Multi-agent based industrial equipment maintenance scheme automatic generation method and system

Abstract

This application provides a method and system for automatically generating maintenance plans for industrial equipment based on multi-agent technology, relating to the field of industrial equipment operation and maintenance technology. The method includes: in response to fault information of industrial equipment, acquiring fault data and a pre-set task scheduling diagram of the industrial equipment; scheduling multiple agents based on the task scheduling diagram and fault data to collaboratively generate a maintenance plan for the industrial equipment; the task scheduling diagram includes multiple task nodes required to generate the maintenance plan, the agents corresponding to the multiple task nodes, and the input-output relationships between the multiple task nodes. The solution of this application improves the accuracy and reliability of automatically generating maintenance plans for industrial equipment.

Description

Technical Field

[0001] This application relates to the field of industrial equipment operation and maintenance technology, and in particular to a method and system for automatically generating industrial equipment maintenance plans based on multi-agent systems. Background Technology

[0002] In the field of industrial equipment operation and maintenance, rapid diagnosis of equipment faults and generation of repair solutions are the core links to ensure production continuity and equipment reliability.

[0003] In traditional methods, after industrial equipment malfunctions, maintenance personnel typically need to manually write a repair plan based on fault records, on-site phenomena, alarm information, inspection results, repair measures, and re-inspection conclusions. This process is not only time-consuming but also highly dependent on the experience and writing habits of the maintenance personnel. With the development of artificial intelligence technology, related technologies can automatically generate repair plans using large language models or intelligent agents. While this improves the efficiency of repair plan generation, the accuracy and reliability of the repair plans remain relatively low.

[0004] Therefore, a solution is needed to improve the accuracy and reliability of automatically generating maintenance plans for industrial equipment. Summary of the Invention

[0005] This application provides a method and system for automatically generating industrial equipment maintenance plans based on multi-agent systems, which can improve the accuracy and reliability of automatically generating industrial equipment maintenance plans.

[0006] In a first aspect, embodiments of this application provide a method for automatically generating industrial equipment maintenance solutions based on multi-agent systems, including:

[0007] In response to fault information of industrial equipment, acquire fault data of the industrial equipment and a pre-set task scheduling diagram;

[0008] Based on the task orchestration diagram and the fault data, multiple intelligent agents are scheduled to collaboratively generate a maintenance plan for the industrial equipment.

[0009] The task orchestration diagram includes multiple task nodes required to generate the maintenance plan, the smart agents corresponding to the multiple task nodes, and the input-output relationships between the multiple task nodes.

[0010] In one possible implementation, the multiple task nodes include fault analysis tasks, evidence retrieval tasks, fault location tasks, maintenance measure generation tasks, and maintenance plan writing tasks; the step of scheduling multiple intelligent agents based on the task orchestration diagram and the fault data to collaboratively generate a maintenance plan for the industrial equipment includes:

[0011] The fault data is standardized to obtain standardized data;

[0012] Based on the task orchestration graph, multiple agents are scheduled to collaboratively execute the following tasks:

[0013] The intelligent agent corresponding to the scheduled fault analysis task performs slot-based analysis on the standardized data to obtain a set of fault phenomenon slots.

[0014] The retrieval agent corresponding to the scheduled evidence retrieval task performs evidence retrieval on the set of fault phenomenon slots to obtain a set of valid evidence.

[0015] The location agent corresponding to the fault location task is scheduled, and the fault location result is determined based on the fault phenomenon slot set and the valid evidence set.

[0016] The scheduling and maintenance measures generate a corresponding action agent, and the target maintenance measures are determined based on the fault location results and the set of valid evidence.

[0017] The writing agent corresponding to the scheduling and maintenance plan writing task generates a structured maintenance plan for the industrial equipment based on the fault location results and the target maintenance measures.

[0018] When executing each task, a task record corresponding to each task is generated; the task record includes one or more of the following: task node, agent, task input, task output, callable tool, quality score, execution time, and version identifier.

[0019] In one possible implementation, the step of performing slot-based parsing on the standardized data to obtain a set of fault phenomenon slots includes:

[0020] The standardized data is parsed in a slot-based manner to obtain a structured set of fault phenomenon slots. The set of fault phenomenon slots includes one or more of the following: equipment object slots, component slots, alarm information slots, operating status slots, abnormal phenomenon slots, trend change slots, severity slots, and accompanying phenomenon slots.

[0021] Based on the slot semantic vector and weight vector corresponding to each slot in the fault phenomenon slot set, the slot credibility of each slot is determined.

[0022] Determine whether the slot confidence level of each slot is greater than or equal to the confidence level threshold;

[0023] If none of the data is greater than or equal to the credibility threshold, the step of slot-based parsing of the standardized data is repeated, or a manual review prompt is output.

[0024] In one possible implementation, the step of performing evidence retrieval on the set of fault phenomenon slots to obtain a valid evidence set includes:

[0025] Retrieval evidence corresponding to the set of fault phenomenon slots is obtained from the preset knowledge retrieval base;

[0026] The relevance score of the retrieved evidence is determined based on the text semantic relevance score, component relevance score, alarm relevance score, and historical maintenance similarity score.

[0027] The set of valid evidence is determined based on the search evidence whose relevance scores are greater than or equal to the relevance score threshold.

[0028] In one possible implementation, determining the fault location result based on the set of fault phenomenon slots and the set of valid evidence includes:

[0029] Based on the set of fault phenomena slots and the set of valid evidence, a set of candidate faults is constructed. Each candidate fault in the set of candidate faults includes: candidate fault location, candidate fault type, and candidate fault cause.

[0030] The fault confidence of each candidate fault is determined based on semantic matching score, rule matching score, historical maintenance similarity score, and fault evidence support score.

[0031] The fault location result is determined based on the candidate faults corresponding to the maximum fault confidence.

[0032] In one possible implementation, determining the target maintenance measures based on the fault location results and the set of valid evidence includes:

[0033] Based on the fault location results and the set of valid evidence, a set of candidate maintenance measures is generated;

[0034] Based on the component matching score, fault type matching score, fault cause matching score, and safety constraint score, determine the matching degree score between each candidate maintenance measure and the fault location result;

[0035] The target maintenance measure is determined based on the candidate maintenance measures corresponding to the maximum matching score.

[0036] In one possible implementation, generating a structured maintenance plan for the industrial equipment based on the fault location results and the target maintenance measures includes:

[0037] Based on the fault phenomena in the standardized data, the fault location, fault type, fault cause corresponding to the fault location result, the target maintenance measures, inspection steps, re-inspection steps, restoration operating conditions, and solution field set, a fault-measure-solution chain constraint is generated.

[0038] Based on the set of valid evidence, the fault-measure-solution chain constraint, and the preset solution template constraint, a structured maintenance solution for the industrial equipment is generated and output.

[0039] In one possible implementation, the plurality of task nodes further includes a quality verification task, and the method further includes:

[0040] The verification agent corresponding to the quality verification task is scheduled to perform the following operations:

[0041] Based on the set of field evidence corresponding to each field in the structured maintenance solution, determine the field evidence support score of the structured maintenance solution;

[0042] Determine the field completeness score, fault location and repair measure accuracy score, and project executability score of the structured maintenance plan;

[0043] Determine the logical consistency score between each field in the structured maintenance plan and the fault-measure-plan chain constraint;

[0044] The overall quality score of the structured maintenance plan is determined based on the accuracy score of the fault location and maintenance measures, the field integrity score, the logical consistency score, the field evidence support score, and the project executability score.

[0045] In one possible implementation, the plurality of task nodes further includes a local rewrite task, and the method further includes:

[0046] The rewriting agent corresponding to the local rewriting task is scheduled to perform the following operations:

[0047] Determine whether the structured maintenance solution contains a target field that satisfies any of the following conditions:

[0048] The field is empty;

[0049] The field integrity score is less than the integrity score threshold;

[0050] The field of evidence support scored 0.

[0051] If a target field exists, the target task node corresponding to the target field is determined according to the task orchestration diagram, and the task execution record corresponding to the target task node is determined.

[0052] Based on the task execution record, the field evidence set corresponding to the target field, and the fault-measure-solution chain constraint, a partial rewrite operation is performed to obtain the rewritten field.

[0053] Determine whether the rewritten field meets the following conditions: the field integrity score is greater than or equal to the integrity score threshold, and the field evidence support score is 1;

[0054] If the conditions are met, the rewritten field is used to replace the target field, and the updated structured maintenance solution is obtained and output.

[0055] Secondly, embodiments of this application provide an automatic generation system for industrial equipment maintenance solutions based on multi-agent systems, comprising:

[0056] The acquisition module is used to acquire fault data and a pre-set task scheduling diagram of the industrial equipment in response to fault occurrence information of the industrial equipment;

[0057] The processing module is used to schedule multiple agents based on the task orchestration diagram and the fault data to collaboratively generate a maintenance plan for the industrial equipment; the task orchestration diagram includes multiple task nodes required to generate the maintenance plan, the agents corresponding to the multiple task nodes, and the input-output relationships between the multiple task nodes.

[0058] Thirdly, embodiments of this application provide another automatic generation system for industrial equipment maintenance solutions based on multi-agent systems, including:

[0059] The processor, and the memory that is in communication with the processor;

[0060] Memory is used to store instructions that the computer executes;

[0061] The processor is configured to execute computer execution instructions stored in memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above.

[0062] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the first aspect and / or various possible implementations of the first aspect described above.

[0063] Fifthly, embodiments of this application provide a computer program product, including a computer program, which, when executed by a processor, is used to implement the first aspect and / or various possible implementations of the first aspect as described above.

[0064] This application provides a method and system for automatically generating maintenance plans for industrial equipment based on multi-agent technology. When an industrial equipment failure occurs, it can acquire fault data and a pre-set task orchestration diagram, and then schedule multiple agents to collaboratively generate a maintenance plan based on the task orchestration diagram and fault data. Relying on the pre-set task orchestration diagram, the task nodes, dedicated executing agents, and the coupling logic of input and output between nodes in the entire maintenance plan generation process can be clearly defined. This breaks down the maintenance plan generation process into multiple task stages with clear relationships, each handled by a corresponding agent. This avoids problems such as disordered scheduling of multi-agent operations, ambiguous task responsibilities, and data interaction errors, ensuring the orderliness of the maintenance plan generation process and the stability of the output. Through this setup, the maintenance plan generation process can be scheduled to be completed in stages by agents under the constraints of the task orchestration diagram. While achieving automatic generation of maintenance plans for industrial equipment, it also improves the accuracy and reliability of maintenance plan generation, thereby enhancing the overall safety and efficiency of industrial equipment operation and maintenance. Attached Figure Description

[0065] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0066] Figure 1 This is a system architecture diagram of an embodiment of this application;

[0067] Figure 2 This is a flowchart of an embodiment of the automatic generation method for industrial equipment maintenance schemes based on multi-agent systems according to this application;

[0068] Figure 3 This is a schematic diagram of a task arrangement according to an embodiment of this application;

[0069] Figure 4 This is a schematic diagram of the structure of an automatic generation system for industrial equipment maintenance solutions based on multi-agent technology, according to an embodiment of this application.

[0070] Figure 5 This is a schematic diagram of the structure of an automatic generation system for industrial equipment maintenance solutions based on multiple agents, according to another embodiment of this application.

[0071] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0072] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application.

[0073] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0074] The collection, storage, use, processing, transmission, provision, and disclosure of financial data or user data involved in the technical solution of this application all comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0075] It should be noted that in the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, they do not mean that the applicant has used or necessarily used the solution.

[0076] It should also be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with relevant laws, regulations and standards, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0077] The method and system for automatically generating industrial equipment maintenance plans based on multi-agent technology of this application can be used in the field of industrial equipment operation and maintenance, and can also be used in any field other than industrial equipment operation and maintenance, such as automation. The application field of the method and system for automatically generating industrial equipment maintenance plans based on multi-agent technology of this application is not limited.

[0078] The multi-agent-based automatic generation method and system for industrial equipment maintenance solutions of this application can be applied to scenarios where maintenance solutions are generated when industrial equipment fails. The automatic generation of maintenance solutions for any industrial equipment can be achieved using the multi-agent-based automatic generation method and system for industrial equipment maintenance solutions of this application.

[0079] In the field of industrial equipment operation and maintenance, rapid diagnosis of equipment faults and generation of repair solutions are the core links to ensure production continuity and equipment reliability.

[0080] In traditional methods, after industrial equipment malfunctions, maintenance personnel typically need to manually draft a repair plan based on fault records, on-site phenomena, alarm information, inspection results, repair measures, and re-inspection conclusions. This process is not only time-consuming but also highly dependent on the experience level and writing habits of the maintenance personnel.

[0081] With the development of artificial intelligence technology, related technologies can automatically generate maintenance plans using large language models or intelligent agents. Although this improves the efficiency of maintenance plan generation, the following problems still exist, resulting in lower accuracy and reliability of the maintenance plans:

[0082] (i) Large language models have strong natural language generation capabilities, but they often lack strict constraints on equipment structure, fault mechanisms, maintenance procedures, safety constraints, and historical maintenance evidence when generating maintenance plans for industrial equipment. When fault descriptions are directly input into large models to generate maintenance plans, the models may generate faulty components, fault causes, maintenance actions, spare parts names, or re-inspection conclusions that do not exist in the input data.

[0083] Such unfounded generation may only manifest as inaccurate content in ordinary text generation scenarios, but in industrial maintenance scenarios, it may lead to incorrect repairs, incorrect replacement of parts, insufficient re-inspection, or missing safety risk warnings, thereby affecting the reliability of equipment recovery.

[0084] (ii) Multi-agent methods typically assign roles to different agents to accomplish complex tasks. For example, one agent might be responsible for analyzing faults, another for generating solutions, and yet another for verifying the content. While this approach achieves a superficial division of labor, it often relies on prompt words and lacks a structured task orchestration mechanism tailored to industrial maintenance processes.

[0085] In actual implementation, the input-output relationships, task order, intermediate result transmission methods, and error rollback paths between different agents are often unclear, which can easily lead to problems such as duplicate analysis, conflicting conclusions, task omissions, and verification failures. For example, the solution generation agent may directly generate maintenance measures before the fault location results are fully confirmed; and after the verification agent discovers an error, it is difficult to pinpoint whether the problem occurred in the fault analysis, fault location, measure generation, or solution writing stage.

[0086] Based on the above-mentioned technical problems, the inventive concept of this application is: how to provide a solution that can improve the accuracy and reliability of automatic generation of maintenance plans for industrial equipment.

[0087] This application provides a method and system for automatically generating maintenance plans for industrial equipment based on multi-agent technology. When an industrial equipment failure occurs, it can acquire fault data and a pre-set task orchestration diagram, and then schedule multiple agents to collaboratively generate a maintenance plan based on the task orchestration diagram and fault data. Relying on the pre-set task orchestration diagram, the task nodes, dedicated executing agents, and the coupling logic of input and output between nodes in the entire maintenance plan generation process can be clearly defined. This breaks down the maintenance plan generation process into multiple task stages with clear relationships, each handled by a corresponding agent. This avoids problems such as disordered scheduling of multi-agent operations, ambiguous task responsibilities, and data interaction errors, ensuring the orderliness of the maintenance plan generation process and the stability of the output. Through this setup, the maintenance plan generation process can be scheduled to be completed in stages by agents under the constraints of the task orchestration diagram. While achieving automatic generation of maintenance plans for industrial equipment, it also improves the accuracy and reliability of maintenance plan generation, thereby enhancing the overall safety and efficiency of industrial equipment operation and maintenance.

[0088] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0089] Figure 1 This is a system architecture diagram of an embodiment of this application, such as... Figure 1 As shown, when industrial equipment malfunctions, fault information is sent to the multi-agent-based industrial equipment maintenance solution automatic generation system. In response to the fault information, the system obtains the fault data of the industrial equipment and a pre-set task orchestration diagram. Based on the task orchestration diagram and fault data, multiple agents are scheduled to collaboratively generate a maintenance solution for the industrial equipment. The task orchestration diagram includes multiple task nodes required to generate the maintenance solution, the agents corresponding to the multiple task nodes, and the input-output relationships between the multiple task nodes.

[0090] Figure 2 This is a flowchart illustrating an embodiment of the automatic generation method for industrial equipment maintenance plans based on multi-agent systems according to this application. This embodiment describes the automatic generation method for industrial equipment maintenance plans based on multi-agent systems using a multi-agent system as the executing entity (hereinafter referred to as the "system"). Figure 2 As shown, the automatic generation method for industrial equipment maintenance solutions based on multi-agent systems may include the following steps:

[0091] S201: In response to information about industrial equipment failures, acquire fault data and a pre-set task scheduling diagram of the industrial equipment.

[0092] In this embodiment, the multi-agent-based automatic generation system for industrial equipment maintenance solutions can be a central server deployed on an industrial operation and maintenance platform, an edge computing node, or a processing system composed of an edge computing node and a central server working together.

[0093] In this embodiment, the fault information of industrial equipment can be alarm events pushed by the equipment monitoring system, abnormal reports entered by maintenance personnel in the work order system, downtime status changes uploaded by the programmable logic controller, or fault confirmation signals uploaded by the field acquisition terminal, etc., without any restrictions.

[0094] In this embodiment, the fault information of the industrial equipment is the start event of the automatic maintenance plan generation process, which triggers the system to establish a task instance corresponding to the fault event and obtain fault data according to the data source associated with the industrial equipment.

[0095] In this embodiment, the fault data may include, but is not limited to: original fault description, equipment name, equipment number, component name, alarm code, operating condition, sensor abnormal information, on-site maintenance personnel remarks, historical maintenance records, equipment manual, maintenance procedures, and retrieval evidence, etc. There are no restrictions here, as long as it is related to the fault of industrial equipment.

[0096] For example, during the fault data acquisition process, the system can aggregate data from multiple data interfaces. For natural language data, it can acquire shift logs, inspection logs, maintenance request texts, and operator descriptions; for structured alarm data, it can acquire alarm codes, alarm levels, trigger times, release times, and associated control unit identifiers; for time-series data, it can extract sensor curves, key operating parameters, and control command sequences within a preset time window before the fault; for historical records, it can extract historical faults, handling measures, and results corresponding to the same equipment, the same components, or the same alarm codes; and for equipment static data, it can extract equipment models, component composition, process locations, maintenance plans, and maintenance manual indexes.

[0097] In this embodiment, after acquiring the fault data, a data integrity check can be performed. The check items include whether the device identifier exists, whether the fault time is valid, whether at least one piece of fault evidence is available, and whether the input fields required by the task orchestration diagram are complete. For missing fields, the system can fill in the identifiable content based on the device ledger, the most recent running snapshot, or historical similar events, and add a missing mark to fields that cannot be filled in, writing them along with the relevant identifiers into the subsequent context.

[0098] In this embodiment, the task orchestration diagram may include multiple task nodes required to generate a maintenance plan, intelligent agents corresponding to the multiple task nodes, and input-output relationships between the multiple task nodes.

[0099] In this embodiment, while acquiring fault data, the system invokes a pre-set task orchestration graph. This task orchestration graph can be pre-stored in the orchestration configuration library of the system or operations and maintenance platform. Each task node in the task orchestration graph is configured with a node identifier, input field definitions, output field definitions, and a calling agent identifier. The input-output relationships between nodes are described using directed edges, which are used to limit the specific fields in which the output results of preceding nodes enter subsequent nodes. After reading the task orchestration graph, the system can write the fault data into the input context of the starting node in the graph.

[0100] In this embodiment, the task orchestration diagram can be generated using the following methods:

[0101] ① Determine the multiple task nodes corresponding to the maintenance plan generation task, as well as the input-output relationships between the multiple task nodes (described using directed dependency edges).

[0102] ② Determine the agent node, intermediate products, and callable tools (third-party tools required to execute the task) corresponding to each task node.

[0103] ③ Generate the corresponding task orchestration diagram based on the task nodes, agent nodes, input-output relationships between task nodes, intermediate products, and callable tools.

[0104] For example, a task orchestration diagram can be represented using the following formula (1):

[0105] in, This represents a task arrangement diagram. Represents a set of nodes. Represents the set of directed dependent edges. Represents a set of intelligent agents. Represents a set of tools.

[0106] The set of nodes can be represented by the following formula (2):

[0107] in, Represents a set of task nodes. Represents a set of agent nodes. Represents a set of tool nodes. This represents the set of intermediate product nodes.

[0108] In a preferred embodiment, the set of task nodes can be represented by the following formula (3):

[0109] in, Indicates the fault analysis task. Indicates the evidence retrieval task. Indicates the fault location task. This indicates the task of generating maintenance measures. This indicates the task of writing a repair plan. Indicates a quality verification task. This indicates a partial rewrite task.

[0110] If task node The output can be used as a task node. Given the input, the directed edge is constructed as shown in the following formula (4):

[0111] The execution status of each task node can be defined using the following formula (5):

[0112] in, Indicates the first The status of each task node.

[0113] S202: Based on task orchestration diagrams and fault data, multiple intelligent agents are scheduled to collaboratively generate maintenance plans for industrial equipment.

[0114] In this embodiment, the system can construct a node scheduling queue based on a task orchestration diagram and start execution from the starting node. For each task node in the task orchestration diagram, the system schedules the corresponding agent to read the corresponding input data and output the corresponding processing result; the output of the previous task node serves as the input of the next task node, enabling multiple agents to collaboratively process the same fault event according to a determined data dependency order. The relevant agents can respectively execute one or more of the following processing functions: fault information analysis, maintenance basis acquisition, fault cause analysis, maintenance content generation, and solution integration output, thereby gradually generating a maintenance plan for the industrial equipment.

[0115] In this embodiment, the processing results of different agents can be transmitted and integrated according to the input-output relationships defined in the task orchestration diagram to form a maintenance plan for the current fault event. Throughout the scheduling process, the system can maintain a node-level context container to record the input data, output data, execution time, and exception information of each task node. After a preceding node completes its task, its output is written to the input area of ​​subsequent nodes after field adaptation. When a node experiences an execution exception or its output is empty, the system can handle the situation according to the exception branch in the task orchestration diagram. Based on this scheduling mechanism, multiple agents collaboratively process the same fault event within the same task instance to generate a maintenance plan for the industrial equipment. This maintenance plan can be stored in the work order system, pushed to the field terminal, or entered into the version management module for further tracking.

[0116] In this embodiment, by incorporating fault data acquisition and the scheduling of multiple agents into the same orchestration framework, the maintenance plan generation process is completed in stages under the constraints of the task orchestration graph, thereby enabling the output of maintenance plans in industrial equipment fault scenarios to have better synergy and manageability.

[0117] In this embodiment, fault data and a pre-set task orchestration diagram can be acquired when industrial equipment malfunctions. Based on the task orchestration diagram and fault data, multiple intelligent agents are scheduled to collaboratively generate a maintenance plan for the industrial equipment. Relying on the pre-set task orchestration diagram, the task nodes, dedicated executing intelligent agents, and the coupling logic of input and output between nodes in the entire maintenance plan generation process can be clearly defined. This breaks down the maintenance plan generation process into multiple task stages with clear relationships, each handled specifically by a corresponding intelligent agent. This avoids problems such as disordered scheduling of multiple intelligent agents, ambiguous task responsibilities, and data interaction errors, ensuring the orderliness of the maintenance plan generation process and the stability of the output. Through this setup, the maintenance plan generation process can be scheduled to be completed in stages by intelligent agents under the constraints of the task orchestration diagram. This not only achieves automatic generation of industrial equipment maintenance plans but also improves the accuracy and reliability of the generated plans, thereby enhancing the overall safety and efficiency of industrial equipment operation and maintenance.

[0118] In one possible implementation, step S202 above, which schedules multiple agents based on a task orchestration diagram and fault data to collaboratively generate a maintenance plan for industrial equipment, may include:

[0119] S11: Standardize the fault data to obtain standardized data.

[0120] S12: Based on the task orchestration graph, schedule multiple agents to collaboratively execute the following tasks:

[0121] S121: Schedule the parsing agent corresponding to the fault parsing task, perform slot-based parsing on the standardized data, and obtain a set of fault phenomenon slots.

[0122] S122: Schedule the retrieval agent corresponding to the evidence retrieval task to perform evidence retrieval on the fault phenomenon slot set and obtain a valid evidence set.

[0123] S123: Schedule the location agent corresponding to the fault location task, and determine the fault location result based on the fault phenomenon slot set and the valid evidence set.

[0124] S124: The intelligent agent that generates the maintenance measures for the scheduled maintenance measures task determines the target maintenance measures based on the fault location results and the set of valid evidence.

[0125] S125: The writing agent corresponding to the scheduling and maintenance plan writing task generates a structured maintenance plan for industrial equipment based on the fault location results and target maintenance measures.

[0126] During the execution of each task, a corresponding task record can be generated. The task record includes one or more of the following: task node, agent, task input, task output, callable tools, quality score, execution time, and version identifier.

[0127] In this embodiment, multiple task nodes may include fault analysis tasks, evidence retrieval tasks, fault location tasks, maintenance measure generation tasks, and maintenance plan writing tasks. The intelligent agents corresponding to each task node may include a scheduling intelligent agent, a parsing intelligent agent, a retrieval intelligent agent, a location intelligent agent, a measure intelligent agent, and a writing intelligent agent. The intermediate products corresponding to each task node include a fault phenomenon slot set, an evidence set, a candidate fault set, and a candidate maintenance measure set.

[0128] For example, Figure 3 This is a schematic diagram of a task orchestration diagram according to an embodiment of this application, as shown below. Figure 3 As shown, the task orchestration diagram can include task nodes, agent nodes, and intermediate product nodes. Task nodes include fault analysis tasks, evidence retrieval tasks, fault location tasks, maintenance measure generation tasks, maintenance plan writing tasks, quality verification tasks, and partial rewriting tasks. Agent nodes include scheduling agents, analysis agents, retrieval agents, location agents, measure agents, writing agents, verification agents, and rewriting agents. Intermediate product nodes include fault slots, evidence sets, candidate fault sets, candidate measure sets, maintenance plans, and verification results.

[0129] In this embodiment, the fault data can be represented using the following formula (6):

[0130]

[0131] in, This indicates the original fault description. Represents device object information. Indicates component information, This indicates an alarm message. Indicates running status information. This indicates historical maintenance records. This represents information from the industrial knowledge base. This indicates the retrieval of evidence.

[0132] The system can standardize the fault data to obtain the standardized data shown in the following formula (7):

[0133]

[0134] in, This represents a standardized processing function used to perform text cleaning, equipment number unification, component name normalization, alarm code mapping, unit standardization, time sequence organization, and invalid field removal.

[0135] In this embodiment, the scheduling agent of the task orchestration graph can assign a corresponding agent to each task node based on the task node type, input data status, and historical execution records.

[0136] The agent scheduling function can be expressed using the following formula (8):

[0137]

[0138] in, Indicates assignment to task nodes intelligent agents, This represents the agent scheduling function. Represents standardized data. Represents task nodes Related historical execution records and intermediate products.

[0139] The intermediate results generated by each agent after performing a task can be represented by the following formula (9):

[0140]

[0141] in, Represents task node The task output, Represents intelligent agents The execution function, Indicates task input, Indicates context constraints, This indicates a callable tool.

[0142] The task record saved by each agent can be represented by the following formula (10):

[0143]

[0144] in, Represents task node Task records, This indicates the quality score of the output. Indicates the execution time or version identifier.

[0145] In this embodiment, the scheduling agent can be used to parse the task status and control the execution order of the task orchestration diagram; the parsing agent can be used to extract equipment objects, fault phenomena, alarm information and operating status from fault data; the retrieval agent can be used to retrieve relevant evidence from the historical maintenance database, equipment knowledge base, fault rule base and maintenance procedures; the location agent can be used to calculate the confidence level of candidate fault locations, fault types and fault causes; the action agent can be used to generate inspection steps, maintenance measures and required tools and spare parts based on the fault location results; and the writing agent can be used to generate maintenance plans according to structured templates.

[0146] In this implementation, by transforming multi-source, heterogeneous fault information into a unified format, the problems of information dispersion, inconsistent fields, and difficulty in organizing maintenance data can be solved, providing standardized input for subsequent agent collaboration and automatic solution generation. The task orchestration diagram constrains the input-output relationships of each agent. Standardized data first enters the parsing agent and then is passed level by level to the retrieval, location, measure generation, and maintenance solution writing stages. This ensures consistent processing results across task nodes, allowing the maintenance solution generation process to be scheduled by agents in stages under the constraints of the task orchestration diagram, improving the accuracy and reliability of automatic maintenance solution generation. Furthermore, the automatic generation of task records each time a task is executed leaves a trace of the task execution process, solving the problems of opaque agent execution, untraceable intermediate results, and difficulty in locating fault responsibility nodes, thus improving data traceability.

[0147] In one possible implementation, the process of performing slot-based parsing of the standardized data in step S121 above to obtain a set of fault phenomenon slots may include:

[0148] S21: Perform slot-based parsing on the standardized data to obtain a structured set of fault phenomenon slots. The set of fault phenomenon slots includes one or more of the following: equipment object slots, component slots, alarm information slots, operating status slots, abnormal phenomenon slots, trend change slots, severity slots, and accompanying phenomenon slots.

[0149] S22: Determine the slot credibility of each slot based on the slot semantic vector and weight vector corresponding to each slot in the fault phenomenon slot set.

[0150] S23: Determine whether the slot credibility of each slot is greater than or equal to the credibility threshold.

[0151] S24: If none of them are greater than or equal to the credibility threshold, then repeat the above steps of slot parsing the standardized data, or output a manual review prompt message.

[0152] In this embodiment, the parsing agent can map entity, attribute, and state information from standardized data to a preset slot template and generate a corresponding semantic vector for each slot. Subsequently, based on the semantic similarity between the slot content and the preset slot template, and combined with the weight vector, a weighted calculation is performed to obtain the slot credibility of each slot.

[0153] In this embodiment, the weight vectors corresponding to each slot can be flexibly set by those skilled in the art according to actual conditions, and no restrictions are imposed here.

[0154] In this embodiment, the confidence threshold can be pre-configured as a fixed value (the specific value can be flexibly set), or dynamically determined based on factors such as device type, fault category, and data source stability.

[0155] In this implementation, when the slot credibility of any slot is lower than the credibility threshold, the system can re-parse the standardized data into slots or output a manual review prompt message so that maintenance personnel can confirm and correct the original text and structured results.

[0156] In this embodiment, if the slot reliability of each slot is greater than or equal to the reliability threshold, then step S122 can be executed.

[0157] In this embodiment, the set of fault phenomenon slots can be represented by the following formula (11):

[0158]

[0159] in, Indicates the slot of the device object. Indicates the component slot. The slot indicates the alarm information. Slot indicating operating status Slots indicating abnormal phenomena Indicates the slot for trend changes. Slots indicating severity The slot indicates the accompanying phenomenon.

[0160] The slot allocation analysis process can be represented by the following formula (12):

[0161]

[0162] in, This represents the slot-based analytical function.

[0163] For each slot The corresponding slot reliability can be expressed using the following formula (13):

[0164]

[0165] in, Indicates slot ( The credibility of ) Represents the semantic vector of the slot. Represents the weight vector. Indicates the bias term. This represents the normalization function.

[0166] In this embodiment, by performing slot-based parsing on standardized data, vague and fragmented fault descriptions can be transformed into structured input, solving the problem of unstable understanding of the original fault descriptions. After slot-based parsing of the standardized data, a credibility threshold can be used to filter the credibility of the parsed slots. For slots with insufficient credibility, corrections can be made through re-parsing or manual review, thereby ensuring consistency and stability of subsequent inputs. Through this setting, reliability screening can be completed before the fault phenomenon slot set enters the subsequent fault location and repair plan generation, resulting in a more complete structured expression of fault phenomena, improving the accuracy and reliability of fault parsing, and thus enhancing the usability of fault analysis results and the stability of plan generation.

[0167] In one possible implementation, the step S122 above, which involves retrieving evidence from the set of faulty slots to obtain a set of valid evidence, may include:

[0168] S31: Retrieve retrieval evidence corresponding to the fault phenomenon slot set from the preset knowledge retrieval base.

[0169] S32: Determine the relevance score of the retrieved evidence based on the text semantic relevance score, component relevance score, alarm relevance score, and historical maintenance similarity score.

[0170] S33: Determine the set of valid evidence based on the search evidence whose relevance scores are greater than or equal to the relevance score threshold.

[0171] In this embodiment, the preset knowledge retrieval base may include, but is not limited to: historical maintenance records, equipment manuals, maintenance procedures, fault knowledge base, alarm rule base, and enterprise knowledge base.

[0172] In this embodiment, the evidence to be retrieved can be text fragments, structured records, or related entries.

[0173] In this embodiment, after obtaining the retrieval evidence, the retrieval evidence can be matched using the fault phenomenon slot set, and text semantic relevance score, component relevance score, alarm relevance score, and historical maintenance similarity score can be generated respectively. Among them, the historical maintenance similarity score can be used to characterize the degree of consistency between the current fault phenomenon and historical maintenance records in terms of phenomenon combination, processing results, and component association.

[0174] In this embodiment, the relevance score can be obtained by weighting the above scores according to preset weights, which can be configured based on factors such as device type, knowledge base distribution, and the stability of the retrieval task.

[0175] In this embodiment, when the relevance score of the retrieved evidence reaches a threshold, it is included in the set of valid evidence and used as the basis for generating subsequent fault location and maintenance measures.

[0176] In this embodiment, the evidence set can be represented by the following formula (14):

[0177]

[0178] in, Indicates the first 1 search result.

[0179] The relevance score can be represented using the following formula (15):

[0180]

[0181] in, Indicates the semantic relevance score of the text. Indicates the component relevance score. This indicates the alarm relevance score. This indicates the score for historical repair similarity. Represents the weighting coefficients, satisfying: .

[0182] The system retains valid evidence with relevance scores greater than or equal to a relevance score threshold (which can be flexibly set), forming a set of valid evidence. .

[0183] In this implementation, by introducing a searchable, recordable, and traceable external knowledge base to generate fault-related retrieval evidence, the risk of large model illusion can be reduced, and the usability and traceability of the retrieval evidence can be improved. After obtaining the retrieval evidence, the evidence is filtered using a relevance score threshold to obtain a set of valid evidence. This ensures that the evidence entering subsequent processing is consistent with and relevant to the current fault phenomenon, improving the reliability and accuracy of the set of valid evidence, thereby providing a more stable supporting foundation for the output results.

[0184] In one possible implementation, the step S123 above, determining the fault location result based on the fault phenomenon slot set and the valid evidence set, may include:

[0185] S41: Based on the set of fault phenomenon slots and the set of valid evidence, construct a set of candidate faults. Each candidate fault in the set of candidate faults includes: candidate fault location, candidate fault type, and candidate fault cause.

[0186] S42: Determine the fault confidence of each candidate fault based on semantic matching score, rule matching score, historical maintenance similarity score, and fault evidence support score.

[0187] S43: Determine the fault location result based on the candidate faults corresponding to the maximum fault confidence.

[0188] In this embodiment, the candidate fault location can be used to indicate the equipment sub-component or functional unit where the abnormality occurs, the candidate fault type can be used to describe whether the abnormality belongs to the category of blockage, wear, looseness, failure or offset, and the candidate fault cause can be used to explain the cause chain that caused the abnormality.

[0189] In this embodiment, each slot in the fault phenomenon slot set can be combined and mapped with each piece of retrieved evidence in the valid evidence set to generate multiple candidate fault entries. Each candidate fault entry can be instantiated from a fault knowledge template in the knowledge base and associated with the corresponding fault location, fault type, and fault cause.

[0190] In this embodiment, the semantic matching score can be used to measure the consistency between the candidate fault and the slot semantics, the rule matching score can be used to measure the conformity between the candidate fault and the preset fault rules, the historical maintenance similarity score can be used to measure the similarity between the candidate fault and historical maintenance cases, and the fault evidence support score can be used to measure the strength of the corroboration of the candidate fault by valid evidence.

[0191] In this embodiment, the scores can be weighted and summarized according to preset weights (which can be flexibly set) to form the fault confidence of the candidate faults, and the candidate fault with the highest confidence is output as the fault location result.

[0192] In this embodiment, the candidate fault set can be represented by the following formula (16):

[0193]

[0194] in, Indicates the first There are 10 candidate faults. Each candidate fault can be represented by the following formula (17):

[0195]

[0196] in, Indicates the candidate fault location. Indicates the candidate fault type. Indicates the candidate fault cause.

[0197] The fault confidence of a candidate fault can be expressed using the following formula (18):

[0198]

[0199] in, Indicates candidate fault Fault confidence, Indicates the semantic matching score. This indicates the score for matching the rule. This represents the score for historical repair similarity. The score indicates that the evidence supports the fault. Represents the weighting coefficients, satisfying: .

[0200] The semantic matching score can be expressed using the following formula (19):

[0201]

[0202] in, Represents the semantic vector of the candidate fault. The semantic vector representing the fault phenomenon slot.

[0203] The rule matching score can be expressed using the following formula (20):

[0204]

[0205] in, Indicates candidate fault The corresponding set of rules, Representation rules Is it a slot set? If satisfied, take the value 1; otherwise, take the value 0.

[0206] In this embodiment, by using structured fault phenomena and retrieved valid evidence together to construct candidate faults, the candidate faults are given clear locations, types and causes. Then, the candidate faults are uniformly quantified through multi-dimensional scoring. Finally, the candidate fault corresponding to the highest fault confidence is selected as the location result, so that the fault location result is consistent with the on-site abnormal information and evidence chain, and provides a stable basis for the generation of subsequent maintenance measures.

[0207] In one possible implementation, the step S124 above, which involves determining the target maintenance measures based on the fault location results and the set of valid evidence, may include:

[0208] S51: Generate a set of candidate maintenance measures based on the fault location results and the set of valid evidence.

[0209] S52: Determine the matching score between each candidate maintenance measure and the fault location result based on the component matching score, fault type matching score, fault cause matching score, and safety constraint score.

[0210] S53: Determine the target maintenance measure based on the candidate maintenance measures corresponding to the maximum matching score.

[0211] In this embodiment, the candidate maintenance measures set can gather multiple alternative maintenance methods related to the fault location results. The methods can correspond to different maintenance actions such as component replacement, parameter adjustment, tightening, cleaning, or shutdown inspection.

[0212] In this embodiment, the component matching score can be used to reflect the degree of consistency between the component acted upon by the candidate maintenance measure and the target component in the fault location result; the fault type matching score can be used to reflect the degree of correspondence between the candidate maintenance measure and the fault type; the fault cause matching score can be used to reflect the degree of targeting of the candidate maintenance measure to the fault cause; and the safety constraint score can be used to reflect whether the candidate maintenance measure meets the constraints such as equipment shutdown, power supply, pressure, temperature, and work permit.

[0213] In this embodiment, after receiving the fault location result and the set of valid evidence, the action agent can call the maintenance action library to generate a set of candidate maintenance actions. The maintenance action library can pre-store component-level maintenance templates, fault type-related handling templates, and fault cause-related maintenance action templates. Based on the fault location result, the action agent retrieves maintenance templates that are associated with the current component, type, and cause from the maintenance action library, and combines the operation items, tool items, and review items in the templates to form multiple candidate maintenance actions.

[0214] Subsequently, the intelligent agent can perform semantic comparison with the target component name, disposal object, action description and fault location result in the candidate maintenance measures, calculate the component matching score, fault type matching score and fault cause matching score, and calculate the safety constraint score by combining the strength of evidence supporting the maintenance action in the effective evidence set. Then, the matching degree score is obtained by comprehensively combining the scores.

[0215] In this embodiment, the comprehensive calculation of the matching score can be achieved by weighted summation. The weights are given by a preset configuration file and can be dynamically adjusted according to device category, fault level, or scenario risk.

[0216] In this implementation, after the matching scores of each candidate maintenance measure are calculated, the action agent can select the candidate maintenance measure with the highest score as the target maintenance measure and output the target maintenance measure to the subsequent maintenance plan writing task. If multiple candidate maintenance measures have the same score, further judgment can be made according to the safety constraint score, evidence support strength, or template priority, so as to ensure that the output result is consistent with the fault location result and the set of valid evidence.

[0217] In this embodiment, the set of candidate maintenance measures can be represented by the following formula (21):

[0218]

[0219] in, Indicates the first One candidate maintenance measure.

[0220] The matching degree of candidate maintenance measures can be expressed by the following formula (22):

[0221]

[0222] in, Indicates candidate maintenance measures Compared with fault location results The degree of matching, Indicates the component matching score. This indicates the score for matching fault types. This indicates the score for matching the cause of the fault. Indicates the safety constraint score. Represents the weighting coefficients, satisfying: .

[0223] When calculating the safety constraint score, if the candidate maintenance measure If the safety constraints are not satisfied, then let: And remove it from the set of optional maintenance measures.

[0224] In this embodiment, the target maintenance measures can be generated by the joint constraints of fault location results, search evidence, and safety conditions, so that the maintenance measures can directly correspond to the faulty component, fault type, and fault cause, and match the on-site operation constraints, thereby improving the accuracy and feasibility of the maintenance plan.

[0225] In one possible implementation, the step S125 above, which generates a structured maintenance plan for industrial equipment based on the fault location results and target maintenance measures, may include:

[0226] S61: Based on the fault phenomena in the standardized data, the fault location results, corresponding fault location, fault type, fault cause, target maintenance measures, inspection steps, re-inspection steps, restoration of operating conditions, and solution field set, generate a fault-measure-solution chain constraint.

[0227] S62: Generate and output structured maintenance solutions for industrial equipment based on the set of valid evidence, the fault-measure-solution chain constraint, and the preset solution template constraint.

[0228] In this embodiment, the fault-measure-solution chain constraint can be used to establish the fault phenomenon, fault location, fault type, fault cause, target maintenance measure, inspection steps, re-inspection steps, and recovery operation conditions into the same logical link, and limit the solution field set to the field range corresponding to the structured output.

[0229] In this embodiment, the inspection step can correspond to the verification of the status of the faulty part, the connection status, or the operating parameters; the re-inspection step can correspond to the functional confirmation or parameter retesting after the maintenance is completed; and the restoration of operating conditions can correspond to the safety status, interlock status, or parameter threshold conditions that the equipment should meet before it is put back into operation.

[0230] In this embodiment, after receiving the fault location result, target maintenance measures, and standardized data, the writing agent reads a preset set of solution fields and organizes the content to be written according to the field names, field order, and field constraints defined in the solution field set. The agent performs semantic extraction on the fault phenomena in the standardized data and consistently associates the fault location, fault type, fault cause, and target maintenance measures. Then, it constructs a fault-measure-solution chain constraint by combining the inspection steps, re-inspection steps, and recovery conditions. Subsequently, the agent performs evidence mapping on the content of each field based on the valid evidence set, filters expressions that match the fault phenomena, location conclusions, and maintenance measures, and forms a structured maintenance plan that can be directly output.

[0231] In this embodiment, the structured maintenance scheme can be written using a fixed field template or output using a key-value structure to facilitate subsequent retrieval, verification, and version management.

[0232] In this embodiment, the fault-response-solution chain constraint can be represented by the following formula (23):

[0233]

[0234] in, Indicates the fault phenomenon. Indicates the location of the fault. Indicates the fault type. Indicate the cause of the malfunction. Indicates the inspection steps, Indicates maintenance measures, This indicates the re-inspection steps. Indicates the conditions for resuming operation. Represents the set of scheme fields.

[0235] The set of scheme fields can be represented by the following formula (24):

[0236]

[0237] in, to These fields respectively represent equipment and component information, fault phenomenon description, fault location and fault type, cause analysis, inspection steps, maintenance and handling measures, required tools or spare parts, re-inspection steps, conditions for resuming operation, risk warnings, and sources of evidence.

[0238] The set of scheme fields for structured maintenance schemes can also be represented by the following formula (25):

[0239]

[0240] in, This represents the scheme generation function. This represents a chain constraint of fault-response-solution. Indicates the set of valid evidence. Indicates the constraints of the scheme template.

[0241] In this embodiment, the writing agent can first form a chain of constraints between fields, and then combine it with a set of valid evidence to complete the structured output. This structured organization method keeps the fields connected and ensures that the content of the solution corresponds one-to-one with the source of evidence, so that the maintenance solution remains stable in terms of field integrity, content consistency and evidence support, thereby outputting a complete, traceable and consistent maintenance solution.

[0242] In one possible implementation, the aforementioned plurality of task nodes may further include a quality verification task, and the method may further include:

[0243] S71: The verification agent corresponding to the scheduling quality verification task performs the following operations: Based on the set of field evidence corresponding to each field in the structured maintenance plan, determine the field evidence support score of the structured maintenance plan.

[0244] S72: Determine the field completeness score, fault location and repair measure accuracy score, and project executability score for the structured maintenance plan.

[0245] S73: Determine the logical consistency score between each field in the structured maintenance plan and the fault-measure-plan chain constraint.

[0246] S74: Determine the overall quality score of the structured maintenance plan based on the accuracy score of fault location and maintenance measures, the field integrity score, the logical consistency score, the field evidence support score, and the engineering executability score.

[0247] In this embodiment, the task nodes in the task orchestration graph may also include quality verification tasks, which may correspond to verification agents. When the verification agent executes the task, it also generates a task record corresponding to the task.

[0248] In this implementation, the field evidence support score can be quantified by the verification agent based on the degree of matching between the field content and the evidence content. The field completeness score can be used to reflect whether the solution fields cover the preset maintenance solution field set. The fault location and maintenance measure accuracy score can be used to reflect whether the fault location, fault type, fault cause, and maintenance measures are consistent with the confirmed fault facts. The engineering executability score can be used to reflect whether the inspection conditions, re-inspection conditions, restoration operation conditions, and resource constraints in the solution meet the on-site execution requirements. The logical consistency score can be used to evaluate whether each field conforms to the chain constraint relationship between the fault, measures, and solution, ensuring that the field content is consistent throughout.

[0249] In this embodiment, the overall quality score can be obtained by normalizing the scores of each item and then weighting and summing them. The weights can be preset or dynamically adjusted according to the type of industrial equipment, the type of fault, and the solution version strategy.

[0250] In this implementation, the verification agent receives the structured maintenance plan output by the writing agent and its corresponding set of field evidence. It extracts the evidence source, content, and constraint relationships for each field, and quantitatively evaluates cases of missing fields, insufficient evidence, conflicting conclusions, or unmet execution conditions, ultimately forming an overall quality score. This score serves as the basis for subsequent plan release, manual review, or partial rewriting, enabling the quality verification task to form a closed-loop collaboration with the preceding fault location, measure generation, and maintenance plan writing tasks.

[0251] In this embodiment, the set of field evidence can be represented using the following formula (26):

[0252]

[0253] in, Representation field The corresponding set of evidence can come from original fault records, alarm information, historical maintenance records, search evidence, fault location results, and intermediate inference results.

[0254] The field evidence support score can be expressed using the following formula (27):

[0255]

[0256] in, Represents the semantic vector of field content. Represents the semantic vector of evidence.

[0257] The overall quality score can be expressed using the following formula (28):

[0258]

[0259] in, The score indicates the accuracy of fault location and repair measures. Indicates the field completeness score. Indicates the logical consistency score. The field indicates the evidence supporting the score. This indicates the score for project feasibility. Represents the weighting coefficients, satisfying: .

[0260] In this embodiment, by jointly verifying the evidence support, completeness, accuracy, logical consistency, and engineering feasibility, the quality results of the structured maintenance plan can be output in a unified scoring format, thereby improving the traceability and reliability of the plan verification.

[0261] In one possible implementation, the aforementioned multiple task nodes may further include a local rewrite task, and the method may further include:

[0262] S81: Schedule the rewriting agent corresponding to the local rewriting task to perform the following operations: Determine whether there exists a target field in the structured maintenance scheme that satisfies any of the following AC conditions:

[0263] A: The field is empty;

[0264] B: The field integrity score is less than the integrity score threshold;

[0265] C: Field evidence support score is 0.

[0266] S82: If the target field does not exist, output the structured maintenance solution directly.

[0267] S83: If a target field exists, determine the target task node corresponding to the target field based on the task orchestration diagram, and determine the task execution record corresponding to the target task node.

[0268] S84: Based on the task execution record, the field evidence set corresponding to the target field, and the fault-measure-solution chain constraint, perform a partial rewrite operation to obtain the rewritten field.

[0269] S85: Determine whether the rewritten field meets the following conditions: the field integrity score is greater than or equal to the integrity score threshold, and the field evidence support score is 1.

[0270] S86: If satisfied, replace the target field with the rewritten field to obtain and output the updated structured maintenance solution.

[0271] S86: If not satisfied, repeat step S83 above, or output a manual review prompt message.

[0272] In this embodiment, the field integrity score can be represented using the following formula (29):

[0273]

[0274] in, Indicates the field completeness score. Indicates the total number of fields. Representation field Whether it is empty, take 1 if it is not empty, otherwise take 0.

[0275] In this embodiment, the integrity score threshold can be flexibly set by those skilled in the art according to actual conditions, and no restrictions are imposed here.

[0276] In this implementation, a field being empty means that no text content has been written into the field; a field completeness score less than the completeness score threshold means that the field lacks necessary semantic elements; and a field evidence support score of 0 means that no effective support relationship has been established between the field and the field evidence set.

[0277] In this embodiment, the rewritten agent can trace back the information on which the target field was generated based on the task execution record, and then combine the field evidence set to complete, correct or restate the field content, and make the rewritten field consistent with the chain constraint.

[0278] In this implementation, the rewritten fields need to be validated again before replacement. The field integrity score is determined by the number of semantic elements covered by the field, the coverage of required information, and the degree of expression closure. The field evidence support score is determined by the degree of matching between the field and the field evidence set. When the rewritten field meets the integrity threshold and the evidence support score is 1, the system writes the field back to the corresponding position in the structured maintenance solution and outputs the updated structured maintenance solution.

[0279] In this embodiment, the rewritten field can be represented using the following formula (30):

[0280]

[0281] in, This indicates the rewritten field. This indicates a locally rewritten function. Represents a set of field evidence. This represents a chain constraint of fault-response-solution. This indicates the task execution record related to the field.

[0282] In this implementation, if the overall quality score is greater than or equal to the preset quality score threshold (which can be flexibly set), and all required fields meet the requirements of completeness, evidence support, and consistency, the system outputs the final industrial equipment maintenance plan.

[0283] In this implementation, when the quality verification agent detects missing fields, insufficient evidence, mismatched measures, or logical conflicts in the solution, the system does not regenerate the entire solution. Instead, it locates the task nodes, intermediate products, and evidence sources corresponding to the erroneous fields based on the task orchestration diagram and calls the local rewriting agent for targeted correction. This setting ensures that local correction only affects the target field, without impacting existing results in other fields. This maintains the original orchestration relationships and traceability of the updated solution and reduces the computational overhead of overall reconstruction.

[0284] The following is a specific embodiment illustrating the automatic generation method for industrial equipment maintenance schemes based on multi-agent systems of this application.

[0285] In a specific embodiment, when an industrial piece of equipment malfunctions, the automatic generation process of a repair plan for that equipment is as follows:

[0286] The first step is to respond to the fault information of industrial equipment and obtain the fault data and pre-set task arrangement diagram of the industrial equipment based on the multi-agent industrial equipment maintenance plan automatic generation system (hereinafter referred to as: the system).

[0287] The second step is to standardize the fault data to obtain standardized data.

[0288] The third step involves the system scheduling the parsing agent corresponding to the fault parsing task based on the task orchestration diagram, performing slot-based parsing on the standardized data, and obtaining a set of fault phenomenon slots.

[0289] The fourth step involves the system scheduling the retrieval agents corresponding to the evidence retrieval tasks based on the task orchestration diagram, performing evidence retrieval on the fault phenomenon slot set, and obtaining a valid evidence set.

[0290] Fifth, the system schedules the location agent corresponding to the fault location task based on the task orchestration diagram, and determines the fault location result based on the fault phenomenon slot set and the valid evidence set.

[0291] The sixth step involves the system generating a corresponding action agent based on the task orchestration diagram and scheduling maintenance measures. The system then determines the target maintenance measures based on the fault location results and the set of valid evidence.

[0292] The seventh step involves the system writing the corresponding intelligent agent for the task based on the task orchestration diagram and the corresponding fault location results and target maintenance measures, thereby generating a structured maintenance plan for the industrial equipment.

[0293] Step 8: The system schedules the verification agents corresponding to the quality verification tasks based on the task orchestration diagram, and determines the overall quality score of the structured maintenance plan based on the accuracy score of fault location and maintenance measures, the field integrity score, the logical consistency score, the field evidence support score, and the engineering executability score.

[0294] In the ninth step, the system determines that the overall quality score is greater than the quality score threshold, and all required fields meet the requirements of evidence support and consistency, but the field integrity score of field A is less than the integrity score threshold.

[0295] Step 10: The system schedules the rewriting agent corresponding to the partial rewriting task based on the task orchestration diagram, determines the target task node corresponding to the target field according to the task orchestration diagram, and determines the task execution record corresponding to the target task node; based on the task execution record, the field evidence set corresponding to the target field, and the fault-measure-solution chain constraint, the partial rewriting operation is performed to obtain the rewritten field; the rewritten field is used to replace the target field to obtain and output the updated structured maintenance solution.

[0296] Figure 4 This is a schematic diagram of the structure of an automatic generation system for industrial equipment maintenance solutions based on multi-agent technology, according to an embodiment of this application. Figure 4 As shown, the automatic generation system for industrial equipment maintenance solutions based on multi-agent technology includes: an acquisition module 41, used to acquire fault data and a pre-set task scheduling diagram in response to fault information of industrial equipment; and a processing module 42, used to schedule multiple agents based on the task scheduling diagram and fault data to collaboratively generate maintenance solutions for industrial equipment; the task scheduling diagram includes multiple task nodes required to generate maintenance solutions, the agents corresponding to the multiple task nodes, and the input-output relationships between the multiple task nodes.

[0297] The automatic generation system for industrial equipment maintenance solutions based on multi-agent technology provided in this application can execute the technical solutions shown in the above method embodiments. Its implementation principle and beneficial effects are similar, and will not be repeated here.

[0298] Figure 5 This is a schematic diagram of the structure of an automatic generation system for industrial equipment maintenance solutions based on multi-agent technology, according to another embodiment of this application. Figure 5As shown, the automatic generation system for industrial equipment maintenance solutions based on multi-agent technology includes: a processor 501 and a memory 502 communicatively connected to the processor 501; the memory 502 stores computer execution instructions; the processor 501 executes the computer execution instructions stored in the memory 502 to implement the steps of the automatic generation method for industrial equipment maintenance solutions based on multi-agent technology in the above-described method embodiments.

[0299] In the aforementioned multi-agent-based automatic generation system for industrial equipment maintenance solutions, the memory 502 and the processor 501 are electrically connected directly or indirectly to enable data transmission or interaction. For example, these components can be electrically connected to each other via one or more communication buses or signal lines, such as a bus connection. The memory 502 stores computer-executable instructions that implement data access control methods, including at least one software functional module that can be stored in the memory 502 in the form of software or firmware. The processor 501 executes various functional applications and data processing by running the software programs and modules stored in the memory 502.

[0300] The memory 502 may be, but is not limited to, Random Access Memory (RAM), Read Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), etc. The memory 502 stores programs, which are then executed by the processor 501 upon receiving execution instructions. Furthermore, the software programs and modules within the memory 502 may include an operating system, which may include various software components and / or drivers for managing system tasks (e.g., memory management, storage device control, power management, etc.) and can communicate with various hardware or software components to provide an operating environment for other software components.

[0301] Processor 501 can be an integrated circuit chip with signal processing capabilities. The aforementioned processor 501 can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor.

[0302] An embodiment of this application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the steps of the various method embodiments of this application.

[0303] An embodiment of this application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the various method embodiments of this application.

[0304] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily essential to this application.

[0305] It should be further noted that although the steps in the flowchart 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 flowchart 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 some of the sub-steps or stages of other steps.

[0306] It should be understood that the above-described device embodiments are merely illustrative, and the device of this application can also be implemented in other ways. For example, the division of units / modules in the above embodiments is only a logical functional division, and there may be other division methods in actual implementation. For example, multiple units, modules, or components may be combined, or integrated into another system, or some features may be ignored or not executed.

[0307] Furthermore, unless otherwise specified, the functional units / modules in the various embodiments of this application can be integrated into one unit / module, or each unit / module can exist physically separately, or two or more units / modules can be integrated together. The integrated units / modules described above can be implemented in hardware or as software program modules.

[0308] In the above embodiments, the descriptions of each embodiment have their own emphasis. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments. The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification.

[0309] Finally, it should be noted that other embodiments of this application will readily conceive of by those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes may be made without departing from its scope.

Claims

1. A method for automatically generating maintenance plans for industrial equipment based on multi-agent systems, characterized in that, include: In response to fault information of industrial equipment, acquire fault data of the industrial equipment and a pre-set task scheduling diagram; Based on the task orchestration diagram and the fault data, multiple intelligent agents are scheduled to collaboratively generate a maintenance plan for the industrial equipment. The task orchestration diagram includes multiple task nodes required to generate the maintenance plan, the intelligent agents corresponding to the multiple task nodes, and the input-output relationships between the multiple task nodes. The multiple task nodes include fault analysis tasks, evidence retrieval tasks, fault location tasks, maintenance measure generation tasks, and maintenance plan writing tasks; the process of scheduling multiple intelligent agents based on the task orchestration diagram and the fault data to collaboratively generate a maintenance plan for the industrial equipment includes: The fault data is standardized to obtain standardized data; Based on the task orchestration graph, multiple agents are scheduled to collaboratively execute the following tasks: The intelligent agent corresponding to the scheduled fault analysis task performs slot-based analysis on the standardized data to obtain a set of fault phenomenon slots. The retrieval agent corresponding to the scheduled evidence retrieval task performs evidence retrieval on the set of fault phenomenon slots to obtain a set of valid evidence. The location agent corresponding to the fault location task is scheduled, and the fault location result is determined based on the fault phenomenon slot set and the valid evidence set. The scheduling and maintenance measures generate a corresponding action agent, and the target maintenance measures are determined based on the fault location results and the set of valid evidence. The writing agent corresponding to the scheduling and maintenance plan writing task generates a structured maintenance plan for the industrial equipment based on the fault location results and the target maintenance measures.

2. The method for automatically generating industrial equipment maintenance plans based on multi-agent systems according to claim 1, characterized in that, Also includes: When executing each task, a task record corresponding to each task is generated; the task record includes one or more of the following: task node, agent, task input, task output, callable tool, quality score, execution time, and version identifier.

3. The method for automatically generating industrial equipment maintenance plans based on multi-agent systems according to claim 1, characterized in that, The standardized data is parsed in a slot-based manner to obtain a set of fault phenomenon slots, including: The standardized data is parsed in a slot-based manner to obtain a structured set of fault phenomenon slots. The set of fault phenomenon slots includes one or more of the following: equipment object slots, component slots, alarm information slots, operating status slots, abnormal phenomenon slots, trend change slots, severity slots, and accompanying phenomenon slots. Based on the slot semantic vector and weight vector corresponding to each slot in the fault phenomenon slot set, the slot credibility of each slot is determined. Determine whether the slot confidence level of each slot is greater than or equal to the confidence level threshold; If none of the data is greater than or equal to the credibility threshold, the step of slot-based parsing of the standardized data is repeated, or a manual review prompt is output.

4. The method for automatically generating industrial equipment maintenance plans based on multi-agent systems according to claim 3, characterized in that, The evidence retrieval of the faulty slot set to obtain a valid evidence set includes: Retrieval evidence corresponding to the set of fault phenomenon slots is obtained from the preset knowledge retrieval base; The relevance score of the retrieved evidence is determined based on the text semantic relevance score, component relevance score, alarm relevance score, and historical maintenance similarity score. The set of valid evidence is determined based on the search evidence whose relevance scores are greater than or equal to the relevance score threshold.

5. The method for automatically generating industrial equipment maintenance plans based on multi-agent systems according to claim 1, characterized in that, The step of determining the fault location result based on the set of faulty slots and the set of valid evidence includes: Based on the set of fault phenomena slots and the set of valid evidence, a set of candidate faults is constructed. Each candidate fault in the set of candidate faults includes: candidate fault location, candidate fault type, and candidate fault cause. The fault confidence of each candidate fault is determined based on semantic matching score, rule matching score, historical maintenance similarity score, and fault evidence support score. The fault location result is determined based on the candidate faults corresponding to the maximum fault confidence.

6. The method for automatically generating industrial equipment maintenance plans based on multi-agent systems according to claim 5, characterized in that, The step of determining the target maintenance measures based on the fault location results and the set of valid evidence includes: Based on the fault location results and the set of valid evidence, a set of candidate maintenance measures is generated; Based on the component matching score, fault type matching score, fault cause matching score, and safety constraint score, determine the matching degree score between each candidate maintenance measure and the fault location result; The target maintenance measure is determined based on the candidate maintenance measures corresponding to the maximum matching score.

7. The method for automatically generating industrial equipment maintenance plans based on multi-agent systems according to claim 6, characterized in that, The step of generating a structured maintenance plan for the industrial equipment based on the fault location results and the target maintenance measures includes: Based on the fault phenomena in the standardized data, the fault location, fault type, fault cause corresponding to the fault location result, the target maintenance measures, inspection steps, re-inspection steps, restoration operating conditions, and solution field set, a fault-measure-solution chain constraint is generated. Based on the set of valid evidence, the fault-measure-solution chain constraint, and the preset solution template constraint, a structured maintenance solution for the industrial equipment is generated and output.

8. The method for automatically generating industrial equipment maintenance plans based on multi-agent systems according to claim 7, characterized in that, The plurality of task nodes also include a quality verification task, and the method further includes: The verification agent corresponding to the quality verification task is scheduled to perform the following operations: Based on the set of field evidence corresponding to each field in the structured maintenance solution, determine the field evidence support score of the structured maintenance solution; Determine the field completeness score, fault location and repair measure accuracy score, and project executability score of the structured maintenance plan; Determine the logical consistency score between each field in the structured maintenance plan and the fault-measure-plan chain constraint; The overall quality score of the structured maintenance plan is determined based on the accuracy score of the fault location and maintenance measures, the field integrity score, the logical consistency score, the field evidence support score, and the project executability score.

9. The method for automatically generating industrial equipment maintenance plans based on multi-agent systems according to claim 8, characterized in that, The multiple task nodes also include a partial rewrite task, and the method further includes: The rewriting agent corresponding to the local rewriting task is scheduled to perform the following operations: Determine whether the structured maintenance solution contains a target field that satisfies any of the following conditions: The field is empty; The field integrity score is less than the integrity score threshold; The field of evidence support scored 0. If a target field exists, the target task node corresponding to the target field is determined according to the task orchestration diagram, and the task execution record corresponding to the target task node is determined. Based on the task execution record, the field evidence set corresponding to the target field, and the fault-measure-solution chain constraint, a partial rewrite operation is performed to obtain the rewritten field. Determine whether the rewritten field meets the following conditions: the field integrity score is greater than or equal to the integrity score threshold, and the field evidence support score is 1; If the conditions are met, the rewritten field is used to replace the target field, and the updated structured maintenance solution is obtained and output.

10. An automatic generation system for industrial equipment maintenance solutions based on multi-agent technology, characterized in that, include: A processor, and a memory communicatively connected to the processor; The memory is used to store computer-executed instructions; The processor is used to execute computer execution instructions stored in the memory, causing the processor to execute the automatic generation method for industrial equipment maintenance scheme based on multiple agents as described in any one of claims 1 to 9.

Images