Nuclear Power Plant Maintenance Procedures, Inspection Methods, and Equipment

Abstract

This application provides a method and apparatus for inspecting nuclear power plant maintenance procedures. The method includes: acquiring the text of the nuclear power plant maintenance procedure to be inspected; extracting structured procedure content information from the maintenance procedure text; performing rule compliance verification on key information in the procedure content information to obtain a first problem list; performing logical reasoning verification on the procedure content information based on a nuclear power domain knowledge base to obtain a second problem list; and generating a quality inspection report based on the first and second problem lists. This embodiment combines rule verification with deep semantic reasoning based on domain knowledge, which helps to improve the automation level of nuclear power plant maintenance procedure inspection, reduce the time consumption of manual review and individual judgment differences; by introducing a nuclear power domain knowledge base for logical reasoning verification, it can effectively supplement the shortcomings of rule verification in identifying implicit logical risks across paragraphs and chapters; thereby improving the efficiency and reliability of nuclear power plant maintenance procedure quality control.

Description

Technical Field

[0001] This application belongs to the field of nuclear power operation and maintenance technology, and in particular relates to a method and device for checking nuclear power maintenance procedures. Background Technology

[0002] Currently, quality inspection of nuclear power plant maintenance procedures mainly relies on a "human-led + traditional tool-assisted" model. Manual review is the mainstream method, with domain experts meticulously reviewing procedural documents word by word to ensure compliance with technical standards and safety regulations. In some scenarios, rule engines based on fixed keyword matching are used to supplement this approach, intercepting basic terminology errors.

[0003] However, manual review methods are time-consuming and inconsistent. Different experts may disagree on the same term or security threshold, making standardization difficult. Traditional rule engine-based review methods can only handle explicit semantic issues and lack the ability to identify implicit logical risks (such as causal inversion, missing conditions, and conflicting actions) across paragraphs and chapters. Summary of the Invention

[0004] This application provides a method and apparatus for checking nuclear power plant maintenance procedures, which can solve the current problems of long review times and inability to detect deep-seated logical risks in nuclear power plant maintenance procedures.

[0005] In a first aspect, embodiments of this application provide a method for inspecting nuclear power plant maintenance procedures, including: Obtain the text of the nuclear power plant maintenance procedures to be inspected; Structured procedure content information is extracted from the maintenance procedure text; the procedure content information includes at least one of the following: operating steps, prerequisites, safety constraints, equipment associations, process parameters, and terminology; The key information in the procedure content is checked for rule compliance to obtain a first problem list; the key information includes at least one of the following: target terminology, process description information, safety step identifier, equipment name, and operation action keywords; Logical reasoning and verification of the procedure content information are performed based on a nuclear power field knowledge base to obtain a second question list; the nuclear power field knowledge base includes at least one of the following: historical cases from the work experience feedback system, nuclear power industry standards and specifications, program feedback records from the program feedback system, and important feedback knowledge. A quality inspection report is generated based on the first problem list and the second problem list.

[0006] In one possible implementation of the first aspect, the rule compliance verification of key information in the procedure content information is performed to obtain a first problem list, including: The procedure content information is input into the first large model to extract the key information; The first large model is used to analyze the context information of the key information in the procedure content information to determine the rule entry corresponding to the key information in the business rule base. The key information is matched with the corresponding rule entries, and the key information that violates the rules is included in the first problem list.

[0007] In one possible implementation of the first aspect, the method further includes: A nuclear power plant maintenance procedure knowledge graph is obtained, which includes multiple nodes. Each node represents a specific operation step, prerequisite, safety constraint, equipment, process parameter or term in a verified maintenance procedure. The logical relationship between the nodes is represented by edges. The first large model queries the nuclear power plant maintenance procedure knowledge graph to find the target nodes associated with the procedure content information, and determines the logical conflict issues of the procedure content based on the logical relationships between the target nodes, and includes the logical conflict issues in the first problem list.

[0008] In one possible implementation of the first aspect, the method further includes: Acquire multimodal data associated with the maintenance procedure, the multimodal data including at least one of equipment drawings, operation and maintenance videos, and process flow diagrams; Feature extraction and structured parsing are performed on the multimodal data. The parsed multimodal data and the procedure content information are semantically aligned and logically cross-validated through a multimodal large model to identify cross-modal issues that are inconsistent between the procedure content information and the multimodal data. The cross-modal issues are then included in the first issue list.

[0009] In one possible implementation of the first aspect, the logical reasoning verification of the procedure content information based on a nuclear power field knowledge base yields a second problem list, including: Input the process parameters and their contextual information in the maintenance procedure text into the second major model; The second major model is used to perform semantic parsing on the process parameters to obtain parsing prompt words; different parsing prompt words correspond to different nuclear power field knowledge bases. In the nuclear power field knowledge base corresponding to the parsing prompt words, domain knowledge related to the context information corresponding to the currently parsed set process is retrieved, and the retrieved domain knowledge is input into the second large model to assist the second large model in performing logical reasoning verification to obtain the second question list.

[0010] In one possible implementation of the first aspect, generating a quality inspection report for the maintenance procedure based on the first problem list and the second problem list includes: The third major model is used to determine the related issues between the first question list and the second question list; If there is no contradiction among the related issues, include any one or all of the related issues in the comprehensive issue list; In the event of contradictions in the related issues, the third major model is used to analyze the contextual information of the related issues in the maintenance procedure text, determine the root cause of the related issues, and include it in the comprehensive issue list. Based on the comprehensive problem list, the quality inspection report is generated.

[0011] In one possible implementation of the first aspect, the quality inspection report further includes: the risk level of each problem in the comprehensive problem list, the rectification priority corresponding to the risk level, and the standardized rectification strategy.

[0012] In one possible implementation of the first aspect, the risk level of each problem in the comprehensive problem list is determined based on a risk quantification assessment system, which includes multi-dimensional quantitative indicators, a weighted scoring algorithm, and risk level classification standards; the multi-dimensional quantitative indicators include the degree of safety impact, probability of operational failure, compliance and non-compliance level, and on-site execution risk. Based on the aforementioned risk quantification assessment system, the risk level of each issue in the comprehensive issue list is determined, specifically including: Based on the multidimensional quantitative indicators and the type, location and scope of influence of each problem, each problem is scored item by item to obtain the sub-score of each problem on each quantitative indicator; The individual scores are input into the weighted scoring algorithm to calculate the comprehensive risk score for each question. The comprehensive risk score is compared with the risk level classification standard to obtain the risk level corresponding to each problem.

[0013] In one possible implementation of the first aspect, the procedure content information further includes chapters, paragraphs, and line numbers; the method further includes: Mark the chapter, paragraph, and line number of each question in the comprehensive question list to obtain the question location information; The problem location information will be included in the quality inspection report.

[0014] Secondly, embodiments of this application provide a nuclear power plant maintenance procedure inspection device, comprising: The acquisition module is used to acquire the text of the nuclear power plant maintenance procedures to be inspected; The extraction module extracts structured procedure content information from the maintenance procedure text; the procedure content information includes at least one of the following: operating steps, prerequisites, safety constraints, equipment associations, process parameters, and terminology; The rule verification module is used to perform rule compliance verification on key information in the procedure content information to obtain a first problem list; the key information includes at least one of the following: target terminology, process description information, safety step identifier, equipment name, and operation action keywords. The knowledge base verification module is used to perform logical reasoning verification on the content information of the procedure based on the nuclear power field knowledge base to obtain a second question list; the nuclear power field knowledge base includes at least one of the following: historical cases from the work experience feedback system, nuclear power industry standards and specifications, program feedback records from the program feedback system, and important feedback knowledge. The report generation module is used to generate a quality inspection report based on the first problem list and the second problem list.

[0015] Thirdly, embodiments of this application provide an electronic device, including: a processor coupled to a memory, wherein the processor executes a computer program or instructions stored in the memory to implement the method as described in the first aspect.

[0016] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method as described in any one of the first aspects.

[0017] Fifthly, embodiments of this application provide a computer program product that, when run on an electronic device, causes the electronic device to perform the method described in any one of the first aspects above.

[0018] It is understood that the beneficial effects of the second to fifth aspects mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here.

[0019] The beneficial effects of this embodiment compared to the prior art are as follows: This embodiment extracts structured procedure content information from the maintenance procedure text and performs rule compliance verification and logical reasoning verification on the procedure content information respectively, thereby obtaining an explicit first problem list and an implicit second problem list, and then generating a quality inspection report. This method combines traditional rule verification with deep semantic reasoning based on domain knowledge, which helps to improve the automation level of nuclear power maintenance procedure inspection, reduce the time consumption of manual review and individual judgment differences; in addition, by introducing a nuclear power domain knowledge base for logical reasoning verification, it can effectively supplement the insufficient ability of rule verification to identify implicit logical risks across paragraphs and chapters, and reduce operational safety hazards caused by program defects; thus improving the efficiency and reliability of nuclear power maintenance procedure quality control. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a software architecture diagram of a nuclear power plant maintenance procedure inspection method system provided in an embodiment of this application; Figure 2 This is a flowchart illustrating a nuclear power plant maintenance procedure inspection method provided in one embodiment of this application; Figure 3 This is a flowchart illustrating a nuclear power plant maintenance procedure inspection method provided in another embodiment of this application; Figure 4 This is a flowchart illustrating a nuclear power plant maintenance procedure inspection method provided in another embodiment of this application; Figure 5 This is a flowchart illustrating a nuclear power plant maintenance procedure inspection method provided in another embodiment of this application; Figure 6 This is a flowchart illustrating a nuclear power plant maintenance procedure inspection method provided in another embodiment of this application; Figure 7 This is a flowchart illustrating a nuclear power plant maintenance procedure inspection method provided in another embodiment of this application; Figure 8 This is a flowchart illustrating a nuclear power plant maintenance procedure inspection method provided in another embodiment of this application; Figure 9 This is a flowchart illustrating a nuclear power plant maintenance procedure inspection method provided in another embodiment of this application; Figure 10 This is a schematic diagram of the structure of the nuclear power plant maintenance procedure inspection device provided in the embodiments of this application. Detailed Implementation

[0022] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0023] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0024] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0025] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrases "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0026] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0027] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0028] Nuclear power plant maintenance procedures are detailed technical documents used to guide maintenance personnel in performing inspections, maintenance, repairs, and tests of nuclear power plant equipment in a safe, standardized, and traceable manner. They typically include operating procedures, prerequisites, safety constraints, equipment relationships, process parameter standards, and definitions of technical terms. The aim is to ensure that maintenance activities are completed using standardized processes while meeting nuclear safety requirements, preventing equipment damage or safety accidents caused by human error or procedural defects.

[0029] In the nuclear power industry, the quality of maintenance procedures is directly related to nuclear safety. Due to the wide variety of equipment, stringent process requirements, and strict safety level classifications in nuclear power plants, each procedure needs to undergo rigorous review to ensure that it is logically correct, the parameters are quantifiable, the safety procedures are complete, and it is consistent with on-site drawings and operation videos.

[0030] Currently, quality inspection of nuclear power plant maintenance procedures mainly relies on a "human-led + traditional tool-assisted" model. Manual review is the mainstream method, with domain experts meticulously reviewing procedural documents word by word to ensure compliance with technical standards and safety regulations. In some scenarios, rule engines based on fixed keyword matching are used to supplement this approach, intercepting basic terminology errors.

[0031] However, existing technologies have significant and systemic shortcomings: (1) Efficiency and consistency bottlenecks: With thousands of existing maintenance procedures, manual review takes a very long time (an average of 2 hours per procedure). More importantly, due to differences in experience and understanding, different experts may have inconsistent judgments on the same terminology or safety threshold, resulting in inconsistent procedure quality and making it difficult to achieve true standardization.

[0032] (2) Blind spots in identifying implicit logical risks: Manual review and traditional rule engines mainly focus on explicit semantic issues (such as terminology errors) and lack the ability to understand the deep logical relationships between operation steps. For implicit logical contradictions such as "reversal of cause and effect", "missing conditions" or "action conflicts" across paragraphs and chapters, existing technologies are almost unable to effectively identify them, which constitutes a potential security risk.

[0033] (3) Low efficiency of problem rectification loop: The existing inspection process stops at "problem discovery" and cannot provide "precise positioning" and "executable optimization solutions". Experts need to manually mark the problem location and rely on their own experience to replace terms and reorganize logic, which is a high cost of secondary manual labor and is easy to introduce new errors.

[0034] (4) Lack of multimodal data verification: Existing technologies only check text-based procedures and cannot cross-verify with multimodal data such as equipment drawings, operation and maintenance videos, and process flow diagrams. This can easily lead to inconsistencies between text descriptions and actual equipment and processes.

[0035] (5) No quantitative standard for risk level: There is no unified risk quantification model after defect identification, and the priority of rectification depends on manual judgment, which is prone to problems such as delayed handling of major risks and unreasonable allocation of resources.

[0036] (6) Weak cross-document logical connections: There is a lack of structured relational modeling of procedural elements, and it is difficult to visualize and systematically identify logical conflicts and dependencies across chapters and documents.

[0037] The purpose of this application's embodiments is to overcome the shortcomings of the prior art described above. See also... Figure 1 , Figure 1 The diagram illustrates a software architecture of a nuclear power plant maintenance procedure inspection system disclosed in an embodiment of this application. The system includes an input layer, a multimodal rule engine layer, a knowledge base and RAG layer, a large model collaborative processing center, and an output layer.

[0038] (a) Input layer The input layer is used to acquire the nuclear power plant maintenance procedure text to be inspected (supporting Word, PDF, and other formats) and parse it: removing irrelevant information such as document format, approval process, and record forms, and extracting structured procedure content information, including operating steps, prerequisites, safety constraints, equipment associations, process parameters, terminology, and chapter / paragraph / line numbers. Simultaneously, if related equipment drawings, maintenance videos, or process flow diagrams exist, they are also collected and transmitted to the next layer.

[0039] (ii) Multimodal rule engine layer This layer is responsible for performing deterministic verification based on preset verification rules and multimodal fusion verification, and outputting the first problem list (basic problem list). It specifically includes three sub-modules: The rule verification submodule inputs the procedure content information into the first main model, extracts key information (target terminology, process description information, safety step identifiers, equipment names, operation action keywords, etc.), determines the corresponding rule entries based on the context, and matches them, adding key information with violations to the first issue list. For example, whether the inspection terminology is standard, whether the process description contains unquantifiable terms such as "installed in place," whether safety steps are missing, whether the equipment name is standardized, and whether the operation action includes quantifiable parameters.

[0040] Knowledge Graph Query Submodule: Obtains a pre-built nuclear power plant maintenance procedure knowledge graph (nodes map to operational elements in verified procedures, edges represent logical relationships), queries the graph for target nodes associated with the current procedure content information through the first main model, detects logical conflicts across chapters and documents based on the logical relationships between nodes, and includes the conflict in the first problem list.

[0041] Multimodal fusion verification submodule: Acquires multimodal data (equipment drawings, operation and maintenance videos, process flow diagrams) associated with maintenance procedures, performs feature extraction and structured parsing, uses a multimodal large model to perform semantic alignment and logical cross-validation between the parsed multimodal data and the procedure content information, identifies inconsistencies between text and multimodal data (such as bolt head type in text not matching the drawing annotation), and includes such cross-modal issues in the first issue list.

[0042] (III) Knowledge Base and RAG Layer This layer is responsible for logical reasoning verification based on the nuclear power domain knowledge base, outputting a second problem list (a list of implicit problems). The nuclear power domain knowledge base pre-integrates historical cases from the work experience feedback system, nuclear power industry standards and specifications, program feedback records from the program feedback system, and important feedback knowledge. During operation, the process parameters and their context from the procedure content information are input into the second main model. The second main model performs semantic parsing on the set procedures (such as bolt tightening, impedance measurement, etc.) in the process parameters, generating parsing prompt words. Based on the prompt words, it locates the corresponding knowledge base partition (such as mechanical and electrical sub-libraries), retrieves the domain knowledge (standard values, historical cases, etc.) most relevant to the current context, and injects the retrieved domain knowledge into the second main model to assist in logical reasoning verification, ultimately generating a second problem list, which includes implicit risk issues such as causal inversion, missing conditions, and parameter deviations.

[0043] (iv) Large Model Collaborative Processing Center The center employs a chain-based call and step-by-step reasoning mechanism to perform a weighted comprehensive judgment on the first and second problem lists, outputting a comprehensive problem list and the confidence level of each problem. Specifically, the first and second problem lists, along with the original maintenance procedure text, are input into a third major model. The third major model first identifies related problems between the two lists (problem pairs pointing to the same operation or equipment). If there are no contradictions among the related problems, any one (or all) of them is included in the comprehensive problem list. If there are contradictions, the contextual information of the related problems in the procedure text is analyzed to determine the root cause and include it in the comprehensive problem list. Simultaneously, the confidence level (e.g., high, medium, low) of each problem in the list is marked to reflect the reliability of the model's judgment.

[0044] (v) Output layer The output layer is used to generate the final quality inspection report and drive the rectification closed loop. Specifically, it includes: Report Generation: Based on a comprehensive list of issues, a structured report is generated. Each issue in the report includes three-dimensional coordinate location information (chapter, paragraph, line number), and the risk level (high / medium / low / alert) of each issue is automatically calculated according to a pre-built risk quantification assessment system, outputting the corresponding rectification priority and standardized rectification strategy.

[0045] Automatic rectification interface: For standardized rectification items such as terminology replacement, the rectification suggestions are automatically executed in the production system (such as the work package system or experience feedback system) through RPA scripts or API interfaces to achieve automated rectification.

[0046] Knowledge accumulation loop: New solutions and rules that are confirmed to be effective during the rectification process are injected into the nuclear power field knowledge base through a preset process, so that the system has the ability to continuously self-optimize.

[0047] Through the above five layers of collaborative work, this system realizes a closed-loop process from document input, multi-dimensional verification, comprehensive judgment, report generation to automatic rectification and knowledge feedback, improving the efficiency, comprehensiveness and automation of nuclear power plant maintenance procedure inspection.

[0048] See Figure 2 , Figure 2 A flowchart illustrating the nuclear power plant maintenance procedure inspection method disclosed in this application is shown. The method includes the following steps: S110: Obtain the text of the nuclear power plant maintenance procedure to be inspected.

[0049] Specifically, the nuclear power plant maintenance procedure text to be inspected can be a text file in Word or PDF format.

[0050] S120: Extract structured procedure content information from maintenance procedure text; the procedure content information includes at least one of the following: operating procedures, prerequisites, safety constraints, equipment associations, process parameters, and terminology.

[0051] The maintenance procedure text is parsed and processed to remove irrelevant information such as document format, approval process, and record form, and the structured procedure content information is extracted from it.

[0052] The procedure content includes at least one of the following: operating steps, prerequisites, safety constraints, equipment associations, process parameters, and terminology. Among them, equipment associations refer to the names, identifiers, or codes of specific equipment or components that are clearly recorded in the maintenance procedure and are related to the operating steps. These are used to determine the logical relationship between the operating objects and the equipment, such as "fan casing", "relay K1", and "pipeline valve V-01".

[0053] For example, the operating steps extracted from a wind turbine maintenance procedure include "turn off the wind turbine power supply" and "measure the motor winding resistance", with the prerequisite being "the wind turbine has been stopped for more than 10 minutes", the safety constraint being "wear insulated gloves", the equipment association being "wind turbine motor M-101", the process parameter being "torque 55 N·m", and the terminology including "cold commissioning".

[0054] S130: Perform rule compliance verification on key information in the procedure content to obtain the first problem list; key information includes at least one of the following: target terminology, process description information, safety procedure identification, equipment name, and operation action keywords.

[0055] This embodiment can be run based on preset verification rules (such as standard terminology comparison, keyword matching of unquantifiable process requirements, comparison of essential safety steps list, and equipment name standardization verification).

[0056] Specifically, key information in the procedure content is matched against these preset rules: for example, whether the testing terminology is a standard expression; checking whether the process description contains non-quantifiable expressions such as "installed in place" or "connected securely"; confirming whether essential safety steps (such as "power off" or "pressure relief") are missing; verifying whether the equipment name is consistent with the specification; and determining whether the key words of the operation action (such as "tighten") are accompanied by necessary quantitative parameters. Based on the matching results, key information containing violations is recorded as the first problem list. For example, if the procedure states "tighten bolts" but does not provide a torque value, then this problem is included in the first problem list.

[0057] S140: Logical reasoning and verification of the procedure content information based on the nuclear power field knowledge base to obtain the second question list.

[0058] The nuclear power knowledge base includes at least one of the following: historical cases from the work experience feedback system, nuclear power industry standards and specifications, process feedback records from the process feedback system, and important feedback knowledge.

[0059] This step involves comparing and analyzing the procedure's content with historical cases and standards in the knowledge base to identify potential hidden logical risks, such as causal inversions across paragraphs or chapters, missing conditions, and conflicting actions. For example, for a procedure that includes bolt tightening operations, the knowledge base may store standard values ​​for tightening torque of similar equipment and historical cases of malfunctions caused by substandard torque. By logically comparing the description in the procedure with this knowledge, it can be determined whether there is a risk of missing torque parameters or deviations from the standard. These identified hidden risks are recorded in the second problem list.

[0060] S150: Generate a quality inspection report based on the first and second problem lists.

[0061] Specifically, based on the first and second problem lists mentioned above, a quality inspection report for the maintenance procedure is generated. This report summarizes the problems found in the two different dimensions of verification and can be used for subsequent review or rectification.

[0062] This embodiment automates the inspection process, significantly reducing the inspection time for a single nuclear power plant maintenance procedure from 2 hours to less than 10 minutes, improving efficiency by more than 12 times.

[0063] This embodiment extracts structured procedure content information from the maintenance procedure text and performs rule compliance and logical reasoning checks on the procedure content information to obtain an explicit first problem list and an implicit second problem list, thereby generating a quality inspection report. This method combines traditional rule verification with domain-knowledge-based deep semantic reasoning, which helps improve the automation level of nuclear power maintenance procedure inspection, reduces the time consumption of manual review and individual judgment differences. Furthermore, by introducing a nuclear power domain knowledge base for logical reasoning verification, it effectively supplements the shortcomings of rule verification in identifying implicit logical risks across paragraphs and chapters, reducing operational safety hazards caused by program defects; thus improving the efficiency and reliability of nuclear power maintenance procedure quality control.

[0064] In some embodiments, see Figure 3 S130 specifically includes the following steps: S131: Input the procedure content information into the first major model to extract key information.

[0065] For example, the first major model could be an existing large language model such as Deepseek or GPT, which uses prompt words to extract key information from the procedure content.

[0066] S132: Analyze the context of key information within the procedure content information using the first major model to determine the corresponding rule entries in the business rule base.

[0067] A business rule base is a set of predefined rules used to verify the compliance of maintenance procedures. It is primarily used to efficiently identify explicit issues such as non-standard terminology, missing or incorrect essential safety procedures, unquantified process parameter descriptions, and incorrect equipment names. Its content typically includes a standard terminology library, a non-quantifiable process requirement rule library, an essential safety procedure rule library, an equipment name specification library, and a format specification sub-library.

[0068] After extracting the key information, the first major model further analyzes the contextual information of each key information in the procedure content information (such as the sentence containing the key information and several words before and after it) to determine the corresponding rule entry in the preset business rule base.

[0069] A rule entry is the smallest unit of judgment in the business rule base. It is used to define a specific verification condition and judgment criterion, such as "the terminology must be a standard terminology" or "the tightening operation should provide a quantified torque value".

[0070] For example, when the key information "tightening" is extracted, the first big model combines the contextual words such as "torque wrench" and "55 N·m" to determine that the rule entry corresponding to the key information is "tightening operation requires a quantified torque value"; if the procedure only writes "tightening" without giving a torque value, then the contextual information lacks a quantified parameter, and the first big model will match the rule entry "torque parameter missing".

[0071] Table 1 below shows examples of keywords for unquantifiable process requirements in the business rule base and their corresponding validation logic, which are used to help the rule engine identify problems where specific quantitative parameters are lacking in the procedures.

[0072] Table 1: Example of an event rule engine

[0073] S133: Match key information with corresponding rule entries and include key information that violates the rules in the first problem list.

[0074] The first major model matches key information with corresponding rule entries to determine if any violations exist. If a match fails (e.g., terminology inconsistent with the standard terminology database, missing safety procedure identifiers, or non-standard equipment names), the key information of the violation and its contextual location are added to the first problem list. For example, for a description of "tightening bolts" without a torque value, the first major language model marks this key information as "process parameter description not quantified," records the problem and its location, and ultimately forms the first problem list.

[0075] This embodiment uses a first-level model to sequentially extract key information, analyze context to determine rule entries, match and output violations. This process, currently based on manual rule verification, is handled by the first-level model, improving execution efficiency and automation of the entire inspection process. Furthermore, the large model's strong generalization ability allows rule verification to adapt to different versions or similarly worded procedural texts, reducing missed detections due to terminology variations or differences in expression.

[0076] In some embodiments, the nuclear power plant maintenance procedure inspection method may also incorporate a nuclear power plant maintenance procedure knowledge graph for logical conflict detection. For details, see [link to documentation]. Figure 4 The method also includes the following steps: S134: Obtain a nuclear power plant maintenance procedure knowledge graph. The nuclear power plant maintenance procedure knowledge graph includes multiple nodes. Each node represents a specific operation step, prerequisite, safety constraint, equipment, process parameter or term in a verified maintenance procedure. The logical relationship between nodes is represented by edges. Specifically, the knowledge graph construction process includes: extracting core elements such as operation steps, prerequisites, safety constraints, equipment associations, and process parameters from the already inspected maintenance procedures, and abstracting them into nodes of the knowledge graph; secondly, determining the logical dependencies between nodes (such as sequence, conditions, conflicts, etc.) through relational modeling, and establishing corresponding edges; finally, performing attribute annotation, adding necessary attribute information (such as the chapter to which the node belongs, document source, etc.) to the nodes, thereby forming a cross-chapter and cross-document full-domain procedure logical association network.

[0077] Thus, each node in the knowledge graph represents a specific operational element in a verified historical maintenance procedure, such as the operation step "turn off the fan power", the precondition "the system is powered off", the safety constraint "wear insulated gloves", the equipment "fan motor M-101", the process parameter "torque 55 N·m", or the term "cold commissioning". The logical relationships between nodes are represented by directed edges, such as sequential dependency (step A → step B), conditional triggering (precondition → operation step), and conflict / mutual exclusion (two operations cannot be executed simultaneously).

[0078] S135: Using the first major model, query the target nodes associated with the content information of the nuclear power maintenance procedures from the knowledge graph, determine the logical conflict issues of the procedure content based on the logical relationships between the target nodes, and include the logical conflict issues in the first problem list.

[0079] When performing rule compliance checks, the first main model uses key information as query conditions to retrieve the associated target nodes in the knowledge graph. For example, if there is an operation step "tightening bolts" in the current procedure, the model uses it as a query term, finds the corresponding target node "tightening bolts" in the knowledge graph, and further obtains related nodes that have logical relationships with this node, such as the precondition "bolts have been placed in place" and the subsequent step "check torque".

[0080] The first major model compares the retrieved target nodes and their logical relationships with the corresponding elements in the current procedure's content. If it finds that the current procedure lacks necessary preconditions, has an inverted order of steps, or matches a known conflict pattern in the knowledge graph, a logical conflict is identified. For example, if the knowledge graph records that "cleaning threads" must be performed before "tightening bolts," but the current procedure does not include this step, it is identified as a logical conflict. Finally, this logical conflict is added to the first problem list.

[0081] This embodiment introduces a pre-constructed nuclear power plant maintenance procedure knowledge graph. Since the knowledge graph stores a large number of operational elements and their logical relationships from verified procedures, it provides rich reference benchmarks for the current procedure, helping to improve the ability to identify cross-step and cross-chapter logical conflicts. The first major language model is used to query target nodes associated with the current procedure content information, and logical conflict issues are identified based on the logical relationships between nodes. This expands the detection capability of the first problem list, which originally could only detect explicit rule issues, to include some logical problems, thereby broadening the coverage of rule-level verification. Finally, because the nodes and relationships in the knowledge graph are explicitly defined, the identified logical conflict issues have good interpretability, making them easier for reviewers to understand and confirm.

[0082] In some embodiments, see Figure 5 The nuclear power plant maintenance procedure inspection method also includes the following steps: S136: Obtain multimodal data associated with maintenance procedures, including at least one of equipment drawings, maintenance videos, and process flow diagrams.

[0083] Optionally, multimodal data includes at least one of the following: two-dimensional or three-dimensional drawings of the equipment (such as fan structure diagrams and piping / instrumentation diagrams), operation and maintenance videos (such as maintenance videos of similar equipment), and process flow diagrams. For example, for a fan bearing replacement procedure, the system simultaneously acquires the fan's three-dimensional assembly drawings and a standard operation video.

[0084] Optionally, the multimodal data and nuclear power plant maintenance procedure text can be derived from a nuclear power plant maintenance procedure document, which can be in Word or PDF format.

[0085] S137: Perform feature extraction and structured parsing on multimodal data, and use a multimodal large model to perform semantic alignment and logical cross-validation between the parsed multimodal data and the procedure content information to identify cross-modal issues that are inconsistent between the procedure content information and the multimodal data, and include the cross-modal issues in the first problem list.

[0086] Optionally, large multimodal models can use the QwenVL open-source model.

[0087] After the parsed multimodal data and procedure content information are input into the multimodal large model, the multimodal large model performs semantic alignment and logical cross-validation. Specifically, the multimodal large model performs cross-modal matching between the operation steps, equipment names, and process parameters in the procedure text and the component annotations in the drawings, action clips in the videos, and process nodes in the flowcharts.

[0088] For example, the specification text states "use an internal hex wrench to loosen the fan housing bolts," while the drawings indicate that the fan housing bolts are external hex heads, resulting in inconsistencies in the multimodal large model's judgment. Another example is that the specification requires "tightening torque of 55 N·m," while the actual operation in the video shows the use of a regular wrench with no torque control, leading to a cross-modal conflict identified by the multimodal large model.

[0089] Finally, the identified inconsistencies between the procedural text and the multimodal data are included in the first issue list as supplementary basic issues for subsequent report generation. For example, the aforementioned issue of inconsistent bolt head type will be recorded as "Inconsistent between equipment drawings and text descriptions," and the issue of missing torque control will be recorded as "Inconsistent between practical operation videos and text requirements."

[0090] Traditional checks are limited to internal text consistency, while this embodiment extends the verification scope to cross-modal comparison between text and multimodal data. This helps to identify potential inconsistencies caused by drawing revisions, delayed video updates, or textual errors, thereby reducing the risk of misoperation due to information conflicts during on-site operations. Secondly, using a large multimodal model for semantic alignment can overcome the comprehension barriers caused by differences in expression forms between different modalities to a certain extent, improving the accuracy of cross-modal problem identification. Therefore, by introducing semantic alignment and logical cross-validation between multimodal data (drawings, videos, flowcharts) and specification text, this embodiment can identify inconsistencies between specification text and drawing annotations, actual operation videos, and process flows.

[0091] In some embodiments, see Figure 6 S140 specifically includes the following steps: S141: Input the process parameters and their contextual information in the maintenance procedure text into the second major model.

[0092] Optionally, the second major model can be the same as the first major model, both using the deepseek model or the GPT model, etc.

[0093] S142: The second major model performs semantic parsing on the process parameters to obtain parsing prompts; different parsing prompts correspond to different nuclear power knowledge bases.

[0094] "Designated procedures" refer to specific technological processes that require specialized knowledge verification, typically with clearly defined quantitative parameters and industry standards. Examples include torque verification for bolt tightening and impedance verification for electrical testing. These designated procedures have a cue word list (different fields have different cue word lists). When the second-level model parses these cue words, it triggers corresponding knowledge base retrieval and logical reasoning verification. Procedures like "on-site cleanup," which do not involve quantitative parameters or complex specifications, are not considered designated procedures.

[0095] The second major model performs semantic parsing on the parts of the process parameters that belong to the set procedures, generating corresponding parsing prompts. Different categories of set procedures correspond to different parsing prompts; for example, the "tightening" operation generates the prompt "search torque standard," and "impedance measurement" generates the prompt "search electrical specifications."

[0096] The corresponding nuclear power knowledge base section is determined based on the parsing prompts. The nuclear power knowledge base is pre-divided into multiple sub-bases according to process categories, such as a mechanical sub-base (stores standards for torque, clearance, etc.), an electrical sub-base (stores standards for impedance, insulation, etc.), and an instrumentation and control sub-base. In this step, the parsing prompt "search for torque standards" will point to the mechanical sub-base.

[0097] S143: In the nuclear power field knowledge base corresponding to the parsing prompt words, retrieve the field knowledge related to the context information corresponding to the currently parsed set process, and input the retrieved field knowledge into the second large model to assist the second large model in logical reasoning and verification, and obtain the second question list.

[0098] Within the corresponding knowledge base sub-base, domain knowledge related to the context information of the currently parsed set process is retrieved. For example, if the current context information includes "wind turbine casing bolt tightening," then historical cases, standards, or experience feedback related to "wind turbine casing," "bolt tightening," and "torque" are retrieved from the mechanical sub-base, such as "the standard torque for tightening bolts on a certain type of wind turbine casing is 55±5 N·m" or "feedback records of insufficient torque causing excessive wind turbine vibration." The retrieved domain knowledge is then input as auxiliary information into the second main model.

[0099] The second major model combines the input procedure content (process parameters and context) with retrieved domain knowledge to perform logical reasoning verification. For example, if the procedure requires a torque of 55 N·m, but the retrieved standard specifies 50 N·m, the model determines that there is a torque value deviation and records this issue as an item in the second problem list. If the procedure does not provide a torque value at all, but the context indicates that the operation belongs to the fastening category, and the retrieved standard includes a torque requirement, the model determines that the torque parameter is missing and also outputs it to the second problem list. Through the above methods, a complete second problem list is finally obtained.

[0100] This embodiment can realize a closed loop of knowledge accumulation: new solutions and rules that are confirmed to be effective during the rectification process can be injected into the knowledge base through a preset process, so that the system has the ability to continuously self-optimize.

[0101] This embodiment inputs process parameters and their context into a second main model, generates parsing prompts through semantic parsing, and retrieves relevant domain knowledge from the corresponding knowledge base partitions based on these prompts to assist the model in reasoning and verification. First, this embodiment avoids traversing the entire knowledge base for each search, reducing retrieval time and improving the overall efficiency of reasoning and verification. Second, different knowledge base partitions correspond to different process steps, making the retrieved domain knowledge more targeted and providing the model with standards, cases, or feedback highly relevant to the current process, thereby enhancing the accuracy of reasoning and verification. Third, inputting the retrieved domain knowledge as auxiliary information into the model, compared to relying solely on the model's own parameterized knowledge, can reduce misjudgments caused by knowledge lag or "illusions," improving the credibility of implicit logical risk identification results. Finally, this scheme refines the verification of process parameters to specific process categories, helping to improve the detection rate of missing or deviated key process parameters such as torque and impedance, thereby reducing operational risks caused by parameter errors.

[0102] In some embodiments, see Figure 7 S150 specifically includes the following steps: S151: Using the third major model, determine the related issues between the first and second problem lists.

[0103] Optionally, the third major model can be the same as the first and second major models, all of which use the deepseek model or the GPT model, etc.

[0104] Related issues refer to pairs of issues in the first issue list (issues discovered by rule validation) and the second issue list (issues discovered by logical reasoning validation) that, based on semantic analysis, belong to the same operational object, the same process step, or the same equipment scenario. Related issues may have different relationships, such as consistency, complementarity, or contradiction, requiring comprehensive analysis to determine their actual root cause.

[0105] Here's a specific example: Suppose there's a wind turbine bolt tightening maintenance procedure. After rule compliance verification, the first issue list contains one question: "Operation sequence does not meet requirements—tighten before inspecting" (the rule engine detected that the tightening step in the procedure is written before the inspection step, while the standard order should be inspection before tightening). After logical reasoning verification, the second issue list contains one question: "Tightening torque parameter does not match standard specification—the torque in the procedure is 55 N·m, while the standard torque for this type of wind turbine in the knowledge base is 50 N·m." Both of these questions are related to bolt tightening operations and are considered related issues.

[0106] S152: If there is no contradiction among the related issues, include any one or all of the related issues in the comprehensive issue list.

[0107] Based on the above example, the two issues relate to the operation sequence and torque parameters, respectively, and belong to different dimensions. There is no logical contradiction (i.e., the correction sequence and the correction torque can be processed in parallel). Therefore, assuming there is no contradiction between the related issues, the third major model includes either (or both) of these two issues in the comprehensive problem list.

[0108] S153: When there are contradictions in the related issues, the third major model is used to analyze the contextual information of the related issues in the maintenance procedure text, determine the root cause of the related issues, and include them in the comprehensive issue list.

[0109] This embodiment provides another example: the first problem list records "thread cleaning was not performed before bolt tightening," and the second problem list records "the thread cleaning step exists, but the cleaning method is incorrect (it requires the use of a cleaning agent, but actually states the use of a dry cloth)." These two problems pertain to the same operation (thread cleaning), but one points out a missing step, while the other points out an incorrect method, creating a contradiction. In this case, the third major model needs to analyze the contextual information of the related problems in the maintenance procedure text, such as checking whether the procedure actually lacks a cleaning step, or whether the step exists but is described incorrectly. Based on the context, the root cause problem is determined (e.g., the missing step is the fundamental cause, and the incorrect method is a derivative of the missing step) and added to the comprehensive problem list.

[0110] S154: Generate a quality inspection report based on a comprehensive problem list.

[0111] This embodiment uses a third major model to identify related issues, determine if contradictions exist, directly add issues to the comprehensive list when there are no contradictions, and then add issues after determining the root cause through contextual analysis when contradictions exist, ultimately generating a report based on the comprehensive list. This unifies and integrates explicit issues generated by rule validation and implicit issues generated by logical reasoning, helping to reduce information fragmentation caused by the different sources of these two types of issues, making the quality inspection report more complete and coherent. Secondly, for contradictory related issues, finding the root cause through contextual analysis rather than simply listing all issues avoids inconsistent modification suggestions in the report, thus reducing the difficulty of decision-making during subsequent rectification. Furthermore, the third major model automatically completes the identification of correlations and the resolution of contradictions, which improves processing efficiency and reduces misjudgments due to human oversight compared to manually comparing two lists and then judging contradictions. Finally, the report generated based on the comprehensive issue list is more focused, providing procedure writers with clearer and more easily implementable improvement guidelines, which helps to accelerate the speed of problem location and correction.

[0112] In some implementations, the quality inspection report also includes: the risk level of each issue in the comprehensive issue list, the corresponding rectification priority, and the standardized rectification strategy.

[0113] For details, see Figure 8 The risk level of each problem, the corresponding rectification priority, and the standardized rectification strategy can be obtained through the following methods: S161: Obtain a pre-built risk quantification assessment system for nuclear power plant maintenance procedures.

[0114] The risk quantification assessment system includes multi-dimensional quantitative indicators, weighted scoring algorithms, and risk level classification standards. The multi-dimensional quantitative indicators include the degree of safety impact (e.g., the level of impact on nuclear safety, equipment, and personnel), operational failure probability (estimated based on historical data the possibility that the problem will lead to operational failure), compliance and non-compliance level (the degree of deviation from standards and regulations), and on-site execution risk (the potential possibility of causing accidents or delays in actual operation).

[0115] S162: Based on the risk quantification assessment system, determine the risk level of each problem in the comprehensive problem list, and output the rectification priority and standardized rectification strategy corresponding to the risk level.

[0116] Optionally, the risk level of each problem in the comprehensive problem list can be determined by a risk level quantification model. The risk level quantification model needs to be trained using risk classification data related to nuclear power plant maintenance procedures and fine-tuned by low-rank adaptation (LORA) on an 8b model of Qwen.

[0117] For details, see Figure 9 S162 specifically includes the following steps: S1621: Based on multidimensional quantitative indicators and the type, location and scope of influence of each problem, each problem is scored item by item to obtain the sub-score of each problem on each quantitative indicator.

[0118] For each issue in the comprehensive problem list (e.g., "bolt tightening torque of 55 N·m does not match the standard value of 50 N·m"), the risk level quantification model first scores each issue based on its type (process parameter deviation), location (step in which it occurs), and scope of impact (involving equipment and subsequent processes), combined with multi-dimensional quantitative indicators. For example: the safety impact level is rated as 3 points (out of 5, it may cause increased vibration but does not directly endanger safety), the probability of operational failure is rated as 2 points (parameter deviation of 5 N·m, low probability of failure), the compliance and non-compliance level is rated as 4 points (significant deviation from the standard), and the on-site execution risk is rated as 3 points (if the operator applies the incorrect torque, it may cause the bolt to loosen).

[0119] S1622: Input the sub-item scores into the weighted scoring algorithm to calculate the comprehensive risk score for each question.

[0120] Based on the example above, after obtaining the individual scores, input these four scores into a weighted scoring algorithm (e.g., weights of 0.4, 0.2, 0.3, and 0.1 for each indicator) to calculate the overall risk score. Assume the calculated result is 3.2 points.

[0121] S1623: Compare the comprehensive risk score with the risk level classification standard to obtain the risk level corresponding to each problem.

[0122] The system compares the overall risk score with a pre-defined risk level classification standard. For example, high risk: 4.0-5.0, medium risk: 2.5-3.9, low risk: 1.0-2.4, warning: 0-0.9. In this example, a score of 3.2 corresponds to medium risk. Based on the risk level, the system outputs the corresponding rectification priority (e.g., "Rectification is recommended in the near future" for medium risk) and standardized rectification strategies (e.g., "The torque parameter deviation issue requires reviewing the design documents and correcting the values ​​in the standard procedure").

[0123] S163: Include risk level, rectification priority, and standardized rectification strategy in the quality inspection report.

[0124] Optionally, the system is designed with external RPA / API automation channels. For standardized rectification items such as terminology replacement, the system-generated suggestions can be automatically executed directly in production systems such as NG forms / work packages via RPA scripts or API interfaces, greatly improving rectification efficiency.

[0125] This embodiment quantifies and grades the issues in the comprehensive problem list, prioritizing them by severity. This helps procedure reviewers focus resources on issues with significant safety impact or high failure probability, thereby improving rectification efficiency. Employing a unified quantitative assessment system (including multi-dimensional indicators such as safety impact, operational failure, compliance violations, and on-site execution) reduces discrepancies in judgment among different personnel compared to subjective grading based on human experience, improving the stability and repeatability of risk classification. The output standardized rectification strategies provide pre-set corrective solutions for common problems, helping writers quickly obtain improvement suggestions and reducing the time spent on solution conception and verification. Including risk levels, rectification priorities, and standardized rectification strategies in the report ensures that it not only provides a problem list but also includes the basis for handling and action guidelines, helping to shorten the cycle from problem discovery to rectification decision-making and improving the effectiveness of rectification measures.

[0126] In some embodiments, when extracting structured procedure content information, the system simultaneously records the chapter number, paragraph number, and line number of each information fragment (such as operating steps, prerequisites, etc.) in the original maintenance procedure text. For example, if the parsed procedure content information includes "Step 3.2: Tighten the bolts with a torque wrench," the system also records that this information is located in Chapter 3, Paragraph 2, Line 15.

[0127] The nuclear power plant maintenance procedure inspection method also includes the following steps: S171: Mark the chapter, paragraph, and line number of each question in the comprehensive question list to obtain the question location information.

[0128] S172: Include problem location information in the quality inspection report.

[0129] In this embodiment, the problem location information is presented in the form of three-dimensional coordinates (chapter-paragraph-line number), which allows reviewers or writers to quickly locate the precise location of the problem without having to read through the entire procedure, thereby reducing search time and improving rectification efficiency.

[0130] In summary, this application's embodiments deeply integrate large-model semantic understanding technology with expertise in the nuclear power field, achieving a leap from "surface text matching" to "deep semantic parsing and causal reasoning." Specific innovations are reflected in: Innovative Hybrid Intelligent Architecture: A pioneering three-layer hybrid intelligent architecture combining a "rule engine + large language model semantic parsing + RAG knowledge enhancement" approach. This architecture organically integrates deterministic verification based on explicit business rules with understanding and reasoning capabilities based on large models and enhanced generation capabilities based on massive domain knowledge retrieval, achieving an optimal balance between efficiency, accuracy, and coverage.

[0131] Enhanced verification based on multidimensional knowledge retrieval (RAG): A dedicated knowledge base for nuclear power plant maintenance was constructed, integrating "experience feedback," "standard specifications," and "important feedback." This knowledge base was dynamically injected into the reasoning process of the large model through RAG technology. This enabled the system to not only check textual logic but also to perform conformity verification based on real industry experience and standards, effectively solving the "illusion" problem and knowledge lag of the large model.

[0132] End-to-end automated closed-loop processing flow: A full-link automated solution was designed, from document parsing, intelligent inspection, and report generation to the automatic execution of rectification suggestions via RPA / interface. The system can not only identify problems and provide suggestions, but also directly drive downstream business systems (such as NG form / work package systems, experience feedback systems IOE / FA / CQ) to complete the automatic processing of some standardized rectification items, realizing the full lifecycle management of quality issues from "discovery-location-suggestion-execution-retention".

[0133] Innovation in multimodal fusion verification mechanism: Introducing multimodal data fusion and alignment technology to build a multimodal verification system covering maintenance procedure texts, equipment drawings, operation and maintenance videos, and process flow diagrams, achieving cross-modal semantic understanding and logical consistency verification, breaking through the limitations of single text verification.

[0134] Innovation in intelligent quantification and graded handling of risk levels: Design a risk level quantification model exclusive to nuclear power plant maintenance procedures. Based on logical defects and safety hazards identified by the large model, automatically complete the risk quantification rating and priority classification, match standardized rectification plans, and achieve precise and standardized risk handling.

[0135] Construction of a knowledge graph-based logical association network for procedures: Building a knowledge graph for nuclear power plant maintenance procedures, mapping operational elements to graph nodes, forming a logical association network across chapters and documents, and realizing visualization of deep logical relationships and automatic identification of conflicts across the entire domain.

[0136] It should be noted that the method in this embodiment has been piloted in the group factory master program development and program improvement project. Currently, a total of 6,000 programs have been inspected for quality and are being continuously optimized and iterated.

[0137] The method in this embodiment contributes to the productivity of nuclear power plants as follows: Direct savings in human resources costs: Based on an average of 1,000 procedures inspected per base per year, approximately 1,980 hours of expert manpower can be saved (from 2,000 hours to 20 hours).

[0138] Accelerate the program iteration cycle: The efficient automatic inspection and intelligent optimization capabilities can significantly shorten the entire process time from the compilation and review of maintenance procedures to their effectiveness, ensuring that on-site maintenance work can obtain high-quality program document support more quickly, and indirectly improving the overall efficiency of key activities such as overhauls.

[0139] Promote the implementation of standardization strategies: Use technical means to make "quality standards quantifiable and traceable", reduce compliance audit risks and potential rectification costs, and provide solid technical support for the construction of nuclear safety culture.

[0140] Enhancing risk management: The expansion of three major technologies—multimodal verification, knowledge graph, and risk quantification—has upgraded the quality control of nuclear power maintenance procedures from "manual qualitative" to "intelligent quantitative," thereby comprehensively improving the inherent safety level of nuclear power operation and maintenance.

[0141] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0142] Corresponding to the nuclear power plant maintenance procedure inspection method described in the above embodiments, Figure 10 A structural block diagram of the nuclear power plant maintenance procedure inspection device provided in the embodiments of this application is shown. For ease of explanation, only the parts related to the embodiments of this application are shown.

[0143] Reference Figure 10 The device includes: The acquisition module 210 is used to acquire the text of the nuclear power plant maintenance procedure to be inspected; The extraction module 220 extracts structured procedure content information from the maintenance procedure text; the procedure content information includes at least one of the following: operating steps, prerequisites, safety constraints, equipment associations, process parameters, and terminology; The rule verification module 230 is used to verify the rule compliance of key information in the procedure content information and obtain the first problem list; the key information includes at least one of the following: target terminology, process description information, safety step identifier, equipment name, and operation action keywords; The knowledge base verification module 240 is used to perform logical reasoning verification on the content information of the procedures based on the nuclear power field knowledge base to obtain a second question list; the nuclear power field knowledge base includes at least one of the following: historical cases from the work experience feedback system, nuclear power industry standards and specifications, program feedback records from the program feedback system, and important feedback knowledge. The report generation module 250 is used to generate a quality inspection report based on the first problem list and the second problem list.

[0144] It should be noted that the information interaction and execution process between the above-mentioned devices / units are based on the same concept as the method embodiments of this application. For details on their specific functions and technical effects, please refer to the method embodiments section, and they will not be repeated here.

[0145] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0146] This application also provides an electronic device, which includes: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the processor executes the computer program to implement the steps in any of the above method embodiments.

[0147] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps described in the various method embodiments above.

[0148] This application provides a computer program product that, when run on an electronic device, enables the electronic device to perform the steps described in the various method embodiments above.

[0149] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include at least: any entity or device capable of carrying computer program code to an electronic device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electrical carrier signals or telecommunication signals.

[0150] Program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0151] Computer program code for performing the operations of the embodiments of this application can be written in one or more programming languages ​​or a combination thereof. These programming languages ​​include object-oriented programming languages ​​such as Python, Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0152] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0153] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0154] In the embodiments provided in this application, it should be understood that the disclosed devices / electronic devices and methods can be implemented in other ways. For example, the device / electronic device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual couplings or direct couplings or communication connections may be through some interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0155] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0156] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for inspecting nuclear power plant maintenance procedures, characterized in that, include: Obtain the text of the nuclear power plant maintenance procedures to be inspected; Extract structured procedure content information from the maintenance procedure text; The procedure content information includes at least one of the following: operating steps, prerequisites, safety constraints, equipment associations, process parameters, and terminology; The key information in the procedure content is checked for rule compliance to obtain a first problem list; the key information includes at least one of the following: target terminology, process description information, safety step identifier, equipment name, and operation action keywords; Logical reasoning and verification of the procedure content information are performed based on a nuclear power field knowledge base to obtain a second question list; the nuclear power field knowledge base includes at least one of the following: historical cases from the work experience feedback system, nuclear power industry standards and specifications, program feedback records from the program feedback system, and important feedback knowledge. A quality inspection report is generated based on the first problem list and the second problem list.

2. The nuclear power plant maintenance procedure inspection method as described in claim 1, characterized in that, The rule compliance verification of key information in the procedure content information is performed to obtain a first problem list, including: The procedure content information is input into the first large model to extract the key information; The first large model is used to analyze the context information of the key information in the procedure content information to determine the rule entry corresponding to the key information in the business rule base. The key information is matched with the corresponding rule entries, and the key information that violates the rules is included in the first problem list.

3. The nuclear power plant maintenance procedure inspection method as described in claim 2, characterized in that, The method further includes: A nuclear power plant maintenance procedure knowledge graph is obtained, which includes multiple nodes. Each node represents a specific operation step, prerequisite, safety constraint, equipment, process parameter or term in a verified maintenance procedure. The logical relationship between the nodes is represented by edges. The first large model queries the nuclear power plant maintenance procedure knowledge graph to find the target nodes associated with the procedure content information, and determines the logical conflict issues of the procedure content based on the logical relationships between the target nodes, and includes the logical conflict issues in the first problem list.

4. The nuclear power plant maintenance procedure inspection method as described in claim 3, characterized in that, The method further includes: Acquire multimodal data associated with the maintenance procedure, the multimodal data including at least one of equipment drawings, operation and maintenance videos, and process flow diagrams; Feature extraction and structured parsing are performed on the multimodal data. The parsed multimodal data and the procedure content information are semantically aligned and logically cross-validated through a multimodal large model to identify cross-modal issues that are inconsistent between the procedure content information and the multimodal data. The cross-modal issues are then included in the first issue list.

5. The nuclear power plant maintenance procedure inspection method as described in claim 1, characterized in that, The process of logically verifying the content of the regulations based on a nuclear power knowledge base yields a second list of questions, including: Input the process parameters and their contextual information in the maintenance procedure text into the second major model; The second major model is used to perform semantic parsing on the process parameters to obtain parsing prompt words; different parsing prompt words correspond to different nuclear power field knowledge bases. In the nuclear power field knowledge base corresponding to the parsing prompt words, domain knowledge related to the context information corresponding to the currently parsed set process is retrieved, and the retrieved domain knowledge is input into the second large model to assist the second large model in performing logical reasoning verification to obtain the second question list.

6. The nuclear power plant maintenance procedure inspection method as described in claim 1, characterized in that, The process of generating a quality inspection report for the maintenance procedure based on the first problem list and the second problem list includes: The third major model is used to determine the related issues between the first question list and the second question list; If there is no contradiction among the related issues, include any one or all of the related issues in the comprehensive issue list; In the event of contradictions in the related issues, the third major model is used to analyze the contextual information of the related issues in the maintenance procedure text, determine the root cause of the related issues, and include it in the comprehensive issue list. Based on the comprehensive problem list, the quality inspection report is generated.

7. The nuclear power plant maintenance procedure inspection method as described in claim 6, characterized in that, The quality inspection report also includes: the risk level of each problem in the comprehensive problem list, the rectification priority corresponding to the risk level, and the standardized rectification strategy.

8. The nuclear power plant maintenance procedure inspection method as described in claim 7, characterized in that, The risk level of each question in the comprehensive question list is determined based on a risk quantification assessment system, which includes multi-dimensional quantitative indicators, a weighted scoring algorithm, and risk level classification standards. The multidimensional quantitative indicators include the degree of security impact, probability of operational failure, level of compliance and non-compliance, and on-site execution risk; Based on the aforementioned risk quantification assessment system, the risk level of each issue in the comprehensive issue list is determined, specifically including: Based on the multidimensional quantitative indicators and the type, location and scope of influence of each problem, each problem is scored item by item to obtain the sub-score of each problem on each quantitative indicator; The individual scores are input into the weighted scoring algorithm to calculate the comprehensive risk score for each question. The comprehensive risk score is compared with the risk level classification standard to obtain the risk level corresponding to each problem.

9. The nuclear power plant maintenance procedure inspection method as described in claim 6, characterized in that, The procedure content information also includes chapters, paragraphs, and line numbers; the method also includes: Mark the chapter, paragraph, and line number of each question in the comprehensive question list to obtain the question location information; The problem location information will be included in the quality inspection report.

10. A nuclear power plant maintenance procedure inspection device, characterized in that, include: The acquisition module is used to acquire the text of the nuclear power plant maintenance procedures to be inspected; The extraction module extracts structured procedure content information from the maintenance procedure text; The procedure content information includes at least one of the following: operating steps, prerequisites, safety constraints, equipment associations, process parameters, and terminology; The rule verification module is used to perform rule compliance verification on key information in the procedure content information to obtain a first problem list; the key information includes at least one of the following: target terminology, process description information, safety step identifier, equipment name, and operation action keywords. The knowledge base verification module is used to perform logical reasoning verification on the content information of the procedure based on the nuclear power field knowledge base to obtain a second question list; the nuclear power field knowledge base includes at least one of the following: historical cases from the work experience feedback system, nuclear power industry standards and specifications, program feedback records from the program feedback system, and important feedback knowledge. The report generation module is used to generate a quality inspection report based on the first problem list and the second problem list.

Images