Tunnel convergence deformation monitoring method

Abstract

The invention discloses a tunnel convergence deformation monitoring method, and particularly relates to the technical field of tunnel monitoring, through integrating equipment such as a thermal infrared imager, a visible light camera, a distributed optical fiber sensor and a laser radar, a self-adaptive inspection system is constructed, and high-precision appearance and temperature monitoring is realized by adopting an improved ORB algorithm and a temperature field calibration technology. Performing multi-source data fusion through space-time registration and wavelet packet decomposition in combination with optical fiber strain data; and predicting a deformation trend based on an LSTM neural network, and establishing a three-level early warning mechanism. According to the tunnel convergence deformation monitoring method disclosed by the invention, the automation and intelligence level of tunnel structure safety monitoring is remarkably improved, 'perception-analysis-decision 'closed-loop management of the tunnel health state is realized, compared with a traditional method, the efficiency is remarkably improved, meanwhile, the maintenance cost can be greatly reduced, and the tunnel operation safety and economy are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of tunnel monitoring, and particularly to a method for monitoring tunnel convergence deformation. Background Technique

[0002] During the construction and operation stages of a tunnel, accurately grasping the convergence deformation of the tunnel structure is crucial for ensuring tunnel safety. Traditional monitoring methods, such as manual inspections, not only have low efficiency and strong subjectivity but also are difficult to detect hidden dangers deep in the tunnel. Although fixed sensor monitoring can obtain some data, its coverage is limited, and its adaptability is poor in the face of complex and changeable tunnel environments. These deficiencies may lead to the failure to detect and handle tunnel safety hazards in a timely manner, resulting in serious safety accidents.

[0003] During long-term operation, tunnels are prone to convergence deformation due to geological activities, load changes, and material aging. Traditional monitoring methods have the following defects: Limitation of single-point measurement: Contact monitoring such as total stations can only obtain discrete point data and cannot cover the overall deformation trend; Sensitivity to environmental interference: Distributed fiber optic sensors are affected by insufficient temperature compensation, and the strain measurement error can reach ±50 με; Warning lag: Existing systems mostly rely on static threshold alarms and cannot identify early progressive damage.

[0004] Therefore, it is urgent to develop an efficient, accurate, and adaptable method for monitoring tunnel convergence deformation. Summary of the Invention

[0005] [[ID=A19]]The present invention aims to provide a comprehensive, accurate, and intelligent method for monitoring tunnel convergence deformation. By leveraging the flexible mobility of intelligent drones, the high sensitivity of thermal infrared imaging, and the collaboration of multiple cutting-edge technologies, it realizes the efficient monitoring of tunnel convergence deformation, promptly detects potential risks, provides a reliable basis for tunnel maintenance management, and ensures the safe and stable operation of the tunnel.

[0006] To achieve the above objective, the technical solution adopted by the present invention is as follows:

[0007] A method for monitoring tunnel convergence deformation, including the following specific steps:

[0008] S1. Construction of an intelligent monitoring system: Integrate a multi-spectral sensing unit, a deformation sensing unit, a gas sensor, a temperature and humidity sensor, and a combined navigation system on the drone. Based on the tunnel three-dimensional point cloud model and historical deformation data, generate inspection flight routes covering the vault, side walls, and invert through a path planning algorithm, and configure an anti-collision buffer mechanism;

[0009] S2. Multi-source data collaborative acquisition: When the UAV flies along the adaptive flight path, the following acquisition modules are synchronously activated: (a) The thermal infrared imager obtains the thermal radiation distribution on the surface of the structure through the temperature field calibration module; (b) The visible light camera obtains the sequence of apparent images of the structure through the feature point matching algorithm; (c) The distributed optical fiber sensor obtains the structural strain field data through Brillouin scattering analysis;

[0010] S3. Data fusion analysis: Establish a temperature-strain coupling analysis model, perform spatio-temporal registration on the thermal infrared temperature data and the optical fiber strain data, extract the structural abnormal response characteristics by wavelet packet decomposition, and input them into the LSTM neural network trained by historical data for deformation trend prediction;

[0011] S4. Intelligent early warning decision-making: Based on the structural safety evaluation index system, when the temperature gradient exceeds 2°C / m and the strain change rate is greater than 50 με / d, trigger the three-level early warning mechanism, and automatically generate an evaluation report including defect location maps, deformation evolution animations, and repair plan suggestions.

[0012] Preferably, in the step S1, the multi-spectral sensing unit includes a thermal infrared imager and a high-definition visible light camera.

[0013] Preferably, in the step S1, the deformation sensing unit includes a distributed optical fiber sensor and a lidar.

[0014] Preferably, in the step S2, the feature point matching algorithm adopts an improved ORB algorithm based on depthwise separable convolution, introduces an attention mechanism in the feature description stage, reduces the feature point detection time-consuming to 67% of the traditional algorithm, and the temperature field calibration module includes a blackbody radiation reference source. By arranging multiple graphene film reference points in the monitoring area, the spatial calibration of the temperature value of the thermal infrared image is realized, and the temperature measurement accuracy reaches ±0.3°C.

[0015] Preferably, in the step S3, the spatio-temporal registration adopts an improved ICP algorithm. By introducing the curvature feature constraint condition, the spatial registration error between the thermal infrared image and the optical fiber strain data is controlled within 3 mm.

[0016] Preferably, in the step S3, the LSTM neural network includes a bidirectional memory unit and an attention mechanism layer. The network input layer receives a 17-dimensional feature vector composed of wavelet packet energy entropy, strain mean, and temperature variance, and the output layer provides the predicted deformation value for the next 72 hours.

[0017] Preferably, the structural safety evaluation index system includes three-level early warning thresholds:

[0018] Level 1 early warning: Local temperature anomaly > 5°C and strain mutation > 200 με, trigger an audible and visual alarm;

[0019] Secondary warning: If the deformation rate is > 1 mm / d for three consecutive monitoring cycles, start the emergency retest procedure;

[0020] Tertiary warning: If the convergence displacement exceeds 85% of the design value, automatically push the structural reinforcement plan.

[0021] Preferably, in step S4, digital twin technology is introduced to map real-time monitoring data to the BIM model, and a visual safety assessment interface including stress redistribution cloud maps and plastic zone development predictions is generated through finite element simulation calculations.

[0022] Compared with the prior art, the present invention has the following beneficial effects:

[0023] 1. This solution integrates multi-source sensors such as multi-spectral, deformation, temperature, and humidity through drones, and combines intelligent path planning algorithms to achieve full-section automatic inspection of tunnels. The efficiency is significantly higher than that of traditional manual inspection. In addition, in the data fusion stage, spatio-temporal registration technology and wavelet packet energy entropy analysis are used to accurately separate the temperature and strain coupling effects and avoid misjudgment.

[0024] 2. This solution realizes the active prevention and control of tunnel deformation trends through LSTM neural network prediction and a tertiary warning mechanism. Its core advantages are: (1) The prediction model is based on bidirectional LSTM and attention mechanism, inputs a 17-dimensional feature vector, and outputs the deformation amount in the next 72 hours, providing a scientific basis for maintenance decisions; (2) The tertiary warning threshold system combines digital twin technology and can automatically trigger differential responses; (3) The evaluation report integrates defect location maps and repair plan libraries. This solution can shorten the response time of structural safety hazards from 5-7 days in traditional methods to within 4 hours, while reducing maintenance costs and significantly improving the operational safety and economy of tunnels. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is a flowchart of a tunnel convergence deformation monitoring method. DETAILED DESCRIPTION OF THE INVENTION

[0026] In order to make the technical means, creative features, achieved purposes, and effects of the present invention easy to understand, the present invention will be further described below in conjunction with specific embodiments.

[0027] As Figure 1 shown, the present invention discloses a tunnel convergence deformation monitoring method, which includes the following specific steps:

[0028] I. Construction of an intelligent monitoring system:

[0029] Integrate a multi-spectral sensing unit, a deformation sensing unit, a gas sensor, a temperature and humidity sensor, and a combined navigation system on the unmanned aerial vehicle (UAV). Based on the 3D point cloud model of the tunnel and historical deformation data, generate an inspection flight path covering the vault, side walls, and invert through a path planning algorithm, and configure an anti-collision buffer mechanism.

[0030] Specifically, the hardware configuration scheme of the UAV:

[0031] Multi-spectral sensing unit: Adopt a FLIR A655sc thermal imager (resolution 640×512, temperature measurement range -40°C to 150°C) paired with a Sony α7R V visible light camera (61 million pixels), and realize dual-spectrum synchronous acquisition through a carbon fiber gimbal.

[0032] Deformation sensing unit: Integrate a RIEGL VUX-240 lidar (scanning frequency 550kHz, accuracy ±5mm) and an Omnisens DITEST fiber optic demodulator (sampling rate 100Hz, strain resolution 1με).

[0033] Anti-collision mechanism: Install ToF laser ranging sensors (maximum detection distance 30m) around the UAV, and combine an obstacle avoidance algorithm for real-time point cloud processing (based on OctoMap dynamic update) to ensure safe flight in the narrow space of the tunnel.

[0034] Implementation of the path planning algorithm:

[0035] 3D modeling stage: Use ContextCapture software to perform voxelization segmentation (mesh size 5cm) on the tunnel BIM model, and extract the feature lines of key areas such as the vault and side walls.

[0036] Flight path generation: Based on the improved 3D coverage path planning of the A* algorithm, set the flight height 1.5m (side wall) and 2m (vault) from the structure surface, with a heading overlap rate of 80% and a lateral overlap rate of 60%.

[0037] Navigation enhancement: The combined navigation system adopts a NovAtel SPAN-IGM-S1 (GNSS+IMU), and realizes a positioning error <10cm through UWB positioning base stations (arrangement spacing 50m) in the absence of GPS environment.

[0038] II. Technical details of multi-source data collaborative acquisition:

[0039] 1. Steps of the temperature calibration experiment

[0040] Layout of reference points: Paste graphene film calibration points (emissivity 0.95±0.01) at intervals of 10m in the tunnel, and use a blackbody radiation source (model M340, temperature stability ±0.1°C) for on-site calibration.

[0041] Temperature Compensation Algorithm: Establish the radiative transfer equation:

[0042] T real = T meas + k1·e -ad + k2·(RH - 50%)

[0043] where k1 is the air attenuation coefficient and RH is the relative humidity, which are dynamically corrected through the measured data of the sensor.

[0044] 2. Improvement and Optimization of the ORB Algorithm

[0045] Accelerated Feature Extraction: Replace the traditional convolutional layer with depthwise separable convolution (the number of parameters is reduced to 1 / 8), and the feature extraction time on the NVIDIA Jetson AGX Orin platform is < 15 ms / frame.

[0046] Attention Mechanism: Add the CBAM module in the descriptor generation stage, assign higher weights to key areas such as tunnel joints and cracks, and the feature matching accuracy rate reaches 92%.

[0047] 3. Deployment of Fiber Optic Strain Monitoring

[0048] Fiber Optic Layout Process: Paste OFSS fiber along the circumferential direction of the tunnel (spacing 0.5 m), and adopt the prestressed encapsulation technology to ensure that the strain transfer efficiency > 90%.

[0049] Brillouin Frequency Shift Analysis: Use the frequency domain decomposition algorithm (resolution 1 MHz), and the strain-temperature decoupling formula:

[0050] Δs = C e ·Δv B - C T ·ΔT

[0051] where C e = 0.048 MHz / με, C T = 1.2 MHz / °C.

[0052] III. Core Technologies of Data Fusion Analysis:

[0053] 1. Time Registration Technical Solution

[0054] Improved CP Algorithm: Introduce curvature feature constraints, and the registration process includes:

[0055] Extract the extreme points of the isotherm curvature from the thermal infrared image;

[0056] Perform KD-tree nearest neighbor search with the fiber optic strain electric field;

[0057] Calculate the transformation matrix using SVD decomposition, and set the iteration termination threshold to 3 mm.

[0058] Time synchronization: The PTPv2 protocol (accuracy = 1 us) is adopted to realize the synchronization of multi-sensor data through hardware triggering.

[0059] 2. Construction of LSTM prediction model

[0060] Network structure:

[0061] Input layer: 17-dimensional vector (including wavelet packet energy entropy, strain mean, etc.)

[0062] Hidden layer: 2 groups of BiLSTM (128 nodes in each layer) + Attention layer (additive model is used for weight calculation)

[0063] Output layer: 3 Dense nodes (corresponding to the predicted values of 24h / 48h / 72h)

[0064] Training data: The transfer learning strategy is adopted, and the pre-training data set contains 3-year monitoring data (about 2.6 TB) of 10 similar tunnels.

[0065] 3. Wavelet packet feature extraction

[0066] Select the db4 wavelet basis for 5-layer decomposition

[0067] Energy entropy calculation formula:

[0068]

[0069] where W j,k is the coefficient of the kth node in the jth frequency band, and the anomaly criterion: the entropy value increase rate > 15%.

[0070] IV. Intelligent early warning decision-making:

[0071] 1. Based on the structural safety evaluation index system, when the temperature gradient exceeds 2 °C / m and the strain change rate is greater than 50 με / d, trigger the three-level early warning mechanism:

[0072] Level 1 early warning: Local temperature anomaly > 5 °C and strain mutation > 200 με, trigger audible and visual alarms;

[0073] Level 2 early warning: The deformation rate in 3 consecutive monitoring periods > 1 mm / d, start the emergency re-measurement procedure;

[0074] Level 3 early warning: The convergence displacement exceeds 85% of the design value, and automatically push the structural reinforcement plan.

[0075] In step S4, digital twin technology is introduced to map the real-time monitoring data to the BIM model, and a visual safety evaluation interface including stress redistribution cloud maps and plastic zone development predictions is generated through finite element simulation calculations.

[0076] 2. Digital Twin System Integration

[0077] BIM Model Update: Real-time import monitoring data in IFC format to generate a dynamic deformed body model in Revit.

[0078] Finite Element Simulation: Use ANSYS Workbench for coupled analysis (thermal - mechanical - seepage multi-physics field), set the calculation step length to 1 hour, and output: maximum principal stress nephogram (threshold 2 MPa marked in red), plastic zone development animation (based on D-P criterion).

[0079] 3. Automatic Generation of Evaluation Report

[0080] The report template includes:

[0081] Defect Location Map: Superimposed display of thermal infrared and strain (RGB channels correspond to temperature, strain, and displacement respectively);

[0082] Repair Solution Library: Associated with the expert knowledge graph.

[0083] System Maintenance and Optimization

[0084] Every time 10 tunnel monitoring tasks are completed, conduct a comprehensive inspection and calibration of the drone, thermal infrared imager, visible light camera, distributed optical fiber sensor, and data processing system. Check the battery capacity, motor performance, flight control system, etc. of the drone to ensure its normal operation. Calibrate the thermal infrared imager and visible light camera to ensure the accuracy of image acquisition. Conduct performance tests on the distributed optical fiber sensor to ensure the reliability of its monitoring data. Analyze the monitoring effect based on the data obtained from each monitoring task, and optimize the flight path planning algorithm, data analysis algorithm, and warning threshold. For example, if it is found that there are errors in the monitoring data of a certain area, adjust the flight path planning to increase the flight coverage times of that area; if the recognition accuracy of some deformation features by the data analysis algorithm is low, optimize the algorithm through training data to improve its recognition ability. Introduce new data processing technologies and algorithms, such as a tunnel convergence deformation prediction model based on Bayesian optimized long short-term memory network (Bayes-LSTM), to predict the future deformation trend of the tunnel and take preventive measures in advance.

[0085] The present invention will be further disclosed below with specific examples:

[0086] Example 1. Convergence Deformation Monitoring of a Straight Tunnel

[0087] Select the DJI M300 RTK drone, which has a flight time of 2 hours and a positioning accuracy of centimeter level. Equip it with a FLIR T1040 thermal infrared imager with a resolution of 640×512 pixels, a Hasselblad visible light camera with a resolution of 48 million pixels, and a distributed optical fiber sensor.

[0088] The three-dimensional model of the tunnel is obtained by using laser scanning technology. Assume that the tunnel is 1000 meters long, 8 meters wide, and 6 meters high. Based on the model and the characteristics of the tunnel structure, the flight route is planned with the help of Pix4Dmapper software. The safety distance between the UAV and the tunnel wall is set to 1 meter, the flight altitude is 3 meters, the flight route is designed as a round-trip parallel route, and the overlap rate of adjacent flight strips is 70%. Place the UAV at the tunnel entrance, check the equipment status through the ground control station, and send the take-off command after confirming that there is no error.

[0089] The UAV flies according to the preset route. The thermal infrared imager collects the surface temperature data of the tunnel lining at a frequency of 10 Hz, the visible light camera takes an image every 2 seconds, and the distributed fiber optic sensor monitors the strain information of the tunnel structure in real time. During the flight process, the ground control station monitors the flight parameters of the UAV in real time, such as speed, altitude, attitude, etc., to ensure stable flight. The collected data is transmitted to the ground control station in real time through the 5G communication module. At the ground control station, the histogram equalization algorithm is used to enhance the contrast of the thermal infrared image, and the wavelet denoising algorithm is used to remove the image noise. For visible light images, the Canny edge detection algorithm is used to detect cracks, and the threshold segmentation technology is used to identify spalling areas. The strain data collected by the distributed fiber optic sensor is analyzed by Brillouin frequency shift monitoring technology. Combining parameters such as the thermal conductivity coefficient and specific heat capacity of the tunnel lining material, a heat conduction model is established to analyze the temperature change rate and temperature gradient of the thermal infrared image. According to the design standards of the tunnel and historical monitoring data, the displacement threshold is set to 5 mm, the strain threshold is set to 0.05%, and the temperature change rate threshold is set to 0.5 °C / s. When the monitored data exceeds the threshold, the system automatically triggers the early warning mechanism. For example, if the temperature change rate reaches 0.6 °C / s in a certain area, and there are cracks in the visible light image of this area, and the distributed fiber optic sensor monitors abnormal strain, the system immediately sends early warning information to relevant personnel through means such as audible and visual alarms, SMS notifications, and platform pushes.

[0090] Example 2. Convergence deformation monitoring of a curved tunnel

[0091] Select the Phantom 4 RTK UAV, equipped with a centimeter-level positioning system, a thermal infrared imager with a resolution of 384×288 pixels, a visible light camera with a resolution of 20 million pixels, and a distributed fiber optic sensor.

[0092] Obtain parameters such as the radius and length of the curved tunnel through the tunnel design drawings and on-site surveys, and use Litchi software to plan the flight route. Considering the characteristics of the curved tunnel, the safety distance between the UAV and the tunnel wall is set to 1.2 meters, and the flight altitude is 3.5 meters. The flight route is designed as a spiral route along the curve of the tunnel, and the overlap rate of adjacent flight strips is 65%.

[0093] Take off the UAV at a suitable position at the tunnel entrance, and the UAV flies according to the preset route. The thermal infrared imager collects temperature data at a frequency of 8 Hz, the visible light camera takes an image every 3 seconds, and the distributed optical fiber sensor synchronously monitors the strain information. During the flight, the flight control system of the UAV adjusts the flight attitude in real time according to the positioning information and route planning to ensure that the acquisition equipment can accurately obtain the data on the surface of the tunnel lining. The collected data is transmitted to the ground control station through a dedicated wireless transmission module.

[0094] In the data processing stage, the adaptive histogram equalization algorithm is used to enhance the contrast of the thermal infrared image, and the median filtering algorithm is used to remove the image noise. For the visible light image, the Sobel operator is used to detect cracks, and the peeling area is identified by the region growing method. The improved principal component analysis (PCA) method is used to analyze the strain data collected by the distributed optical fiber sensor, and combined with the thermal characteristic parameters of the tunnel lining material, a heat conduction model is established to analyze the relationship between temperature change and tunnel deformation.

[0095] According to the design requirements and safety standards of the tunnel, the displacement threshold is set at 6 mm, the strain threshold is set at 0.06%, and the temperature gradient threshold is set at 0.8 °C / m. When the monitored data exceeds the threshold, the system automatically issues a warning. For example, if the temperature gradient in a certain area reaches 0.9 °C / m, and signs of peeling are found in the visible light image, and the strain monitored by the distributed optical fiber sensor exceeds the threshold, the system quickly notifies the relevant personnel through various warning methods so that timely measures can be taken for treatment.

[0096] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. What is described in the above embodiments and the specification only illustrates the principle of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A tunnel convergence deformation monitoring method, characterized in that, The specific steps are as follows: S1. Construction of intelligent monitoring system: Integrate multi-spectral sensing unit, deformation sensing unit, gas sensor, temperature and humidity sensor, and combined navigation system on the UAV. Based on the tunnel three-dimensional point cloud model and historical deformation data, generate inspection routes covering the vault, side walls, and invert through path planning algorithm, and configure anti-collision buffer mechanism; S2. Cooperative acquisition of multi-source data: When the UAV flies along the adaptive route, synchronously activate the following acquisition modules: (a) The thermal infrared imager obtains the thermal radiation distribution on the surface of the structure through the temperature field calibration module; (b) The visible light camera obtains the apparent image sequence of the structure through the feature point matching algorithm; (c) The distributed optical fiber sensor obtains the structural strain field data through Brillouin scattering analysis; S3. Data fusion analysis: Establish a temperature-strain coupling analysis model, perform spatio-temporal registration on the thermal infrared temperature data and the fiber strain data, adopt wavelet packet decomposition to extract the abnormal response characteristics of the structure, and input them into the LSTM neural network trained by historical data for deformation trend prediction; S4. Intelligent early warning decision-making: Based on the structural safety evaluation index system, when the temperature gradient exceeds 2°C / m and the strain change rate is greater than 50 με / d, trigger the three-level early warning mechanism, and automatically generate an evaluation report including defect location maps, deformation evolution animations, and repair plan suggestions.

2. The tunnel convergence deformation monitoring method according to claim 1, characterized in that: In step S1, the multi-spectral sensing unit includes a thermal infrared imager and a high-definition visible light camera.

3. A tunnel convergence deformation monitoring method according to claim 1, characterized in that: In step S1, the deformation sensing unit includes a distributed optical fiber sensor and a lidar.

4. A tunnel convergence deformation monitoring method according to claim 1, characterized in that: In step S2, the feature point matching algorithm adopts an improved ORB algorithm based on depthwise separable convolution, introduces an attention mechanism in the feature description stage, reduces the feature point detection time-consuming to 67% of the traditional algorithm. The temperature field calibration module includes a blackbody radiation reference source, and realizes the spatial calibration of the temperature value of the thermal infrared image by arranging multiple graphene film reference points in the monitoring area, and the temperature measurement accuracy reaches ±0.3°C.

5. A tunnel convergence deformation monitoring method according to claim 1, characterized in that: In step S3, the spatio-temporal registration adopts an improved ICP algorithm, and controls the spatial registration error between the thermal infrared image and the fiber strain data within 3 mm by introducing curvature feature constraint conditions.

6. A tunnel convergence deformation monitoring method according to claim 1, characterized in that: In step S3, the LSTM neural network includes a bidirectional memory unit and an attention mechanism layer. The network input layer receives a 17-dimensional feature vector composed of wavelet packet energy entropy, strain mean, and temperature variance, and the output layer provides the predicted deformation value for the next 72 hours.

7. A tunnel convergence deformation monitoring method according to claim 1, characterized in that: The structural safety evaluation index system includes three-level early warning thresholds: First-level early warning: Local temperature anomaly > 5°C and strain mutation > 200 με, trigger audible and visual alarms; Second-level early warning: Deformation rate > 1 mm / d for three consecutive monitoring periods, start the emergency re-measurement procedure; Third-level early warning: Convergence displacement exceeds 85% of the design value, automatically push the structural reinforcement plan.

8. A tunnel convergence deformation monitoring method according to claim 1, characterized in that: Introduce digital twin technology in step S4, map the real-time monitoring data to the BIM model, and generate a visual safety evaluation interface including stress redistribution cloud maps and plastic zone development predictions through finite element simulation calculations.

Images