Multifunctional subway tunnel disease intelligent inspection rail car and detection method

Abstract

The invention discloses a multifunctional subway tunnel disease intelligent inspection rail car and a detection method, and belongs to the technical field of rail detection. Aiming at the problems of low manual detection efficiency and single equipment function in the prior art, the invention provides an intelligent inspection vehicle integrating laser scanning, ground penetrating radar and artificial intelligence. The inspection vehicle is provided with a support telescopic system, and camera height self-adaptive adjustment is achieved through a laser distance measuring sensor and an XGBoost algorithm. The disease image acquisition system and the three-dimensional laser scanner cooperate to obtain surface disease data; the internal disease data acquisition system detects internal cracks of the lining through a ground penetrating radar; the disease analysis system deploys an improved YOLOv11 model to realize intelligent identification; and the global control system is integrated with a multi-source data visual interface. The detection method comprises the steps of adaptive adjustment, multi-source data acquisition, preprocessing, intelligent identification and positioning feedback. The system constructs an Internet of Things architecture, significantly improves the detection efficiency and precision, and is suitable for health monitoring of the subway tunnel structure.

Description

Technical Field

[0001] The present invention belongs to the field of track detection technology, and in particular relates to a multifunctional intelligent inspection rail vehicle for subway tunnel defects and a detection method. Background Art

[0002] Existing technologies often rely on manual inspection using handheld detectors to detect subway tunnel defects. This method is not only labor-intensive but also highly subjective, easily leading to significant errors in test results. With the transformation of rail transit toward information technology and intelligent systems, new detection methods are being developed through the application of intelligent innovation and digitalization, building a digital ecosystem for rail transit and effectively addressing the shortcomings of manual inspection. However, domestic tunnel inspection technology is currently a late starter, and railway inspections are still based on traditional manual inspections. Tunnel inspection vehicles are also only used sparingly in highway tunnel inspections, lagging behind international standards.

[0003] Existing tunnel inspection vehicles suffer from numerous deficiencies. First, most can only inspect the surface of the tunnel's inner wall and are unable to effectively detect defects such as the depth of cracks in the lining, missing rebar, and voids. Second, the inspection vehicles' movement speed is generally slow, especially in railway tunnel inspections, which severely impacts inspection efficiency. The uneven brightness of the light source within the tunnel makes it difficult to guarantee the uniform quality of images captured by moving data acquisition equipment. Furthermore, existing tunnel inspection technology lacks equipment that can rapidly detect and effectively integrate inspection feedback information, and the data sources and information chain for each indicator are insufficiently comprehensive. Furthermore, a comprehensive set of inspection and monitoring technologies has yet to be systematically implemented, and an integrated platform for data collection, analysis, prediction, evaluation, and early warning functions has not yet been established. Finally, an imperfect system and the lack of unified technical standards also limit the promotion and application of inspection technology. Summary of the Invention

[0004] In order to solve the above problems, the present invention discloses a multifunctional subway tunnel intelligent inspection vehicle and detection method. The subway tunnel disease intelligent detection vehicle of the present invention is intended to technically realize comprehensive, efficient and accurate tunnel disease detection. Through automated and intelligent detection means, it replaces traditional manual detection, reduces the impact on subway operations, and conducts comprehensive detection of various diseases in-depth inside the tunnel. Through the intelligent disease recognition system and the vehicle intelligent motion analysis system, the disease intelligent recognition system is used to establish a multi-disease intelligent recognition system, and the vehicle intelligent motion system is used to determine the time and location of the disease, thereby realizing real-time disease recognition, positioning, recording and analysis, as well as later fixed-point inspections. The inspection vehicle can integrate multiple advanced technologies, build an intelligent platform, and improve the level of detection technology to solve the problems of low efficiency of manual inspections, bulky intelligent inspection equipment and single functions in the past, and promote the innovation and development of subway tunnel disease detection technology.

[0005] To solve the problems existing in the above-mentioned prior art, the present invention proposes an intelligent inspection vehicle and detection method for subway tunnel defects. The specific scheme is as follows: A multifunctional intelligent inspection vehicle for subway tunnel defects comprises a vehicle body; an incremental encoder, a lighting system, a high-power power supply, a defect analysis system, a global control system, a bracket extension system, a defect image acquisition system, a GPS positioning device, a three-dimensional laser scanner, an internal defect data acquisition system, and a power system; the defect analysis system and the global control system are fixed to reserved positions on the vehicle body, the defect analysis system and the defect image acquisition system are connected by a data transmission line, and the global control system is connected to the defect analysis system; the bracket extension system is fixed to the vehicle body, the top panel of the bracket extension system is bolted to the defect image acquisition system, an incremental encoder is installed on the wheel hub of the vehicle body, lighting equipment is bolted to the left and right sides and the front of the vehicle body, and a ring light strip is installed on the top of the extension bracket; the internal defect data acquisition system and the disease analysis system are connected via a data transmission line; and the three-dimensional laser scanner and the disease analysis system are connected via a data transmission line. The high-power power supply is used for the incremental encoder, lighting equipment, disease analysis system, global control system, bracket extension system, disease image acquisition system, GPS positioning device, 3D laser scanner, internal disease data acquisition system, and power system, and is powered by circuit lines.

[0006] The bracket's telescopic system adjusts its height via a servo motor system. Six holes are designed for a 50cm adjustment range, with holes spaced every 10cm. A laser rangefinder is installed on the bracket's top to measure the tunnel's top height in real time. Based on the measured data and the pre-set optimal standard position, the controller adjusts the bracket's height to ensure it is always in the optimal position for defect collection. Adjusting the telescopic length allows the acquisition device to capture defect images in tunnels of varying diameters, adapting to tunneling methods such as shield or tunnel boring.

[0007] The disease image acquisition system includes 5 high-precision area array industrial cameras, 5 data transmission lines, 5 camera power supply data lines, high-precision area array industrial cameras and a lighting equipment fixing panel; the 5 high-precision area array industrial cameras are connected to the arc-shaped holes of the fixing panel that can be used to adjust the camera angle through bolts.

[0008] The GPS positioning device is fixedly connected to the welding frame of the trolley body; The internal disease data collection system includes: 1) Ground Penetrating Radar System: Storing and Processing Data 2) The immovable part of the intelligent robotic arm, which supports the rotation axis of the intelligent robotic arm and the free extension and retraction part of the intelligent robotic arm 3) Intelligent robotic arm rotation axis: rotates according to the rotation angle issued by the global control system.

[0009] 3) The free extension and retraction part of the robotic arm: It retracts and retracts according to the extension and retraction distance issued by the global control system.

[0010] 4) Intelligent detection sensor: transmits the detection distance data to the global control system through the data line in real time. 5) Geological radar detection head: carries out detection tasks after receiving the operation command from the global control system and sends data to the disease image acquisition system.

[0011] The global control system is human-computer interaction, and the interface design function modules include the following: 1) Real-time video monitoring area: displays real-time images of the tunnel surface 2) Sensor data display area: GPS coordinates, crack width, leakage area, geological radar data and other values 3) Disease distribution map: Dynamically display the location density of diseases 4) Historical data line chart: shows the trend of crack changes 5) Equipment status indicator area: power, communication, hardware status lights 6) Control panel: start / stop button, parameter settings 7) Power system display area: speed, power of each power supply device and estimated remaining time 8) Route planning display area: route planning, geological radar determination position setting, etc.

[0012] The detection method based on the multifunctional subway tunnel disease intelligent inspection vehicle specifically includes the following steps: The inspection vehicle's global control system begins an automated inspection based on a pre-loaded route. During the inspection, the internal defect data collection system, the support extension and retraction system, and the defect analysis system are enabled. The support extension and retraction system uses the measured distance between the support tip and the tunnel lining and a pre-trained AI model for extension and retraction distance using the XGBoost algorithm to perform adaptive intelligent adjustments. Defect collection also begins, with data transmitted to the defect analysis system. The captured image data is captured at 0.05-second intervals, generating an image and assessing image resolution. If the resolution falls below 150 dpi, the generated information is transmitted to the motion control system, which controls the vehicle's speed reduction by 2 km / h increments. When the resolution exceeds 150 dpi, tunnel surface defect data is generated. Intelligent recognition methods are then used to identify surface leaks and cracks in the subway tunnel. The size of the defect and its location in the lining are determined. Based on the route's location and time, GPS positioning information is simultaneously read and fed into the global control system to generate a return route for the inspection. This route includes the location of the internal defect inspection stop and the optimal speed, at which the vehicle will stop on the return trip. The three-dimensional laser scanning equipment and internal disease data collection system are started, and the commands are issued uniformly by the global control system.

[0013] The spatial coordinate system of the tunnel section of the defective section is scanned by 3D laser, and the coordinate information of the defect is transmitted to the intelligent robotic arm. Based on the coordinate information, the robotic arm touches the cross-section segment by segment. After touching, the geological radar is turned on to detect the situation inside the lining and transmit the detection data to the defect analysis system. The data of the surface defects of the subway tunnel are collected by automatic inspection vehicles. The collected data is filtered, denoised, screenshotted with sliding windows, and contrast stretched to form the defect data of the tunnel surface. Based on the YOLOv11 intelligent detection model for image recognition, the apparent water seepage and cracks in the subway tunnel are identified.

[0014] The specific method of the intelligent lifting and angle adjustment is as follows: 1. Start the control system laser ranging module and servo motor system to ensure that all equipment is working properly; 2. Real-time monitoring: The sensor on the laser ranging module monitors and feeds back the height data of the acquisition module from the tunnel top in real time to ensure the best acquisition effect; 3. Data transmission: The laser ranging module sensor transmits the data of the distance to the tunnel height to the control system; 4. Data processing: The control system processes data through algorithms and determines whether the current height is suitable for collecting disease image data based on the diameter of the shield tunnel or mining method tunnel. 5. Generate control signal: If the height is not appropriate, the control system generates a lifting control signal; 6. Execution adjustment: The control signal drives the servo motor system to automatically adjust the height of the bracket; 7. Loop monitoring: Repeat the above steps and adjust the height of the bracket in real time to ensure that the camera is always in the best shooting position and achieve the best acquisition range and effect; 8. Angle adjustment: After determining the optimal height for acquisition, the overlap range of camera acquisition is set to 5% based on the camera angle overlap data fed back by the camera acquisition software. When the overlap range is greater than 5%, the camera acquisition software issues an automatic adjustment command, and the control system sends the automatic adjustment command to the servo motor and starts adjusting the angle. After repeated adjustments, the angle with the best acquisition effect and field of view is determined.

[0015] This invention, through a modular design approach, separates the analytical and computational equipment from traditional inspection carts. Through an integrated design approach, it simplifies the vehicle structure, enhancing the system's flexibility and maintainability. Furthermore, through integrated design, it streamlines the onboard detection equipment and optimizes the overall layout. Based on technologies such as Internet of Things sensing, three-dimensional laser scanning imaging, ground-penetrating radar detection, artificial intelligence, and wireless ad hoc networks, this invention integrates laser detection, electromagnetic wave detection, power storage, and wireless high-throughput data transmission. It also designs data transmission paths and remote control signal configuration paths to achieve the integrated application of multiple functional devices.

[0016] The inspection vehicle of the present invention combines laser scanning, ground-penetrating radar, and artificial intelligence technologies, enabling both data collection and intelligent identification of subway tunnel defects. Surface water leakage can be further assessed using infrared imaging to determine the presence of significant water-bearing structures, while ground-penetrating radar image data can be used to identify potentially hazardous cracks behind the lining. Adjustable telescopic supports adapt the data collection equipment to varying tunnel diameters. Furthermore, the vehicle can be unmanned and capable of self-navigating inspections. Compared to existing single-function intelligent inspection vehicles, the present invention provides a more comprehensive approach to tunnel defect detection, boasts a higher degree of automation, and offers superior identification results.

[0017] This invention innovatively combines the XGBoost algorithm with a mechanical locking device. Through the AI ​​model, it dynamically predicts the optimal hole position, and cooperates with high-precision screw transmission and double locking mechanism to achieve rapid and stable adjustment of the bracket height, solving the technical bottleneck of traditional telescopic systems that rely on manual calibration and are susceptible to vibration displacement.

[0018] This paper proposes a modular track inspection system consisting of an electromechanical component (EMC) and an intelligent control system. The EMC component is responsible for the operation of the trolley on the track and the efficient operation of various data acquisition devices, while the intelligent system adjusts the EMC operating parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is a schematic diagram of the overall architecture and components of the multifunctional intelligent subway tunnel defect inspection vehicle of the present invention.

[0020] Figure 2 This is a system architecture topology diagram of the subway tunnel defect detection method described in the present invention.

[0021] Figure 3 This is a structural diagram of the internal disease detection system of the present invention.

[0022] Figure 4 It is a structural schematic diagram of the support telescopic system of the present invention.

[0023] The numbers in the figure are as follows: 1. Carriage body; 2. Incremental encoder; 3. Lighting system; 4. High-power power supply; 5. Fault analysis system; 6. Global control system; 7. Bracket extension system; 8. Fault image acquisition system; 9. GPS positioning device; 10. 3D laser scanner; 11. Internal fault data acquisition system; 12. Power system; 13. Wireless data transmission system; 7-1. Telescopic hole; 7-2. Laser ranging sensor; 7-3. Controller; 7-4. Telescopic rod end; 8-1. Area scan industrial camera 1; 8-2. Area scan industrial camera 2; 8-3. Area scan industrial camera 3; 8-4. Area scan industrial camera 4; 8-5. Area scan industrial camera 5; 8-6. Fixed panel; 3-4. Ring light strip; 11-1. Ground-penetrating radar system; 11-2. Intelligent detection sensor; 11-3. Geological radar probe; 11-4. Retractable portion; 11-5. Immovable portion; 11-6. Data cable; 11-7. Rotation axis. DETAILED DESCRIPTION

[0024] The present invention relates to a multifunctional intelligent inspection vehicle for subway tunnel defects, the structural composition and functional characteristics of which are as follows: The inspection vehicle comprises a vehicle body 1 and an integrated intelligent detection system. The front end of the vehicle body 1 is equipped with an adjustable telescopic bracket 7, driven by a servo motor. This bracket has a continuously adjustable height range of 125cm-175cm and a 50cm travel range controlled by six telescopic holes spaced 10cm apart. The bracket's top is integrated with a collection system 8 consisting of a laser rangefinder and five high-resolution area array industrial cameras. This system automatically adjusts the camera to the optimal shooting position by measuring the tunnel top height in real time and providing feedback to the global control system 6. This allows for the acquisition of 270° circumferential disease images, adapting to the cross-sectional shape of tunnels constructed using different construction methods (shield tunneling / excavation), and ensuring full circumferential disease image capture.

[0025] The vehicle's travel system features two wheel-mounted incremental encoders for precise distance and speed measurement, combined with the four-wheel drive structure to ensure stable movement. The vehicle's lighting system utilizes a dual-mode fill-light design: adjustable LED light sources are installed on both sides of the vehicle, while a high-brightness light strip is arranged in a circular pattern behind the acquisition module. This dual-mode design ensures high-definition imaging in dark tunnel environments.

[0026] The inspection equipment utilizes a multi-sensor fusion solution: a top robotic arm carries a 3D laser scanner 10 and an internal defect data acquisition system 11. The former uses point cloud modeling to identify water-bearing structures behind the lining, while the latter detects internal cracks. Integrated temperature and humidity sensors monitor environmental parameters in real time, providing multi-dimensional data support for defect analysis. All collected data is transmitted in real time via dedicated cables to the onboard defect analysis system 5, which is equipped with an industrial computer for online processing, enabling simultaneous diagnosis of both external and internal tunnel defects.

[0027] The equipment installation utilizes a modular design with standardized removable sensor interfaces. The 3D laser scanner, ground-penetrating radar, and camera utilize standardized quick-swap interfaces with the host computer. Combined with the modular bracket design, equipment replacement can be completed in less than 10 minutes. The laser ranging sensor supports rapid switching between TOF and triangulation ranging modes, adapting to the differentiated requirements of shield tunneling and bored tunneling. The power supply and control system are isolated in a layered layout, with the mobile power supply and control system arranged in upper and lower layers with an electromagnetic shielding partition in between. The power supply layer integrates a cooling fan and an independent temperature control module to prevent interference with the control circuit caused by heating of the lithium battery pack. Hot-swap replacement is also supported to ensure battery life.

[0028] Reconfigurable design of the car body The car body 1 is made of aluminum alloy profiles spliced ​​by mortise and tenon structure. Maintenance-oriented component isolation technology includes a magnetic quick-release plate for the universal adjustment mechanism of industrial cameras 8-1 through 5, and redundant positioning pins at the robotic arm joints of the internal disease data acquisition system 11. This eliminates the need to disassemble related modules in the event of a single component failure. The IP65 protective housing utilizes a snap-on seal structure, enabling core components to be disassembled and cleaned within 30 seconds, improving bolt-on efficiency.

[0029] This design innovatively combines intelligent height adjustment, multi-source data fusion detection and an adaptive fill-light system. Through the coordinated optimization of mechanical-electrical control-algorithm, it significantly improves the efficiency and accuracy of tunnel disease detection, providing an intelligent solution for subway tunnel structural health monitoring. Example

[0030] Step 1: Solution Design Based on the size of subway tunnel defects during the operation period, a preliminary design of the trolley's main frame, color, size, motor type, battery capacity, tire model, steering form, overall mass, load capacity, maximum speed, endurance time, and material type is carried out.

[0031] According to on-site investigations, the subway track spacing is 1435mm, and the distance between platforms is generally around 5km. To ensure a 40km range, the vehicle was chosen. To ensure safety during inspections, the vehicle body is affixed with reflective tape and equipped with railings. The inspection vehicle primarily consists of the vehicle body, a power unit mounted on the underside, and inspection equipment including 3D laser scanning and ground-penetrating radar. This equipment is connected to a remote analysis and processing system, which includes an intelligent defect detection system and a vehicle positioning system. The intelligent defect recognition system establishes intelligent recognition models for cracks and water leaks, and identifies them. The vehicle's motion system uses incremental encoder readings to determine the real-time location of defects, enabling the positioning, recording, and analysis of subway tunnel defects.

[0032] Step 2: Main structure The collection vehicle is constructed from patterned plate and steel, with a wheelbase of 1435mm based on the track's wheelbase. To facilitate intelligent fault identification and repair at the site of a defect, the vehicle is 2m long and 1.7m wide, conveniently equipped with the collection and detection systems. It features a seven-piece steel frame constructed from 4x8 square tubes with a thickness of 275mm and a load capacity of 1 ton, taking into account safety, durability, and environmental considerations. The tires are four T-shaped nylon wheels. To ensure long-range data collection, a brushless DC motor drives the front wheels. The vehicle's battery uses a Chaowei 62V48A rechargeable battery manufactured in Texas, capable of a range of 40km on a full charge. The vehicle can be operated either remotely or manually, making it suitable for tunnel environments. Reflective tape is affixed to the vehicle's steel plates.

[0033] Step 3: Acquisition Equipment The collection vehicle, area array industrial camera, incremental encoder, LED lighting, temperature and humidity sensor, expansion dock, etc. are integrated together, and all components are detachable. The internal layout of the collection system is standardized, and the data cables and power cables are neatly routed. The different modules within the entire inspection vehicle and image acquisition system are designed and selected separately to meet the needs of actual subway tunnel inspections. The machine vision system first needs to calculate the camera's visual acquisition range (FOV field of view), that is, the range of the tunnel lining section that needs to be collected. The diameter of the shield tunnel is D = 5400mm, and the acquisition range L of the tunnel section is: The acquisition system needs to capture a 260° range of the entire tunnel, which is calculated to be 12252.2mm. To avoid missing images and ensure subsequent stitching of full-section images, a 20% overlap is set within the areas captured by adjacent cameras. Five area array cameras are considered for the acquisition task, and subsequent calculations will determine whether this meets the requirements. Therefore, the total range to be captured is x: According to calculations, the FOV of each area array industrial camera is x = 3603.6 mm, and the acquisition accuracy requirement is R = 1 mm / pixel. The resolution of each camera is: An area scan camera with a resolution of approximately 4000 can meet the above acquisition requirements. The industrial area scan camera developed by Medvision was selected, with a resolution of 4112×3088. The 4112 resolution meets the acquisition accuracy requirements for the circumferential direction of the tunnel lining. To ensure that no frame drops occur in the lining image during acquisition, the sensor's maximum frame rate is 30 fps. The maximum speed for rapid inspections, vscan, is: Calculations show that the maximum speed for rapid inspections is 92 m / s. This plan utilizes five industrial cameras, which can simultaneously record video and automatically take photos at set times. Five area scan cameras will capture defects on the tunnel's vault, haunches, and sidewalls. These five cameras record the video, facilitating subsequent defect location and identification. To balance inspection efficiency and defect detection accuracy, the inspection platform uses the MindVision-MV-SUC1206GC area scan camera.

[0034] The focal length of the lens is selected by determining the object distance. Considering that the camera integration module is fixed on a rectangular disk with a radius r = 500mm at the center of the tunnel, the object distance WD of the camera is WD = D / 2-r = 2200mm, and the pixel size of the IMX226 is m = 1.85μm, the focal length of the camera is: Therefore, a lens with a focal length of 12mm is selected. The horizontal distance of the 12mm lens at a distance of 2.8m from the tunnel is 3.36m. The industrial camera can be set to take pictures at intervals. Calculating that 3 pictures are taken per second, the maximum speed of the inspection vehicle that can stitch together a complete picture of the tunnel is 36.29km / h: During shield tunnel image acquisition, blind spots exist due to the lighting angles within the tunnel and obstructions from some ancillary facilities. In these blind spots, the CMOS sensor cannot receive sufficient light, resulting in a loss of image detail and insufficient contrast between defects and the lining. This hinders subsequent image preprocessing and defect identification. Therefore, supplemental lighting is necessary to ensure stable image acquisition and efficient defect identification.

[0035] During the actual data collection process, since there is no GNSS signal in the tunnel, an encoder is required to locate the testing platform. Encoders are classified into two types: absolute and incremental. Incremental encoders offer a simpler structure, greater immunity to interference, and a service life exceeding 50,000 hours. Since absolute position coordinates are not required during the actual data collection process, an incremental encoder was selected and installed on the testing platform's wheel hub. The encoder generates pulses as the wheel hub rotates. These pulse signals are fed into a pulse counter, which calculates displacement based on the number of signals counted by the counter and the encoder parameters.

[0036] Detection method This invention provides a multifunctional, integrated intelligent detection method for subway tunnel defects. It primarily includes the following steps: During an automatic or manual inspection, a telescopic robotic arm scans the tunnel lining with a 3D laser scanner and a ground-penetrating radar, uploading the scanned data to a remote analysis system. A positioning system also uploads motion information to the remote analysis and processing system. Before the inspection, the five area array industrial cameras on the inspection cart are adjusted to capture data images of the entire tunnel cross-section. The acquisition system's laser ranging module and servo motors enable intelligent lifting and angle adjustment of the industrial cameras, ensuring that the images align with the tunnel contours.

[0037] The operating steps for intelligent lifting and angle adjustment are: 1. Start the control system laser ranging module and servo motor system to ensure that all equipment is working properly.

[0038] 2. Real-time monitoring: The sensor on the laser ranging module monitors and feeds back the height data of the acquisition module from the tunnel top in real time, ensuring optimal acquisition results.

[0039] 3. Data transmission: The laser ranging module sensor transmits the data of the distance to the tunnel height to the control system.

[0040] 4. Data processing: The control system processes data through algorithms and determines whether the current height is suitable for collecting disease image data based on the diameter of the shield or mining tunnel.

[0041] 5. Generate control signal: If the height is not appropriate, the control system generates a lifting control signal.

[0042] 6. Execute adjustment: The control signal drives the servo motor system to automatically adjust the height of the bracket.

[0043] 7. Loop monitoring: Repeat the above steps and adjust the bracket height in real time to ensure that the camera is always in the best shooting position to achieve the best acquisition range and effect.

[0044] 8. Angle Adjustment: After determining the optimal acquisition height, use the camera angle overlap data fed back by the camera acquisition software to set the camera acquisition overlap range to 5%. When the overlap range exceeds 5%, the camera acquisition software issues an automatic adjustment command, and the control system sends the automatic adjustment command to the servo motor and begins to adjust the angle. After repeated adjustments, determine the angle that optimizes the acquisition effect and field of view.

[0045] The adjustment mechanism of the support telescopic system specifically includes: Hole Design and Locking: The telescopic bracket is equipped with six telescopic holes spaced 10 cm apart, ranging from 0 cm to 50 cm. Each hole is integrated with a mechanical locking device, which uses a combination of spring clips and tapered locking pins to ensure that the telescopic rod is automatically locked when in place. Intelligent adjustment process: A laser ranging sensor collects real-time distance data between the top of the support and the tunnel lining. An AI model trained with the XGBoost algorithm combines the tunnel diameter shield method / dig method with the optimal shooting posture requirements to output the target hole position. A controller drives the servo motor, which moves the telescopic rod to the target hole position through a screw-nut transmission system. Once in position, the self-locking feature of the screw and the mechanical locking device work together to prevent the load from sliding. Secondary adjustment mechanism: When the height needs to be adjusted, the controller sends a signal to release the locking device, the motor drives the screw to rotate in the opposite direction, the telescopic rod is disengaged from the current hole position, and then repositioned to the new target hole position and locked; Redundant design: The transmission system adopts a double-nut anti-backlash structure to eliminate the impact of backlash on positioning accuracy; the locking pin adopts a redundant contact surface design to ensure stability under extreme working conditions.

[0046] The process of data processing by the remote analysis and processing system is as follows: Collect data on surface defects in subway tunnels using automated or manually driven inspection vehicles; The collected data is subjected to a series of processing such as filtering and denoising, sliding window screenshot, and contrast stretching to form tunnel surface disease data. As the database of the artificial intelligence model, the disease image data is manually annotated and converted using the open source annotation tool Labelme to generate the YOLO dataset format, which is divided into training set, test set, and validation set in a ratio of 7:2:1.

[0047] Based on the divided training, test, and validation sets, a YOLOv11 intelligent detection model for image recognition was established. This model was used to identify apparent water leaks and cracks in subway tunnels. The YOLOv11 model was improved by adding the SimAM lightweight attention mechanism to the residual block of the C2PSA module in the YOLOv11 network, unlocking new potential in convolutional neural networks. The detection model also adopted the WioU loss function.

[0048] The real-time position of the inspection vehicle is calculated based on the counting information of the incremental encoder.

[0049] Target detection identifies cracks or water leaks, focusing on these locations and recording their locations. First, 3D laser scanning is used to extract characteristic points and geometric information from the tunnel surface. Using a high-resolution industrial camera, the tunnel surface is imaged at high resolution. The high-precision point cloud and image data acquired through laser scanning, combined with an improved YOLOv11 defect detection algorithm, automatically identifies cracks and water leaks on the tunnel lining surface, measures parameters such as the length and width of the defects, and monitors the distribution of water leaks and crack development trends, classifying their hazard levels. This system quickly and accurately identifies the location, distribution, and area of ​​defects, enabling quantitative detection and visualization of cracks and water leaks. Laser scanning technology is then used to rapidly acquire 3D point cloud data of the tunnel and construct a high-precision 3D model of the tunnel, including geometric features such as the tunnel outline, cracks, and water leaks. This truly reflects the tunnel's geometry and spatial structure.

[0050] After laser scanning to obtain 3D information about tunnel surface defects, combined with ground-penetrating radar (GPR) image data, a comprehensive assessment is made of the presence of potentially hazardous cracks behind the lining. This system relies on the reflection and scattering properties of electromagnetic waves. A transmitting antenna emits high-frequency, broadband, short-pulse electromagnetic waves toward a target within the lining. These waves reflect off the target and are then received by a receiving antenna. As the electromagnetic waves propagate through the lining, they reflect and scatter when encountering interfaces with different dielectric constants, such as cracks, voids, or aquifers. Cracks cause changes in the reflected electromagnetic wave signal. The GPR system's receiving antenna receives the reflected electromagnetic wave signal, processes it, and generates a radar image. Analysis of this radar image allows the specific location, depth, and size of the cracks behind the lining to be determined. The GPR system can be used to inspect the internal damage of the segment structure surrounding the cracks, detecting damage within the segment structure at the crack site. GPR has a high sensitivity to changes in the concrete dielectric, making it suitable for early detection of water seepage in structures. Images can be used to analyze the cause and extent of water seepage. Both data types are converted to the same coordinate system, using the tunnel's geographic or engineering coordinate system as a reference for spatial registration and fusion. The filtered and denoised laser scanning data is used to generate a 3D model of the tunnel surface, including geometric features such as the tunnel outline and cracks. The ground-penetrating radar (GPR) data is then filtered and gain-adjusted to extract electromagnetic reflection characteristics and generate a radar image. The geometric features of the defect point cloud data acquired by the laser scanning are then matched with the electromagnetic reflection characteristics of the GPR data to establish a comprehensive analysis model. This model accurately identifies both surface defects and latent defects hidden behind the tunnel lining, which are invisible to the human eye. This comprehensive model covers the tunnel surface and interior, precisely locating the occurrence of various defects. Based on defect type and severity, a defect distribution map is generated, providing an intuitive basis for tunnel safety assessment and maintenance decisions. Data acquired by laser scanning and GPR are stored and managed digitally, facilitating the establishment of a tunnel defect database and information management system. This allows for long-term data preservation, querying, and updating, providing data support for the full lifecycle management of tunnels.

[0051] The inspection vehicle and remote analysis and processing system of the present invention constitute an Internet of Things (IoT) for subway tunnel inspection. The system mainly consists of the inspection vehicle perception layer, the wireless data transmission system network layer, and the remote analysis and processing system application layer.

[0052] Perception layer: The inspection car is equipped with a variety of sensors to collect data: 3D laser scanner: collects infrared field intensity data and extracts the geometric features of lining surface defects.

[0053] Ground Penetrating Radar: Collects radar data to detect the size and location of cracks behind the tunnel lining.

[0054] Incremental encoder: reads the number of wheel hub laps of the inspection trolley in real time to achieve precise positioning and data recording.

[0055] Network layer: Through the wireless data transmission system, various data collected by the perception layer are transmitted to the remote analysis and processing system in real time.

[0056] Application layer: After receiving the data, the remote analysis and processing system performs the following analysis: Car motion analysis system: Calculates the precise position of the inspection car based on positioning data.

[0057] Intelligent defect identification system: Identifies tunnel defects as the vehicle travels; combines infrared field strength and radar data to further distinguish water leakage and crack defects to determine whether the defects exist.

[0058] Wireless data transmission system: The temperature and humidity sensors transmit the collected data to the computer via Bluetooth. As part of the Internet of Things, the computer uploads the results of the disease image acquisition system, internal disease data acquisition system, and disease analysis system to the Internet of Things cloud for storage.

Claims

1. A multifunctional subway tunnel disease intelligent inspection vehicle, characterized by: include: The vehicle body (1) is equipped with a wheel hub, a power system (12) and a four-wheel drive structure, and the power system (12) provides energy for the inspection vehicle; The bracket telescopic system (7) is driven by a servo motor, and a laser distance sensor (7-2) and a plurality of high-precision area array industrial cameras are arranged on the top of the bracket for adaptively and intelligently adjusting the height and angle of the camera; Disease image acquisition system (8), including a multi-table array industrial camera on the top of the bracket and an adjustable angle fixed panel (8-6), The three-dimensional laser scanner (10) is connected to the fixed panel (8-6) and the end of the telescopic rod through bolts and is used to scan the three-dimensional structure of the tunnel lining surface; The internal disease data acquisition system (11) includes a ground penetrating radar system (11-1), an intelligent robotic arm, an intelligent detection sensor (11-2), and a geological radar detection head (11-3), and is used to collect characteristics of crack diseases inside the tunnel lining; An incremental encoder (2), mounted on the wheel hub, for real-time measurement of displacement and speed; The disease analysis system (5) is connected to the image acquisition system (8), the three-dimensional laser scanner (10), and the internal disease data acquisition system (11) in real time, and deploys an improved YOLOv11 intelligent detection model, which integrates the SimAM attention mechanism in the C2PSA residual block and adopts the WIOU loss function; The global control system (6) integrates a human-computer interaction interface, including a window for the disease analysis system (5) and a window for real-time monitoring of the monitoring data of all other sensors, while also viewing disease distribution, equipment status, and path planning; Lighting system (3), including LED light sources with adjustable angles on both sides of the vehicle body, a light sensor controller and a ring light strip at the rear of the acquisition module, supporting dual-mode automatic fill light; The wireless data transmission system (13) includes sensors that transmit data to a computer via Bluetooth. The computer transmits the processing results of the collected information to the signal base station in the tunnel via 5G technology, thereby realizing real-time communication between the perception layer data and the remote analysis and processing system.

2. The inspection vehicle according to claim 1, characterized in that: The bracket telescopic system (7) achieves height adjustment through the telescopic hole; based on the real-time distance between the bracket top and the tunnel lining fed back by the laser ranging sensor, combined with the telescopic distance AI model pre-trained by the XGBoost algorithm, the target hole position is calculated; the controller generates instructions according to the target hole position, drives the servo motor to drive the screw or nut transmission component to rotate, converts the rotational motion into linear motion, and enables the telescopic rod to move accurately to the target hole position; each hole position is equipped with a mechanical locking device, including a buckle and a locking pin, and the telescopic rod is doubly fixed by the self-locking feature and the locking device after it is in place; during the secondary adjustment, the controller sends a signal to release the lock, and the motor drives the transmission component to adjust to the new hole position and then re-lock it.

3. The inspection vehicle according to claim 1, characterized in that: Improvements to the YOLOv11 model in the disease analysis system (5) include: The SimAM lightweight attention mechanism is embedded in the residual block of the C2PSA module to enhance the local feature extraction capability of cracks and water seepage; The WIOU loss function is used to optimize the bounding box regression accuracy and reduce the false detection rate caused by complex background interference.

4. The inspection vehicle according to claim 1, characterized in that: The three-dimensional laser scanner (10) is based on laser interference imaging technology. By analyzing the attenuation and offset signals of laser beams, it scans the geometric morphology and spatial structure of lining surface diseases and judges the development trend. The ground penetrating radar (11) detects the depth and position of internal cracks in the lining through the reflection signals of high-frequency broadband short-pulse electromagnetic waves.

5. The inspection vehicle according to claim 1, characterized in that: In the internal disease data acquisition system (11), the ground penetrating radar system (11-1) is connected to the immovable part (11-5) of the intelligent robotic arm through bolts. The immovable part (11-5) of the robotic arm is connected to the freely telescopic part (11-4) of the intelligent robotic arm through the rotating shaft (11-7) of intelligent rotation. The intelligent detection sensor is connected to the head of the freely telescopic part (11-4) of the intelligent robotic arm through bolts.

6. The inspection vehicle according to claim 1, characterized in that: The annular light strip (3-4) of the lighting system (3) synchronously adjusts the supplementary light angle with the area array industrial camera to ensure uniform illumination for image acquisition in the dark tunnel environment, and adaptively adjusts the brightness dynamically to adapt to different environmental temperatures and humidities.

7. The inspection vehicle according to claim 1, characterized in that: The human-machine interaction interface of the global control system (6) includes: A real-time video monitoring area and a sensor data display area; a dynamic heat map of disease distribution throughout the tunnel, a line graph of historical data, an equipment status indicator and a path planning display area; a disease feature fusion area where the AI-labeled crack map, three-dimensional surface features of cracks, and internal features of cracks are separately separated, and a fused three-dimensional crack feature display area; the sensor data display area includes GPS coordinates, power of power equipment, vehicle speed, brightness and angle of lighting equipment. The control panel supports start / stop buttons, parameter settings, and real-time switching of model weights.

8. An intelligent tunnel disease detection method based on the inspection vehicle according to any one of claims 1 to 6, characterized in that: It includes the following steps: Step 1: Adaptive intelligent adjustment. Start the laser ranging module and the servo motor system, and monitor and adjust the bracket height and camera angle in real time to ensure that the image acquisition overlap rate is 5%. Step 2: Multi-source data acquisition. Collect tunnel surface images through the area array industrial camera, obtain lining water-bearing structure data by the three-dimensional laser scanner (10), and detect internal crack diseases by the ground penetrating radar. Step 3: Data preprocessing. Perform filtering and denoising, sliding window screenshot, and block contrast stretching on the image data, construct a YOLO format data set, and divide it into a training set, a validation set, and a test set according to 7:2:

1. Step 4: Intelligent recognition. Based on the improved YOLOv11 model, detect water leakage and cracks, combine laser scanning and radar data, construct three-dimensional features of crack diseases, and comprehensively determine the disease risk level. Step 5: Positioning and feedback. Generate displacement coordinates through the incremental encoder (2), associate with the BIM / GIS map to mark the disease position, and plan a 3D maintenance path.

9. The method according to claim 7, characterized in that The risk level determination in Step 4 includes: The water leakage level is divided into slight when A < 0.5 m², general when 0.5 m² ≤ A ≤ 2 m², severe when 2 m² < A ≤ 5 m², and critical when A > 5 m² based on the water-bearing structure area scanned by the three-dimensional laser. The crack level is divided into grade I shallow micro cracks, grade II middle-layer cracks, and grade III penetrating cracks based on the depth and length detected by the ground penetrating radar.

Images