Autonomous charging and cooperative operation and maintenance system and method for tunnel wall-climbing robot

Abstract

The invention discloses an autonomous charging and cooperative operation and maintenance system and method for a wall-climbing robot of a closed structure such as a tunnel. The system is composed of a tunnel wall-climbing robot, an intelligent charging pile, an edge computing gateway and a cloud operation and maintenance platform. The core of the method is that a cloud platform operates a task and charging cooperative scheduling algorithm based on multi-target rolling optimization, and an action instruction of a robot is dynamically generated; the edge gateway adopts a lightweight screening-precise quantification two-stage AI processing pipeline to perform real-time grading and compression uploading on the detection data; the robot realizes high-fault-tolerance wall surface magnetic attraction butt joint based on fusion positioning and visual servo control; by defining a standardized data interaction protocol, a closed loop of state perception-intelligent decision-accurate execution is realized. According to the invention, full-process automation and intelligentization of tunnel detection from sensing, decision making, charging to operation and maintenance response are realized, and long-term unattended operation in a real sense is achieved.

Description

Technical Field

[0001] This invention relates to the field of road maintenance equipment technology, and in particular to an autonomous charging and collaborative operation and maintenance system and method for a wall-climbing robot used in enclosed structures such as tunnels. Background Technology

[0002] As a crucial transportation infrastructure, the structural health of tunnels directly impacts operational safety. Traditional manual inspection methods suffer from drawbacks such as low efficiency, high risk, strong subjectivity, and difficulty in digitizing and managing data. In recent years, wall-climbing robots have been gradually applied to tunnel inspection due to their ability to replace manual labor on vertical walls. However, existing tunnel inspection robot systems still have significant shortcomings: most robots have short battery life, unable to support complete inspection of long-distance tunnels, requiring manual intervention for charging or battery replacement, making it difficult to achieve truly long-term unmanned operation. Moreover, robots typically only handle data collection, with data analysis heavily reliant on manual interpretation at the back end, resulting in long response cycles from problem detection to maintenance initiation, failing to meet the need for rapid handling of urgent defects such as cracks and leaks. Furthermore, the lack of effective linkage between robots, charging facilities, and operation and maintenance management platforms creates isolated "information islands," failing to form an automated closed loop from perception and decision-making to execution.

[0003] Therefore, there is an urgent need to design a complete system that integrates autonomous replenishment, intelligent analysis, and collaborative operation and maintenance in order to break through the bottlenecks of existing technologies. Summary of the Invention

[0004] This invention provides an autonomous charging and collaborative operation and maintenance system for a tunnel climbing robot, comprising: The tunnel climbing robot is used to perform inspection tasks on the tunnel wall. It includes a robot mobile platform, a central processing unit and wireless communication module, as well as a sensing module integrating multiple sensors and a magnetic charging interface. The sensing module includes a lidar and an IMU, a digital beam radar, a high-definition camera and a wireless communication module.

[0005] The autonomous charging pile is fixedly deployed on the inner wall of the tunnel. It includes a pile body, a pile body controller, a pile end charging connector that cooperates with the magnetic charging interface, a charging power module, and a pile end communication module. An edge computing gateway, deployed at the tunnel site, communicates with the robot's wireless communication module and the pile-end communication module to perform local data fusion and real-time processing. The cloud-based operation and maintenance platform is connected to the edge computing gateway via a wide area network for macro-task management, data storage and analysis, and maintenance task dispatch. On the other hand, an operation and maintenance method based on the above system is provided. The core of this method is to seamlessly integrate the robot’s autonomous survival capability (charging) with the intelligent operation and maintenance process (detection-analysis-alarm-dispatch). Through the “end-edge-cloud” collaborative architecture, the computing load is reasonably allocated to achieve efficient and real-time transaction processing.

[0006] An autonomous charging and collaborative operation and maintenance system for a tunnel climbing robot includes: At least one tunnel-climbing robot is used to perform detection tasks on the tunnel wall. It includes a robot mobile platform, a central processing unit and a wireless communication module, as well as a sensing module integrating multiple sensors and a magnetic charging interface. The sensing module includes a lidar and an IMU, a digital beam radar, a high-definition camera and a wireless communication module. At least one autonomous charging pile is fixedly deployed on the inner wall of the tunnel, which includes a pile controller with an internally integrated charging power module, a pile end docking mechanism that cooperates with the magnetic charging interface, and a pile end communication module. The edge computing gateway, deployed at the tunnel site, communicates with the robot's wireless communication module and the pile-end communication module to perform local data fusion and real-time processing. The cloud-based operations and maintenance platform is connected to the edge computing gateway via a wide area network for macro-level task management, data storage and analysis, and maintenance task dispatch. The tunnel climbing robot is configured to: autonomously plan a path to the nearest available autonomous charging station and dock for charging when its own battery level is below a first threshold during the detection task; during the charging docking phase, it generates pose adjustment commands by combining UWB coarse positioning and image servo control based on visual guidance markers, driving the mobile platform to complete the final fault-tolerant docking with the charging station; after the detection data collected by the defect detection sensor module is preliminarily analyzed by the edge computing gateway, if an emergency defect is identified, an alarm is immediately triggered and uploaded to the cloud operation and maintenance platform, which automatically generates and dispatches a maintenance work order. The task scheduling module of the cloud-based operation and maintenance platform is configured to run a cooperative scheduling algorithm based on model predictive control, with the goal of maximizing the system's task completion efficiency. It takes the real-time collected robot status, task queue, and charging pile status as input, solves the problem through rolling optimization, and dynamically outputs a global task package containing the detection path point sequence and predictive charging instructions to the robot. The edge computing gateway's intelligent defect processing module is configured to: sequentially perform rapid suspected region screening implemented by a first lightweight neural network model and defect classification and parameter calculation implemented by a second high-precision neural network model on the received robot image data, and dynamically trigger differentiated data upload strategies based on the comparison of the calculation results with preset thresholds.

[0007] Furthermore, the specific steps of the cooperative scheduling algorithm based on model predictive control include: a) Status Acquisition and Prediction: Real-time acquisition of the battery level, location, task progress of all robots, and the status of all charging stations; based on the robot motion and power consumption model, predict its state trajectory within a future time window; b) Optimization problem modeling: Construct a mixed integer optimization objective function with the total task completion rate of the system as the core, while penalizing total energy consumption, task delay time and charging cost; c) Rolling time domain solution: In each decision cycle, with the current system state as the initial value, the above optimization problem is solved in the prediction time domain to obtain the optimal scheduling sequence for a future period of time; d) Command issuance and feedback: The optimal command in the current time of the scheduling sequence is issued to the corresponding robot, and new status feedback is received in the next cycle, and this process is repeated.

[0008] Furthermore, the visual servo control process is as follows: the robot captures the image of the guide sign through the camera, calculates the pixel deviation of the sign center in the image coordinate system, converts the deviation into the position and angle error in the robot's own system through a pre-calibrated hand-eye matrix, and then generates a PID control quantity to drive the movement of the mobile platform until the pixel error converges to the allowable range.

[0009] Furthermore, the magnetic charging interface and the charging connector at the charging pile end are connected by a five-degree-of-freedom passive compliant structure: the connector at the charging pile end is installed through a cross slide and ball joint structure, which can translate and deflect in the plane to absorb the robot's final docking posture error. The navigation and positioning module includes a combined navigation unit that integrates at least two technologies from visual SLAM, laser SLAM, and UWB ultra-wideband positioning; The autonomous charging station also includes a guidance device to assist the robot in the final docking calibration, which can be any one of an infrared beacon, a QR code, or a visual identifier.

[0010] Furthermore, the edge computing gateway has a built-in lightweight defect recognition AI model, which is used to analyze images or data collected by the defect detection sensor module in real time, filter out suspected defect data, and prioritize uploading high-confidence emergency defect data to the cloud operation and maintenance platform. The edge computing gateway has a built-in defect processing module, which is configured to use a two-level cascaded neural network model to perform real-time analysis of the image stream and dynamically distinguish the defect level based on the comparison of the quantitative parameters of the analysis results with preset thresholds, thereby triggering differentiated data upload strategies.

[0011] Furthermore, the cloud-based operations and maintenance platform includes: The task scheduling module is used to assign periodic or temporary inspection tasks to multiple tunnel climbing robots; The digital twin module is used to construct a three-dimensional virtual model of the tunnel and the robot, enabling real-time mapping of the operating status; The maintenance management module is used to receive emergency defect alarms, automatically generate maintenance work orders containing the defect location, type, and image, and dispatch them to the terminal devices of designated maintenance personnel.

[0012] The above-mentioned method for autonomous charging and collaborative operation and maintenance of tunnel climbing robots includes the following steps: Task execution and data collection steps: The cloud-based operation and maintenance platform sends inspection task instructions to the tunnel wall climbing robot; the robot moves along the tunnel wall according to the preset path, while continuously collecting tunnel appearance status data through the defect detection sensor module; Autonomous charging decision and execution steps: The robot's main controller monitors its own power level in real time; when the power level is lower than the first preset threshold, it pauses or plans to interrupt the current task and queries the edge computing gateway for the location information of the nearest available autonomous charging station through the robot's wireless communication module; the robot autonomously navigates to the target charging station based on the location information, completes docking, and starts charging until the power level is restored to above the second preset threshold. Defect identification and classification processing steps: The edge computing gateway receives the detection data from the robot and runs the local defect identification algorithm for real-time analysis; the identification results are divided into normal data, general defects and emergency defects; for general defects, they are recorded and uploaded to the cloud in batches on a regular basis; for emergency defects, an alarm signal is generated immediately and packaged together with key data and sent to the cloud operation and maintenance platform. Collaborative Operation and Maintenance Response Steps: After receiving an alarm signal, the cloud-based operation and maintenance platform automatically triggers the operation and maintenance process: generating a detailed maintenance work order, which includes at least the defect location, type, image, and time information; dispatching the maintenance work order to the mobile terminal of the corresponding maintenance personnel through the interface; and updating the defect status to pending processing in the system's digital twin model.

[0013] Furthermore, the dynamic threshold is dynamically adjusted based on the distance of the robot's current task from the nearest charging station. The calculation formula is: Threshold = Basic threshold + Safety factor × (Energy consumption required to reach the nearest charging station / Total battery capacity).

[0014] Furthermore, in the edge intelligence processing step, the second high-precision neural network model adopts a ResNet architecture enhanced with an attention mechanism, and introduces regression loss terms for geometric parameters such as crack width and peeling area into the loss function to achieve end-to-end defect identification and quantification.

[0015] Furthermore, in the autonomous charging decision-making and execution steps, the robot performs global path planning through the navigation and positioning module to approach the charging pile. After entering the communication range of the charging pile, it performs precise positioning and posture adjustment through the identification and guidance device, and finally achieves physical docking.

[0016] Furthermore, in the defect identification and grading process, the edge computing gateway uses a deep learning-based image recognition model for defect identification and predefines the types of emergency defects. The predefined types include any one of the following: cracks exceeding the width threshold, large-area peeling, or water leakage.

[0017] Furthermore, the method also includes a system self-check and status synchronization step: the cloud-based operation and maintenance platform periodically sends heartbeat detection to the robot and the edge computing gateway, and receives the device status information returned by them. The device status information includes the robot's battery level, location, health status, and the working status of the charging pile.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention achieves true unmanned operation and maintenance: By leveraging the robot's autonomous charging capability, the bottleneck of battery life is broken, enabling the system to perform long-term, periodic tunnel monitoring without relying on human intervention.

[0019] 2. Improved emergency response speed: By leveraging the real-time capabilities of edge computing, emergency defects can be identified locally and alerted instantly, reducing the "discovery-reporting" delay from hours to minutes or even seconds, greatly enhancing the safety of tunnel operations.

[0020] 3. A closed-loop operation and maintenance management system has been formed: Robot detection, data analysis, task assignment and maintenance feedback are integrated into a unified platform, realizing digital and automated closed-loop management from problem discovery to problem handling, which significantly improves operation and maintenance efficiency.

[0021] 4. Optimized resource utilization: Through unified scheduling and multi-robot collaborative management on the cloud platform, detection tasks and charging resources can be rationally allocated to avoid conflicts and improve the overall system efficiency. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the deployment of an autonomous charging and collaborative operation and maintenance system for wall-climbing robots used in enclosed structures such as tunnels within a tunnel. Figure 2 This is a schematic diagram of the structural modules of a tunnel-climbing robot; Figure 3 This is a schematic diagram of the structural modules of an autonomous charging station; Figure 4 This is a schematic diagram of the functional modules of the cloud-based operations and maintenance platform; Figure 5 This is a flowchart of the collaborative operation and maintenance method; Figure 6 This is a schematic diagram of the operating logic of an autonomous charging and collaborative operation and maintenance system for a wall-climbing robot in enclosed structures such as tunnels.

[0023] In the diagram: 100: Tunnel climbing robot; 200: Autonomous charging pile; 300: Edge computing network management; 400: Cloud-based operation and maintenance platform; 101: Robot mobile platform; 102: Central processing unit and wireless access module; 103: Sensing module integrating multiple sensors; 104: Magnetic charging interface; 201: Pile controller; 202: Pile-end charging connector; 203: Charging power module; 204: Pile-end communication module; 205: Guiding sign; 401: Task scheduling module; 402: Digital twin module; 403: Maintenance management module. Detailed Implementation

[0024] This invention discloses an autonomous charging and collaborative operation and maintenance system and method for a tunnel wall-climbing robot. The system consists of a tunnel wall-climbing robot, an intelligent charging pile, an edge computing gateway, and a cloud-based operation and maintenance platform. The core of the method lies in: 1) The cloud platform runs a task and charging collaborative scheduling algorithm based on multi-objective rolling optimization to dynamically generate the robot's action commands; 2) The edge gateway adopts a two-level AI processing pipeline of "lightweight screening - precise quantification" to classify and compress the detection data in real time before uploading; 3) The robot achieves high-fault-tolerant magnetic docking with the wall based on fusion positioning and visual servo control; 4) By defining a standardized data interaction protocol, a closed loop of "state perception - intelligent decision-making - precise execution" is achieved. This invention realizes the full-process automation and intelligence of tunnel detection from perception, decision-making, charging to operation and maintenance response, achieving true long-term unmanned operation.

[0025] The following detailed description, in conjunction with the accompanying drawings and specific embodiments, provides a more comprehensive overview of the autonomous charging and collaborative operation and maintenance system and method for a wall-climbing robot used in enclosed structures such as tunnels. Example

[0026] (i) Construction of an autonomous charging and collaborative operation and maintenance system for wall-climbing robots used in enclosed structures such as tunnels like Figure 1 As shown, this system is deployed within a tunnel. An autonomous charging station 200 is deployed at regular intervals (e.g., every 500 meters) along the tunnel wall. One or more tunnel-climbing robots 100 are performing tasks. Edge computing gateways 300 are deployed at suitable locations at the tunnel entrance or inside. A cloud-based operations and maintenance platform 400 is deployed in a remote data center.

[0027] 1. Tunnel Climbing Robot 100 (see...) Figure 2 ) Main Controller 101: The main controller uses NVIDIA Jetson AGX Orin and runs the ROS 2 Humble system, which is responsible for overall decision control.

[0028] Navigation and positioning module 102: Employs a tightly coupled LiDAR-IMU odometry system. The robot achieves its own localization and real-time tunnel map construction through LiDAR SLAM, while simultaneously deploying UWB tags near each charging station 200. When the robot approaches a charging station, it performs precise localization via UWB (accuracy up to 10-30 cm), overcoming the cumulative errors that may occur in long, straight, and repetitive tunnel environments. Specifically, the LOAM algorithm is used for front-end scanning and matching, and the back-end uses factor graph optimization to fuse IMU pre-integrated data to construct a globally consistent map. Simultaneously, UWB anchor points are deployed near each charging station. The robot obtains centimeter-level localization in the anchor point coordinate system by calculating the time difference of arrival (TDOA). This localization result is injected into the aforementioned factor graph as an external observation factor for closed-loop correction.

[0029] The defect detection sensor module 103 includes a 20-megapixel industrial camera (for high-definition visible light imaging) and an uncooled infrared thermal imager (for detecting leaks and internal cavities). Data fusion allows for a comprehensive assessment of the tunnel's health status. Data collected by the camera and thermal imager, after H.265 encoding, along with synchronized timestamps and pose tags, is streamed via the RTSP protocol to the media server address specified by the edge computing gateway 300.

[0030] Magnetic charging interface 104: Employs a magnetic electrical connector. The robot end features a gold-plated spring probe, while the charging pile end uses copper-plated electrode plates. During docking, an electromagnet generates an attractive force to ensure stable and reliable contact. The docking control program subscribes to pose topics published by the navigation and positioning module and visual servo error topics published by the camera, managing six states—"approaching," "coarse alignment," "fine alignment," "contact," "charging," and "disconnect"—through a finite state machine (FSM).

[0031] 2. 200 self-service charging stations in the tunnel (see...) Figure 2 ) The main body of the tunnel self-charging pile includes the pile body 201, charging connector 202, charging power supply 203, pile end communication equipment 204, and guidance sign 205.

[0032] Pile 201: The main body is a steel cabinet with a protective coating on the surface. A maintenance door is provided on the back of the cabinet. The pile is fixed to the embedded parts of the tunnel sidewall by fasteners and is used to provide structural support and physical protection for each unit.

[0033] Charging connector 202: As a five-DOF passive compliant structure, it is mounted on the pile body 201. The connecting end extends outwards, and the outer shell is made of high-strength engineering plastic. The core is a copper-plated electrode plate, used to connect with the charging interface 104 of the robot requiring charging to form a charging circuit. Specifically, the five-DOF passive compliant structure consists of: a connector base mounted on the pile body via a pair of orthogonal linear slide rails, each slide rail equipped with a return spring; the base and the electrode plate are connected by a spherical bearing, enabling pitch and yaw oscillation.

[0034] Charging power supply 203: Located inside the pile body 201, it is connected to the charging connector 202, the pile end communication device 204, and the tunnel power supply system. Its internal circuit mainly consists of two parts: one part is a tunnel power supply adapter unit with a rectifier circuit and a DC / DC converter as its core, and the other part is an intelligent control unit mainly composed of a main control chip, a power switch circuit, and a communication circuit. The above circuits work together to exchange charging operation information with the pile end communication device 204 and provide power to the robot 100 in a timely manner. In addition, it is also equipped with a safety protection circuit.

[0035] The pile-end communication device 204 is installed on the pile body 201. Centered on a main control chip, it also includes a power supply regulator circuit and a communication module. The communication module is responsible for interacting with the edge computing gateway 300 via Wi-Fi 6 and communicating with the charging power supply 203 through a serial interface. The pile-end communication module subscribes to and publishes its own status (idle / occupied / faulty) via the MQTT protocol and receives instructions from the edge gateway (e.g., reserve space for robot X).

[0036] Guiding Label 205: The main body is a QR code label used to help the robot identify the specific location of the charging station. It uses a specific ID tag from the AprilTag 36h11 family. This tag has strong resistance to occlusion and lighting interference, and can provide 6-DOF pose estimation.

[0037] Edge computing gateway 300 The edge computing gateway 300 is fixed to pre-embedded parts in the tunnel sidewall using fasteners and is strategically deployed throughout the tunnel to achieve full coverage of interactive signals. Internally, the edge computing gateway 300 mainly consists of a main control computing module, a communication module, and a power module. The core of the main control computing module is the main control chip, which is primarily responsible for processing received and transmitted data and analyzing tunnel monitoring results using a built-in defect identification algorithm. The communication module receives and transmits data from the tunnel climbing robot 100, the autonomous charging pile 200, and the cloud-based operation and maintenance platform 400. This module consists of two parts: a WIFI-6 module for downlink communication with field equipment and a 4G / 5G cellular network module for uplink communication with the cloud-based operation and maintenance platform. The power module provides power to the edge computing gateway 300 using the tunnel's power distribution system.

[0038] The edge computing gateway 300 is configured to execute the following defect processing framework: receive and decode real-time video streams from the robot 100; call a first lightweight neural network model to perform initial screening of each frame of image, the first model being optimized for tunnel wall texture features and outputting the coordinates of suspected defect areas at a speed higher than 10 FPS; for the suspected areas, call a second high-precision neural network model for classification and geometric parameter calculation; dynamically classify the results into 'urgent defects' and 'general defects' according to preset threshold rules; for 'urgent defects', immediately extract keyframe images and metadata, and upload them to the cloud platform 400 through a high-priority message queue; for 'general defects', only generate logs locally and upload summaries in batches at regular intervals. The first and second models are periodically updated using model parameters trained based on historical tunnel defect data issued by the cloud platform 400. (Specific details are as follows:) Input: Pull a video stream from a media server and decode it into a sequence of image frames {It}.

[0039] First-level fast filtering: Scale It to 416x416 resolution and input the lightweight model YOLO-Fastest-XL. This model has been retrained on the tunnel defect dataset for this scene, optimizing the anchor box size. The output is a list of bounding boxes, Bt. If Bt is empty, this frame is marked as "normal" and enters the sampling queue.

[0040] The second level of precise quantification: For each bounding box b in Bt, the corresponding region is cropped from the original image It, standardized, and input into the CBAM-Attention-ResNet34 model. This model outputs two branches: a) the probability distribution of defect categories; b) for the "crack" category, its sub-pixel level edge is additionally output, and then the maximum width is calculated; for "peeling", its pixel area is output.

[0041] Leveling and Triggering: A rule engine is set up: IF (Category == Crack AND Calculated Width > 5mm) OR (Category == Peeling AND Pixel Area > 5000px) THEN Level = Urgent. Urgent data immediately triggers an upload operation.

[0042] Model Update: The cloud platform collects difficult example samples (such as misjudged "leakage") from each gateway monthly, trains them centrally, and then encrypts and distributes the new model parameters Wnew. The gateway uses a federated averaging algorithm to perform a weighted average of Wnew and the local parameter Wlocal, which is fine-tuned based on recent data, to update the local model: W = 0.9 * Wnew + 0.1 * Wlocal.

[0043] 4. Cloud-based operations and maintenance platform 400 The cloud-based operations and maintenance platform mainly consists of a four-layer framework: the physical resource layer, the data interaction layer, the core service layer, and the application presentation layer. The physical resource layer essentially comprises the hardware devices of the cloud-based operations and maintenance platform, with the core being servers located in a remote data center. The data interaction layer is responsible for receiving, parsing, and calibrating the massive data packets sent by the edge computing gateway 300, facilitating applications in the service layer. Furthermore, the data interaction layer is also responsible for packaging the instruction data streams sent from the upper layer according to the MQTT protocol and transmitting them to the edge computing gateway 300 via the wide area network. The core service layer is responsible for executing business logic based on the received data; its main functions can be categorized into three modules (see...). Figure 4 The core service layer consists of a task scheduling module 401, a digital twin module 402, and a maintenance management module 403. In addition, the core service layer is responsible for generating control data for user-initiated operation commands and sending it to the data interaction layer. The application presentation layer is responsible for direct information interaction with users, mainly including service interfaces such as WeChat robot API, mobile mini-programs, and 3D model of digital twin tunnel.

[0044] The collaborative scheduling algorithm (Model Predictive Control, MPC) of the task scheduling module 401 is implemented through the following method: System modeling: Each robot is regarded as an intelligent agent, and its discrete-time state equation is: s_k+1 = f(s_k, a_k), where the state s includes position, battery level, and current task ID, and the action a includes movement direction, speed, and whether to go to the charging station.

[0045] Prediction time domain: set to N steps (e.g., 5 minutes per step, predicting the next hour in total).

[0046] Rolling optimization: At each time t, solve the following optimization problem: Minimize: Σ_{k=t}^{t+N}[-α*R(s_k)+β*E(s_k,a_k)+γ*D(s_k)+δ*C(s_k,a_k)] Constraint condition: s_k+1=f(s_k,a_k), Energy (s_k) >= 0, Charging pile capacity constraints.

[0047] Where R represents the task reward, E represents energy consumption, D represents task delay, and C represents charging cost. This mixed integer programming problem is solved efficiently using the OR-Tools solver.

[0048] Command execution: Only the a_t in the solution sequence (i.e. the optimal action at the current moment) is converted into a specific "go to coordinate (X,Y) to perform inspection" or "go to charging pile P to charge" command, and sent via MQTT.

[0049] The digital twin module 402 is implemented using the following method: a BIM (Building Information Modeling) or high-precision laser point cloud model based on the tunnel forms the geometric skeleton of the twin. The twin module uses engines such as CesiumJS or Three.js for Web3D rendering. Each robot and charging station is instantiated as an intelligent agent object in the twin. Its position, posture, and color (e.g., yellow for low battery) are updated driven by real-time data streams, achieving synchronization with "millisecond-level latency". The module also integrates a lightweight physics engine (such as Cannon.js). When simulating future robot movements, its trajectory, energy consumption prediction, and even the safety of suction force at specific slopes can be calculated.

[0050] The maintenance management module 403 is implemented through the following method: Rule Engine: The module has a built-in configurable rule engine (such as a lightweight engine configured using Drools or JSON).

[0051] Maintenance resource modeling: Create digital profiles for each maintenance team or individual in the system, including: skill tags (proficient in crack repair, waterproofing, etc.), real-time location (via mobile phone GPS), and current task load.

[0052] Order dispatch optimization algorithm: When a new work order arrives, the module runs a "multi-target matching algorithm": Filtering: Filter out a candidate set of skill matches from idle or lightly loaded personnel.

[0053] Scoring: Calculate a comprehensive score for each candidate: Score = α * (1 / Distance to Arrival) + β * Skill Match + γ * (1 / Current Load). Where α, β, and γ are adjustable weights.

[0054] Decision-making and notification: Select the highest scorer and push the work order details and navigation link (connected to the map app) to their mobile device with one click via the "Enterprise WeChat Robot API" or "SMS Gateway". At the same time, update the person's status to "Work Order Assigned" in the digital twin.

[0055] (II) Method and Flow Taking a complete "detection-charging-alarm" cycle as an example, the implementation process of this method is illustrated: 1. Task execution and data collection: (1) The maintenance personnel set up a periodic full-line inspection task on the cloud platform 400. The platform sends the task instructions to the robot 100 through the edge gateway 300 via the 4G / 5G network.

[0056] (2) Robot 100 starts from charging pile (200-A) and travels at a constant speed along the preset path. At the same time, the camera takes pictures of the tunnel wall at a speed of 1 frame per second and transmits them to the edge computing gateway 300 in real time via Wi-Fi 6.

[0057] 2. Autonomous charging decision-making and execution: (1) Assume that when the robot travels to 1 kilometer, the main controller 101 detects that the remaining power is 20% (first threshold) and triggers the autonomous charging process.

[0058] (2) The robot sends a charging request to the edge gateway 300. The gateway finds that the charging pile (200-B) 500 meters ahead is idle, and sends its coordinates to the robot.

[0059] (3) The robot calls up the laser SLAM map and plans the shortest path to 200-B. When it is about 10 meters away from 200-B, the UWB receiver on the robot starts to work and accurately calculates the relative position with the charging station.

[0060] (4) When the distance is 2 meters, the robot activates the front camera, identifies the QR code on the charging pile (guide device 205), and makes the final posture fine adjustment.

[0061] (5) The robot slowly approaches, and the probe of the magnetic charging interface 104 contacts the electrode plate of the charging connector 202 at the pile end. The electromagnet is energized and attracted, the charging circuit is established, and high-current charging begins. Until the power reaches 95% (second threshold), the robot automatically detaches and reports to the platform "charging complete, ready to continue the task".

[0062] 3. Defect identification and graded processing: (1) During the robot detection process, the edge computing gateway 300 continuously receives image data and runs a lightweight YOLO-Fastest model adapted from MobileNet-V3 for real-time inference.

[0063] (2) The model identifies a longitudinal crack in the image with a width exceeding 5 mm and predefines it as an "urgent defect". The gateway immediately (within 1 second) performs the following actions: ① Record the defect in the local database.

[0064] ② Extract the keyframe image containing the defect from the video stream.

[0065] ③ Generate an alarm data packet in JSON format, including: robot ID, precise coordinates fused from GPS / UWB, timestamp, defect type "emergency crack", confidence level 0.92, and keyframe image.

[0066] ④ The data packet is uploaded to the cloud-based operations and maintenance platform 400 via the MQTT protocol.

[0067] 4. Collaborative Operation and Maintenance Response (see...) Figure 5 ): (1) The maintenance management module 403 of the cloud-based operation and maintenance platform 400 receives the alarm data packet.

[0068] (2) The module automatically creates a new maintenance work order, generates the work order number automatically, and fills in all alarm information.

[0069] (3) In the 3D model of the tunnel, the digital twin module 402 highlights a red alarm icon at the corresponding position.

[0070] (4) The platform pushes the work order link and brief information (such as: "Urgent! A wide crack was found at K25+130 of Tunnel No. 3. Please deal with it in time!") to the WeChat group of "Tunnel Maintenance Team" through the enterprise's internal WeChat robot API, and assigns and notifies the relevant responsible persons.

[0071] (5) The maintenance personnel can click the link to view the work order details and defect pictures on the mobile app and proceed to handle them. After the handling is completed, click "Completed" on the app, and the platform will update the status, forming a closed loop.

[0072] By combining the above systems and methods, this invention successfully constructs an intelligent, autonomous, and efficient tunnel operation and maintenance ecosystem.

[0073] The embodiments of the present invention have been described in detail above. However, the present invention is not limited to the above embodiments. Various changes that can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. An autonomous charging and collaborative operation and maintenance system for a tunnel climbing robot, characterized in that, include: At least one tunnel-climbing robot is used to perform detection tasks on the tunnel wall. It includes a robot mobile platform, a central processing unit and a wireless communication module, as well as a sensing module integrating multiple sensors and a magnetic charging interface. The sensing module includes a lidar and an IMU, a digital beam radar, a high-definition camera and a wireless communication module. At least one autonomous charging pile is fixedly deployed on the inner wall of the tunnel, which includes a pile controller with an internally integrated charging power module, a pile end docking mechanism that cooperates with the magnetic charging interface, and a pile end communication module. The edge computing gateway, deployed at the tunnel site, communicates with the robot's wireless communication module and the pile-end communication module to perform local data fusion and real-time processing. The cloud-based operations and maintenance platform is connected to the edge computing gateway via a wide area network for macro-level task management, data storage and analysis, and maintenance task dispatch. The tunnel climbing robot is configured to: autonomously plan a path to the nearest available autonomous charging station and dock for charging when its own battery level is below a first threshold during the detection task; during the charging docking phase, it generates pose adjustment commands by combining UWB coarse positioning and image servo control based on visual guidance markers, driving the mobile platform to complete the final fault-tolerant docking with the charging station; after the detection data collected by the defect detection sensor module is preliminarily analyzed by the edge computing gateway, if an emergency defect is identified, an alarm is immediately triggered and uploaded to the cloud operation and maintenance platform, which automatically generates and dispatches a maintenance work order. The task scheduling module of the cloud-based operation and maintenance platform is configured to run a cooperative scheduling algorithm based on model predictive control, with the goal of maximizing the system's task completion efficiency. It takes the real-time collected robot status, task queue, and charging pile status as input, solves the problem through rolling optimization, and dynamically outputs a global task package containing the detection path point sequence and predictive charging instructions to the robot. The edge computing gateway's intelligent defect processing module is configured to: sequentially perform rapid suspected region screening implemented by a first lightweight neural network model and defect classification and parameter calculation implemented by a second high-precision neural network model on the received robot image data, and dynamically trigger differentiated data upload strategies based on the comparison of the calculation results with preset thresholds.

2. The autonomous charging and collaborative operation and maintenance system for the tunnel climbing robot according to claim 1, characterized in that, The specific steps of the cooperative scheduling algorithm based on model predictive control include: a) Status Acquisition and Prediction: Real-time acquisition of the battery level, location, task progress of all robots, and the status of all charging stations; based on the robot motion and power consumption model, predict its state trajectory within a future time window; b) Optimization problem modeling: Construct a mixed integer optimization objective function with the total task completion rate of the system as the core, while penalizing total energy consumption, task delay time and charging cost; c) Rolling time domain solution: In each decision cycle, with the current system state as the initial value, the above optimization problem is solved in the prediction time domain to obtain the optimal scheduling sequence for a future period of time; d) Command issuance and feedback: The optimal command in the current time of the scheduling sequence is issued to the corresponding robot, and new status feedback is received in the next cycle, and this process is repeated; The visual servo control process is as follows: the robot captures the image of the guide sign through the camera, calculates the pixel deviation of the sign center in the image coordinate system, converts the deviation into the position and angle error in the robot's own system through the pre-calibrated hand-eye matrix, and then generates PID control quantity to drive the movement of the mobile platform until the pixel error converges to the allowable range.

3. The autonomous charging and collaborative operation and maintenance system for the tunnel climbing robot according to claim 1, characterized in that, The magnetic charging interface and the charging connector at the charging pile end are connected by a five-degree-of-freedom passive compliant structure: the connector at the charging pile end is installed through a cross slide and ball joint structure, which can translate and deflect in the plane to absorb the robot's final docking posture error. The navigation and positioning module includes a combined navigation unit that integrates at least two technologies from visual SLAM, laser SLAM, and UWB ultra-wideband positioning; The autonomous charging station also includes a guidance device to assist the robot in the final docking calibration, which can be any one of an infrared beacon, a QR code, or a visual identifier.

4. The autonomous charging and collaborative operation and maintenance system for the tunnel climbing robot according to claim 1, characterized in that, The edge computing gateway has a built-in lightweight defect recognition AI model, which is used to analyze images or data collected by defect detection sensor modules in real time, filter out suspected defect data, and prioritize uploading high-confidence emergency defect data to the cloud operation and maintenance platform. The edge computing gateway has a built-in defect processing module, which is configured to use a two-level cascaded neural network model to perform real-time analysis of the image stream and dynamically distinguish the defect level based on the comparison of the quantitative parameters of the analysis results with preset thresholds, thereby triggering differentiated data upload strategies.

5. The autonomous charging and collaborative operation and maintenance system for the tunnel climbing robot according to claim 1, characterized in that, The cloud-based operations and maintenance platform includes: The task scheduling module is used to assign periodic or temporary inspection tasks to multiple tunnel climbing robots; The digital twin module is used to construct a three-dimensional virtual model of the tunnel and the robot, enabling real-time mapping of the operating status; The maintenance management module is used to receive emergency defect alarms, automatically generate maintenance work orders containing the defect location, type, and image, and dispatch them to the terminal devices of designated maintenance personnel.

6. A method for autonomous charging and collaborative operation and maintenance of a tunnel-climbing robot, characterized in that, Using the system according to any one of claims 1 to 5 includes the following steps: Task execution and data collection steps: The cloud-based operation and maintenance platform sends inspection task instructions to the tunnel wall climbing robot; the robot moves along the tunnel wall according to the preset path, while continuously collecting tunnel appearance status data through the defect detection sensor module; Autonomous charging decision and execution steps: The robot's main controller monitors its own power level in real time; when the power level is lower than the first preset threshold, it pauses or plans to interrupt the current task and queries the edge computing gateway for the location information of the nearest available autonomous charging station through the robot's wireless communication module; the robot autonomously navigates to the target charging station based on the location information, completes docking, and starts charging until the power level is restored to above the second preset threshold. Defect identification and classification processing steps: The edge computing gateway receives the detection data from the robot and runs the local defect identification algorithm for real-time analysis; the identification results are divided into normal data, general defects and emergency defects; for general defects, they are recorded and uploaded to the cloud in batches on a regular basis; for emergency defects, an alarm signal is generated immediately and packaged together with key data and sent to the cloud operation and maintenance platform. Collaborative Operation and Maintenance Response Steps: After receiving an alarm signal, the cloud-based operation and maintenance platform automatically triggers the operation and maintenance process: generating a detailed maintenance work order, which includes at least the defect location, type, image, and time information; dispatching the maintenance work order to the mobile terminal of the corresponding maintenance personnel through the interface; and updating the defect status to pending processing in the system's digital twin model.

7. The autonomous charging and collaborative operation and maintenance method for a tunnel climbing robot according to claim 6, characterized in that, The dynamic threshold is dynamically adjusted based on the distance of the robot's current task from the nearest charging station. The calculation formula is: Threshold = Base threshold + Safety factor × (Energy consumption required to reach the nearest charging station / Total battery capacity).

8. The autonomous charging and collaborative operation and maintenance method for a tunnel climbing robot according to claim 6, characterized in that, The second high-precision neural network model within the edge computing gateway employs an attention-enhanced ResNet architecture and introduces regression loss terms for the geometric parameters of crack width and peeling area into the loss function to achieve end-to-end defect identification and quantification.

9. The autonomous charging and collaborative operation and maintenance method for a tunnel climbing robot according to claim 6, characterized in that, In the autonomous charging decision-making and execution process, the robot uses the navigation and positioning module to plan a global path to approach the charging pile. After entering the communication range of the charging pile, it uses the identification and guidance device to perform precise positioning and posture adjustment, and finally achieves physical docking. In the defect identification and hierarchical processing steps, the edge computing gateway uses a deep learning-based image recognition model for defect identification and predefines the types of emergency defects. The predefined types include any one of the following: cracks exceeding the width threshold, large-area peeling, or water leakage.

10. The autonomous charging and collaborative operation and maintenance method for a tunnel climbing robot according to claim 6, characterized in that, It also includes a system self-check and status synchronization step: the cloud-based operation and maintenance platform periodically sends heartbeat detection to the robot and the edge computing gateway, and receives the device status information returned by them. The device status information includes the robot's battery level, location, health status, and the working status of the charging pile.

Images