the number of defects. .

[0113] Step S55. Based on all preliminary risk score vectors and their defect location relationships, reconstruct the risk intensity field to obtain a risk intensity distribution map;

[0114] Specifically, based on all preliminary risk score vectors and their defect location relationships, a risk intensity field is reconstructed. The reconstruction process, by considering the location of defects and their impact on the surrounding area, yields a global risk intensity distribution map. This map represents the risk intensity of different areas within the tunnel, effectively showing which areas' defects pose a greater threat to the overall safety of the tunnel.

[0115] Specifically, risk intensity field reconstruction is typically accomplished through spatial interpolation, weighted summation, or simulation based on physical models. The reconstructed risk intensity field can be presented visually using color coding or contour maps to intuitively display the risk distribution.

[0116] The final risk intensity distribution map can be represented as follows:

[0117]

[0118] in, This is a risk intensity distribution map; For the first Risk score for each defect; For the first The spatial interpolation function corresponding to the location of each defect; This represents the number of defects. .

[0119] Furthermore, step S60 includes steps S61 to S64:

[0120] Step S61. Normalize the fourth information to obtain the tunnel operating environment feature matrix;

[0121] Specifically, the tunnel's operating environment typically includes factors such as temperature, humidity, air pressure, and wind speed. These environmental factors directly affect the tunnel's dynamic response and defect development. Therefore, when processing this information, a normalization operation is required to ensure that different environmental factors are compared and calculated on the same scale. After normalization, a standardized tunnel operating environment characteristic matrix is ​​obtained.

[0122] Step S62. Perform matching analysis between the operating environment feature matrix and the risk intensity distribution map to generate an environmental risk corrected weight map;

[0123] Specifically, the normalized tunnel operating environment characteristic matrix will be matched with the risk intensity distribution map for analysis. This will correlate changes in the tunnel's environment with the existing risk intensity distribution map, assessing the impact of different environmental factors on defect risk. In particular, different operating environments (such as temperature and humidity) may have different effects on the weaknesses of the tunnel structure, thereby altering the risk intensity of defects. By combining environmental characteristics with the risk intensity distribution map, an environmental risk correction weight map is generated.

[0124] Step S63. Input the environmental risk corrected weight map and risk intensity distribution map into the preset multilayer perceptron to perform risk value correction prediction and obtain the corrected risk intensity distribution map;

[0125] Specifically, the environmental risk-corrected weight map and the risk intensity distribution map are input into a multilayer perceptron (MPB) to predict and correct risk values. An MPB is a common type of artificial neural network capable of learning the complex relationship between input features and output targets. Here, the MPB is used to combine environmental factors with defect risk data for risk prediction and correction, thereby generating a corrected risk intensity distribution map.

[0126] The input to the multilayer perceptron neural network is an environmental risk-corrected weight map and a risk intensity distribution map, and the output is the corrected defect risk intensity value. The training process of the neural network adjusts the model parameters by optimizing the loss function.

[0127] Step S64. Generate tunnel defect information for the regional units in the corrected defect risk intensity map that have a risk level higher than the preset safety threshold;

[0128] Specifically, the system compares the revised risk intensity distribution map with preset safety thresholds, identifying areas where the risk level exceeds the threshold. These areas represent higher-risk sections of the tunnel that require further inspection, maintenance, or reinforcement. The system extracts the defect information from these high-risk areas to generate the final tunnel defect information.

[0129] Furthermore, step S63 specifically includes steps S631 to S634:

[0130] Step S631. Based on the environmental risk modified weight map, perform feature weight extraction processing to construct a set of feature vectors that reflect the degree of impact of environmental factors on structural risk;

[0131] Specifically, the system performs feature weight extraction processing based on the environmental risk modified weight map, with the aim of identifying and extracting environmental factors affecting structural risk. Specifically, environmental factors (such as temperature, humidity, and air pressure) have different degrees of influence on structural risk, and these influences are reflected through the environmental risk modified weight map. By analyzing each environmental factor in the map, the weight of each environmental feature in structural risk is extracted, and these weights are formed into a feature vector set.

[0132] The set of feature vectors is:

[0133]

[0134] in, It is a set of feature vectors; For the first The weight of environmental factors.

[0135] Step S632. The feature vector set and the risk intensity distribution map are fused and encoded to obtain a multidimensional risk vector;

[0136] Specifically, the system will set the feature vectors Risk intensity distribution map A fusion encoding process is performed. The feature vector set reflects the impact of environmental factors on structural risks, while the risk intensity distribution map reflects the defect risk intensity of different parts of the tunnel. To integrate these two pieces of information, the system needs to fuse and encode them to obtain a multi-dimensional risk vector. This vector contains the combined influence between environmental factors and defect risks, providing a more comprehensive risk description.

[0137] The fusion process can typically be achieved through splicing, weighted averaging, or other algorithms. The formula is expressed as:

[0138]

[0139] in, This is the fused multidimensional risk vector; This is the fusion function; It is a set of feature vectors; This is a risk intensity distribution map.

[0140] Step S633. Input the multidimensional risk vector into the multilayer perceptron neural network model, perform nonlinear mapping and residual fitting processing, and obtain the corrected defect risk intensity prediction result;

[0141] Specifically, the fused multidimensional risk vector will be input into a multilayer perceptron neural network model. This model will perform nonlinear mapping and residual fitting to generate a revised defect risk intensity prediction. The multilayer perceptron can learn the complex relationship between input features and output targets through training. Specifically, the neural network performs nonlinear transformations on the input through multiple activation functions, thereby obtaining more accurate risk predictions.

[0142] The nonlinear mapping process involves weighted inputs, activation function applications, and output calculations for each layer in a multi-layer neural network. Residual fitting is used to improve the training performance of the neural network by introducing residual terms to reduce prediction errors.

[0143] Step S634. The corrected prediction results are partitioned according to the preset risk classification threshold to obtain the corrected risk intensity distribution map;

[0144] Specifically, the system compares the corrected defect risk intensity prediction results with preset risk classification thresholds and performs zoning processing. The risk level of each region is determined based on its prediction result and the preset threshold. If the risk intensity of a region exceeds the threshold, the region is marked as a high-risk region; otherwise, it is marked as a low-risk region. Through this processing, the system generates a corrected risk intensity distribution map. The risk classification thresholds are set according to specific applications and safety standards. Through the above process, a tunnel defect risk distribution map containing the corrected risk intensity is finally obtained, providing a detailed reference for tunnel safety assessment and maintenance.

[0145] Example 2:

[0146] like Figure 2 As shown, this embodiment provides a tunnel defect detection system, the system including:

[0147] The acquisition unit 10 is used to acquire first information, second information, third information and fourth information. The first information is tunnel structure sensor data, the second information is tunnel image data, the third information is train operation data, and the fourth information is tunnel operating environment data.

[0148] The fusion modeling unit 20 is used to perform fusion modeling processing on the first information to obtain the tunnel structure response matrix;

[0149] The first analysis unit 30 is used to perform dynamic coupling analysis based on the structural response matrix and third information to generate a tunnel dynamic response model.

[0150] Detection unit 40 is used to perform defect detection processing on the second information to obtain a defect distribution map;

[0151] The second analysis unit 50 is used to jointly analyze the tunnel dynamic response model and the defect distribution map to form a risk intensity distribution map;

[0152] The correction unit 60 is used to perform environmental correction analysis on the risk intensity distribution map based on the fourth information to determine the tunnel defect information.

[0153] In one specific embodiment disclosed in this application, the fusion modeling unit 20 includes:

[0154] The first partitioning unit is used to divide the tunnel into multiple regional units based on the sensor deployment locations corresponding to the first information, and to cluster and integrate the sensor data in each regional unit to obtain the structural state response vector of each regional unit.

[0155] The mapping unit is used to map the structural state response vector of each region unit to three-dimensional spatial coordinates to generate a structural response mapping matrix;

[0156] The third analysis unit is used to analyze and process the structural response mapping matrix based on the tunnel structure information to obtain the structural response feature matrix;

[0157] The integration unit is used to integrate the structural response feature matrices of all regional units according to the tunnel structural information to obtain the tunnel structural response matrix.

[0158] In one specific embodiment disclosed in this application, the first analysis unit 30 includes:

[0159] The parsing unit is used to parse the third information and construct the train operation feature vector;

[0160] The input unit is used to input the train operation feature vector and the tunnel structure response matrix into a preset temporal convolutional neural network model in a one-to-one correspondence, and generate a coupling weight matrix.

[0161] The first building unit is used to construct a chain graph model representing the relationship between train operation status and tunnel structural response based on the coupling weight matrix;

[0162] The first calculation unit is used to calculate the degree of mutual influence between different regional units under each operating state based on the chain diagram model, and generate the regional coupling strength matrix.

[0163] The modeling unit is used to jointly model the tunnel dynamic response model based on the tunnel structure response matrix, train operation feature vector, and regional coupling strength matrix.

[0164] In one specific embodiment disclosed in this application, the detection unit 40 includes:

[0165] The preprocessing unit is used to preprocess the second information to obtain the target image;

[0166] The second partitioning unit is used to divide the target image into structural regions based on tunnel structure information to obtain a structural region mapping matrix.

[0167] The extraction unit is used to extract and identify the defect image features of each structural region based on a preset convolutional neural network model, and obtain the defect detection results. The defect detection results include at least the defect detection coordinates.

[0168] The transformation unit is used to convert the defect detection coordinates into the defect location in the tunnel structure coordinate system and to build a defect location mapping model.

[0169] The first fusion unit is used to perform spatial fusion processing on the defect detection results in each structural region based on the defect location mapping relationship model to construct a defect distribution map.

[0170] In one specific embodiment disclosed in this application, the second analysis unit 50 includes:

[0171] The second building unit is used to construct the defect location and structural response coupling correspondence matrix based on the defect distribution map and the tunnel dynamic response model;

[0172] The second fusion unit is used to perform feature vector fusion on the structural response data corresponding to each defect location in the tunnel dynamic response model to obtain the local response feature set at each defect location.

[0173] The setting unit is used to set the corresponding risk weight for each defect location based on the defect distribution map;

[0174] The scoring unit is used to score risk based on the local response feature set and risk weights to obtain a preliminary risk level score vector.

[0175] The reconstruction unit is used to reconstruct the risk intensity field based on all preliminary risk score vectors and their defect location relationships, and obtain the risk intensity distribution map.

[0176] In one specific embodiment disclosed in this application, the correction unit 60 includes:

[0177] The normalization unit is used to normalize the fourth information to obtain the tunnel operating environment feature matrix;

[0178] The matching unit is used to perform matching analysis between the operating environment feature matrix and the risk intensity distribution map to generate an environmental risk corrected weight map.

[0179] The input unit is used to input the environmental risk corrected weight map and the risk intensity distribution map into the preset multilayer perceptron for risk value correction prediction, and obtain the corrected risk intensity distribution map.

[0180] The generation unit is used to generate tunnel defect information from the regions in the corrected defect risk intensity map that have a risk level higher than a preset safety threshold.

[0181] In one specific embodiment disclosed in this application, the substitution unit includes:

[0182] The third construction unit is used to extract feature weights based on the environmental risk modified weight map and construct a set of feature vectors that reflect the degree of impact of environmental factors on structural risks.

[0183] The third fusion unit is used to fuse the feature vector set with the risk intensity distribution map to obtain a multidimensional risk vector.

[0184] The execution unit is used to input the multidimensional risk vector into the multilayer perceptron neural network model, perform nonlinear mapping and residual fitting processing, and obtain the corrected defect risk intensity prediction result.

[0185] The partitioning unit is used to partition the corrected prediction results according to the preset risk classification threshold to obtain the corrected risk intensity distribution map.

[0186] It should be noted that the specific methods by which each module performs operations in the system described in the above embodiments have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0187] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0188] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for detecting tunnel defects, characterized in that, include: Acquire first information, second information, third information and fourth information, wherein the first information is tunnel structure sensor data, the second information is tunnel image data, the third information is train operation data, and the fourth information is tunnel operating environment data; The first information is fused and modeled to obtain the tunnel structure response matrix; Dynamic coupling analysis based on structural response matrix and third information is used to generate tunnel dynamic response model; The second piece of information is processed for defect detection to obtain a defect distribution map; By jointly analyzing the tunnel dynamic response model and the defect distribution map, a risk intensity distribution map is generated; Based on the fourth information, an environmental correction analysis was performed on the risk intensity distribution map to determine the tunnel defect information; The generation of the tunnel dynamic response model includes: The third piece of information is parsed to construct a train operation feature vector; The train operation feature vector and the tunnel structure response matrix are input into a preset temporal convolutional neural network model in a one-to-one correspondence to generate a coupling weight matrix, which represents the correlation between the train operation features and the response of each area of ​​the tunnel. Based on the coupling weight matrix, a chain graph model is constructed to characterize the relationship between train operation status and tunnel structural response. The nodes of the chain graph model are composed of regional units, and each node carries the response characteristics of the corresponding regional unit. Directed edges between nodes are established according to the weight values ​​in the coupling weight matrix. The weight of the edge is determined by the magnitude of the correlation between the nodes. The regional units are obtained by dividing the tunnel based on the sensor deployment positions corresponding to the first information. The degree of mutual influence between different regional units under each operating state is calculated based on the chain diagram model, and a regional coupling strength matrix is ​​generated. Based on the joint modeling of the tunnel structure response matrix, train operation feature vector, and regional coupling strength matrix, the tunnel dynamic response model is constructed. The tunnel dynamic response model reflects the influence of train dynamic operation on the structural state of different areas of the tunnel.

2. The tunnel defect detection method according to claim 1, characterized in that... The first information is fused and modeled to obtain the tunnel structure response matrix, including: The sensor data within each regional unit are clustered and integrated to obtain the structural state response vector of each regional unit; The structural state response vector of each regional unit is mapped to three-dimensional spatial coordinates to generate a structural response mapping matrix; The structural response mapping matrix is ​​analyzed and processed based on tunnel structural information to obtain the structural response feature matrix. The structural response feature matrices of all regional units are integrated according to the tunnel structural information to obtain the tunnel structural response matrix.

3. The tunnel defect detection method according to claim 2, characterized in that... The second piece of information is processed for defect detection to obtain a defect distribution map, including: The second information is preprocessed to obtain the target image; Based on tunnel structure information, the target image is divided into structural regions to obtain a structural region mapping matrix; Based on a preset convolutional neural network model, defect image features of each structural region are extracted and identified to obtain defect detection results, which at least include defect detection coordinates. The defect detection coordinates are converted into the defect location in the tunnel structure coordinate system, and a defect location mapping model is constructed. The defect detection results within each structural region are spatially fused based on a defect location mapping model to construct the defect distribution map.

4. The tunnel defect detection method according to claim 3, characterized in that... The tunnel dynamic response model and defect distribution map are jointly analyzed to form a risk intensity distribution map, including: Based on the aforementioned defect distribution map and tunnel dynamic response model, a coupling correspondence matrix between defect location and structural response is constructed. Feature vector fusion is performed on the structural response data corresponding to each defect location in the tunnel dynamic response model to obtain the local response feature set at each defect location. A corresponding risk weight is assigned to each defect location based on the defect distribution map; Risk scoring is performed based on local response feature sets and risk weights to obtain a preliminary risk level score vector. The risk intensity field is reconstructed based on all preliminary risk score vectors and their defect location relationships to obtain the risk intensity distribution map.

5. A tunnel defect detection system, characterized in that, include: The acquisition unit is used to acquire first information, second information, third information and fourth information, wherein the first information is tunnel structure sensor data, the second information is tunnel image data, the third information is train operation data and the fourth information is tunnel operating environment data; A fusion modeling unit is used to perform fusion modeling processing on the first information to obtain the tunnel structure response matrix; The first analysis unit is used to perform dynamic coupling analysis based on the structural response matrix and third information to generate a tunnel dynamic response model. The detection unit is used to perform defect detection processing on the second information to obtain a defect distribution map. The second analysis unit is used to jointly analyze the tunnel dynamic response model and the defect distribution map to form a risk intensity distribution map; The correction unit is used to perform environmental correction analysis on the risk intensity distribution map based on the fourth information to determine the tunnel defect information; The first analysis unit includes: The parsing unit is used to parse the third information and construct a train operation feature vector; The input unit is used to input the train operation feature vector and the tunnel structure response matrix into a preset temporal convolutional neural network model in a one-to-one correspondence, and generate a coupling weight matrix, which represents the correlation between the train operation features and the response of each area of ​​the tunnel. The first construction unit is used to construct a chain graph model representing the relationship between train operation status and tunnel structure response based on the coupling weight matrix. The nodes of the chain graph model are composed of region units. Each node carries the response characteristics of the corresponding region unit. Directed edges between nodes are established according to the weight values ​​in the coupling weight matrix. The weight of the edge is determined by the magnitude of the correlation between the nodes. The region units are obtained by dividing the tunnel according to the sensor deployment positions corresponding to the first information. The first calculation unit is used to calculate the degree of mutual influence between different regional units under each operating state based on the chain diagram model, and generate the regional coupling strength matrix. The modeling unit is used to jointly model the tunnel dynamic response model based on the tunnel structure response matrix, train operation feature vector and regional coupling strength matrix. The tunnel dynamic response model reflects the influence of train dynamic operation on the structural state of different areas of the tunnel.

6. The tunnel defect detection system according to claim 5, characterized in that, The fusion modeling unit includes: The first partitioning unit is used to divide the tunnel into multiple regional units based on the sensor deployment locations corresponding to the first information, and to cluster and integrate the sensor data in each regional unit to obtain the structural state response vector of each regional unit. The mapping unit is used to map the structural state response vector of each region unit to three-dimensional spatial coordinates to generate a structural response mapping matrix; The third analysis unit is used to analyze and process the structural response mapping matrix based on the tunnel structure information to obtain the structural response feature matrix; An integration unit is used to integrate the structural response feature matrices of all regional units according to the tunnel structural information to obtain the tunnel structural response matrix.

7. The tunnel defect detection system according to claim 5, characterized in that, The detection unit includes: The preprocessing unit is used to preprocess the second information to obtain the target image; The second partitioning unit is used to divide the target image into structural regions based on tunnel structure information to obtain a structural region mapping matrix. An extraction unit is used to extract and identify defect image features of each structural region based on a preset convolutional neural network model to obtain defect detection results, wherein the defect detection results include at least defect detection coordinates. The conversion unit is used to convert the defect detection coordinates into the defect location in the tunnel structure coordinate system and to construct a defect location mapping model. The first fusion unit is used to perform spatial fusion processing on the defect detection results in each structural region based on the defect location mapping relationship model to construct the defect distribution map.

8. The tunnel defect detection system according to claim 7, characterized in that, The second analysis unit includes: The second construction unit is used to construct a defect location and structural response coupling correspondence matrix based on the defect distribution map and the tunnel dynamic response model; The second fusion unit is used to perform feature vector fusion on the structural response data corresponding to each defect location in the tunnel dynamic response model to obtain the local response feature set at each defect location. The setting unit is used to set the corresponding risk weight for each defect location based on the defect distribution map; The scoring unit is used to score risk based on the local response feature set and risk weights to obtain a preliminary risk level score vector. The reconstruction unit is used to reconstruct the risk intensity field based on all preliminary risk score vectors and their defect location relationships, thereby obtaining the risk intensity distribution map.

Images