Intelligent detection method and system for tunnel health

Abstract

The invention relates to the technical field of tunnel traffic, in particular to an intelligent detection method for tunnel health, which comprises the following steps: acquiring tunnel monitoring data of a target position in a tunnel; obtaining stress balance data and thermal coupling data corresponding to the target position according to the strain force data and the temperature and humidity data of each node of the target position and preset tunnel basic data, and determining whether the current tunnel monitoring data is abnormal or not based on the stress balance data and the thermal coupling data; when determining that the current tunnel monitoring data is not abnormal, inputting the image data into a pre-trained tunnel health prediction model to obtain a health state prediction result of the target position; wherein the tunnel health prediction model is a convolutional neural network model trained based on the marked tunnel image. The tunnel health detection method has the beneficial effects that the accuracy and the reliability of tunnel health detection are greatly improved, an accurate basis is provided for tunnel operation and maintenance decisions, and the tunnel traffic safety is guaranteed.

Description

Technical Field

[0001] This application relates to the field of tunnel traffic technology, and in particular to an intelligent method and system for detecting tunnel health. Background Technology

[0002] As a key component of urban transportation, the structural health of rail transit tunnels directly affects operational safety. Over the long term, tunnel structure inspection and monitoring technologies have evolved from purely manual methods to automation and intelligent systems; however, current mainstream technologies still have many limitations.

[0003] Currently, tunnel inspection is mainly divided into traditional manual inspection and intelligent inspection. Traditional manual inspection relies on on-site observation and handheld instrument measurement, which is inefficient, labor-intensive, and prone to missed or false detections due to tunnel lighting and dust environment. In addition, it poses personnel safety risks and is difficult to adapt to large-scale routine tunnel monitoring.

[0004] Existing intelligent detection methods often rely on single types of data, which has certain limitations. Specifically, relying solely on image data is susceptible to environmental interference, leading to misjudgments; while relying solely on strain data cannot intuitively reflect the location and morphology of apparent defects, nor can it distinguish between strain anomalies caused by environmental factors and structural damage. Some multi-data fusion methods lack pre-verification of data validity, and predictions based on distorted data reduce the accuracy of results; furthermore, image detection results often remain at the pixel level, failing to accurately map to the physical space of the tunnel, requiring manual conversion and thus lacking practical applicability. Summary of the Invention

[0005] (a) Technical problems to be solved

[0006] In view of the above-mentioned shortcomings and deficiencies of the prior art, this application provides an intelligent detection method and system for tunnel health, which solves the technical problems of low efficiency and easy to miss detection in traditional manual detection methods, insufficient reliability due to environmental interference and difficulty in distinguishing between environmental and damage factors in single data-driven intelligent detection, and low prediction accuracy due to the lack of effective pre-verification in multi-data fusion monitoring methods.

[0007] (II) Technical Solution

[0008] To achieve the above objectives, the main technical solutions adopted in this application include:

[0009] In a first aspect, embodiments of this application provide an intelligent detection method for tunnel health, including:

[0010] Acquire tunnel monitoring data at the target location within the tunnel; the tunnel monitoring data includes image data of the target location and strain data and temperature and humidity data of multiple nodes at the target location;

[0011] Based on the strain and temperature and humidity data of each node at the target location, as well as the pre-set tunnel foundation data, obtain the stress balance data and thermo-coupling data corresponding to the target location, and determine whether there are any anomalies in the current tunnel monitoring data based on the stress balance data and thermo-coupling data.

[0012] When it is determined that there are no anomalies in the current tunnel monitoring data, the image data is input into a pre-trained tunnel health prediction model to obtain the health status prediction result of the target location; the tunnel health prediction model is a convolutional neural network model trained based on the labeled tunnel image.

[0013] Optionally, in one specific embodiment, the tunnel foundation data includes: tunnel elastic modulus, tunnel Poisson's ratio, tunnel foundation temperature, tunnel foundation humidity, and tunnel thermal expansion coefficient;

[0014] Based on the strain and temperature / humidity data of each node at the target location, as well as the pre-set tunnel foundation data, obtain the stress balance and thermo-coupling data corresponding to the target location, including:

[0015] Based on the strain data of each node of the target, as well as the pre-set tunnel elastic modulus and tunnel Poisson's ratio, the stress balance data corresponding to the target location is obtained.

[0016] Based on the temperature, humidity, and strain data of each node at the target location, as well as the pre-set tunnel foundation temperature, tunnel foundation humidity, and tunnel thermal expansion coefficient, the corresponding thermal coupling data of the target location are obtained.

[0017] Optionally, in a specific embodiment, stress balance data corresponding to the target location is obtained based on the strain data of each node of the target, as well as the pre-set tunnel elastic modulus and tunnel Poisson's ratio, including:

[0018] Based on the strain data of each node, the pre-set tunnel elastic modulus and tunnel Poisson's ratio, and the pre-set Formula 1, the radial stress data of each node is obtained; Formula 1 is:

[0019] ;

[0020] in, Here, E represents the radial stress data corresponding to the node at index i, and μ represents the tunnel's elastic modulus and Poisson's ratio. The strain force data corresponding to the node with index i;

[0021] When the radial stress data of each node is determined to be qualified, the stress distribution residual between the node and the next adjacent node is obtained based on the radial stress data of each node; the stress distribution residual is the difference between the radial stress data of two adjacent nodes.

[0022] Once the radial stress data for each node is confirmed to be acceptable, stress balance data is obtained based on all stress distribution residuals.

[0023] Optionally, in one specific embodiment, stress balance data is obtained based on all stress distribution residuals, including:

[0024] Based on all stress distribution residuals and a pre-set formula (Formula 2), stress balance data are obtained; formula 2 is:

[0025] ;

[0026] Among them, R y This is stress equilibrium data, where n is the number of nodes, i is the node index, and R... i+1 The stress distribution residual between the node at index i+1 and its next adjacent node.

[0027] Optionally, in one specific embodiment, the temperature and humidity data includes temperature data and humidity data;

[0028] Based on the temperature, humidity, and strain data of each node at the target location, as well as the pre-set tunnel foundation temperature, tunnel foundation humidity, and tunnel thermal expansion coefficient, the corresponding thermal coupling data for the target location is obtained, including:

[0029] Based on the temperature and humidity data of each node at the target location, as well as the pre-set tunnel foundation humidity and Formula 3, humidity correction parameters are obtained; Formula 3 is:

[0030] k i =1+0.002×(RH i -RH0);

[0031] Where, k i RH is the humidity correction parameter corresponding to the node with index i. i Here is the humidity data corresponding to the node with index i, where RH0 is the basic humidity of the tunnel.

[0032] Based on the temperature data, humidity data, humidity correction parameters, and radial stress data of each node at the target location, as well as the pre-set tunnel foundation humidity, tunnel foundation temperature, tunnel thermal expansion coefficient, tunnel elastic modulus, and tunnel Poisson's ratio, the mechanical strain data, temperature strain data, and humidity strain data corresponding to each node are obtained.

[0033] The mechanical strain data, temperature strain data, and humidity strain data of each node are weighted and summed to obtain the temperature and humidity strain data corresponding to each node. Based on each strain data, it is confirmed whether the temperature and humidity strain data of each node are abnormal.

[0034] When there are no anomalies in the temperature, humidity and strain data of each node, the thermal coupling data of the target location is obtained based on the temperature, humidity and strain data of all nodes.

[0035] Optionally, in a specific embodiment, based on the temperature data, humidity data, humidity correction parameters, and radial stress data of each node at the target location, as well as pre-set parameters such as tunnel foundation humidity, tunnel foundation temperature, tunnel thermal expansion coefficient, tunnel elastic modulus, and tunnel Poisson's ratio, the mechanical strain data, temperature strain data, and humidity strain data corresponding to each node are obtained, including:

[0036] Based on the temperature data and humidity correction parameters of each node at the target location, as well as the pre-set tunnel foundation temperature and tunnel thermal expansion coefficient, the temperature strain data corresponding to each node is obtained; the temperature strain data of any node is the product of the tunnel thermal expansion coefficient and humidity correction parameters, and the product of the difference between the temperature data of that node and the tunnel foundation temperature.

[0037] Based on the humidity data of each node at the target location and the pre-set humidity of the tunnel foundation, obtain the humidity strain data corresponding to each node; the humidity strain data of any node is the product of the difference between the humidity data of the electrolysis and the humidity of the tunnel foundation, and the pre-set humidity strain coefficient.

[0038] Based on the radial stress data of each node, the pre-set tunnel elastic modulus and tunnel Poisson's ratio, and the pre-set Formula 4, the mechanical strain data corresponding to each node is obtained; Formula 4 is:

[0039] ;

[0040] in, The mechanical strain data corresponding to the node with index i.

[0041] Optionally, in a specific embodiment, based on the temperature, humidity, and strain data of all nodes, thermal coupling data at the target location is obtained, including:

[0042] Based on the temperature, humidity, and strain data of all nodes, and the pre-set Formula 5, thermo-coupling data is obtained; Formula 5 is:

[0043] ;

[0044] Among them, L y This is thermally coupled data, where n is the number of nodes and i is the index of the node. The difference in temperature and humidity strain data between the node at index i and the next adjacent node.

[0045] Optionally, in one specific embodiment, image data is input into a pre-trained tunnel health prediction model to obtain a health status prediction result for the target location, including:

[0046] The image data is preprocessed to convert it into corresponding image tensor data; wherein, the preprocessing includes ROI cropping, distortion correction, image denoising, and pixel value normalization.

[0047] Feature extraction is performed on the image tensor data using multiple convolution kernels of different sizes to obtain corresponding multi-scale features;

[0048] Based on a pre-set channel attention mechanism, attention features corresponding to multi-scale features are obtained;

[0049] Attention features are input into pre-set classification fully connected layers and regression fully connected layers respectively to obtain corresponding damage types and regression results; the damage types include circumferential cracks, longitudinal cracks, lining spalling or no damage, and the regression results include damage location and damage size.

[0050] Based on the injury type and regression results, the health status prediction results at the target location are obtained.

[0051] Optionally, in a specific embodiment, before preprocessing the image data, a first mapping strategy between the image pixel coordinate system and the actual spatial coordinate system of the tunnel, and a second mapping strategy between the tunnel cross-section size and the image pixel size are established based on the pre-set camera calibration parameters and tunnel cross-section size parameters.

[0052] Then, based on the damage type and regression results, the health status prediction result at the target location is obtained, including:

[0053] Based on the regression results, the first mapping strategy, and the second mapping strategy, the actual damage location and actual damage size of the defect are confirmed.

[0054] Based on the damage type, actual damage location, and actual damage size, the health status prediction results for the target location are obtained.

[0055] Secondly, embodiments of this application provide an intelligent detection system for tunnel health, including: a sensor array and an edge processor deployed in the tunnel, and a control platform set at a remote end;

[0056] A sensor array is used to acquire tunnel monitoring data at target locations within the tunnel and send the tunnel monitoring data to an edge processor; the tunnel monitoring data includes image data of the target location and strain data and temperature and humidity data of multiple nodes at the target location;

[0057] The edge processor is used to obtain stress balance data and thermal coupling data corresponding to the target location based on the strain and temperature and humidity data of each node at the target location, as well as the pre-set tunnel foundation data. Based on the stress balance data and thermal coupling data, if it is determined that there are no abnormal monitoring data in the current tunnel monitoring data, the image data is sent to the control platform.

[0058] A control platform is used to input image data into a pre-trained tunnel health prediction model to obtain the health status prediction result of the target location; the tunnel health prediction model is a convolutional neural network model trained based on the labeled tunnel image.

[0059] (III) Beneficial Effects

[0060] This application presents an intelligent tunnel health detection method that acquires multi-dimensional monitoring data at the target location, encompassing both apparent damage characteristics and structural mechanics and environmental influencing factors, overcoming the limitations of single-data detection. Furthermore, by calculating stress balance data and thermo-coupling data to determine anomalies in the monitoring data, it effectively filters distorted data caused by sensor malfunctions and environmental interference, providing a reliable data foundation for subsequent model predictions and significantly reducing the risk of misjudgment. Only valid data is input into a pre-trained convolutional neural network model, enabling accurate identification of damage types, locations, and sizes while reducing the burden of processing invalid data and improving detection efficiency. Simultaneously, the combination of multi-modal data fusion and physical constraint verification significantly improves the accuracy and reliability of tunnel health detection, providing precise basis for tunnel operation and maintenance decisions and ensuring tunnel safety. Attached Figure Description

[0061] Figure 1 This is a schematic flowchart of an intelligent detection method for tunnel health provided in an embodiment of this application;

[0062] Figure 2 This is a schematic diagram of an image recognition flowchart provided in an embodiment of this application;

[0063] Figure 3 This is a schematic diagram of the sensor array arrangement provided in an embodiment of this application;

[0064] Figure 4 This is a schematic diagram of a sensor node provided in an embodiment of this application. Detailed Implementation

[0065] To better explain and facilitate understanding of this application, the following detailed description of the application is provided in conjunction with the accompanying drawings and specific embodiments.

[0066] As a crucial component of urban transportation, the structural health of rail transit tunnels directly impacts operational safety. Tunnel inspection technology has evolved from purely manual methods to automated and intelligent systems, but current mainstream solutions still have many limitations. Traditional manual inspection relies on on-site observation and handheld instrument measurements, resulting in low efficiency, high labor intensity, susceptibility to tunnel lighting and dust environments leading to missed or false detections, and posing personnel safety risks. It is ill-suited to the needs of large-scale, routine tunnel monitoring. Existing intelligent inspection methods largely rely on single data points, exhibiting significant limitations. Some multi-data fusion methods lack prior verification of data validity, leading to poor applicability of image inspection results.

[0067] The intelligent tunnel health detection method proposed in this application acquires multi-dimensional monitoring data covering apparent damage, structural mechanics, and environmental factors, overcoming the limitations of single-data detection. It identifies data anomalies by calculating stress balance and thermo-coupling data, filtering distorted data to provide a reliable foundation for model prediction and reduce the risk of misjudgment. Valid data is input into a pre-trained convolutional neural network model to accurately identify damage information and improve detection efficiency. The combination of multi-modal data fusion and physical constraint verification significantly improves detection accuracy and reliability, providing precise basis for operation and maintenance decisions and ensuring traffic safety.

[0068] To better understand the above technical solutions, exemplary embodiments of this application will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of this application are shown in the drawings, it should be understood that this application can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this application can be understood more clearly and thoroughly, and that the scope of this application can be fully conveyed to those skilled in the art.

[0069] This application provides an intelligent method for detecting tunnel health, such as... Figure 1 As shown, it includes:

[0070] S1. Acquire tunnel monitoring data at the target location within the tunnel; the tunnel monitoring data includes image data of the target location and strain data and temperature and humidity data of multiple nodes at the target location;

[0071] S2. Based on the strain force data and temperature and humidity data of each node at the target location, as well as the pre-set tunnel foundation data, obtain the stress balance data and thermo-coupling data corresponding to the target location, and determine whether there are any abnormalities in the current tunnel monitoring data based on the stress balance data and thermo-coupling data.

[0072] S3. When it is determined that there are no abnormalities in the current tunnel monitoring data, the image data is input into the pre-trained tunnel health prediction model to obtain the health status prediction result of the target location; the tunnel health prediction model is a convolutional neural network model trained based on the labeled tunnel image.

[0073] The embodiments provided in this application acquire multi-dimensional monitoring data of the target location, covering both apparent damage characteristics and structural mechanics and environmental influencing factors, thus overcoming the limitations of single-data detection. By calculating stress balance data and thermo-coupling data to determine anomalies in the monitoring data, distorted data caused by sensor failures and environmental interference can be effectively filtered out, providing a reliable data foundation for subsequent model predictions and significantly reducing the risk of misjudgment. Only valid data is input into the pre-trained convolutional neural network model, which can accurately identify the damage type, location, and size, while reducing the processing pressure of invalid data and improving detection efficiency. At the same time, the combination of multimodal data fusion and physical constraint verification greatly improves the accuracy and reliability of tunnel health detection, providing accurate basis for tunnel operation and maintenance decisions and ensuring tunnel passage safety.

[0074] Optionally, in a specific embodiment, the tunnel foundation data includes: tunnel elastic modulus, tunnel Poisson's ratio, tunnel foundation temperature, tunnel foundation humidity, and tunnel thermal expansion coefficient;

[0075] Based on the strain and temperature / humidity data of each node at the target location, as well as the pre-set tunnel foundation data, obtain the stress balance and thermo-coupling data corresponding to the target location, including:

[0076] Based on the strain data of each node of the target, as well as the pre-set tunnel elastic modulus and tunnel Poisson's ratio, the stress balance data corresponding to the target location is obtained.

[0077] Based on the temperature, humidity, and strain data of each node at the target location, as well as the pre-set tunnel foundation temperature, tunnel foundation humidity, and tunnel thermal expansion coefficient, the corresponding thermal coupling data of the target location are obtained.

[0078] Furthermore, based on the strain data of each node of the target, and the pre-set tunnel elastic modulus and tunnel Poisson's ratio, the stress balance data corresponding to the target location is obtained, including:

[0079] Based on the strain data of each node, the pre-set tunnel elastic modulus and tunnel Poisson's ratio, and the pre-set Formula 1, the radial stress data of each node is obtained; Formula 1 is:

[0080] ;

[0081] in, Here, E represents the radial stress data corresponding to the node at index i, and μ represents the tunnel's elastic modulus and Poisson's ratio. The strain force data corresponding to the node with index i;

[0082] When the radial stress data of each node is deemed acceptable, the stress distribution residual between the node and the next adjacent node is obtained based on the radial stress data of each node; the stress distribution residual is the difference between the radial stress data of two adjacent nodes.

[0083] Once the radial stress data for each node is confirmed to be acceptable, stress balance data is obtained based on all stress distribution residuals and a pre-set formula (Formula 2). Formula 2 is:

[0084] ;

[0085] Among them, R y This is stress equilibrium data, where n is the number of nodes, i is the node index, and R... i+1 The stress distribution residual between the node at index i+1 and its next adjacent node.

[0086] Specifically, pre-set parameters are retrieved from the tunnel foundation database, including the tunnel's elastic modulus E (e.g., 3.45 × 10⁻⁶ for C50 concrete). 4 MPa), tunnel Poisson's ratio μ (typically taken as 0.2), and simultaneously acquire raw strain data ε1, ε2...ε at n monitoring nodes at the target location. n (Unit: με), and perform preliminary data processing to remove obvious outliers (such as mutation data exceeding ±500 με).

[0087] Based on Formula 1, pay attention to calculating the radial stress data for each node. Ensure units are consistent during calculation; strain data needs to be converted to dimensionless form, such as (120με = 120 × 10⁻⁶). -6 Finally, the stress value (unit: MPa) of each node is obtained.

[0088] Based on tunnel operation and maintenance standards, the acceptable range for radial stress data is set (e.g., 0.5MPa≤). (≤8MPa), carefully verify that the radial stress data of each node is within the acceptable range. If the stress at any node exceeds the range, the data for this batch is directly deemed abnormal, and subsequent calculations are terminated; if the stress at all nodes is acceptable, proceed to the next step.

[0089] Based on valid radial stress data, the stress distribution residual R for each pair of adjacent nodes is calculated. i The residual is the difference between the radial stress data of two adjacent nodes, i.e. , where i ranges from 1 to n-1, ensuring that the residual is non-negative and reflects the fluctuation range of stress at adjacent nodes.

[0090] Based on Formula 2, the global stress balance data R at the target location is obtained by summing all stress distribution residuals and taking the average value. yR y The smaller the value, the more uniform the stress distribution in the region, and the more stable the structure under stress.

[0091] Taking three detection nodes (n=3) at a certain tunnel cross-section as an example, the following example illustrates the point:

[0092] Obtain tunnel foundation data: E = 3.45 × 10 4 MPa, μ=0.2;

[0093] The preprocessed strain data are ε1=120με, ε2=115με, and ε3=110με;

[0094] First, based on the pre-set Formula 1, calculate the coefficient term: the coefficient is (3.45 × 10 4 ×(1-0.2)) / ((1+0.2)(1-0.4))≈38333.3MPa;

[0095] Note the calculation of radial stress data for each node, the first node being... =38333.3×120×10 -6 ≈4.60MPa, the second node is =38333.3×115×10 -6 ≈4.41MPa. Similarly, the third node is 4.22MPa.

[0096] The radial stress data for the three nodes are 4.60 MPa, 4.41 MPa and 4.22 MPa, respectively, all within the acceptable range of 0.5~8 MPa, and are therefore judged to be fully qualified.

[0097] Calculate the stress distribution residual: =0.19MPa, =0.19MPa;

[0098] Then R y =0.19MPa, meaning the final stress balance data for this section is 0.19MPa. Combined with the preset threshold (e.g., ≤0.5MPa is considered acceptable), it is determined that the stress distribution in this area is uniform, the data is valid, and it is preliminarily deemed acceptable.

[0099] The embodiments of this application first retrieve standardized tunnel foundation parameters and preprocess raw strain data to eliminate abnormal data in advance, ensuring the reliability of the calculation basis. Based on Formula 1, which conforms to the characteristics of tunnel lining materials, the radial stress of a single node is accurately calculated. Combined with a clear qualified range verification, node data with abnormal stress can be effectively screened out, preventing distorted data from entering subsequent processes. By calculating the stress distribution residual of adjacent nodes and combining it with Formula 2 to obtain global stress balance data, the uniformity of stress distribution in the target area can be intuitively reflected, and the structural stress stability can be accurately judged. The entire process is logically rigorous, with clear steps and quantifiable execution. It not only breaks through the subjective limitations of traditional stress analysis, but also improves the accuracy of data validity judgment through multi-stage verification, providing a reliable mechanical basis for tunnel health detection. At the same time, the illustrative examples reduce the difficulty of implementing the solution and facilitate practical application by engineers.

[0100] Furthermore, when anomalies are identified in the strain force data (insufficient number of valid data points after removing outliers from the original data / significant abrupt changes still exist in the remaining data), it indicates a fault in the sensor acquisition process or severe environmental interference, leading to distorted strain force data. Possible causes include: loose sensor wiring, electromagnetic interference, sudden severe vibrations within the tunnel (such as train derailment or construction blasting), etc., resulting in collected data that cannot reflect the true stress state of the tunnel structure.

[0101] Immediately mark this batch of tunnel monitoring data as "invalid," do not upload it to the control platform, and do not proceed to the subsequent stress calculation process; then, trigger a sensor fault warning, and push the maintenance personnel to check the strain sensor at the corresponding node, check the wiring, power supply, and installation and fixation, and retrieve the tunnel environmental data from the same period (such as train operation records and construction logs) to check for any sudden interference factors; after the sensor fault is resolved / the interference factors are eliminated, re-collect the strain data at the target location, and then start the subsequent process.

[0102] Furthermore, when the radial stress data of any node is unqualified (the radial stress data of any node exceeds the qualified range of 0.5MPa~8MPa), it indicates that the structure of the corresponding node is under abnormal stress, and there are two core possibilities: First, the sensor data is distorted (such as sensor drift or failure), causing the calculated stress value to deviate from the actual value; second, the tunnel structure corresponding to the node has local damage (such as crack penetration or lining voids), causing stress concentration or stress drop, which exceeds the normal stress range.

[0103] Terminate the stress balance calculation for this batch of data and mark it as "radial stress anomaly data"; prioritize checking the strain sensors of the corresponding non-conforming nodes: recalibrate the sensors, compare with historical data, and determine if there is a sensor malfunction; if the sensor is normal, immediately retrieve the image data of the node and manually check whether there is apparent damage (such as cracks or spalling) at the corresponding tunnel location; if the image data shows damage, trigger a structural damage warning, push it to maintenance personnel for on-site inspection, and further assess the severity of the damage; if the image is normal, combine temperature and humidity data to investigate whether the stress anomaly is caused by environmental factors (such as extreme temperature and humidity), and continuously monitor the data change trend of the node.

[0104] If the global stress balance data is unqualified (e.g., exceeding a preset threshold), it indicates that the stress distribution at the target location is uneven, and the stress fluctuations of adjacent nodes exceed a reasonable range, reflecting insufficient structural stability in that area of ​​the tunnel. Possible causes include: multi-node sensor cluster failure, continuous damage in local areas of the tunnel (e.g., circumferential cracking), and stress redistribution caused by tunnel foundation settlement.

[0105] Mark the data in this area as having an abnormal stress distribution and suspend the input of data from this area into the tunnel health prediction model; conduct a comprehensive check of strain sensors at all nodes at this target location to confirm whether there is a cluster failure (such as power supply abnormalities or signal interference); retrieve image data of this area and check the apparent damage node by node, focusing on the locations of adjacent nodes with large stress residuals; combine historical tunnel monitoring data to analyze whether the abnormal stress distribution is a long-term trend change (such as caused by foundation settlement). If it is a sudden anomaly, immediately initiate on-site special testing; if it is a sensor failure, repair it and re-collect data; if it is structural damage, formulate targeted treatment plans (such as grouting reinforcement or lining repair) according to the severity of the damage.

[0106] This embodiment provides a clear direction for operation and maintenance troubleshooting by accurately locating the core causes of anomalies in raw data, single-node radial stress anomalies, and global stress balance anomalies, avoiding blind handling. It establishes a closed-loop handling mechanism of "marking invalid data - early warning push - multi-dimensional verification - problem resolution - re-collection," which can quickly filter distorted data and prevent unqualified data from entering subsequent testing processes, ensuring the accuracy of tunnel health testing results. It utilizes multiple methods, including sensor verification, image inspection, environmental data tracing, and historical data comparison, to achieve efficient source tracing and precise handling of anomalies for different anomaly types. Simultaneously, it promotes precise allocation of operation and maintenance resources through tiered early warnings (sensor fault warning, structural damage warning). The entire handling logic considers issues at the data, equipment, and structural levels, reducing the interference of invalid data on testing results while promptly identifying potential safety hazards in the tunnel structure, providing reliable support for tunnel operation and maintenance decisions, and ensuring tunnel operational safety.

[0107] Optionally, in one specific embodiment, the temperature and humidity data includes temperature data and humidity data;

[0108] Based on the temperature, humidity, and strain data of each node at the target location, as well as the pre-set tunnel foundation temperature, tunnel foundation humidity, and tunnel thermal expansion coefficient, the corresponding thermal coupling data for the target location is obtained, including:

[0109] Based on the temperature and humidity data of each node at the target location, as well as the pre-set humidity of the tunnel foundation and Formula 3, the humidity correction parameters are obtained; Formula 3 is:

[0110] k i =1+0.002×(RH i -RH0);

[0111] Where, k i RH is the humidity correction parameter corresponding to the node with index i. i Here is the humidity data corresponding to the node with index i, where RH0 is the basic humidity of the tunnel.

[0112] Based on the temperature data, humidity data, humidity correction parameters, and radial stress data of each node at the target location, as well as the pre-set tunnel foundation humidity, tunnel foundation temperature, tunnel thermal expansion coefficient, tunnel elastic modulus, and tunnel Poisson's ratio, the mechanical strain data, temperature strain data, and humidity strain data corresponding to each node are obtained.

[0113] The mechanical strain data, temperature strain data, and humidity strain data of each node are weighted and summed to obtain the temperature and humidity strain data corresponding to each node. Based on each strain data, it is confirmed whether the temperature and humidity strain data of each node are abnormal.

[0114] When there are no anomalies in the temperature, humidity and strain data of each node, the thermal coupling data of the target location is obtained based on the temperature, humidity and strain data of all nodes.

[0115] Furthermore, based on the temperature data, humidity data, humidity correction parameters, and radial stress data of each node at the target location, as well as pre-set parameters such as tunnel foundation humidity, tunnel foundation temperature, tunnel thermal expansion coefficient, tunnel elastic modulus, and tunnel Poisson's ratio, the mechanical strain data, temperature strain data, and humidity strain data corresponding to each node are obtained, including:

[0116] Based on the temperature data and humidity correction parameters of each node at the target location, as well as the pre-set tunnel foundation temperature and tunnel thermal expansion coefficient, the temperature strain data corresponding to each node is obtained; the temperature strain data of any node is the product of the tunnel thermal expansion coefficient and humidity correction parameters, and the product of the difference between the temperature data of that node and the tunnel foundation temperature.

[0117] Based on the humidity data of each node at the target location and the pre-set humidity of the tunnel foundation, obtain the humidity strain data corresponding to each node; the humidity strain data of any node is the product of the difference between the humidity data of the electrolysis and the humidity of the tunnel foundation, and the pre-set humidity strain coefficient.

[0118] Based on the radial stress data of each node, the pre-set tunnel elastic modulus and tunnel Poisson's ratio, and the pre-set Formula 4, the mechanical strain data corresponding to each node is obtained; Formula 4 is:

[0119] ;

[0120] in, The mechanical strain data corresponding to the node with index i.

[0121] Furthermore, based on the temperature, humidity, and strain data of all nodes, thermal coupling data at the target location is obtained, including:

[0122] Based on the temperature, humidity, and strain data of all nodes, and the pre-set Formula 5, thermo-mechanical coupling data are obtained; Formula 5 is:

[0123] ;

[0124] Among them, L y This is thermally coupled data, where n is the number of nodes and i is the index of the node. The difference in temperature and humidity strain data between the node at index i and the next adjacent node.

[0125] Specifically, preset parameters are retrieved from the tunnel foundation database: tunnel foundation temperature T0 (taken as 20℃), tunnel foundation humidity RH0 (taken as 60%RH), and tunnel thermal expansion coefficient α (e.g., 1.2×10 for C50 concrete). -5 / ℃), humidity strain coefficient k h (Take 1με / %RH), tunnel elastic modulus E (3.45×10) 4 MPa), tunnel Poisson's ratio μ (taken as 0.2); simultaneously acquire temperature and humidity data (T1, T2, ..., T) at m monitoring nodes at the target location. n RH1, RH2, ..., RH n ), raw strain data ε1, ε2...ε n , and the radial stress data calculated in the above steps.

[0126] Based on Formula 3, the humidity correction parameter for each node is calculated one by one. This parameter is used to correct the influence of humidity on the thermal expansion coefficient of concrete, so that the temperature strain calculation is more in line with the actual environment.

[0127] Subsequently, the temperature strain, humidity strain, and mechanical strain of each node were calculated separately, ensuring that the units were consistent (the strain unit is με):

[0128] Temperature strain reflects the structural strain caused by temperature changes combined with humidity effects, and it is calculated as α×k. i ×(T i -T0), where k i T is the humidity correction parameter for node i. i For the temperature data of node i, humidity strain reflects the shrinkage or expansion strain of concrete caused by humidity changes, and its calculation formula is k. h ×(RH i -RH0), RH i For the humidity data of node i, mechanical strain reflects the pure mechanical strain caused by structural stress, and its calculation formula is Formula 4.

[0129] The three strain components of each node are weighted and summed to obtain temperature and humidity strain data. The weights are set according to tunnel operation experience (e.g., temperature strain weight 0.4, humidity strain weight 0.2, mechanical strain weight 0.4).

[0130] Calculate the deviation between the actual strain data and the temperature and humidity strain data, and set an appropriate threshold (e.g., ≤5με). If the temperature and humidity strain data of any node exceeds the threshold, the temperature and humidity strain of that node is determined to be abnormal; if all nodes meet the standard, proceed to the next step.

[0131] Based on qualified temperature and humidity strain data, the temperature and humidity strain difference ΔLi between adjacent nodes is calculated. Based on Formula 5, all temperature and humidity strain differences are summed and averaged to obtain the global thermo-mechanical coupling data of the target location. The smaller the global thermo-mechanical coupling data, the more stable the environment-structure coupling relationship in the area, and the more reliable the data.

[0132] The embodiments of this application retrieve standardized basic parameters that conform to the characteristics of tunnel lining materials, and combine them with humidity correction parameters to accurately correct the influence of the environment on the coefficient of thermal expansion, ensuring that the temperature strain calculation conforms to the actual working conditions. The strain data is subdivided into three components: temperature, humidity, and mechanical components, and then weighted and fused to comprehensively quantify the combined influence of environmental and structural stress on tunnel strain, breaking through the limitations of single strain analysis. By comparing the deviation between actual strain and theoretical coupled strain and determining the threshold, abnormal data caused by environmental interference or sensor failure can be effectively screened out, ensuring the reliability of input data. The global thermo-mechanical coupling data is calculated based on the average value of the coupling strain difference between adjacent nodes, which can intuitively reflect the stability of the environmental-structural coupling relationship in the target area, providing an accurate basis for environmental-structural collaborative analysis for tunnel health detection. The entire process is logically rigorous and quantifiable, taking into account both data accuracy and engineering practicality, which not only improves the scientific nature of thermo-mechanical coupling analysis, but also lays a solid data foundation for subsequent tunnel health status determination.

[0133] Furthermore, when the temperature, humidity, and strain data of a single node are unqualified, it indicates that the actual collected strain data deviates significantly from the theoretically calculated temperature, humidity, and strain data, suggesting an imbalance in the environment-structure coupling relationship. The core reasons fall into three categories: first, malfunction of the temperature, humidity, or strain sensors, leading to distorted data; second, extreme abrupt changes in the local tunnel environment (such as sudden high temperatures or water seepage), exceeding the adaptability range of conventional coupling models; and third, damage to the tunnel structure corresponding to that node (such as cracks or lining peeling), resulting in abnormal strain that cannot be explained by environmental and mechanical factors.

[0134] Immediately mark this node as an abnormal coupling node and suspend its data from participating in global thermal coupling data calculations; prioritize checking the temperature and humidity sensors and strain sensors of this node: calibrate sensor accuracy, compare historical data fluctuation trends, and check for faults such as sensor drift or loose wiring; if the sensors are normal, retrieve the synchronous image data of this node and manually check whether there is any apparent damage or sudden environmental anomaly (such as water seepage or lining spalling) at the corresponding tunnel location; if damage is found, trigger a structural damage warning and push it to maintenance personnel for on-site inspection and assessment; if it is a sudden environmental change, continue monitoring the node's data until the environment returns to stability; if it is a sensor failure, repair it and re-collect data, then recalculate the temperature, humidity, and strain data.

[0135] When the global thermo-coupling data is unqualified, it indicates that the temperature, humidity, and strain data of adjacent nodes at the target location fluctuate beyond the reasonable range, reflecting poor stability of the environment-structure coupling relationship in that area. Possible causes include: multi-node sensor cluster failure (such as abnormal power supply or signal interference), uneven environmental distribution in local areas of the tunnel (such as local water seepage or ventilation dead zones), and continuous damage to the tunnel structure (such as circumferential cracking or uneven foundation settlement), leading to significant differences in the coupled strain of adjacent nodes.

[0136] Mark the target area as an area of ​​coupling anomaly and suspend the input of data from this area into the tunnel health prediction model; comprehensively check the temperature, humidity, and strain sensors of all nodes in this area to confirm whether there is a cluster failure, and simultaneously check the sensor power supply and signal transmission links; retrieve the overall image data of this area, check the apparent damage node by node, pay special attention to the corresponding positions of adjacent nodes with large stress residuals, and determine whether there is continuous structural damage; combine the tunnel's historical monitoring data to analyze whether the coupling anomaly is a long-term trend change (such as caused by foundation settlement). If it is a sudden anomaly, immediately initiate on-site special environmental testing (such as temperature and humidity distribution, water seepage) and structural testing; if it is a sensor failure, repair it and re-collect data and calculate; if it is an uneven environment, optimize the tunnel ventilation and drainage system; if it is structural damage, formulate treatment plans such as grouting reinforcement and lining repair according to the severity of the damage.

[0137] This application provides a clear direction for operation and maintenance troubleshooting by accurately locating the core causes of single-node coupling anomalies and global coupling anomalies, avoiding inefficient blind handling. It establishes a complete process of "marking anomalies - prioritizing verification - multi-dimensional tracing - tiered handling - data closure," which can quickly isolate unqualified data, prevent distorted data from affecting subsequent test results, and ensure the accuracy of tunnel health monitoring. For different anomaly types, it uses multiple methods such as sensor calibration, image inspection, environmental monitoring, and historical data comparison to achieve efficient tracing and precise handling of anomaly problems. Simultaneously, it promotes precise allocation of operation and maintenance resources through tiered early warning. The handling logic considers three core issues: equipment failure, environmental anomalies, and structural damage. This not only solves high-frequency data and equipment problems but also promptly identifies potential safety hazards in the tunnel structure, providing reliable support for tunnel operation and maintenance decisions and effectively ensuring tunnel operation safety and the stability of the monitoring system.

[0138] Optionally, in one specific embodiment, image data is input into a pre-trained tunnel health prediction model to obtain a health status prediction result for the target location, such as... Figure 2 As shown, it includes:

[0139] The image data is preprocessed to transform it into corresponding image tensor data. The preprocessing includes ROI cropping, distortion correction, image denoising, and pixel value normalization.

[0140] Feature extraction is performed on image tensor data using multiple convolutional kernels of different sizes to obtain corresponding multi-scale features;

[0141] Based on a pre-set channel attention mechanism, attention features corresponding to multi-scale features are obtained;

[0142] The attention features are input into the pre-set classification fully connected layer and regression fully connected layer respectively to obtain the corresponding damage type and regression result; the damage type includes circumferential crack, longitudinal crack, lining spalling or no damage, and the regression result includes damage location and damage size.

[0143] Based on the injury type and regression results, the health status prediction results at the target location are obtained.

[0144] Furthermore, before preprocessing the image data, a first mapping strategy between the image pixel coordinate system and the actual spatial coordinate system of the tunnel, and a second mapping strategy between the tunnel cross-section size and the image pixel size are established based on the pre-set camera calibration parameters and tunnel cross-section size parameters.

[0145] Then, based on the injury type and regression results, the predicted health status at the target location is obtained, including:

[0146] Based on the regression results, the first mapping strategy, and the second mapping strategy, the actual damage location and actual damage size of the defect are confirmed.

[0147] Based on the damage type, actual damage location, and actual damage size, the health status prediction results for the target location are obtained.

[0148] Specifically, preset camera calibration parameters (intrinsic parameters: focal length f, principal point coordinates (u0, v0), distortion coefficients q1, q2, p1, p2; extrinsic parameters: camera installation mileage S0, installation angle θ, ground clearance H) and tunnel cross-sectional dimension parameters (actual cross-sectional width W) are retrieved from the tunnel foundation database. real Cross-sectional height H real ).

[0149] Based on the principle of perspective transformation, a mapping equation is established between image pixel coordinates (u,v) and the actual spatial coordinates (X,Y,Z) of the tunnel to realize the conversion from pixel position to the actual mileage and cross-sectional position of the tunnel. The mapping equation is as follows:

[0150] ;

[0151] Where K is the camera intrinsic parameter matrix, and [R|t] is the camera extrinsic parameter matrix (rotation matrix R + translation vector t). The pixel coordinates can be restored to the tunnel space coordinates through inverse transformation. Combined with the camera installation mileage S0, the actual mileage location of the damage can be determined.

[0152] Based on the ratio between the actual tunnel cross-sectional width and the pixel width of the tunnel cross-section in the image, a pixel-to-physical size conversion factor is calculated to convert pixel-level damage dimensions to actual physical dimensions. The formula for calculating the factor is: k size =W real / W pixel , where ksize W is the pixel-to-physical size conversion factor. pixel The width of the tunnel cross-section in the image can be directly used to calculate the size of cracks, spalling area, etc.

[0153] The acquired raw images are sequentially processed as follows: ROI cropping (preserving the effective area of ​​the tunnel lining and removing background redundancy), distortion correction (eliminating lens distortion based on camera distortion coefficient), image denoising (using Gaussian filtering to remove dust and lighting interference), and pixel value normalization (scaling pixel values ​​to the [0,1] range to adapt to the model input format), and finally converted into image tensor data.

[0154] Two different sizes of convolution kernels, 3×3 and 5×5, are used to perform convolution operations on image tensor data. The shallow 3×3 convolution kernel extracts detailed features such as crack edges and peeling boundaries, while the deep 5×5 convolution kernel extracts global features such as damage texture and morphology, forming a multi-dimensional, multi-scale feature set.

[0155] A channel attention mechanism is introduced, which learns the importance weights of different feature channels to perform weighted fusion of multi-scale features. It automatically enhances the feature weights of damaged areas such as cracks and spalling, and suppresses the interference of background noise such as natural textures of tunnel lining and dust, generating attention features that focus on damaged areas.

[0156] Attention features are input into the classification fully connected layer and the regression fully connected layer, respectively. The classification layer outputs the damage type (circumferential crack, longitudinal crack, lining spalling, no damage) through the Softmax activation function; the regression layer outputs the pixel-level location (e.g., bounding box coordinates) and pixel size (e.g., crack pixel width, spalling pixel area) of the damage through the Sigmoid activation function.

[0157] Substituting the pixel coordinates obtained from the regression into the inverse transformation equation of the first mapping strategy, the cross-sectional location of the damage in the actual space of the tunnel (such as the arch crown and arch waist) is calculated. Combined with the camera installation mileage, the actual mileage location of the damage is determined.

[0158] Multiply the pixel-level damage size by the conversion factor of the second mapping strategy to restore it to the actual physical size, such as the actual crack width and the actual peeling area.

[0159] By integrating damage type, actual mileage location, and actual damage size, and combining tunnel operation and maintenance standards (such as crack width ≥0.3mm being medium risk), the system automatically determines the damage risk level (low risk, medium risk, high risk) and ultimately generates a health status prediction report for the target location.

[0160] The embodiments of this application construct a dual-mapping strategy before image preprocessing. Through perspective transformation and size conversion, the pixel-level results output by the model are accurately restored to the actual mileage location and physical size of the tunnel. This solves the problem of the disconnect between traditional image detection results and actual engineering needs, and can be directly used for operation and maintenance positioning without manual conversion. By extracting deep and shallow features through multi-size convolutional kernels and combining channel attention mechanism to enhance the features of the damaged area and suppress background noise, the accuracy of damage identification is greatly improved, and the misjudgment caused by environmental interference is effectively reduced. The dual-branch output of classification and regression can simultaneously realize the qualitative analysis of damage type, the location of damage, and the quantitative analysis of size. The entire process not only improves the automation and accuracy of tunnel health detection, but also provides an intuitive and practical basis for operation and maintenance decisions, significantly reduces operation and maintenance costs, and ensures the safety of tunnel operation.

[0161] In addition, this application provides an intelligent detection system for tunnel health, as shown in the figure, including: multiple sensor arrays deployed in the tunnel and edge processors corresponding to each group of sensor arrays, and a control platform set at a remote end; wherein, each edge processor is set near its corresponding sensor array and is electrically connected to all sensors in the sensor array, while the control platform set at a remote end is communicatively connected to all edge processors and all sensor arrays.

[0162] The placement of any sensor array in the tunnel is as follows: Figure 3 As shown,

[0163] Any sensor array is used to acquire tunnel monitoring data at the target location within the tunnel and send the tunnel monitoring data to the edge processor; the tunnel monitoring data includes image data of the target location and strain data and temperature and humidity data of multiple nodes at the target location;

[0164] Any edge processor is used to obtain stress balance data and thermo-coupling data corresponding to the target location based on the strain and temperature and humidity data of each node at the target location, as well as the pre-set tunnel foundation data. Based on the stress balance data and thermo-coupling data, if it is determined that there are no abnormal monitoring data in the current tunnel monitoring data, the image data is sent to the control platform.

[0165] The control platform is used to input image data into a pre-trained tunnel health prediction model to obtain the health status prediction results of the target location; the tunnel health prediction model is a convolutional neural network model trained based on the labeled tunnel image.

[0166] Furthermore, each sensor structure is as follows: Figure 4As shown, this figure details the appearance and internal structure of the sensor node. The housing features a waterproof and dustproof design, achieving a high protection rating (e.g., IP68), enabling stable operation in humid and dark environments. The sensor connects to external devices via a standard interface, facilitating installation and maintenance. The internal circuit board integrates a microprocessor, memory, and various types of sensor chips, collectively forming a highly sensitive, low-power monitoring unit. Its housing protection rating: Indicates a waterproof and dustproof rating (e.g., IP68), demonstrating its ability to operate in harsh environments. Interface types: Shows the interfaces for connecting the sensor to external devices, such as power interfaces and data transmission interfaces. Built-in components: Briefly introduces the main components on the internal circuit board, such as the microprocessor, memory, and sensor chips.

[0167] Furthermore, the sensor array is evenly distributed along the tunnel axis, covering multiple locations including the top, sidewalls, and bottom. Each sensor can not only operate independently but also collaborate with other adjacent nodes to form a complete monitoring network.

[0168] In the description of this application, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0169] In this application, unless otherwise expressly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0170] In this application, unless otherwise expressly specified and limited, "above" or "below" the second feature can mean that the first and second features are in direct contact, or that they are in indirect contact through an intermediate medium. Furthermore, "above," "on top of," and "over" the second feature can mean that the first feature is directly above or diagonally above the second feature, or simply that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature can mean that the first feature is directly below or diagonally below the second feature, or simply that the first feature is at a lower horizontal level than the second feature.

[0171] In the description of this specification, the terms "one embodiment," "some embodiments," "embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0172] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make modifications, alterations, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A smart method for detecting tunnel health, characterized in that, include: Acquire tunnel monitoring data for the target location within the tunnel; The tunnel monitoring data includes image data of the target location and strain data and temperature and humidity data of multiple nodes at the target location; Based on the strain and temperature and humidity data of each node at the target location, as well as the pre-set tunnel foundation data, obtain the stress balance data and thermo-coupling data corresponding to the target location, and determine whether there are any anomalies in the current tunnel monitoring data based on the stress balance data and thermo-coupling data. When it is determined that there are no anomalies in the current tunnel monitoring data, the image data is input into a pre-trained tunnel health prediction model to obtain the health status prediction result of the target location; the tunnel health prediction model is a convolutional neural network model trained based on the annotated tunnel image.

2. The intelligent detection method for tunnel health according to claim 1, characterized in that, The tunnel foundation data includes: tunnel elastic modulus, tunnel Poisson's ratio, tunnel foundation temperature, tunnel foundation humidity, and tunnel thermal expansion coefficient; Based on the strain and temperature / humidity data of each node at the target location, as well as the pre-set tunnel foundation data, obtain the stress balance and thermo-coupling data corresponding to the target location, including: Based on the strain data of each node of the target, as well as the pre-set tunnel elastic modulus and tunnel Poisson's ratio, the stress balance data corresponding to the target location is obtained. Based on the temperature, humidity, and strain data of each node at the target location, as well as the pre-set tunnel foundation temperature, tunnel foundation humidity, and tunnel thermal expansion coefficient, the corresponding thermal coupling data of the target location are obtained.

3. The intelligent detection method for tunnel health according to claim 2, characterized in that, Based on the strain data of each node of the target, and the pre-set tunnel elastic modulus and tunnel Poisson's ratio, the stress balance data corresponding to the target location is obtained, including: Based on the strain data of each node, the pre-set tunnel elastic modulus and tunnel Poisson's ratio, and the pre-set Formula 1, the radial stress data of each node is obtained; Formula 1 is: ； in, Here, E represents the radial stress data corresponding to the node at index i, and μ represents the tunnel's elastic modulus and Poisson's ratio. The strain force data corresponding to the node with index i; When the radial stress data of each node is determined to be qualified, the stress distribution residual between the node and the next adjacent node is obtained based on the radial stress data of each node; the stress distribution residual is the difference between the radial stress data of two adjacent nodes. Once the radial stress data for each node is confirmed to be acceptable, stress balance data is obtained based on all stress distribution residuals.

4. The intelligent detection method for tunnel health according to claim 3, characterized in that, Based on all stress distribution residuals, stress balance data are obtained, including: Based on all stress distribution residuals and a pre-set formula (Formula 2), stress balance data are obtained; formula 2 is: ； Among them, R y This is stress equilibrium data, where n is the number of nodes, i is the node index, and R... i+1 The stress distribution residual between the node at index i+1 and its next adjacent node.

5. The intelligent detection method for tunnel health according to claim 3, characterized in that, The temperature and humidity data includes temperature data and humidity data; Based on the temperature, humidity, and strain data of each node at the target location, as well as the pre-set tunnel foundation temperature, tunnel foundation humidity, and tunnel thermal expansion coefficient, the corresponding thermal coupling data for the target location is obtained, including: Based on the temperature and humidity data of each node at the target location, as well as the pre-set tunnel foundation humidity and Formula 3, humidity correction parameters are obtained; Formula 3 is: k i =1+0.002×(RH i -RH0); Where, k i RH is the humidity correction parameter corresponding to the node with index i. i Here is the humidity data corresponding to the node with index i, where RH0 is the basic humidity of the tunnel. Based on the temperature data, humidity data, humidity correction parameters, and radial stress data of each node at the target location, as well as the pre-set tunnel foundation humidity, tunnel foundation temperature, tunnel thermal expansion coefficient, tunnel elastic modulus, and tunnel Poisson's ratio, the mechanical strain data, temperature strain data, and humidity strain data corresponding to each node are obtained. The mechanical strain data, temperature strain data and humidity strain data of each node are weighted and summed to obtain the temperature and humidity strain data corresponding to each node. Based on each strain data, it is confirmed whether the temperature and humidity strain data of each node are abnormal. When there are no anomalies in the temperature, humidity and strain data of each node, the thermal coupling data of the target location is obtained based on the temperature, humidity and strain data of all nodes.

6. The intelligent detection method for tunnel health according to claim 5, characterized in that, Based on the temperature data, humidity data, humidity correction parameters, and radial stress data of each node at the target location, as well as pre-set parameters such as tunnel foundation humidity, tunnel foundation temperature, tunnel thermal expansion coefficient, tunnel elastic modulus, and tunnel Poisson's ratio, the mechanical strain data, temperature strain data, and humidity strain data corresponding to each node are obtained, including: Based on the temperature data and humidity correction parameters of each node at the target location, as well as the pre-set tunnel foundation temperature and tunnel thermal expansion coefficient, the temperature strain data corresponding to each node is obtained; the temperature strain data of any node is the product of the tunnel thermal expansion coefficient and humidity correction parameters, and the product of the difference between the temperature data of that node and the tunnel foundation temperature. Based on the humidity data of each node at the target location and the pre-set humidity of the tunnel foundation, obtain the humidity strain data corresponding to each node; the humidity strain data of any node is the product of the difference between the humidity data of the electrolysis and the humidity of the tunnel foundation, and the pre-set humidity strain coefficient. Based on the radial stress data of each node, the pre-set tunnel elastic modulus and tunnel Poisson's ratio, and the pre-set Formula 4, the mechanical strain data corresponding to each node is obtained; Formula 4 is: ； in, The mechanical strain data corresponding to the node with index i.

7. The intelligent detection method for tunnel health according to claim 5, characterized in that, Based on the temperature, humidity, and strain data of all nodes, the thermo-mechanical coupling data of the target location is obtained, including: Based on the temperature, humidity, and strain data of all nodes, and the pre-set Formula 5, thermo-coupling data is obtained; Formula 5 is: ； Among them, L y This is thermally coupled data, where n is the number of nodes and i is the index of the node. The difference in temperature and humidity strain data between the node at index i and the next adjacent node.

8. The intelligent detection method for tunnel health according to claim 1, characterized in that, Image data is input into a pre-trained tunnel health prediction model to obtain health status prediction results for the target location, including: The image data is preprocessed to convert it into corresponding image tensor data; wherein, the preprocessing includes ROI cropping, distortion correction, image denoising, and pixel value normalization. Feature extraction is performed on the image tensor data using multiple convolution kernels of different sizes to obtain corresponding multi-scale features; Based on a pre-set channel attention mechanism, attention features corresponding to multi-scale features are obtained; Attention features are input into pre-set classification fully connected layers and regression fully connected layers respectively to obtain corresponding damage types and regression results; the damage types include circumferential cracks, longitudinal cracks, lining spalling or no damage, and the regression results include damage location and damage size. Based on the injury type and regression results, the health status prediction results at the target location are obtained.

9. The intelligent detection method for tunnel health according to claim 8, characterized in that, Before preprocessing the image data, a first mapping strategy between the image pixel coordinate system and the actual spatial coordinate system of the tunnel, and a second mapping strategy between the tunnel cross-section size and the image pixel size are established based on the pre-set camera calibration parameters and tunnel cross-section size parameters. Then, based on the damage type and regression results, the health status prediction result at the target location is obtained, including: Based on the regression results, the first mapping strategy, and the second mapping strategy, the actual damage location and actual damage size of the defect are confirmed. Based on the damage type, actual damage location, and actual damage size, the health status prediction results for the target location are obtained.

10. An intelligent detection system for tunnel health, characterized in that, include: Sensor arrays and edge processors deployed in the tunnel, and a control platform set up at a remote location; A sensor array is used to acquire tunnel monitoring data of target locations within the tunnel and send the tunnel monitoring data to the edge processor; The tunnel monitoring data includes image data of the target location and strain data and temperature and humidity data of multiple nodes at the target location; The edge processor is used to obtain stress balance data and thermal coupling data corresponding to the target location based on the strain and temperature and humidity data of each node at the target location, as well as the pre-set tunnel foundation data. Based on the stress balance data and thermal coupling data, if it is determined that there are no abnormal monitoring data in the current tunnel monitoring data, the image data is sent to the control platform. A control platform is used to input image data into a pre-trained tunnel health prediction model to obtain the health status prediction result of the target location; the tunnel health prediction model is a convolutional neural network model trained based on the labeled tunnel image.

Images