A tunnel reinforcement system and method based on multi-agent collaboration and digital twinning

Abstract

The application discloses a tunnel reinforcing system and method based on multi-agent cooperation and digital twinning, and belongs to the technical field of tunnel construction, and comprises the following steps: S1: digital twinning initialization: an agent collects target tunnel data and generates a tunnel parameter set, uploads the tunnel parameter set to a cloud platform, and generates a three-dimensional digital twinning model; S2: damage diagnosis and scheme generation: a damage intelligent diagnosis model performs hierarchical evaluation on a damage labeling area in the digital twinning model and automatically generates a reinforcing scheme; S3: multi-machine cooperative parallel construction: after a cooperative scheduling algorithm receives a construction scheme, the task is decomposed into independent subtasks that can be parallelly operated and dynamically allocated to repair the damage labeling area, and the agent triggers and starts in sequence according to the process sequence of drilling, planting, steel plate installation and grouting; the application realizes full-process automation and parallel operation, establishes a closed-loop adaptive mechanism, and solves the problems of low efficiency and lack of feedback.

Description

Technical Field

[0001] This invention belongs to the field of tunnel construction technology, specifically relating to a tunnel reinforcement system and method based on multi-agent collaboration and digital twins. Background Technology

[0002] Currently, tunnel structure reinforcement construction faces two major challenges: first, low automation levels, reliance on manual labor, resulting in safety risks and limited efficiency; second, a lack of adaptive adjustment capabilities during construction, making it difficult to address individualized damage variations in tunnel structures. Existing technologies include a tunnel construction robot that automates the installation of suspended components, but its functionality is limited, failing to cover the entire process from detection to grouting, and its fixed operating modules lack adaptability to complex tunnel environments. Chinese patent CN111396096A discloses a "construction robot for tunnel steel-bonded reinforcement," which, while integrating detection and multiple construction modules, relies on a single robotic arm for sequential operation, resulting in significant efficiency bottlenecks and a lack of real-time closed-loop feedback and autonomous decision-making capabilities during construction.

[0003] A typical process for tunnel reinforcement and repair includes: first, a full-section inspection of the tunnel lining to identify damage such as cracks, voids, and rebar distribution; then, a steel plate anchoring scheme is designed based on the damage level; followed by drilling, implantation of chemical anchors, steel plate installation and bonding, grouting, and sealing. This process has significant unique characteristics when executed within the confined space of a tunnel: the working space is limited in height, the tunnel's inner diameter is fixed, and simultaneous operation of multiple robotic arms can easily lead to spatial interference; the construction procedures have strong time constraints, requiring strict sequential execution of drilling, anchoring, steel plate installation, and grouting, with subsequent procedures waiting for the completion of preceding ones; and there are many environmental interference factors, such as the possibility of touching rebar during drilling requiring real-time adjustment of the hole position, and the need for timely testing of grouting effects to determine the presence of voids, resulting in considerable uncertainty in the actual time consumption of each procedure.

[0004] In summary, existing technologies cannot solve the efficiency bottleneck caused by sequential operations in tunnel reinforcement construction, nor can they cope with the conflicts of parallel operations under strong temporal constraints of multiple processes in confined spaces, and they lack the ability to adaptively adjust based on real-time detection data. Therefore, there is an urgent need for a tunnel reinforcement construction robot swarm system and method that can realize parallel construction by multiple agents, dynamic task matching, local anomaly rescheduling, spatial conflict resolution, and closed-loop feedback throughout the entire process. Summary of the Invention

[0005] This invention provides a tunnel reinforcement system and method based on multi-agent collaboration and digital twins, which solves the technical problems of low automation, low efficiency of serial operation, and lack of real-time closed-loop feedback and adaptive adjustment capabilities.

[0006] The objective of this invention can be achieved through the following technical solutions: A tunnel reinforcement method based on multi-agent collaboration and digital twins includes the following steps: S1, Digital Twin Initialization: The intelligent agent collects target tunnel data and generates a tunnel parameter set, uploads the tunnel parameter set to the cloud platform, and generates a three-dimensional digital twin model; S2, Damage Diagnosis and Solution Generation: The intelligent damage diagnosis model deployed on the cloud platform performs a graded assessment of the damage-marked areas in the three-dimensional digital twin model, and the reinforcement solution generation module deployed on the cloud platform automatically generates a reinforcement solution. S3, Multi-machine collaborative parallel construction: After receiving the reinforcement plan, the collaborative scheduling algorithm module deployed on the cloud platform decomposes the reinforcement plan into independent sub-tasks that can be parallelized and dynamically allocates them to repair the damaged marked areas. After receiving the corresponding independent sub-tasks, the multiple intelligent agents are triggered and started in sequence according to the process order of drilling and bolting, steel plate installation and grouting. The intelligent agent of the next process only starts working after the corresponding sub-task of the previous process is completed.

[0007] Preferably, step S2 includes the following: S201, the intelligent damage diagnosis model performs a graded assessment of the damage-marked areas in the three-dimensional digital twin model and generates a structured diagnosis report; the structured diagnosis report includes damage-marked data and damage grading results generated based on the damage-marked data; the damage-marked data includes crack location coordinates, crack width, segment ellipticity deviation value, and internal reinforcement distribution diagram. S202, the reinforcement scheme generation module receives the structured diagnostic report and performs qualitative reasoning on the damage grading results. The reasoning results and the damage annotation data are used together as input constraints for the finite element simulation model, and the optimal steel plate block size, bolt spacing arrangement and grouting pressure parameters are automatically calculated. S203, the reinforcement scheme generation module generates a reinforcement scheme data package that includes a steel plate block setting diagram, a bolt arrangement diagram, a grouting parameter table, and construction procedure optimization.

[0008] Preferably, step S3, decomposing the reinforcement scheme into parallel independent subtasks, includes the following: parsing the reinforcement scheme data package into a task dependency network in the form of a directed acyclic graph; based on the inherent sequential constraints of tunnel reinforcement construction, namely drilling and bolting, steel plate installation, and grouting are executed sequentially, marking the task nodes corresponding to the above procedures as dependent task nodes that must be executed sequentially; and based on the damage distribution in the tunnel parameter set, marking the same procedure task nodes in different damage areas as independent subtasks that can be executed in parallel.

[0009] Preferably, the dynamic allocation in step S3 to repair the damaged marked area includes: the collaborative scheduling algorithm module, based on the real-time location, remaining power, and sensor status of multiple agents, adopts an improved Hungarian algorithm with the dual objective functions of minimizing completion time and minimizing movement path conflicts, to perform optimal matching between tasks and agents.

[0010] Preferably, step S3 further includes a dynamic rescheduling step: during the operation, when a construction deviation or agent failure is detected, the collaborative scheduling algorithm module triggers local task replanning to reallocate unfinished tasks to other agents.

[0011] Preferably, after the dynamic rescheduling step in step S3, a conflict resolution step is also included: when there is overlap in the work space of adjacent agents, the cooperative scheduling algorithm module automatically allocates time windows or adjusts the work order to eliminate interference based on the collision prediction model of the coordinate system in the tunnel.

[0012] Preferably, it also includes S4, process monitoring and adaptive adjustment: the intelligent agent continuously scans and monitors the construction area in real time to obtain real-time data, and the cloud platform compares the real-time data with the three-dimensional digital twin model in real time and triggers adaptive adjustment commands based on the deviation.

[0013] A tunnel reinforcement system based on multi-agent collaboration and digital twins, comprising a cloud platform and multiple agents; The cloud platform is equipped with a damage intelligent diagnosis model, a repair scheme adaptive generation module, and a collaborative scheduling algorithm module. The input end of the intelligent damage diagnosis model is used to obtain the tunnel parameter set, and the output end of the intelligent damage diagnosis model is connected to the input end of the reinforcement scheme generation module to transmit the damage grading assessment results to the reinforcement scheme generation module. The output of the reinforcement scheme generation module is connected to the input of the collaborative scheduling algorithm module, and is used to transmit the generated reinforcement scheme to the collaborative scheduling algorithm module; Multiple of the aforementioned intelligent agents are deployed within the tunnel and connected to the cloud platform via a wireless communication network; The collaborative scheduling algorithm module communicates with the control terminals of the multiple intelligent agents through the wireless communication network.

[0014] Preferably, the intelligent agent includes: The detection agent is used to collect tunnel structure data and real-time monitoring data of the construction process; A drilling agent used to drill anchor bolt holes and insert anchor bolts; Install intelligent agents to attach steel plates to the pipe segment wall and secure them; The grouting intelligent agent is used to perform grouting, fill grouting, and sealing treatment.

[0015] The beneficial effects of this invention are as follows: 1. Enhance the automation and intelligence of the entire construction process: Through digital twin initialization and intelligent damage diagnosis, realize the automatic acquisition of tunnel structure data, 3D modeling and automatic generation of reinforcement schemes, replace traditional manual inspection and design, and significantly reduce human intervention and safety risks.

[0016] 2. Break through the efficiency bottleneck of serial operation and realize multi-machine parallel collaboration: adopt a multi-cooperative scheduling algorithm to decompose the reinforcement scheme into independent sub-tasks that can be parallelized, and trigger each intelligent agent in sequence according to the time constraints of drilling, bolting, steel plate installation and grouting, so that the same or different processes between different damaged areas can be executed in parallel. Under the premise of strictly following the tunnel reinforcement process sequence, the overall construction efficiency of the cluster is greatly improved.

[0017] 3. Establish closed-loop adaptive control capability throughout the construction process: By continuously comparing real-time monitoring data during the construction process with the cloud-based digital twin model, adaptive adjustment commands are automatically triggered based on deviations, forming a closed-loop feedback mechanism of "perception-decision-execution-verification". This enables the construction system to dynamically respond to individual tunnel damage differences and on-site interference factors, significantly enhancing the robustness of the system and the controllability of construction quality. Attached Figure Description

[0018] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0019] Figure 1 This is a flowchart of a tunnel reinforcement process provided in one embodiment of the present invention.

[0020] Figure 2 This is a global schematic diagram of tunnel reinforcement provided in one embodiment of the present invention.

[0021] Figure 3 This is a schematic diagram of the robot swarm system relationship provided in one embodiment of the present invention.

[0022] Figure 4 This is a schematic diagram of the intelligent agent installation structure provided in one embodiment of the present invention. Legend: 1. Mobile shared base platform; 2. Intelligent agent; 3. Steel plate; 4. Crack. Detailed Implementation

[0023] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided.

[0024] like Figures 1-4 As shown, a tunnel reinforcement method based on multi-agent collaboration and digital twins includes the following steps: S1, Digital Twin Initialization: Agent 2 enters the target tunnel section to collect tunnel data and generate a tunnel parameter set. The tunnel parameter set is then uploaded to the cloud platform to generate a three-dimensional digital twin model.

[0025] S2, Damage Diagnosis and Solution Generation: The built-in intelligent damage diagnosis model on the cloud platform performs graded evaluation of the damage-marked areas in the 3D digital twin model. For example, based on whether the width of crack 4 exceeds 0.3mm or whether the ellipticity deviation exceeds 25‰, the damaged area is divided into different levels, and corresponding reinforcement solutions are automatically generated for each damaged area, including parameters such as the size of the steel plate 3 to be pasted, the number and location of anchor bolts, and the grouting pressure.

[0026] S3, Multi-machine Collaborative Parallel Construction: After receiving the reinforcement plan, the cloud platform's collaborative scheduling algorithm model decomposes the tasks in the plan into independent, parallelizable sub-tasks and dynamically assigns them to each agent 2. Specifically, there are three independent damaged areas within the tunnel, each requiring the sequential completion of four processes: drilling and bolting, steel plate 3 installation, and grouting. The collaborative scheduling algorithm model triggers the activation of each type of agent 2 in the order of drilling and bolting, steel plate 3 installation, and grouting, strictly limiting the operation of an agent 2 for a later process to only begin after the corresponding task of the previous process has been completed.

[0027] For example, the scheduling algorithm first triggers the drilling agent to perform drilling tasks in the first damaged area. After the drilling agent completes all drilling and plugging in the area and reports the completion signal, the collaborative scheduling algorithm module arranges the next agent 2 to enter the first area to start. At the same time, since the second damaged area is independent of the first area, the drilling agent can immediately move to the second area to continue drilling after completing the first area. While the next agent 2 is working in the first area, the drilling agent 2 is drilling in the second area, and the two work in parallel.

[0028] Through this "process-level serial and region-level parallel" triggering mechanism, this method, while strictly adhering to the inherent process sequence of drilling and bolting → steel plate 3 installation → grouting, enables the parallel execution of the same or different processes between different damaged areas, thereby solving the problem of low efficiency of existing single-arm serial intelligent agents.

[0029] In one embodiment, step S2 includes the following: S201, the intelligent damage diagnosis model performs a graded assessment of the damage-annotated areas in the three-dimensional digital twin model and generates a structured diagnosis report; the structured diagnosis report includes damage annotation data and damage grading results generated based on the damage annotation data; the damage annotation data includes the location coordinates of crack 4, the width of crack 4, the ellipticity deviation value of the segment, and the distribution diagram of internal reinforcement.

[0030] S202, in the scheme reasoning stage, using the engineering material library (including performance parameters such as Q345 steel with tensile strength ≥345MPa, M16 chemical anchors, and rigid epoxy resin) as constraints, the reinforcement scheme generation module performs qualitative reasoning on the damage classification results based on the rule reasoning engine. The reasoning results and damage annotation data are used together as input constraints for the finite element simulation model, automatically calculating the optimal steel plate 3-block dimensions, bolt spacing arrangement, and grouting pressure parameters. For example: If the width of crack 4 is greater than 0.3mm and less than 0.5mm, the reinforcement scheme generation module determines it as moderate damage and requires the application of a single layer of steel plate 3. If the width of crack 4 is greater than 0.5mm or the ellipticity deviation is greater than 25‰, it is determined as severe damage and requires the application of a double layer of steel plate 3 and increased bolt spacing. The rule reasoning engine outputs a qualitative reasoning result, such as "moderate damage, requires application of a single layer of steel plate 3, recommended bolt spacing 400mm".

[0031] Then, the qualitative reasoning result is used as the input constraint condition for the finite element simulation model. Based on the actual geometric dimensions of the tunnel lining, material parameters, and the distribution of cracks 4 in the damaged area, the finite element simulation model automatically iterates and calculates under the constraint conditions, and outputs the optimal steel plate 3 segment size. For example, a steel plate 3 with a length of 3m and a width of 1m is divided into two steel plates 3 of 1.5m × 1m, to adapt to the distribution of cracks 4 and facilitate transportation and bolt spacing.

[0032] For example, an M16 anchor bolt is placed every 300mm along the edge of steel plate 3, avoiding the positions of the reinforcing bars marked on the reinforcing bar distribution diagram, and the grouting pressure parameters are adjusted accordingly. For instance, the initial grouting pressure is set to 0.3MPa, and then adjusted to 0.4MPa based on the width of crack 4 and the thickness of the segment. The entire calculation process requires no manual intervention.

[0033] S203, Scheme Output Stage: After the finite element simulation model calculations are completed, the cloud platform packages the results into a reinforcement scheme data package. This data package specifically includes: a CAD or PDF format diagram of the steel plate 3 segmentation, indicating the length, width, and thickness of each steel plate 3, as well as the overlap method between segments; a bolt layout diagram on the tunnel lining unfolded diagram, indicating the coordinates of each bolt hole and the offset of the reinforcement reinforcement; a grouting parameter table listing the grouting pressure, estimated grouting volume, and grouting sequence for each grouting hole; and construction procedure optimization suggestions, such as reinforcing the most severely damaged area first, or grouting the arch top first to prevent grout flow downwards. Finally, the cloud platform uses a 5G communication network to send this reinforcement scheme data package to each intelligent agent 2 within the tunnel with a single click.

[0034] In multi-agent collaborative operations for tunnel reinforcement construction, how to transform the reinforcement scheme into a task sequence that can be executed by the agent 2 is a key challenge. Although existing technologies integrate multiple modules, all modules share the same robotic arm, making true parallelism impossible. In one embodiment, step S3, which decomposes the reinforcement scheme into parallel independent sub-tasks, includes the following: parsing the reinforcement scheme data packet into a task dependency network in the format of a directed acyclic graph.

[0035] Specifically, based on the inherent sequential constraints of tunnel reinforcement construction, namely drilling, bolting, steel plate 3 installation, and grouting must be performed sequentially, the task nodes corresponding to these four processes are marked as dependent task nodes that must be performed sequentially. This means that the installation of any steel plate 3 must wait until the drilling and bolting in its corresponding area are completed before it can begin.

[0036] Meanwhile, based on the damage distribution in the tunnel parameter set, for example, if three independent damage areas are detected in the tunnel at different mileage locations, the same process task nodes in different damage areas are marked as independent task nodes that can be executed in parallel.

[0037] For example, drilling in region A, region B, and region C can be performed simultaneously by multiple drilling agents 2, without interference. Similarly, tack plugging in region A and drilling in region B can be performed concurrently because they belong to different processes in different regions and do not overlap spatially. Through this dependency network modeling, this solution clarifies which tasks must wait and which tasks can be executed simultaneously, providing a clear structured input for subsequent resource matching and dynamic scheduling. This solves the problem that existing technologies cannot identify parallel opportunities under strong temporal constraints.

[0038] When multiple agents 2 simultaneously execute different tasks in different areas, how to allocate tasks to the most suitable agent 2 to achieve the shortest overall completion time and the least path conflict is the core challenge in cluster scheduling. In the existing technology, the fixed allocation method cannot adapt to the changes in the real-time location and power of agent 2, which can easily lead to one agent 2 being overloaded while another is idle, or multiple agents 2 competing for the same work surface in a narrow tunnel. In one embodiment, the dynamic allocation in step S3 to repair the damaged marked area includes: the collaborative scheduling algorithm module, based on the real-time location, remaining power, and sensor status of multiple agents 2, adopts an improved Hungarian algorithm, with minimizing completion time and minimizing movement path conflict as the dual objective functions, to perform optimal matching between tasks and agents 2.

[0039] The Hungarian algorithm was originally used to solve the assignment problem. This scheme improves upon it, using minimizing completion time and minimizing movement path conflicts as the dual objective function. Specific use case: Assume there are three drilling tasks in a tunnel located at mileages K1, K2, and K3 respectively. The three drilling agents are currently located near K1, near K4, and at the tunnel entrance, with remaining battery levels of 80%, 30%, and 100% respectively.

[0040] The improved Hungarian algorithm constructs a 3×3 cost matrix, where the cost of each cell is a weighted sum of three parts: the travel time between the task point and agent 2's current position, whether agent 2's remaining battery power can support the completion of the task, and the probability of path conflicts that may occur if multiple agents 2 move simultaneously (e.g., if K1 and K2 are too close, sending two agents 2 simultaneously will cause a conflict, increasing the cost). The algorithm searches globally for the assignment scheme that minimizes the total cost. The result is that task K1 is assigned to agent 2 located near K1 (shortest travel time), task K2 is assigned to agent 2 at the entrance (sufficient battery power and path does not conflict with K1), and task K3 is assigned to agent 2 near K4 (although slightly farther, moderate battery power). This achieves dynamic optimal matching between tasks and agents 2, maximizing cluster collaboration efficiency while ensuring construction safety.

[0041] Tunnel construction is subject to numerous unforeseen disruptive factors. These include accidental contact with reinforcing bars during drilling, necessitating repositioning of the borehole; prolonged curing time of chemical anchoring agents during bolt installation due to temperature variations; and malfunctions of one of the intelligent agents (agent 2). Existing static scheduling techniques cannot handle these dynamic changes, and deviations often lead to the entire work line stalling. In one embodiment, a dynamic rescheduling step is included: when a construction deviation or intelligent agent 2 malfunction is detected, the collaborative scheduling algorithm module triggers local task replanning, reallocating unfinished tasks to other intelligent agents.

[0042] For example, when a drilling agent is performing a drilling task in a certain area, the torque sensor detects that the drilling resistance suddenly exceeds the threshold, indicating that it has touched a rebar. According to the rebar avoidance strategy, the drilling coordinates need to be shifted and the drilling needs to be restarted, causing the estimated completion time of the task to be extended by 15 minutes compared to the original plan. At this time, the cooperative scheduling algorithm module immediately identifies this deviation and performs local replanning only for the unfinished drilling tasks in that area and the subsequent bolting and installation tasks that depend on that drilled hole, while the parallel operations in other areas are not affected. The replanning involves dispatching another idle drilling agent to assist in completing the remaining holes, or adjusting the time window order of subsequent tasks. If an agent 2 malfunctions and cannot continue working, the cooperative scheduling algorithm module will extract the unfinished task list of agent 2, re-match it to other agents 2 of the same type with sufficient power, and recalculate the optimal path.

[0043] The inner diameter of a tunnel is typically fixed and narrow. When multiple robotic agents operate simultaneously, their range of motion may overlap. For example, if a drilling robotic agent is drilling in a certain area, and another robotic agent is installing steel plates (3), it may need to enter the same area to install the steel plates (3). Without a conflict resolution mechanism, they may collide, causing equipment damage or construction interruption. In existing technologies, multi-agent conflict resolution is mainly applied to AGV path planning, such as using time windows to avoid collisions between mobile robotic agents in a warehouse. However, the interference in the working space of robotic arms within a tunnel is more complex because the posture of the robotic arms can change while the working point is fixed.

[0044] In one embodiment, after the dynamic rescheduling step in step S3, a conflict resolution step is also included: when there is overlap in the work space of adjacent agents 2, the collaborative scheduling algorithm module automatically allocates time windows or adjusts the work order to eliminate interference based on the collision prediction model of the coordinate system in the tunnel. Spatial collision information is mapped to temporal collision information, and time windows are automatically allocated or the work order is adjusted to eliminate interference.

[0045] For example, near the same mileage section of the tunnel, drilling is required on the left side, and steel plate 3 needs to be installed on the right side. The workspaces of the robotic arms of the two agents overlap near the tunnel centerline. The collision prediction model first calculates the possible spatial overlap intervals on the time axis based on the current posture of the two robotic arms, the coordinates of the target work point, and the motion trajectory. Then, the collaborative scheduling algorithm module automatically allocates time windows for this section: for example, it stipulates that within 0 to 10 minutes, the drilling agent on the left occupies the overlapping space, and the installation agent on the right postpones its entry; within 10 to 20 minutes, the installation agent on the right occupies the space, and the drilling agent on the left retreats after completion. If an agent 2 is delayed for any reason, the collaborative scheduling algorithm module will dynamically adjust the subsequent time windows. Another approach is to adjust the work order: if drilling and installation have a process dependency, drilling must be completed and the agent must exit before installation is allowed to enter. Through this collision prediction-based time window allocation and order adjustment, this solution actively eliminates the risk of spatial interference while ensuring parallel efficiency, and solves the safety problem of multiple agents 2 operating in parallel within the confined space of the tunnel.

[0046] In one embodiment, it also includes S4: process monitoring and adaptive adjustment: during the installation of steel plate 3, the visual positioning module monitors the fit gap between steel plate 3 and the wall of the pipe segment in real time. When the cloud platform finds that the gap in a certain local area exceeds 3 mm, it immediately sends an instruction to the installation intelligent body, requiring the hydraulic loader to increase the local pressure in that area until the fit meets the standard.

[0047] During grouting, ultrasonic sensors monitor the filling status in real time. If a void is detected, the cloud platform immediately sends a grouting instruction to the grouting agent, adjusting the grouting pressure from 0.3 MPa to 0.5 MPa until the void is eliminated. If rebar is encountered during drilling, the cloud platform recalculates the rebar avoidance offset coordinates based on the ultrasonic rebar distribution map and sends it to drilling agent 2. All adjustment instructions are executed locally without interrupting the operation of other agents, and the adjusted real-time data is uploaded to the cloud platform again, forming a closed-loop verification.

[0048] In addition, S5 is included: Closed-loop grouting and quality verification. After the initial grouting material has cured, the intelligent agent 2 performs a supplementary grouting process: areas with hollow spots detected by ultrasonic testing are re-grouted until the test is passed; subsequently, sealant is applied to the outer edge of the steel plate 3 and around the bolts to complete the waterproof sealing. The entire grouting process adopts a pressure-flow dual control mode to ensure that "the epoxy resin grout does not shrink after solidification, and achieves good filling of the gap between the pipe segment and the steel plate 3, without hollow spots." After the grouting is qualified, the intelligent agent 2 performs a final tightening operation on all bolts in sequence. Finally, the construction area is scanned for acceptance and the bolt torque is measured, and a construction quality inspection report is generated and uploaded to the cloud.

[0049] A tunnel reinforcement system based on multi-agent collaboration and digital twins includes a cloud platform and multiple agents. The cloud platform is equipped with a damage intelligent diagnosis model, a repair scheme adaptive generation module, and a collaborative scheduling algorithm module. Multiple agents are deployed inside the tunnel and connected to the cloud platform via a wireless communication network. The output of the reinforcement scheme generation module is connected to the input of the collaborative scheduling algorithm module, which communicates with the control terminal of the agents via a wireless communication network.

[0050] In one embodiment, the intelligent agent 2 includes: a detection intelligent agent for collecting tunnel structure data and real-time monitoring data of the construction process; a drilling intelligent agent for drilling anchor bolt holes and inserting anchor bolts; an installation intelligent agent for attaching the steel plate 3 to the tunnel segment wall and fastening it; and a grouting intelligent agent for performing joint grouting, supplementary grouting and sealing treatment.

[0051] It also includes a mobile shared base platform 1 for movement, positioning, and energy sharing in tunnels, using a steel ring reinforcement project in an operating subway tunnel as an example. After system initialization, the detection agent completes the scan, and the cloud platform diagnoses that some tunnel segments have through cracks 4 and an ellipticity exceeding 25%, determining that reinforcement is required. The cloud platform automatically generates a solution based on standard design and performs simulation verification, and then the task is issued.

[0052] On-site, the drilling and installation robots operated in parallel. The drilling robot followed a cloud-based hole location map that had already avoided the reinforcing bars. Simultaneously, the installation robot completed a simulated trial assembly of the first ring steel plate 3 outside the tunnel before transporting the component into the tunnel. During installation, laser tracking was used to correct the position in real time to ensure accuracy.

[0053] When the intelligent agent is installed to temporarily fix the steel plate 3, the detection intelligent agent simultaneously scans the gap between the steel plate 3 and the segment, and the data is transmitted back in real time. The cloud platform calculates and finds that the local gap exceeds the tolerance, and immediately instructs the installation intelligent agent to fine-tune the pressure of the "hydraulic loader" in that part, and optimizes the grouting parameters of the grouting intelligent agent in that area.

[0054] During the grouting process, the grouting intelligent system strictly follows the adjusted parameters, and pressure sensors ensure dense filling. After completion, the final step is to automatically link the construction data to the corresponding positions in the 3D digital twin model, forming a complete as-built digital archive.

[0055] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A tunnel reinforcement method based on multi-agent collaboration and digital twin, characterized in that, Includes the following steps: S1, Digital Twin Initialization: The intelligent agent collects target tunnel data and generates a tunnel parameter set, uploads the tunnel parameter set to the cloud platform, and generates a three-dimensional digital twin model; S2, Damage Diagnosis and Solution Generation: The intelligent damage diagnosis model deployed on the cloud platform performs a graded assessment of the damage-marked areas in the three-dimensional digital twin model, and the reinforcement solution generation module deployed on the cloud platform automatically generates a reinforcement solution. S3: Multi-machine collaborative parallel construction. After receiving the reinforcement plan, the collaborative scheduling algorithm module deployed on the cloud platform decomposes the reinforcement plan into independent sub-tasks that can be parallelized and dynamically allocates them to repair the damaged marked areas. After receiving the corresponding independent sub-tasks, the multiple intelligent agents are triggered and started in sequence according to the process order of drilling and bolting, steel plate installation and grouting. The intelligent agent of the next process only starts working after the corresponding sub-task of the previous process is completed.

2. The tunnel reinforcement method based on multi-agent collaboration and digital twin as described in claim 1, characterized in that, Step S2 includes the following: S201, the intelligent damage diagnosis model performs a graded assessment of the damage-marked areas in the three-dimensional digital twin model and generates a structured diagnosis report; the structured diagnosis report includes damage-marked data and damage grading results generated based on the damage-marked data; the damage-marked data includes crack location coordinates, crack width, segment ellipticity deviation value, and internal reinforcement distribution diagram. S202, the reinforcement scheme generation module receives the structured diagnostic report and performs qualitative reasoning on the damage grading results. The reasoning results and the damage annotation data are used together as input constraints for the finite element simulation model, and the optimal steel plate block size, bolt spacing arrangement and grouting pressure parameters are automatically calculated. S203, the reinforcement scheme generation module generates a reinforcement scheme data package that includes a steel plate block setting diagram, a bolt arrangement diagram, a grouting parameter table, and construction procedure optimization.

3. The tunnel reinforcement method based on multi-agent collaboration and digital twin as described in claim 1, characterized in that, Step S3, which decomposes the reinforcement scheme into parallel independent subtasks, includes the following: parsing the reinforcement scheme data package into a task dependency network in the form of a directed acyclic graph; based on the inherent sequential constraints of tunnel reinforcement construction, namely drilling and bolting, steel plate installation, and grouting, which are executed sequentially, marking the task nodes corresponding to the above procedures as dependent task nodes that must be executed sequentially; and based on the damage distribution in the tunnel parameter set, marking the same procedure task nodes in different damage areas as independent subtasks that can be executed in parallel.

4. The tunnel reinforcement method based on multi-agent collaboration and digital twin as described in claim 3, characterized in that, The dynamic allocation in step S3 to repair the damaged marked area includes: the cooperative scheduling algorithm module, based on the real-time location, remaining power and sensor status of multiple agents, adopts an improved Hungarian algorithm, with minimizing completion time and minimizing movement path conflict as the dual objective functions, to perform optimal matching between tasks and agents.

5. A tunnel reinforcement method based on multi-agent collaboration and digital twins according to claim 4, characterized in that, Step S3 also includes a dynamic rescheduling step: during the operation, when a construction deviation or agent failure is detected, the collaborative scheduling algorithm module triggers local task replanning, and reassigns the unfinished tasks to other agents.

6. A tunnel reinforcement method based on multi-agent collaboration and digital twins according to claim 5, characterized in that, Following the dynamic rescheduling step in step S3, a conflict resolution step is also included: when adjacent agents have overlapping job spaces, the collaborative scheduling algorithm module automatically allocates time windows or adjusts job order to eliminate interference based on the collision prediction model of the tunnel coordinate system.

7. The tunnel reinforcement method based on multi-agent collaboration and digital twin as described in claim 1, characterized in that, It also includes S4, process monitoring and adaptive adjustment: the intelligent agent continuously scans and monitors the construction area in real time to obtain real-time data, and the cloud platform compares the real-time data with the three-dimensional digital twin model in real time and triggers adaptive adjustment commands based on the deviation.

8. A tunnel reinforcement system based on multi-agent collaboration and digital twins, applicable to the tunnel reinforcement method based on multi-agent collaboration and digital twins as described in any one of claims 1-7, characterized in that, Including a cloud platform and multiple intelligent agents; The cloud platform is equipped with a damage intelligent diagnosis model, a repair scheme adaptive generation module, and a collaborative scheduling algorithm module. The input end of the intelligent damage diagnosis model is used to obtain the tunnel parameter set, and the output end of the intelligent damage diagnosis model is connected to the input end of the reinforcement scheme generation module to transmit the damage grading assessment results to the reinforcement scheme generation module. The output of the reinforcement scheme generation module is connected to the input of the collaborative scheduling algorithm module, and is used to transmit the generated reinforcement scheme to the collaborative scheduling algorithm module; Multiple of the aforementioned intelligent agents are deployed within the tunnel and connected to the cloud platform via a wireless communication network; The collaborative scheduling algorithm module communicates with the control terminals of the multiple intelligent agents through the wireless communication network.

9. A tunnel reinforcement system based on multi-agent collaboration and digital twins according to claim 8, characterized in that, The intelligent agent includes: The detection agent is used to collect tunnel structure data and real-time monitoring data of the construction process; A drilling agent used to drill anchor bolt holes and insert anchor bolts; Install intelligent agents to attach steel plates to the pipe segment wall and secure them; The grouting intelligent agent is used to perform grouting, fill grouting, and sealing treatment.

Images