An integrated wheel set maintenance intelligent workshop-based whole-process AI agent management and control system

Abstract

The application is suitable for the field of rail transit technology, and provides a whole-process AI intelligent agent management and control system based on an integrated wheel set maintenance intelligent workshop, which comprises a multi-source fusion intelligent agent module, a role perception decision intelligent agent module, a device conflict arbitration intelligent agent module and a closed-loop optimization intelligent agent module; the multi-source fusion intelligent agent module is used for collecting the state and process data of equipment, realizing cross-modal alignment through an AI model to generate a feature vector, and updating a distributed knowledge graph based on a time series graph convolution network; the role perception decision intelligent agent module is used for analyzing user permission labels and instructions, and generating a scheduling scheme in combination with equipment load; the device conflict arbitration intelligent agent module is used for monitoring the spatial topology of AGV and mechanical hands; and the closed-loop optimization intelligent agent module is used for analyzing wheel set press-fitting torque data, extracting historical cases of the knowledge graph, and generating process adjustment instructions. The application realizes the whole-process intelligent leap from a data fusion intelligent agent, a dynamic decision intelligent agent to a process closed loop.

Description

Technical Field

[0001] This invention relates to the field of rail transit technology, specifically to a full-process AI intelligent control system based on an integrated intelligent workshop for wheelset maintenance. Background Technology

[0002] Wheelset maintenance in rail transit is a crucial maintenance activity to ensure the safe and stable operation of trains. Wheelsets, consisting of wheels and axles, directly affect driving safety, comfort, and energy consumption.

[0003] Wheelset maintenance involves data such as equipment monitoring, process standards, and historical fault databases scattered across PMS (Production Management System), MES (Manufacturing Execution System), and individual sensors, easily creating information silos. Wheel inspection robots, AGVs, and other equipment operate independently, lacking unified scheduling and unable to dynamically adjust priorities based on the overall tasks of the workshop. Wheel axle flaw detection relies on technicians' visual inspection, and existing AI-assisted equipment is not linked to the maintenance knowledge base, making it impossible to optimize key parameters such as pressing torque in real time. In addition, new processes, equipment, and standards are constantly emerging, and current training relies on apprenticeship and static manuals, resulting in low knowledge transfer efficiency and an inability to keep up with the latest industry technical standards.

[0004] In summary, current wheelset maintenance workshops generally suffer from serious information silos, insufficient dynamic collaboration, delayed process control, and low knowledge transfer efficiency. Therefore, a full-process AI intelligent agent management and control system based on an integrated intelligent wheelset maintenance workshop is proposed to address the above-mentioned problems. Summary of the Invention

[0005] In view of the shortcomings of the existing technology, the purpose of this invention is to provide a full-process AI intelligent control system based on an integrated intelligent workshop for wheelset maintenance, so as to solve the problems existing in the background technology.

[0006] This invention is implemented as follows: a full-process AI intelligent agent control system based on an integrated intelligent workshop for wheelset maintenance, the system comprising:

[0007] The multi-source fusion intelligent agent module is used to collect equipment status and process data, generate feature vectors through cross-modal alignment using an AI model, and update a distributed knowledge graph based on a temporal graph convolutional network. The status and process data include flaw detection images and audio data.

[0008] The role-aware decision-making intelligent agent module is used to parse user permission tags and instructions, generate scheduling schemes in combination with device load, and send instructions to the device controller through edge nodes and store them on the blockchain.

[0009] The equipment conflict arbitration intelligent agent module is used to monitor the spatial topology of AGV and robot. When the trajectories overlap, it generates a fusion instruction based on the timestamp verification version number through Raft consensus and updates the permission topology tree. The spatial topology is the relative positional relationship between AGV and robot.

[0010] The closed-loop optimization intelligent agent module is used to analyze wheelset pressing torque data, extract historical cases from the knowledge graph to generate process adjustment instructions, send them to the PLC via blockchain signature, and provide feedback on the optimization effect. The wheelset pressing torque data refers to the torque applied during the pressing process and the torque change trend.

[0011] As a further aspect of the present invention: the multi-source fusion intelligent agent module includes:

[0012] The cross-modal alignment unit is used to input the flaw detection image into the ResNet-50 model to extract crack morphology features. At the same time, the audio data is converted into a 128-dimensional mel spectrum through MFCC, and the industrial-grade CLIP model is used to map to a unified semantic space to generate a fusion feature vector carrying confidence weights.

[0013] The graph dynamic evolution unit is used to construct a three-layer relational topology with wheelset serial numbers as the main entity, predict the association path between bearing wear and crack propagation based on the temporal graph convolutional network, and automatically expand the subgraph structure when new fault cases are added to the database.

[0014] The storage optimization unit is used to store knowledge graph subgraphs by production line workstation partition, dynamically migrate data to edge nodes based on query frequency heatmaps, and realize real-time relationship queries through graph databases.

[0015] As a further aspect of the present invention: the role perception and decision-making intelligent agent module includes:

[0016] The permission-driven parsing unit is used to identify the role tags associated with the user's work badge QR code, parse natural language queries into structured operation chains, and filter out unauthorized instructions.

[0017] The multi-objective optimization unit is used to establish a multi-objective function of equipment utilization, order delay penalty coefficient, and energy consumption threshold, and uses AI optimization algorithm to generate scheduling scheme;

[0018] The edge collaboration unit is used to cache less than 50MB of process standard sub-maps on the PAD, decompose complex requests into edge light computing and cloud heavy inference tasks, and enable the system to directly connect to the controller of high-risk equipment via a protocol.

[0019] As a further aspect of the present invention: the role perception and decision-making intelligent agent module further includes:

[0020] The blockchain trusted evidence storage unit is used to sign the issued equipment emergency stop commands and process parameter adjustment operations using the SM2 national cryptographic algorithm, and synchronize the hash value of the operation log to the Fabric consortium chain after IPFS sharding and storage.

[0021] As a further aspect of the present invention: the device conflict arbitration intelligent agent module includes:

[0022] The conflict prediction unit is used to calculate the path occupancy rate within a set time period based on the spatial topology of the AGV and the robot. When the path overlap rate is greater than 30%, a high-risk conflict is marked.

[0023] The consensus arbitration unit is used to attach a Lamport timestamp vector to the operation log of each device. If the version offset exceeds the dynamic threshold, the master node is elected through the AI-driven dynamic consensus mechanism to generate fusion instructions.

[0024] The permission topology update unit adjusts the weight of the device permission tree based on the arbitration result and binds the AGV task delay record to the knowledge graph.

[0025] As a further aspect of the present invention: the closed-loop optimization agent module includes:

[0026] The process deviation analysis unit is used to collect wheelset pressing torque data in real time according to the set collection frequency. When the detection standard deviation exceeds the process upper limit by 5%, it is marked as a critical deviation event.

[0027] The graph reverse optimization unit is used to extract the historical similar case handling subgraph from the knowledge graph, calculate the matching degree of the new solution through the graph attention mechanism, and inject the confidence rule.

[0028] The equipment parameter adaptive unit is used to generate hydraulic compensation commands with specified accuracy, which are then sent to the PLC after being signed by the blockchain, and the wheelset roundness error is fed back to the knowledge graph.

[0029] As a further aspect of the present invention: the multi-source fusion intelligent agent module also includes an antimagnetic RFID hardware group, which consists of a high-temperature resistant tag embedded in the inner ring of the wheel axle bearing and a reader / writer deployed at the maintenance station. The ID of the high-temperature resistant tag is bound to the unique identification entity of the wheelset in the knowledge graph, and the hardware group is used to collect the status and process data of the equipment.

[0030] Compared with the prior art, the beneficial effects of the present invention are:

[0031] This invention utilizes a multi-source fusion intelligent agent module to deeply integrate equipment status monitoring data and process parameters, and employs cross-modal alignment technology to construct a dynamically evolving knowledge graph, completely eliminating information silos and achieving intelligent association and semantic connectivity of heterogeneous data. A role-aware decision-making intelligent agent module, based on real-time parsing of user permission levels and equipment operating load, collaboratively generates precise scheduling instructions, combining blockchain distributed evidence storage technology to ensure the auditability and security of the entire process, and enhancing anomaly response capabilities and multi-task collaborative efficiency. An innovative equipment conflict arbitration intelligent agent module, through real-time spatial topology perception and a dynamic consensus mechanism, efficiently resolves trajectory overlap conflicts between AGVs and robotic arms, achieving seamless collaboration and resource optimization across multiple agents in complex scenarios. A closed-loop optimization intelligent agent module deeply mines historical case library knowledge, proactively optimizing key process parameters such as wheelset pressing torque, and continuously synchronizes with the latest technical standards through a knowledge graph self-learning mechanism, comprehensively improving assembly process accuracy while constructing an ecosystem for efficient knowledge flow. Overall, this invention uses a systematic approach to overcome the core bottlenecks that have long existed in the field of wheelset maintenance, such as data fragmentation, inefficient collaboration, outdated processes, and knowledge gaps, and achieves a leap in intelligent transformation of the entire process from data fusion and dynamic decision-making to process closure. Attached Figure Description

[0032] Figure 1 This is a schematic diagram of a full-process AI intelligent control system based on an integrated intelligent workshop for wheelset maintenance.

[0033] Figure 2 This is a schematic diagram of the structure of a multi-source fusion intelligent agent module in a full-process AI intelligent agent control system based on an integrated intelligent workshop for wheelset maintenance.

[0034] Figure 3 This is a schematic diagram of the role perception and decision-making intelligent agent module in a full-process AI intelligent agent control system based on an integrated intelligent workshop for wheelset maintenance.

[0035] Figure 4 This is a schematic diagram of the structure of an equipment conflict arbitration intelligent agent module in a full-process AI intelligent agent management and control system based on an integrated wheelset maintenance intelligent workshop.

[0036] Figure 5 This is a schematic diagram of the closed-loop optimized intelligent agent module in a full-process AI intelligent agent control system based on an integrated intelligent workshop for wheelset maintenance.

[0037] Figure 6 This is a flowchart of a whole-process control method based on an integrated intelligent workshop for wheelset maintenance. Detailed Implementation

[0038] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0039] The specific implementation of the present invention will be described in detail below with reference to specific embodiments.

[0040] like Figure 1 As shown, this embodiment of the invention provides a full-process AI intelligent agent control system based on an integrated intelligent workshop for wheelset maintenance. The system includes:

[0041] The multi-source fusion intelligent agent module 100 is used to collect the status and process data of the equipment, generate feature vectors through cross-modal alignment using the Zhongke Zhiyuan AI model, and update the distributed knowledge graph based on the temporal graph convolutional network. The status and process data include flaw detection images and audio data.

[0042] The Role Perception and Decision-Making Intelligent Agent Module 200 is used to parse user permission tags and instructions, generate a scheduling scheme in combination with device load, and send instructions to the device controller through edge nodes and perform blockchain notarization.

[0043] The equipment conflict arbitration intelligent agent module 300 is used to monitor the spatial topology of AGV and robot. When the trajectories overlap, it generates a fusion instruction based on the timestamp verification version number through Raft consensus and updates the permission topology tree. The spatial topology is the relative positional relationship between AGV and robot.

[0044] The closed-loop optimization intelligent agent module 400 is used to analyze wheelset pressing torque data, extract historical cases from the knowledge graph to generate process adjustment instructions, send them to the PLC via blockchain signature, and provide feedback on the optimization effect. The wheelset pressing torque data refers to the torque applied during the pressing process and the torque change trend.

[0045] It should be noted that the multi-source fusion intelligent agent module includes an anti-magnetic RFID hardware group, which consists of a high-temperature resistant tag embedded in the inner ring of the wheel axle bearing and a reader / writer deployed at the maintenance station. The ID of the high-temperature resistant tag is bound to the unique wheel pair identifier entity in the knowledge graph, and the hardware group is used to collect equipment status and process data.

[0046] In this embodiment of the invention, a multi-source fusion intelligent agent module deeply integrates equipment status monitoring data and process parameters, and uses cross-modal alignment technology to construct a dynamically evolving knowledge graph, completely eliminating information silos and realizing intelligent association and semantic connectivity of heterogeneous data; a role-aware decision-making intelligent agent module, based on real-time parsing of user permission levels and equipment operating load, collaboratively generates precise scheduling instructions, and combines blockchain distributed evidence storage technology to ensure the auditability and security of the entire process operation, strengthening anomaly response capabilities and multi-task collaborative efficiency; an equipment conflict arbitration intelligent agent module innovatively resolves trajectory overlap conflicts of AGV and robotic arm equipment through real-time spatial topology perception and dynamic consensus mechanism, realizing seamless collaboration and resource optimization of multiple intelligent agents in complex scenarios; a closed-loop optimization intelligent agent module deeply mines knowledge from historical case databases, proactively optimizes key process parameters such as wheelset pressing torque, and continuously synchronizes with the latest technical standards through a knowledge graph self-learning mechanism, comprehensively improving assembly process accuracy while building an ecosystem for efficient knowledge flow. Overall, this invention uses a systematic approach to overcome the core bottlenecks that have long existed in the field of wheelset maintenance, such as data fragmentation, inefficient collaboration, outdated processes, and knowledge gaps, and achieves a leap in intelligent transformation of the entire process from data fusion and dynamic decision-making to process closure.

[0047] like Figure 2 As shown, in a preferred embodiment of the present invention, the multi-source fusion intelligent agent module 100 includes:

[0048] The cross-modal alignment unit 101 is used to input the flaw detection image into the ResNet-50 model to extract crack morphology features. At the same time, the audio data is converted into a 128-dimensional spectral density using MFCC, and the industrial-grade CLIP model is used to map it to a unified semantic space to generate a fusion feature vector carrying confidence weights.

[0049] The graph dynamic evolution unit 102 is used to construct a three-layer relational topology with wheelset serial numbers as the main entity, predict the association path between bearing wear and crack propagation based on the time-series graph convolutional network, and automatically expand the subgraph structure when new fault cases are added to the database.

[0050] The storage optimization unit 103 is used to store knowledge graph subgraphs by production line workstation partition, dynamically migrate data to edge nodes according to the query frequency heatmap, and realize real-time relationship query through graph database.

[0051] In this embodiment of the invention, the multi-source fusion intelligent agent module 100 achieves deep integration and dynamic knowledge management of heterogeneous industrial data through three core units. The cross-modal alignment unit 101 uses a ResNet-50 model to analyze the crack morphology features (such as crack length, bifurcation angle, and other geometric attributes) in the flaw detection image. Simultaneously, it converts audio data such as bearing noise into a 128-Vimel spectrum representation using MFCC. Based on an industrial scenario-optimized CLIP model, it maps image and audio features to a unified semantic space, generating a fused feature vector carrying confidence weights. For example, if the correlation between a bearing crack image feature and a noise spectrum feature in the semantic space is greater than a set value, a high-weight alarm will be triggered. The graph dynamic evolution unit 102 constructs a three-layer relational topology with wheelset serial numbers as the main entity. The physical layer contains the attribute parameters of the bearing and wheel, the process layer contains the pressing parameters and flaw detection records, and the fault layer contains historical cases. It uses a temporal graph convolutional network (T-GCN) to predict the equipment state evolution path, which can identify the correlation between bearing wear and crack propagation rate. When a new fault case is added, the subgraph structure is automatically expanded. The storage optimization unit 103... Knowledge graph subgraphs are stored in partitions according to production line workstations. High-frequency data is dynamically migrated to edge nodes based on query heatmaps. Millisecond-level relationship queries can be achieved through the Neo4j graph database. For example, it is possible to retrieve all flaw detection records and maintenance work orders within two years associated with a certain wheel pair sequence number.

[0052] like Figure 3 As shown, in a preferred embodiment of the present invention, the role perception and decision-making intelligent agent module 200 includes:

[0053] The permission-driven parsing unit 201 is used to identify the role tags associated with the user's work badge QR code, parse natural language queries into structured operation chains, and filter out unauthorized instructions.

[0054] The multi-objective optimization unit 202 is used to establish a multi-objective function of equipment utilization, order delay penalty coefficient and energy consumption threshold, and to generate a scheduling scheme using AI optimization algorithm;

[0055] The edge collaboration unit 203 is used to cache less than 50MB of process standard sub-maps on the PAD end, decompose complex requests into edge light computing and cloud heavy inference tasks, and enable the system to directly connect to the controller of high-risk equipment through the protocol.

[0056] The blockchain trusted evidence storage unit 204 is used to sign the issued equipment emergency stop command and process parameter adjustment operation with the SM2 national cryptographic algorithm, and synchronize the hash value to the Fabric consortium chain after the operation log is stored in IPFS shards.

[0057] In this embodiment of the invention, the role-aware decision-making intelligent agent module 200 achieves a safe, efficient, and reliable intelligent decision-making closed loop through the collaboration of four core units. The permission-driven parsing unit 201 identifies the user's role label (such as senior technician, quality inspector, or intern) based on their employee ID card QR code, converts natural language instructions into structured operation sequences in real time, and automatically intercepts requests exceeding permissions. For example, if a junior operator attempts to skip the approval process and directly modify bearing press-fit parameters, the system immediately terminates the instruction and triggers a permission alarm, eliminating the risk of misoperation at the source. The multi-objective optimization unit 202 constructs a Pareto optimal model covering multi-dimensional constraints such as real-time equipment utilization, order delivery delay penalty coefficient, and production line energy consumption threshold. It uses AI optimization algorithms to dynamically generate a globally optimal scheduling scheme. For example, when a flaw detection station experiences a sudden equipment failure, it simultaneously optimizes AGV path planning and diverts wheelsets awaiting inspection to idle stations. The edge collaboration unit 203 pre-caches high-frequency process standard sub-maps (such as wheel flaw detection judgment threshold tables) on the PAD end. It uses lightweight container technology to decompose complex requests into edge-end real-time computing and cloud-based deep inference tasks, and is based on OPC. The UA protocol directly connects to the unloading machine's PLC controller to execute high-risk operations such as emergency stops, ensuring low latency in responding to critical commands. The blockchain-based trusted evidence storage unit 204 uses the national cryptographic SM2 algorithm to digitally sign sensitive operations such as equipment emergency stop commands and process parameter adjustments. The complete operation log is then encrypted through IPFS distributed storage fragments and its hash fingerprint is synchronized to each supervisory node on the Fabric consortium blockchain, achieving tamper-proof evidence storage and full lifecycle traceability of the operation process, providing a trusted technological foundation for quality auditing. Therefore, it enhances the maintenance workshop's management capabilities in access control, global resource scheduling, edge response, and operation auditing.

[0058] like Figure 4 As shown, in a preferred embodiment of the present invention, the device conflict arbitration intelligent agent module 300 includes:

[0059] The conflict prediction unit 301 is used to calculate the path occupancy rate within a future set time based on the spatial topology of the AGV and the robot arm, and to mark high-risk conflicts when the path overlap rate is greater than 30%.

[0060] Consensus arbitration unit 302 is used to attach Lamport timestamp vectors to the operation logs of each device. If the version offset exceeds the dynamic threshold, a master node is elected through an AI-driven dynamic consensus mechanism to generate fusion instructions.

[0061] The permission topology update unit 303 adjusts the weight of the device permission tree according to the arbitration result and binds the AGV task delay record to the knowledge graph.

[0062] In this embodiment of the invention, the device conflict arbitration intelligent agent module 300 achieves device collaborative safety in complex scenarios through a triple protection mechanism. The conflict prediction unit 301 constructs a real-time spatial topology map based on the AGV LiDAR point cloud and the robot's kinematic model, calculates the path occupancy rate in the next few seconds using a trajectory prediction algorithm, and immediately marks a high-risk conflict when the path overlap rate exceeds 30%, for example, when an AGV and a robot are about to have their trajectories intersect in a material transfer area. The consensus arbitration unit 302 appends a version vector containing a Lamport timestamp to the operation log of each device. When the version offset between devices exceeds a dynamic threshold (e.g., a 300ms deviation between the robot's emergency stop command and the AGV's avoidance signal), the conflict is resolved. When the Raft consensus algorithm is automatically triggered to elect a master node and generate a fusion command, the master node decides to suspend the robot's operation and authorize the AGV to pass first. The permission topology update unit 303 dynamically adjusts the weight of the device permission tree according to the arbitration result (for example, the path priority coefficient of a certain AGV that frequently triggers conflicts will be lowered in the permission tree), and binds the key parameters of the conflict event (such as delay duration and avoidance path) to the fault mode library of the knowledge graph to provide historical basis for subsequent scheduling decisions. This module significantly shortens the device conflict response time, and at the same time reduces the device collaborative failure rate through dynamic optimization of the permission tree.

[0063] like Figure 5 As shown, in a preferred embodiment of the present invention, the closed-loop optimization agent module 400 includes:

[0064] The process deviation analysis unit 401 is used to collect wheelset pressing torque data in real time according to the set collection frequency. When the detection standard deviation exceeds the process upper limit by 5%, it is marked as a critical deviation event.

[0065] The graph reverse optimization unit 402 is used to extract the historical similar case handling subgraph in the knowledge graph, calculate the matching degree of the new solution through the graph attention mechanism, and inject the confidence rule;

[0066] The equipment parameter adaptive unit 403 is used to generate hydraulic compensation commands with specified accuracy, which are then sent to the PLC after being signed by the blockchain, and the wheelset roundness error is fed back to the knowledge graph.

[0067] In this embodiment of the invention, assuming that the process deviation analysis unit 401 samples wheelset press-fit torque data at a high frequency of 1000 times per second, when it detects that the standard deviation of 10 consecutive sampling points exceeds the process upper limit by 5%, it will automatically mark it as a critical deviation event. For example, if the press-fit torque of a bogie bearing fluctuates by ±8%, a system alarm will be triggered. At this time, the graph reverse optimization unit 402 will extract the historical similar case handling subgraph from the knowledge graph in real time (such as searching the database of cases of the same type of bearing exceeding the torque limit in the past three years), calculate the matching degree between the new handling scheme and the historical optimal solution through the graph attention neural network, and generate rules with injected confidence weights. For example, if the matching degree reaches 92%, it is recommended to increase the hydraulic compensation value by 0.3MPa, and the system will execute accordingly. The equipment parameter adaptive unit 403 will generate a high-precision hydraulic compensation command based on the rules, and after hash signing by the national cryptographic SM3 algorithm, it will be sent to the PLC actuator through the OPC UA protocol, and the wheelset roundness error data after execution will be fed back to the process optimization node of the knowledge graph in real time. In summary, this module improves the wheelset assembly pass rate and reduces similar process deviations annually through a continuous feedback learning mechanism.

[0068] like Figure 6 As shown in the figure, this embodiment of the invention also provides a method for full-process control based on an integrated intelligent workshop for wheelset maintenance, the method comprising the following steps:

[0069] S100 collects the status and process data of the equipment, generates feature vectors through cross-modal alignment using the Zhongke Zhiyuan AI model, and updates the distributed knowledge graph based on the temporal graph convolutional network. The status and process data includes flaw detection images and audio data.

[0070] S200 parses user permission tags and instructions, generates a scheduling scheme based on device load, and sends instructions to the device controller through edge nodes and performs blockchain notarization.

[0071] S300 monitors the spatial topology of the AGV and the robot arm. When the trajectories overlap, it generates a fusion instruction based on the timestamp verification version number through Raft consensus and updates the permission topology tree. The spatial topology is the relative positional relationship between the AGV and the robot arm.

[0072] S400 analyzes the wheelset pressing torque data, extracts historical cases from the knowledge graph to generate process adjustment instructions, sends them to the PLC via blockchain signature, and provides feedback on the optimization effect. The wheelset pressing torque data refers to the torque applied during the pressing process and the torque change trend.

[0073] In this invention, the invention deeply integrates equipment status monitoring data and process parameters, utilizes cross-modal alignment technology to construct a dynamically evolving knowledge graph, completely eliminating information silos and achieving intelligent association and semantic connectivity of heterogeneous data. Based on real-time parsing of user permission levels and equipment operating load, it collaboratively generates precise scheduling instructions, and combines blockchain distributed evidence storage technology to ensure the auditability and security of the entire process, strengthening anomaly response capabilities and multi-task collaborative efficiency. Innovatively, through real-time spatial topology perception and dynamic consensus mechanisms, it efficiently resolves trajectory overlap conflicts of AGVs and robotic arms, achieving seamless collaboration and resource optimization of multiple agents in complex scenarios. It mines knowledge from historical case libraries to proactively optimize key process parameters such as wheelset pressing torque, and continuously synchronizes with the latest technical standards through a knowledge graph self-learning mechanism, comprehensively improving assembly process accuracy while building an ecosystem for efficient knowledge flow. Overall, this method, with a systematic approach, overcomes the core bottlenecks that have long existed in the wheelset maintenance field, such as data fragmentation, inefficient collaboration, process lag, and knowledge gaps, achieving a comprehensive improvement from data fusion and dynamic decision-making to process closed-loop.

[0074] The above description only details the preferred embodiments of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0075] It should be understood that although the steps in the flowcharts of the various embodiments of the present invention are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the various embodiments may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least a portion of the sub-steps or stages of other steps.

[0076] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0077] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the disclosure in the specification and embodiments. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and embodiments are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the claims.

Claims

1. A full-process AI intelligent agent control system based on an integrated intelligent workshop for wheelset maintenance, characterized in that, The system includes: The multi-source fusion intelligent agent module is used to collect equipment status and process data, generate feature vectors through cross-modal alignment using an AI model, and update a distributed knowledge graph based on a temporal graph convolutional network. The status and process data include flaw detection images and audio data. The role-aware decision-making intelligent agent module is used to parse user permission tags and instructions, generate scheduling schemes in combination with device load, and send instructions to the device controller through edge nodes and store them on the blockchain. The equipment conflict arbitration intelligent agent module is used to monitor the spatial topology of AGV and robot. When the trajectories overlap, it generates a fusion instruction based on the timestamp verification version number through Raft consensus and updates the permission topology tree. The spatial topology is the relative positional relationship between AGV and robot. The closed-loop optimization intelligent agent module is used to analyze wheelset pressing torque data, extract historical cases from the knowledge graph to generate process adjustment instructions, send them to the PLC via blockchain signature and provide feedback on the optimization effect. The wheelset pressing torque data refers to the torque applied during the pressing process and the torque change trend. The device conflict arbitration intelligent agent module includes: The conflict prediction unit is used to calculate the path occupancy rate within a set time period based on the spatial topology of the AGV and the robot. When the path overlap rate is greater than 30%, a high-risk conflict is marked. The consensus arbitration unit is used to attach a Lamport timestamp vector to the operation log of each device. If the version offset exceeds the dynamic threshold, the master node is elected through the AI-driven dynamic consensus mechanism to generate fusion instructions. The permission topology update unit adjusts the weight of the device permission tree based on the arbitration result and binds the AGV task delay record to the knowledge graph.

2. The full-process AI intelligent agent control system based on an integrated wheelset maintenance intelligent workshop as described in claim 1, characterized in that, The multi-source fusion intelligent agent module includes: The cross-modal alignment unit is used to input the flaw detection image into the ResNet-50 model to extract crack morphology features. At the same time, the audio data is converted into a 128-dimensional mel spectrum through MFCC, and the industrial-grade CLIP model is used to map to a unified semantic space to generate a fusion feature vector carrying confidence weights. The graph dynamic evolution unit is used to construct a three-layer relational topology with wheelset serial numbers as the main entity, predict the association path between bearing wear and crack propagation based on the temporal graph convolutional network, and automatically expand the subgraph structure when new fault cases are added to the database. The storage optimization unit is used to store knowledge graph subgraphs by production line workstation partition, dynamically migrate data to edge nodes based on query frequency heatmaps, and realize real-time relationship queries through graph databases.

3. The full-process AI intelligent agent control system based on an integrated wheelset maintenance intelligent workshop as described in claim 1, characterized in that, The role-aware decision-making intelligent agent module includes: The permission-driven parsing unit is used to identify the role tags associated with the user's work badge QR code, parse natural language queries into structured operation chains, and filter out unauthorized instructions. The multi-objective optimization unit is used to establish a multi-objective function of equipment utilization, order delay penalty coefficient, and energy consumption threshold, and uses AI optimization algorithm to generate scheduling scheme; The edge collaboration unit is used to cache less than 50MB of process standard sub-maps on the PAD, decompose complex requests into edge light computing and cloud heavy inference tasks, and enable the system to directly connect to the controller of high-risk equipment via a protocol.

4. The full-process AI intelligent agent control system based on an integrated wheelset maintenance intelligent workshop as described in claim 3, characterized in that, The role-aware decision-making agent module also includes: The blockchain trusted evidence storage unit is used to sign the issued equipment emergency stop commands and process parameter adjustment operations using the SM2 national cryptographic algorithm, and synchronize the hash value of the operation log to the Fabric consortium chain after IPFS sharding and storage.

5. The full-process AI intelligent agent control system based on an integrated wheelset maintenance intelligent workshop according to claim 1, characterized in that, The closed-loop optimization agent module includes: The process deviation analysis unit is used to collect wheelset pressing torque data in real time according to the set collection frequency. When the detection standard deviation exceeds the process upper limit by 5%, it is marked as a critical deviation event. The graph reverse optimization unit is used to extract the historical similar case handling subgraph from the knowledge graph, calculate the matching degree of the new solution through the graph attention mechanism, and inject the confidence rule. The equipment parameter adaptive unit is used to generate hydraulic compensation commands with specified accuracy, which are then sent to the PLC after being signed by the blockchain, and the wheelset roundness error is fed back to the knowledge graph.

6. The full-process AI intelligent agent control system based on an integrated wheelset maintenance intelligent workshop according to claim 1, characterized in that, The multi-source fusion intelligent agent module also includes an anti-magnetic RFID hardware group, which consists of a high-temperature resistant tag embedded in the inner ring of the wheel axle bearing and a reader / writer deployed at the maintenance station. The ID of the high-temperature resistant tag is bound to the unique identification entity of the wheelset in the knowledge graph, and the hardware group is used to collect the status and process data of the equipment.

Images