Large-transportation-bridge load automatic checking method based on multi-agent cooperation

Abstract

The application discloses a large piece transportation bridge load automatic checking and calculating method based on multi-agent cooperation, comprising the following steps: receiving and analyzing the natural language macro instruction input by a user; disassembling a standardized task description, formulating an agent cooperation strategy by a core layer and issuing a task execution token; generating a checking bridge parameter and a model list by a GIS spatial analysis and calculation agent; estimating calculation power by an engineering calculation agent and calling a supercomputing center to distribute a virtual machine or a calculation node; generating a single finite element calculation report on each distributed calculation node; monitoring a node, cutting off an abnormal task and returning to a healthy node to execute again; extracting key indicators and determining the safety level of the bridge by a report generation agent; and generating an overall checking report and providing the user with the checking report for review and download. The application makes full use of multi-agent cooperation and supercomputing parallel calculation, automatically executes bridge load checking and risk analysis, and has the advantages of high efficiency, precision and reliability.

Description

Technical Field

[0001] This invention relates to the field of bridge load calculation, and in particular to an automated load calculation method for heavy transport bridges based on multi-agent collaboration. Background Technology

[0002] Existing bridge verification methods largely rely on manual or semi-automated methods to collect bridge structural parameters, establish vehicle load models, and perform finite element analysis. This typically requires manual compilation of route information, bridge lists, and vehicle load data, making the calculation process cumbersome and time-consuming. Especially when dealing with large-scale transport tasks across multiple sections, traditional methods struggle to achieve rapid parallel analysis of multiple bridges and various operating conditions. Furthermore, the integration of parameters from different bridges and vehicles in existing technologies often relies on manual judgment, lacking unified data processing and automated scheduling methods. This results in low computational efficiency, difficulty in ensuring data consistency, and a tendency for omissions or errors in complex large-scale transport scenarios.

[0003] Furthermore, existing bridge load verification systems lack dynamic awareness and adaptive capabilities in allocating computing resources. They cannot intelligently adjust computing power based on the complexity of the bridge structure and the scale of the load. Moreover, the identification and rollback mechanisms for abnormal calculation tasks are inadequate, requiring manual intervention when calculations are interrupted or results are incomplete. This prevents the achievement of fully automated and intelligent management of the entire process. These shortcomings limit the accuracy and efficiency of bridge load verification in cross-regional heavy-lift transportation tasks, making it difficult to meet the actual needs for rapidly generating safety level assessments and risk warnings.

[0004] Therefore, how to provide an automated method for calculating the load of heavy-duty transport bridges based on multi-agent collaboration is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose an automated load calculation method for heavy-duty transport bridges based on multi-agent collaboration. This invention fully utilizes multi-agent collaboration and supercomputing parallel computing to automatically perform bridge load calculations and risk analyses, possessing the advantages of high efficiency, accuracy, and reliability.

[0006] The automated load calculation method for heavy-lift transport bridges based on multi-agent collaboration according to embodiments of the present invention includes the following steps:

[0007] It receives macro-level natural language commands input by users through the interactive interface, and the core layer large language model parses the command content to generate a standardized task description;

[0008] The standardized task description is broken down into an ordered sequence of subtasks, and the core layer formulates the agent collaboration strategy and issues task execution tokens.

[0009] The GIS spatial analysis and computational agent parses the route and verifies vehicle characteristic parameters based on an ordered sub-task sequence, and extracts bridge structural parameters and high-fidelity finite element model file indexes to generate a list of verified bridge parameters and models.

[0010] The engineering computing agent receives and verifies the bridge parameters and model list, estimates the computing power based on the structural complexity of the bridge model, calls the supercomputing center to allocate virtual machines or computing nodes, and generates a grouped parallel computing scheduling table.

[0011] At each assigned computing node, the engineering computing agent loads the finite element model, applies non-standard vehicle loads, sets boundary conditions, and triggers mechanical solution according to the grouped parallel computing schedule, generating a single finite element calculation report.

[0012] During parallel computing, the status of computing nodes is monitored in real time. If a faulty node or non-convergent task is detected, the engineering computing agent automatically truncates the abnormal task, generates abnormal labeling information, and rolls back the abnormal task to a healthy node for re-execution.

[0013] The report-generating agent extracts key indicators from individual finite element calculation reports, determines the safety level of each bridge, and outputs the calculation results for each bridge.

[0014] The report-generating agent statistically analyzes the calculation results and anomaly annotation information of a single bridge to generate a comprehensive calculation report, which is then provided to users for review and download.

[0015] Optionally, the natural language macro-instructions include route range, vehicle information, and verification constraints;

[0016] Optionally, the formulation of the intelligent agent collaboration strategy and the issuance of task execution tokens specifically include:

[0017] The core layer large language model reads fields and extracts task elements from the standardized task description, identifies the route range field, vehicle information field, verification constraint field, timeliness field and report output field, and breaks down the task into route parsing sub-task, bridge list extraction sub-task, computing power estimation and allocation sub-task, load model automatic calculation sub-task and report generation sub-task according to the task processing order.

[0018] Based on the input field requirements and output result requirements of each subtask, establish the task connection relationship to obtain an ordered subtask sequence;

[0019] A collaborative strategy for intelligent agents is formulated based on an ordered sequence of subtasks. The collaborative strategy includes assigning each subtask to a corresponding intelligent agent, setting the calling order of each intelligent agent, determining the data flow path between subtasks, allocating computing resources and priorities for subtasks, and arranging scheduling according to task dependencies and task parallelism.

[0020] According to the agent collaboration strategy, each subtask is written with a task identifier, input parameters, output results, dependencies, execution priority, and resource call items to form a corresponding task execution token, which is then sent to the corresponding agent.

[0021] Optionally, the generation of the bridge parameters and model list specifically includes:

[0022] The GIS spatial analysis and computing agent reads the route range input parameter items and the heavy vehicle parameter input parameter items from the ordered subtask sequence, parses them, and generates a route centerline data set and a vehicle feature parameter set.

[0023] A spatial retrieval area is generated based on the route centerline data, and the spatial retrieval area is a continuous buffer coverage area along the route centerline.

[0024] Search the set of bridge spatial elements within the spatial retrieval area, determine the set of bridge entities along the route based on the spatial intersection relationship between the bridge spatial elements and the route centerline, and write the bridge entity identifier, bridge spatial coordinates and bridge structure type index for each bridge entity.

[0025] Based on the bridge entity identifiers, the sets of bridge entities along the route are deduplicated, and the linear mileage sequence of the bridge spatial coordinates is sorted from smallest to largest according to the route travel direction to generate a list of bridges along the route in sequence.

[0026] According to the list of bridges along the route, the data base is accessed one by one, and the set of bridge structural parameters and the index of high-fidelity finite element model file are extracted for each bridge entity.

[0027] Perform a consistency check based on the vehicle feature parameter set, and output the checked vehicle feature parameter set.

[0028] Based on the bridge structure parameter set, the high-fidelity finite element model file index, and the verified vehicle feature parameter set, a verification bridge parameter item is generated for each bridge entity, and all verification bridge parameter items are summarized and encapsulated to generate a verification bridge parameter and model list.

[0029] Optionally, the generation of the grouped parallel computing schedule specifically includes:

[0030] The engineering computing agent receives the list of bridge parameters and models for verification, reads the bridge structure type index, bridge structure parameter set, high-fidelity finite element model file index, and verified vehicle feature parameter set corresponding to each bridge entity, and generates a set of parallel verification tasks.

[0031] Based on the bridge structure type index, the structural complexity level of the bridge model is determined for each parallel verification task, and a preset complexity weight value is associated with each structural complexity level.

[0032] The computational power requirement for a single bridge is calculated based on the structural complexity level of the bridge model, the finite element model scale factor, and the vehicle load scale factor. The computational power requirement for a single bridge is the product of the complexity weight value, the finite element model scale factor, and the vehicle load scale factor.

[0033] The total computing power requirement is obtained by summing up all single-bridge computing power requirements, and then the total computing power requirement is mapped to the number of compute nodes and the number of virtual machines.

[0034] The set of tasks to be verified in parallel is divided into multiple parallel computing groups according to the number of computing nodes, and a computing group number and node priority index are written for each parallel computing group.

[0035] Call the supercomputing center's interface to submit a computing power application request, and receive the set of computing node identifiers and the set of virtual machine identifiers returned by the supercomputing center;

[0036] The engineering computing agent associates and binds each parallel computing group with the corresponding computing node identifier and virtual machine identifier to generate a group parallel computing scheduling table.

[0037] Optionally, the generation of the single finite element calculation report specifically includes:

[0038] The engineering computing agent reads the parallel computing scheduling table of the groups, obtains the computing node identifier, virtual machine identifier and verification bridge parameter items corresponding to each parallel computing group, and creates a headless mode running instance for the corresponding parallel computing group on each computing node.

[0039] Based on the high-fidelity finite element model file index in the bridge parameter items, locate the storage location of the finite element model file, import the finite element model file into professional finite element software, and generate the corresponding finite element calculation engineering file for the bridge entity.

[0040] A set of non-standard load descriptions for irregularly shaped large vehicles is generated based on the verified set of vehicle characteristic parameters.

[0041] Based on the non-standard load description set of irregular large vehicles, a non-standard load model of the vehicle is generated in professional finite element software, and the non-standard load model of the vehicle is applied to the corresponding positions of the bridge deck element and the main beam element in the finite element calculation engineering file.

[0042] Read the verification constraints and set boundary conditions for the bridge elements in the finite element calculation project file;

[0043] Trigger mechanical solutions in professional finite element software and generate a single finite element calculation report.

[0044] Optionally, the generation of the anomaly annotation information and the rollback of the anomaly task specifically include:

[0045] During parallel computing, the engineering computing agent reads the grouped parallel computing scheduling table to obtain the computing node identifier, virtual machine identifier, verification bridge parameter items and solution status identifier corresponding to each parallel computing group, and establishes a node status monitoring record for each computing node.

[0046] The system collects computing node operating status parameters based on node status monitoring records, and performs node fault determination based on these parameters to obtain a set of faulty nodes.

[0047] Based on the output status of the solution log, the non-convergent task determination of the model is performed, resulting in a set of non-convergent tasks;

[0048] Perform abnormal task truncation on parallel computing groups marked as fault nodes and verification bridge parameter items marked as non-convergent tasks, and generate abnormal labeling information for each truncated abnormal task information.

[0049] Based on the node status monitoring records, a set of healthy node identifiers is identified, and abnormal tasks corresponding to faulty nodes or non-convergent tasks are written into the task rollback queue. According to the parallel computing group number and node priority index, the abnormal tasks are reassigned to the virtual machines corresponding to the set of healthy node identifiers for execution, and the group parallel computing scheduling table is updated.

[0050] Optionally, the output of the calculation results for a single bridge specifically includes:

[0051] The report generating agent receives a set of individual finite element calculation reports generated by parallel computing, reads the anomaly annotation information, and groups the individual finite element calculation reports into single bridge report groups according to bridge entity identifier and load condition identifier;

[0052] Based on the preset prompt word template, the individual finite element calculation reports in the single bridge report group are structured and parsed to extract the bridge entity identifier, vehicle identifier, load condition identifier, solution state identifier, bridge stress extreme value, displacement deflection, deflection distribution and support reaction force, and obtain the set of key indicators for the single bridge.

[0053] Perform unit system consistency processing on the set of key indicators for a single bridge;

[0054] Read the industry standard threshold set and combine it with the processed single bridge key index set to calculate the safety judgment index set, which includes stress utilization coefficient, deflection utilization coefficient and support reaction force utilization coefficient.

[0055] The safety level is determined based on the solution status indicators and the set of safety assessment indicators;

[0056] The report generates an intelligent agent that summarizes, encapsulates, and outputs the calculation results for a single bridge.

[0057] Optionally, the generation of the overall performance evaluation report specifically includes:

[0058] The report-generating agent receives the set of calculation results for a single bridge and the set of anomaly annotation information, and reads the route range field in the standardized task description. It then arranges the set of calculation results for a single bridge in the order of route number, start position, end position, and driving direction to generate a route bridge result sequence.

[0059] The route throughput rate is calculated by statistically analyzing the number of bridges passed and the total number of bridges based on the route bridge result sequence, and then using the ratio of the number of bridges passed to the total number of bridges.

[0060] Risk points are identified based on the route and bridge result sequence, and the risk point set is generated into a risk point sequence according to the order of the route and bridge result sequence;

[0061] A set of risk-specific prompts is generated based on the risk point sequence. The set of risk-specific prompts includes stress prompts for stress utilization coefficients exceeding preset stress warning thresholds, deflection prompts for deflection utilization coefficients exceeding preset deflection warning thresholds, support prompts for support reaction force utilization coefficients exceeding preset reaction force warning thresholds, and abnormal prompts for abnormal markings. A corresponding risk-specific prompt is associated with each risk point.

[0062] A route overview is generated by summarizing the calculation results of individual bridges and the verification constraints.

[0063] Read the approval template from the transportation authority, and map and fill the corresponding layout fields of the approval template with route overview information, route pass rate, risk point sequence, risk-specific prompt set and abnormal annotation information set, and generate the layout result;

[0064] The layout results are appended to the calculation results of individual bridges in the route and bridge result sequence to generate a final calculation report for each bridge, and the final calculation report is provided to users for review and download.

[0065] The beneficial effects of this invention are:

[0066] This invention achieves fully automated bridge verification processes by introducing a multi-agent collaboration mechanism. After the user inputs macro-level natural language commands, the core layer's large language model parses the route range, vehicle information, and verification constraints into standardized task descriptions, breaks them down into ordered sub-task sequences, and generates task execution tokens based on task dependencies and execution order, distributing them to each agent for execution. This achieves automatic task allocation and determination of the invocation order. The GIS spatial analysis and computation agent can automatically retrieve bridge entities along the route based on route centerline data and vehicle characteristic parameters, extracting bridge structural parameters and high-fidelity finite element model file indexes. Simultaneously, it generates a list of bridge parameters and models for verification, ensuring data integrity and structure, and providing closed-loop data support for subsequent parallel computing. In this way, the entire verification process requires no manual intervention, ensuring unified management and efficient processing of information from multiple bridges and various working conditions.

[0067] The engineering computational agent estimates computing power and schedules tasks in parallel based on the structural complexity of the bridge model, the size of the finite element model, and the scale of vehicle loads. It then allocates virtual machines and computing nodes by calling the supercomputing center, generating a parallel computing scheduling table to achieve dynamic allocation and load balancing of computing resources. On each computing node, headless running instances can automatically load the finite element model, apply non-standard vehicle loads, and set boundary conditions, triggering mechanical solutions to generate individual finite element calculation reports, greatly improving the computational efficiency and controllability of large-scale bridge verification. Simultaneously, the system monitors the node running status and solution log output in real time. For faulty nodes or non-convergent tasks, it automatically truncates abnormal tasks, generates abnormal annotation information, and rolls back the abnormal tasks to healthy nodes for re-execution, ensuring the stability and continuity of the computation process, reducing manual intervention, and improving system reliability.

[0068] Based on a single finite element analysis (FEM) report, the report-generating AI can automatically extract key indicators such as bridge stress, displacement deflection, and support reactions, perform unit system consistency processing, and combine industry-standard thresholds to calculate safety assessment indicators and safety levels, generating calculation results for a single bridge. Simultaneously, it statistically analyzes all bridge calculation results, identifies risk points and generates targeted risk alerts, and integrates route overviews and approval templates to complete the overall FEM report output. This technical solution enables intelligent management of the entire process of cross-regional heavy-lift bridge load verification, significantly improving calculation efficiency and data consistency, and providing reliable and traceable safety level and risk alert information for engineering decisions. This significantly enhances the accuracy, automation level, and application value of bridge load verification. Attached Figure Description

[0069] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0070] Figure 1 The flowchart shows the automated load calculation method for heavy transport bridges based on multi-agent collaboration proposed in this invention.

[0071] Figure 2 This is a schematic diagram of the multi-agent collaboration method for automated load calculation of heavy transport bridges based on multi-agent collaboration proposed in this invention.

[0072] Figure 3 This is a flowchart illustrating the abnormal task rollback process of the automated load verification method for large-item transport bridges based on multi-agent collaboration proposed in this invention. Detailed Implementation

[0073] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0074] refer to Figures 1-3 An automated load calculation method for heavy-lift bridges based on multi-agent collaboration includes the following steps:

[0075] It receives macro-level natural language commands input by users through the interactive interface, and the core layer large language model parses the command content to generate a standardized task description;

[0076] The standardized task description is broken down into an ordered sequence of subtasks, and the core layer formulates the agent collaboration strategy and issues task execution tokens.

[0077] The GIS spatial analysis and computational agent parses the route and verifies vehicle characteristic parameters based on an ordered sub-task sequence, and extracts bridge structural parameters and high-fidelity finite element model file indexes to generate a list of verified bridge parameters and models.

[0078] The engineering computing agent receives and verifies the bridge parameters and model list, estimates the computing power based on the structural complexity of the bridge model, calls the supercomputing center to allocate virtual machines or computing nodes, and generates a grouped parallel computing scheduling table.

[0079] At each assigned computing node, the engineering computing agent loads the finite element model, applies non-standard vehicle loads, sets boundary conditions, and triggers mechanical solution according to the grouped parallel computing schedule, generating a single finite element calculation report.

[0080] During parallel computing, the status of computing nodes is monitored in real time. If a faulty node or non-convergent task is detected, the engineering computing agent automatically truncates the abnormal task, generates abnormal labeling information, and rolls back the abnormal task to a healthy node for re-execution.

[0081] The report-generating agent extracts key indicators from individual finite element calculation reports, determines the safety level of each bridge, and outputs the calculation results for each bridge.

[0082] The report-generating agent statistically analyzes the calculation results and anomaly annotation information of a single bridge to generate a comprehensive calculation report, which is then provided to users for review and download.

[0083] In this embodiment, the natural language macro-instructions include route scope, vehicle information, and verification constraints. The route scope includes road name, route number, starting position, ending position, station interval, driving direction, and administrative division information. The vehicle information includes vehicle type, tractor and trailer information, total vehicle weight, number of axles, axle load, wheelbase, track width, total vehicle length, total vehicle width, total vehicle height, and load distribution information. The verification constraints include travel time requirements, speed limits, off-center load limits, lane constraints, meeting constraints, bridge safety level requirements, load combination requirements, boundary condition requirements, and report output requirements. The core layer large language model is an AI-based language understanding unit, which contains a multi-layer semantic understanding network and a context association mechanism. It receives natural language macro-instructions input by the user through the interactive interface, segments and parses the instruction text into independent semantic units, and standardizes and structures the semantic units according to preset semantic slots to ensure that the parsing results are complete and consistent in concept and structure.

[0084] In this embodiment, the formulation of the agent cooperation strategy and the issuance of task execution tokens specifically include:

[0085] The core layer large language model reads fields and extracts task elements from the standardized task description, identifies the route range field, vehicle information field, verification constraint field, timeliness field and report output field, and breaks down the task into route parsing sub-task, bridge list extraction sub-task, computing power estimation and allocation sub-task, load model automatic calculation sub-task and report generation sub-task according to the task processing order.

[0086] Based on the input field requirements and output result requirements of each subtask, a task connection relationship is established. The output result of the route analysis subtask is used as the input of the bridge inventory extraction subtask, the output result of the bridge inventory extraction subtask is used as the input of the computing power estimation and allocation subtask and the load model automatic calculation subtask, and the output result of the load model automatic calculation subtask is used as the input of the report generation subtask, resulting in an ordered subtask sequence. The output result of the route analysis subtask includes the spatial coordinates of bridges along the route, bridge entity identifiers, and bridge structure type indexes; the output result of the bridge inventory extraction subtask includes the bridge finite element model file index, span parameters, structural component parameters, and bridge load-bearing capacity indicators; the output result of the computing power estimation and allocation subtask includes the number of virtual machines required for each computing node, the computing group number, and the node priority index; the output result of the load model automatic calculation subtask includes the non-standard load model file for large vehicles, boundary condition setting records, and a set of mechanical solution parameters.

[0087] A collaborative strategy for agents is formulated based on an ordered sequence of subtasks. This strategy includes assigning subtasks to corresponding agents, setting the calling order of agents, determining data flow paths between subtasks, allocating computational resources and priorities for subtasks, and scheduling based on task dependencies and parallelism. The formulation of this collaborative strategy is executed by the core layer's large language model. First, it identifies the input-output dependencies of each subtask based on the ordered sequence. Then, combining the functional capabilities of each agent and the currently available computational resources, it determines the allocation scheme for each subtask. The route parsing subtask and bridge inventory extraction subtask are assigned to the GIS spatial analysis computation agent; the computational power estimation allocation subtask and load model automatic calculation subtask are assigned to the engineering computation agent; and the report generation subtask is assigned to the report generation agent. After completing the subtask allocation, the core layer's large language model sets the calling order and data flow paths for agents, and adjusts the order based on task parallelism, node load, computational node priority, and task dependency constraints to ensure efficient collaboration between agents and that the execution order of each subtask is consistent with data dependencies.

[0088] According to the agent collaboration strategy, each subtask is written with a task identifier, input parameters, output results, dependencies, execution priority, and resource call items to form a corresponding task execution token, which is then sent to the corresponding agent.

[0089] In this embodiment, the verification of bridge parameters and the generation of the model list specifically include:

[0090] The GIS spatial analysis and computing agent reads and parses the route range input parameters and heavy vehicle parameter input parameters from the ordered subtask sequence, generating a route centerline data set and a vehicle characteristic parameter set. The route centerline data set is obtained based on the route range field, which includes road name, route number, start position, end position, station interval, driving direction, and administrative division information. The vehicle characteristic parameter set includes vehicle gross weight parameters, vehicle axle combination parameters, vehicle axle load parameters, vehicle wheelbase parameters, track width parameters, vehicle external dimensions parameters, and off-center load constraint parameters.

[0091] A spatial retrieval area is generated based on the route centerline data. The spatial retrieval area is a continuous buffer coverage area along the route centerline. The width of the continuous buffer coverage area is determined based on the vehicle outline size parameters and the road lateral safety distance parameters. The vehicle outline size parameters include the vehicle's total length, total width, and total height parameters. The road lateral safety distance parameter is a preset minimum lateral safety distance.

[0092] Search the set of bridge spatial elements within the spatial retrieval area, determine the set of bridge entities along the route based on the spatial intersection relationship between the bridge spatial elements and the route centerline, and write the bridge entity identifier, bridge spatial coordinates and bridge structure type index for each bridge entity.

[0093] Based on the bridge entity identifiers, the sets of bridge entities along the route are deduplicated, and the linear mileage sequence of the bridge spatial coordinates is sorted from smallest to largest according to the route driving direction to generate a list of bridges along the route in sequence. The linear mileage sequence is obtained by projecting the bridge spatial coordinates of each bridge entity onto the route centerline to obtain the projection point, and the cumulative distance from the starting position of the route centerline to the projection point is calculated.

[0094] According to the order of the bridges along the route, the data base is accessed one by one. For each bridge entity, the set of bridge structural parameters and the index of high-fidelity finite element model file are extracted. The set of bridge structural parameters includes bridge name, bridge span parameters, bridge total length parameters, structural system type parameters, main beam section parameters, pier and abutment parameters, support parameters, material parameters, and design load level parameters. The index of high-fidelity finite element model file is used to identify the corresponding finite element model file. The index includes the storage location of the finite element model file, model version identifier, and model unit system identifier.

[0095] Perform consistency checks based on the vehicle feature parameter set and output the checked vehicle feature parameter set. The consistency check includes calculating the sum of the axle loads of the vehicle and comparing it with the total vehicle weight parameter, calculating the sum of the wheelbases of the vehicle and comparing it with the total vehicle length parameter in the vehicle outline dimension parameters, checking whether the vehicle outline dimension parameters meet the spatial retrieval area constraint requirements, verifying whether the off-center load constraint parameters are within the allowable range, and confirming that the vehicle axle number combination parameters and wheelbase parameters match the vehicle total weight parameters and vehicle axle load parameters.

[0096] Based on the bridge structure parameter set, the high-fidelity finite element model file index, and the verified vehicle feature parameter set, a verification bridge parameter item is generated for each bridge entity. All verification bridge parameter items are summarized and encapsulated to generate a verification bridge parameter and model list. The verification bridge parameter items include bridge entity identifier, bridge spatial coordinates, bridge structure type index, bridge structure parameter set, high-fidelity finite element model file index, and verified vehicle feature parameter set.

[0097] In this embodiment, the generation of the grouped parallel computing scheduling table specifically includes:

[0098] The engineering computing agent receives the list of bridge parameters and models for verification, reads the bridge structure type index, bridge structure parameter set, high-fidelity finite element model file index, and verified vehicle feature parameter set corresponding to each bridge entity, and generates a set of parallel verification tasks.

[0099] Based on the bridge structure type index, the structural complexity level of the bridge model is determined for each parallel verification task, and a preset complexity weight value is associated with each structural complexity level. The structural complexity levels of the bridge model include simply supported beam level, continuous beam level and cable-stayed bridge level.

[0100] The computational power requirement for a single bridge is calculated based on the structural complexity level of the bridge model, the finite element model scale factor, and the vehicle load scale factor. The computational power requirement for a single bridge is the product of the complexity weight value, the finite element model scale factor, and the vehicle load scale factor. The finite element model scale factor is determined by the number of nodes and elements in the finite element model, and the vehicle load scale factor is determined by the number of vehicle axles and the number of non-standard load conditions.

[0101] The total computing power requirement is obtained by summing up all single-bridge computing power requirements and mapping the total computing power requirement to the number of compute nodes and virtual machines. The number of compute nodes and virtual machines are determined based on the single-node available computing power threshold and the preset parallelism limit.

[0102] The set of tasks to be verified in parallel is divided into multiple parallel computing groups according to the number of computing nodes. The parallel computing groups are determined according to the principle that the sum of the single-bridge computing power requirements in each group is similar, and a computing group number and node priority index are written for each parallel computing group.

[0103] Call the supercomputing center interface to submit a computing power application request. The computing power application request includes the number of computing nodes, the number of virtual machines, the number of parallel computing groups, the computing group number and node priority index of each parallel computing group, and receive the set of computing node identifiers and the set of virtual machine identifiers returned by the supercomputing center.

[0104] The engineering computing agent associates and binds each parallel computing group with the corresponding computing node identifier and virtual machine identifier to generate a group parallel computing scheduling table.

[0105] In this embodiment, the generation of a single finite element calculation report specifically includes:

[0106] The engineering computing agent reads the parallel computing scheduling table of the groups, obtains the computing node identifier, virtual machine identifier and verification bridge parameter items corresponding to each parallel computing group, and creates a headless mode running instance for the corresponding parallel computing group on each computing node. The headless mode running instance loads and calls the professional finite element software through the preset headless startup instruction set. The headless startup instruction set includes software startup parameter items, engineering file input path items, calculation result output path items and log output path items.

[0107] Based on the high-fidelity finite element model file index in the bridge parameter items, locate the storage location of the finite element model file, import the finite element model file into professional finite element software, and generate the corresponding finite element calculation engineering file for the bridge entity.

[0108] Based on the verified set of vehicle characteristic parameters, a set of non-standard load descriptions for irregularly shaped large vehicles is generated. This set includes load location descriptions, load amplitude descriptions, and load combination descriptions corresponding to vehicle axle number combination parameters, vehicle axle load parameters, vehicle wheelbase parameters, track width parameters, vehicle external dimensions parameters, and off-center load constraint parameters. The load combination description includes load condition identifiers, combination numbers, a set of combination coefficients, and a combination sequence identifier. The set of combination coefficients is used to characterize the coefficient values ​​of each axle load amplitude corresponding to the vehicle axle load parameters under different load conditions. The load combination requirements include the range of allowable load conditions, the range of combination coefficient values, and the upper limit of the number of combinations.

[0109] Based on the non-standard load description set of irregular large vehicles, a non-standard load model of the vehicle is generated in professional finite element software. The non-standard load model of the vehicle is then applied to the corresponding positions of the bridge deck elements and main beam elements in the finite element calculation engineering file. The application position of the non-standard load model of the vehicle is determined by the bridge deck coordinate reference, which includes the bridge deck longitudinal mileage reference along the route driving direction and the bridge deck transverse reference perpendicular to the bridge centerline. The engineering calculation agent maps the action position of each axle to the bridge deck longitudinal coordinate sequence based on the vehicle wheelbase parameters. Based on the wheelbase parameters and the lane constraints and off-center load limits in the verification constraints, the transverse offset range of the bridge deck is determined and the transverse coordinate sequence of the bridge deck is generated, thereby determining the set of applied coordinates of the non-standard load model of the vehicle on the bridge deck elements and main beam elements.

[0110] The process involves reading the verification constraints and setting boundary conditions for the bridge elements in the finite element calculation file. Setting boundary conditions includes determining the support constraint type and constraint degrees of freedom based on the bridge structural parameter set; determining the lateral load offset range based on the off-center load limit and lane constraint in the verification constraints; and determining the quasi-static solution step size parameters and the number of solution cases based on the speed limit and traffic time requirements in the verification constraints. Support constraint types include fixed constraints, unidirectional sliding constraints, bidirectional sliding constraints, and elastic constraints. Constraint degrees of freedom include the longitudinal translational degrees of freedom, lateral translational degrees of freedom, vertical translational degrees of freedom, and rotational degrees of freedom about the longitudinal direction, lateral direction, and vertical direction of the bridge element. The quasi-static solution step size parameters are determined based on the speed limit and traffic time requirements in the verification constraints. The number of solution cases is the product of the number of segments in the time segment sequence and the number of combination numbers in the load combination description.

[0111] In professional finite element software, a mechanical solution is triggered. The mechanical solution includes performing internal force calculation, stress calculation, and displacement calculation on the finite element calculation engineering file after applying the non-standard vehicle load model, and generating a single finite element calculation report. The single finite element calculation report includes the bridge entity identifier, the vehicle identifier carried by the bridge parameter item for verification, the load condition identifier, the extreme value of bridge stress, displacement deflection, deflection distribution, support reaction force, and solution status identifier.

[0112] In this embodiment, the generation of anomaly annotation information and the rollback of anomaly tasks specifically include:

[0113] During parallel computing, the engineering computing agent reads the grouped parallel computing scheduling table to obtain the computing node identifier, virtual machine identifier, verification bridge parameter items and solution status identifier corresponding to each parallel computing group, and establishes a node status monitoring record for each computing node.

[0114] The system collects computing node operating status parameters based on node status monitoring records, and performs node fault determination based on these parameters to obtain a set of faulty nodes. The computing node operating status parameters include computing node connectivity, virtual machine operating status, CPU utilization, GPU utilization, memory utilization, disk read / write status, network transmission status, professional finite element software process status, and solution log output status. The node fault determination includes detecting abnormal computing node connectivity, abnormal virtual machine operating status, abnormal exit of professional finite element software processes, interruption of solution log output, and abnormal duration of computing node operating status parameters exceeding a preset fault threshold exceeding a preset duration threshold. Computing nodes that meet the node fault determination conditions are marked as faulty nodes.

[0115] Based on the output status of the solution log, the model non-convergence task determination is performed to obtain a set of non-convergence tasks. The model non-convergence task determination includes detecting that the number of solution iterations exceeds a preset iteration number threshold, the residual change exceeds a preset residual change threshold, the number of solution step size reductions exceeds a preset reduction threshold, and the solution status is marked as non-convergence. The verification bridge parameter items that meet the model non-convergence task determination conditions are marked as non-convergence tasks.

[0116] Anomaly task truncation is performed on parallel computing groups marked as fault nodes and bridge parameter items marked as non-convergent tasks. Anomaly annotation information is generated for each truncated anomaly task. Anomaly task truncation for parallel computing groups includes stopping the professional finite element software process on the fault node and releasing the corresponding virtual machine resources. Anomaly task truncation for bridge parameter items includes stopping the iterative calculation of the corresponding solution condition and saving the solution log fragment at the truncation time. The anomaly annotation information includes anomaly type identifier, anomaly occurrence time identifier, computing node identifier, virtual machine identifier, parallel computing group number, bridge entity identifier, load condition identifier, solution status identifier, and anomaly log index.

[0117] Based on the node status monitoring records, a set of healthy node identifiers is identified, and abnormal tasks corresponding to faulty nodes or non-convergent tasks are written into the task rollback queue. According to the parallel computing group number and node priority index, the abnormal tasks are reassigned to the virtual machines corresponding to the set of healthy node identifiers for execution, and the group parallel computing scheduling table is updated.

[0118] In this embodiment, the output of the calculation results for a single bridge specifically includes:

[0119] The report generating agent receives a set of individual finite element calculation reports generated by parallel computing, reads the anomaly annotation information, and groups the individual finite element calculation reports into single bridge report groups according to bridge entity identifier and load condition identifier;

[0120] Based on the preset prompt word template, the individual finite element calculation reports in the single bridge report group are structured and parsed to extract the bridge entity identifier, vehicle identifier, load condition identifier, solution state identifier, bridge stress extreme value, displacement deflection, deflection distribution and support reaction force, and obtain the set of key indicators for the single bridge.

[0121] The key indicators of a single bridge are processed to ensure consistency of the unit system. Based on the unit system identifier in the individual finite element calculation report, the extreme values ​​of bridge stress, displacement deflection and support reaction are converted into a preset unified unit system. The deflection distribution is then generated into a deflection distribution sequence according to the preset mileage sampling interval.

[0122] The set of industry-standard thresholds is read and combined with the processed set of key indicators for a single bridge to calculate the set of safety judgment indicators. The set of industry-standard thresholds includes the design allowable stress threshold, allowable deflection threshold, allowable support reaction threshold, and the validity conditions of the solution state. The set of safety judgment indicators includes the stress utilization coefficient, deflection utilization coefficient, and support reaction utilization coefficient. The stress utilization coefficient is the ratio of the extreme stress of the bridge to the design allowable stress threshold, the deflection utilization coefficient is the ratio of the displacement deflection to the allowable deflection threshold, and the support reaction utilization coefficient is the ratio of the support reaction force to the allowable support reaction force threshold.

[0123] The safety level is determined based on the solution status identifier and the set of safety judgment indicators. The safety level determination includes comparing the maximum value of the stress utilization coefficient, deflection utilization coefficient and support reaction force utilization coefficient with the preset classification threshold set when the solution status identifier meets the solution status validity condition to obtain the safety level identifier, and adding an abnormal labeling label to the safety level identifier when abnormal labeling information exists.

[0124] The report-generating agent summarizes and encapsulates the calculation results of a single bridge, which include bridge entity identifiers, vehicle identifiers, load condition identifiers, key indicators for a single bridge, safety judgment indicators, safety level identifiers, and anomaly labeling identifiers.

[0125] In this embodiment, the generation of the overall performance evaluation report specifically includes:

[0126] The report-generating agent receives the set of calculation results for a single bridge and the set of anomaly annotation information, and reads the route range field in the standardized task description. It then arranges the set of calculation results for a single bridge in the order of route number, start position, end position, and driving direction to generate a route bridge result sequence.

[0127] The number of bridges passed and the total number of bridges are statistically analyzed based on the route bridge result sequence. The route pass rate is calculated by the ratio of the number of bridges passed to the total number of bridges. The number of bridges passed is the number of bridges whose safety level meets the preset pass conditions and which have not been marked with an anomaly label.

[0128] Risk points are identified based on the route and bridge result sequence, and the risk point set is generated into a risk point sequence according to the order of the route and bridge result sequence. The risk points include: bridge entities whose safety level indicators do not meet the preset passage conditions, bridge entities with additional abnormal labeling, and the load condition label set corresponding to each bridge entity.

[0129] A set of risk-specific warnings is generated based on the risk point sequence. The set of risk-specific warnings includes stress warnings for stress utilization coefficients exceeding preset stress warning thresholds, deflection warnings for deflection utilization coefficients exceeding preset deflection warning thresholds, support warnings for support reaction force utilization coefficients exceeding preset reaction force warning thresholds, and anomaly warnings for anomaly markings. Each risk point is associated with a corresponding risk-specific warning.

[0130] The route overview information is generated based on the calculation results of individual bridges and the summary of verification constraints. The route overview information includes the route number, route start and end information, total route mileage information, total number of bridges information, heavy vehicle identification information, and summary information of verification constraints.

[0131] Read the approval template from the transportation authority and map the route overview information, route pass rate, risk point sequence, risk-specific prompt set and abnormal annotation information set to the corresponding layout fields of the approval template to generate the layout result. The approval template includes the layout field set and layout rule set of the cover information area, route overview area, bridge calculation summary area, risk point details area, risk-specific prompt area, abnormal annotation area and appendix area.

[0132] The layout results are appended to the calculation results of individual bridges in the route and bridge result sequence to generate a total verification report for each bridge. The total verification report is then provided to users for review and download. The total verification report includes route overview information, summary statistics of bridge verification, route passability, risk point sequence, set of risk-specific prompts, set of anomaly annotation information, and detailed calculation results of individual bridges in the appendix area.

[0133] Example 1: To verify the feasibility of this invention in practice, it was applied to a scenario involving the passage calculation of heavy-duty transport bridges on a provincial highway. This scenario involves multiple national and provincial highways and mountainous bridge groups. The bridge structures are complex, including simply supported beams, continuous beams, and cable-stayed bridges. There are a large number of bridges, and the vehicles are multi-axle heavy transport vehicles with complex axle spacing combinations. Furthermore, there are restrictions on off-center loads and time constraints. Previously, in similar transport tasks, it was necessary to manually retrieve the finite element model for each bridge, manually apply loads, and perform calculations one by one. This was not only time-consuming, but also required manual reallocation of computing resources if a calculation node malfunctioned or the solution failed to converge during parallel computing, leading to a prolonged overall calculation cycle and impacting the efficiency of transport plan approval.

[0134] In this embodiment, the user inputs the route range, vehicle information, and verification constraints through an interactive interface. The macro-level natural language instructions are parsed by the core layer's large language model to generate standardized task descriptions, which are then broken down into an ordered sequence of subtasks. The GIS spatial analysis and computational agent automatically identifies bridge entities along the route, extracts bridge structural parameters and high-fidelity finite element model file indexes, and generates a list of bridge parameters and models for verification. The engineering computational agent performs computational power estimation based on the structural complexity of the bridge model, calls the supercomputing center to allocate computing nodes and virtual machines, and generates a grouped parallel computing schedule. Finite element models are automatically loaded onto each allocated computing node, generating non-standard vehicle load models and applying them to the bridge deck and main beam elements. Support constraints and solution parameters are set based on the verification constraints, triggering mechanical solutions to generate a single finite element calculation report. During parallel computation, the running status of computing nodes and the output status of solution logs are monitored in real time. When a faulty node or non-convergent task is detected, abnormal task truncation is automatically executed, generating abnormal label information. The abnormal task is written to the task rollback queue and reassigned to the virtual machine corresponding to the healthy node identifier set for execution, ensuring computational continuity. The report generation agent performs structured parsing on individual finite element calculation reports, calculates the set of safety judgment indicators and generates calculation results for a single bridge, further calculates route passability, identifies risk points and generates targeted risk warnings, and finally outputs a comprehensive calculation report for users to review and download.

[0135] In this application scenario, the present invention significantly shortens the bridge load verification cycle. Compared with the previous manual scheduling method, the overall calculation time is significantly reduced, the number of manual interventions is significantly decreased, and calculation interruptions caused by abnormal nodes no longer cause the overall process to stall, effectively improving the utilization rate of computing resources. The speed of generating verification results meets the timeliness requirements of transportation approval, and risk identification and prompts are more centralized and accurate, ensuring the continuity and stability of the bridge passage verification for heavy-duty transportation. This verifies the feasibility and engineering application value of the present invention in the parallel verification scenario of complex bridge groups.

[0136] Table 1. Performance Comparison of the Invention and Traditional Load Calculation Methods for Heavy-Duty Transport Bridges

[0137] Total verification time (min) 180 ≤30 Number of manual interventions (count) 36 16 Delay for handling abnormal tasks (min) 90 42 Calculate the average utilization rate of nodes (%) 55 99 Route passability recognition time (s) 600 310 Accuracy rate of risk point identification (%) 85 94 Total experience report generation time (min) 120 58

[0138] The results shown in Table 1 are based on performance tests conducted on 50 bridges, where the method of this invention uses 10 computing nodes to perform the verification in parallel.

[0139] Regarding the total verification time, traditional methods require approximately 180 minutes to complete the verification of 50 bridges under single-node or weakly parallel conditions. In contrast, the method of this invention uses a grouped parallel computing scheduling table to allocate the verification task across 10 computing nodes, controlling the total verification time to within 30 minutes, significantly shortening the overall verification cycle. This effect stems from the engineering computing agent performing computational power estimation based on the structural complexity of the bridge model, and combining this with a supercomputing center to dynamically allocate computing nodes, thereby achieving parallel solution of multiple bridge tasks.

[0140] Regarding the number of manual interventions, the traditional method requires 36 counts, while the method of this invention reduces this to 16 counts; the latency for handling abnormal tasks is reduced from 90 minutes to 42 minutes. These data demonstrate that the method of this invention has a stronger automatic handling capability for abnormal tasks. This is because, during parallel computing, the system continuously monitors the status of computing nodes, virtual machines, and solution log output. Upon detecting a faulty node or a non-convergent task, it can automatically perform abnormal task truncation, abnormal labeling information generation, and task rollback and reallocation, thus reducing the frequency of manual intervention and significantly shortening the anomaly recovery time.

[0141] In terms of resource utilization and result output efficiency, the average utilization rate of computing nodes increased from 55% to 99%, the route passability identification time was shortened from 600s to 310s, and the overall verification report generation time was shortened from 120min to 58min. These results demonstrate that the method of this invention not only significantly improves the average utilization level of computing nodes but also enhances the efficiency of result statistics and report output. This effect is achieved because the method of this invention estimates computing power based on the structural complexity of the bridge model, the finite element model scale factor, and the vehicle load scale factor. It also divides parallel computing groups according to the principle that the sum of the computing power requirements of a single bridge within each group is similar, enabling 10 computing nodes to continuously undertake the corresponding verification tasks. Simultaneously, the report generation agent can directly extract key indicators such as bridge stress extremes, displacement deflection, and support reactions from individual finite element calculation reports and automatically complete safety level determination, risk point identification, and template-based layout.

[0142] In terms of result quality, the accuracy of risk point location improved from 85% to 94%, an increase of 9 percentage points. This result indicates that the method of the present invention is more accurate in identifying bridge traffic risks. The reason for this is that the method of the present invention does not rely solely on a single calculation result for judgment, but rather combines solution status indicators, anomaly annotation information, stress utilization coefficient, deflection utilization coefficient, and support reaction force utilization coefficient for joint analysis. Therefore, it can more accurately locate high-risk bridges and abnormal working conditions.

[0143] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An automated load calculation method for heavy-lift bridges based on multi-agent collaboration, characterized in that, Includes the following steps: It receives macro-level natural language commands input by users through the interactive interface, and the core layer large language model parses the command content to generate a standardized task description; The standardized task description is broken down into an ordered sequence of subtasks, and the core layer formulates the agent collaboration strategy and issues task execution tokens. The GIS spatial analysis and computational agent parses the route and verifies vehicle characteristic parameters based on an ordered sub-task sequence, and extracts bridge structural parameters and high-fidelity finite element model file indexes to generate a list of verified bridge parameters and models. The engineering computing agent receives and verifies the bridge parameters and model list, estimates the computing power based on the structural complexity of the bridge model, calls the supercomputing center to allocate virtual machines or computing nodes, and generates a grouped parallel computing scheduling table. At each assigned computing node, the engineering computing agent loads the finite element model, applies non-standard vehicle loads, sets boundary conditions, and triggers mechanical solution according to the grouped parallel computing schedule, generating a single finite element calculation report. During parallel computing, the status of computing nodes is monitored in real time. If a faulty node or non-convergent task is detected, the engineering computing agent automatically truncates the abnormal task, generates abnormal labeling information, and rolls back the abnormal task to a healthy node for re-execution. The report-generating agent extracts key indicators from individual finite element calculation reports, determines the safety level of each bridge, and outputs the calculation results for each bridge. The report-generating agent statistically analyzes the calculation results and anomaly annotation information of a single bridge to generate a comprehensive calculation report, which is then provided to users for review and download.

2. The automated load calculation method for heavy-lift bridges based on multi-agent collaboration as described in claim 1, characterized in that, The natural language macro-instructions include route range, vehicle information, and verification constraints.

3. The automated load calculation method for heavy-lift bridges based on multi-agent collaboration as described in claim 1, characterized in that, The formulation of the agent collaboration strategy and the issuance of task execution tokens specifically include: The core layer large language model reads fields and extracts task elements from the standardized task description, identifies the route range field, vehicle information field, verification constraint field, timeliness field and report output field, and breaks down the task into route parsing sub-task, bridge list extraction sub-task, computing power estimation and allocation sub-task, load model automatic calculation sub-task and report generation sub-task according to the task processing order. Based on the input field requirements and output result requirements of each subtask, establish the task connection relationship to obtain an ordered subtask sequence; A collaborative strategy for intelligent agents is formulated based on an ordered sequence of subtasks. The collaborative strategy includes assigning each subtask to a corresponding intelligent agent, setting the calling order of each intelligent agent, determining the data flow path between subtasks, allocating computing resources and priorities for subtasks, and arranging scheduling according to task dependencies and task parallelism. According to the agent collaboration strategy, each subtask is written with a task identifier, input parameters, output results, dependencies, execution priority, and resource call items to form a corresponding task execution token, which is then sent to the corresponding agent.

4. The automated load calculation method for heavy-lift bridges based on multi-agent collaboration as described in claim 1, characterized in that, The generation of the bridge parameters and model list specifically includes: The GIS spatial analysis and computing agent reads the route range input parameter items and the heavy vehicle parameter input parameter items from the ordered subtask sequence, parses them, and generates a route centerline data set and a vehicle feature parameter set. A spatial retrieval area is generated based on the route centerline data, and the spatial retrieval area is a continuous buffer coverage area along the route centerline. Search the set of bridge spatial elements within the spatial retrieval area, determine the set of bridge entities along the route based on the spatial intersection relationship between the bridge spatial elements and the route centerline, and write the bridge entity identifier, bridge spatial coordinates and bridge structure type index for each bridge entity. Based on the bridge entity identifiers, the sets of bridge entities along the route are deduplicated, and the linear mileage sequence of the bridge spatial coordinates is sorted from smallest to largest according to the route travel direction to generate a list of bridges along the route in sequence. According to the list of bridges along the route, the data base is accessed one by one, and the set of bridge structural parameters and the index of high-fidelity finite element model file are extracted for each bridge entity. Perform a consistency check based on the vehicle feature parameter set, and output the checked vehicle feature parameter set. Based on the bridge structure parameter set, the high-fidelity finite element model file index, and the verified vehicle feature parameter set, a verification bridge parameter item is generated for each bridge entity, and all verification bridge parameter items are summarized and encapsulated to generate a verification bridge parameter and model list.

5. The automated load calculation method for heavy-lift bridges based on multi-agent collaboration as described in claim 1, characterized in that, The generation of the grouped parallel computing scheduling table specifically includes: The engineering computing agent receives the list of bridge parameters and models for verification, reads the bridge structure type index, bridge structure parameter set, high-fidelity finite element model file index, and verified vehicle feature parameter set corresponding to each bridge entity, and generates a set of parallel verification tasks. Based on the bridge structure type index, the structural complexity level of the bridge model is determined for each parallel verification task, and a preset complexity weight value is associated with each structural complexity level. The computational power requirement for a single bridge is calculated based on the structural complexity level of the bridge model, the finite element model scale factor, and the vehicle load scale factor. The computational power requirement for a single bridge is the product of the complexity weight value, the finite element model scale factor, and the vehicle load scale factor. The total computing power requirement is obtained by summing up all single-bridge computing power requirements, and then the total computing power requirement is mapped to the number of compute nodes and the number of virtual machines. The set of tasks to be verified in parallel is divided into multiple parallel computing groups according to the number of computing nodes, and a computing group number and node priority index are written for each parallel computing group. Call the supercomputing center's interface to submit a computing power application request, and receive the set of computing node identifiers and the set of virtual machine identifiers returned by the supercomputing center; The engineering computing agent associates and binds each parallel computing group with the corresponding computing node identifier and virtual machine identifier to generate a group parallel computing scheduling table.

6. The automated load calculation method for heavy-lift bridges based on multi-agent collaboration as described in claim 1, characterized in that, The generation of the single finite element calculation report specifically includes: The engineering computing agent reads the parallel computing scheduling table of the groups, obtains the computing node identifier, virtual machine identifier and verification bridge parameter items corresponding to each parallel computing group, and creates a headless mode running instance for the corresponding parallel computing group on each computing node. Based on the high-fidelity finite element model file index in the bridge parameter items, locate the storage location of the finite element model file, import the finite element model file into professional finite element software, and generate the corresponding finite element calculation engineering file for the bridge entity. A set of non-standard load descriptions for irregularly shaped large vehicles is generated based on the verified set of vehicle characteristic parameters. Based on the non-standard load description set of irregular large vehicles, a non-standard load model of the vehicle is generated in professional finite element software, and the non-standard load model of the vehicle is applied to the corresponding positions of the bridge deck element and the main beam element in the finite element calculation engineering file. Read the verification constraints and set boundary conditions for the bridge elements in the finite element calculation project file; Trigger mechanical solutions in professional finite element software and generate a single finite element calculation report.

7. The automated load calculation method for heavy-lift bridges based on multi-agent collaboration as described in claim 1, characterized in that, The generation of anomaly annotation information and the rollback of anomaly tasks specifically include: During parallel computing, the engineering computing agent reads the grouped parallel computing scheduling table to obtain the computing node identifier, virtual machine identifier, verification bridge parameter items and solution status identifier corresponding to each parallel computing group, and establishes a node status monitoring record for each computing node. The system collects computing node operating status parameters based on node status monitoring records, and performs node fault determination based on these parameters to obtain a set of faulty nodes. Based on the output status of the solution log, the non-convergent task determination of the model is performed, resulting in a set of non-convergent tasks; Perform abnormal task truncation on parallel computing groups marked as fault nodes and verification bridge parameter items marked as non-convergent tasks, and generate abnormal labeling information for each truncated abnormal task information. Based on the node status monitoring records, a set of healthy node identifiers is identified, and abnormal tasks corresponding to faulty nodes or non-convergent tasks are written into the task rollback queue. According to the parallel computing group number and node priority index, the abnormal tasks are reassigned to the virtual machines corresponding to the set of healthy node identifiers for execution, and the group parallel computing scheduling table is updated.

8. The automated load calculation method for heavy-lift bridges based on multi-agent collaboration as described in claim 1, characterized in that, The specific output of the calculation results for a single bridge includes: The report generating agent receives a set of individual finite element calculation reports generated by parallel computing, reads the anomaly annotation information, and groups the individual finite element calculation reports into single bridge report groups according to bridge entity identifier and load condition identifier; Based on the preset prompt word template, the individual finite element calculation reports in the single bridge report group are structured and parsed to extract the bridge entity identifier, vehicle identifier, load condition identifier, solution state identifier, bridge stress extreme value, displacement deflection, deflection distribution and support reaction force, and obtain the set of key indicators for the single bridge. Perform unit system consistency processing on the set of key indicators for a single bridge; Read the industry standard threshold set and combine it with the processed single bridge key index set to calculate the safety judgment index set, which includes stress utilization coefficient, deflection utilization coefficient and support reaction force utilization coefficient. The safety level is determined based on the solution status indicators and the set of safety assessment indicators; The report generates an intelligent agent that summarizes, encapsulates, and outputs the calculation results for a single bridge.

9. The automated load calculation method for heavy-lift bridges based on multi-agent collaboration as described in claim 1, characterized in that, The generation of the overall performance evaluation report specifically includes: The report-generating agent receives the set of calculation results for a single bridge and the set of anomaly annotation information, and reads the route range field in the standardized task description. It then arranges the set of calculation results for a single bridge in the order of route number, start position, end position, and driving direction to generate a route bridge result sequence. The route throughput rate is calculated by statistically analyzing the number of bridges passed and the total number of bridges based on the route bridge result sequence, and then using the ratio of the number of bridges passed to the total number of bridges. Risk points are identified based on the route and bridge result sequence, and the risk point set is generated into a risk point sequence according to the order of the route and bridge result sequence; A set of risk-specific prompts is generated based on the risk point sequence. The set of risk-specific prompts includes stress prompts for stress utilization coefficients exceeding preset stress warning thresholds, deflection prompts for deflection utilization coefficients exceeding preset deflection warning thresholds, support prompts for support reaction force utilization coefficients exceeding preset reaction force warning thresholds, and abnormal prompts for abnormal markings. A corresponding risk-specific prompt is associated with each risk point. A route overview is generated by summarizing the calculation results of individual bridges and the verification constraints. Read the approval template from the transportation authority, and map and fill the corresponding layout fields of the approval template with route overview information, route pass rate, risk point sequence, risk-specific prompt set and abnormal annotation information set, and generate the layout result; The layout results are appended to the calculation results of individual bridges in the route and bridge result sequence to generate a final calculation report for each bridge, and the final calculation report is provided to users for review and download.

Images