An evidence constraint-based technical material intelligent compiling system and method

Abstract

This application belongs to the interdisciplinary field of computer application technology and aerospace equipment technology, and specifically relates to an intelligent technical document compilation system and method based on evidence constraints. The system includes a data governance module, a compilation constraint graph construction module, an automated generation module, an intelligent review module, an intelligent retrieval module, a change impact analysis module, and an interface integration module. The system converts multi-source equipment technical data into source record objects, chapter unit objects, and evidence objects, constructs a chapter constraint graph, and generates chapter task packages containing template skeletons, mandatory entities, mandatory parameters, and evidence sets. Based on this, it outputs initial drafts of the document with evidence identifiers, and forms candidate release versions through rule review, graph consistency review, and semantic review. When the technical status changes, it identifies affected chapters and recompiles them accordingly. This invention improves the traceability, consistency, and timeliness of technical document compilation, review, retrieval, and updating.

Description

Technical Field

[0001] This application belongs to the interdisciplinary field of computer application technology and aviation equipment technology, and specifically relates to an intelligent compilation system and method for technical data based on evidence constraints. Background Technology

[0002] Equipment user technical documentation typically includes user manuals, maintenance manuals, troubleshooting guides, lists of aviation materials and tools, technical bulletins, and supplementary pages for different configurations. This type of documentation is characterized by its dispersed sources, frequent version changes, numerous cross-references between chapters, dense use of technical terminology, strict parameter constraints, and significant differences in applicable configurations. Especially in the context of aviation equipment, complex electromechanical equipment, and high-reliability equipment, technical documentation not only serves as information dissemination but also directly supports maintenance operations, fault isolation, and configuration control; therefore, its requirements for consistency, traceability, and timeliness are higher than those of general document systems.

[0003] Most existing equipment technical documentation compilation systems already possess functions such as document storage, template retrieval, keyword search, and basic verification. Some systems have even introduced semantic retrieval, knowledge graphs, or text generation models to improve the efficiency of data organization and retrieval. However, most of these solutions still remain at the level of document management, text retrieval, or draft generation, lacking a mechanism for unified constraints around chapter-level objects, evidence, configuration status, and version propagation relationships.

[0004] When technical documents enter the formal compilation and continuous updating stage, there are often numerous overlaps between design basis, standard clauses, test conclusions, maintenance records, failure cases, and published chapters. Relying solely on template filling, free generation, or manual review not only makes it difficult to ensure the consistency of parameters, the dependencies between steps, and the completeness of safety warnings across different chapters, but also makes it difficult to quickly locate the chapters that truly need revision after changes in technical status.

[0005] Meanwhile, when using technical documentation, operations and maintenance personnel typically want to directly obtain valid content corresponding to the target object, operational action, usage conditions, and current status, and be able to clearly identify the applicable configuration scope, current version, and source basis of the content. However, existing systems often struggle to simultaneously achieve chapter-level traceable compilation, controlled generation, joint review, version-linked updates, and interpretable retrieval. This results in documentation that, while producing textual results, still suffers from issues such as scattered sources, delayed updates, coarse review granularity, and insufficient reliability.

[0006] Most existing equipment user technical data compilation systems still rely primarily on "manual data collection + manual drafting + manual proofreading + manual publishing and updating." Even when large models are introduced, they often remain at the level of generating initial text drafts, lacking features such as chapter task breakdown, evidence binding, graph constraint verification, and the ability to disseminate the impact of changes. This can easily lead to the following problems:

[0007] First, during the compilation of the data, there is a lack of a unified structured connection between the design basis, standard clauses, configuration differences, failure cases and existing chapter content, resulting in the generated content being readable but lacking clear basis, making subsequent review difficult;

[0008] Second, there are relationships such as multiple references to the same parameter between chapters, dependencies on the order of steps, and reuse of security warnings. It is difficult to identify cross-chapter conflicts in a timely manner by relying solely on manual review or general text review models.

[0009] Third, equipment design status, service notices and standards and specifications change frequently. Existing systems can usually only be rewritten as a whole or rely on manual chapter-by-chapter checks, which cannot quickly identify the scope of chapters that are truly affected.

[0010] Fourth, while maintenance personnel hope to directly obtain valid operational content related to "object-action-condition-status" when searching, existing systems often cannot simultaneously display the current valid version, applicable configuration, and source basis, affecting the reliability of use.

[0011] Therefore, there is an urgent need for an evidence-constrained intelligent compilation system and method for technical documents, which can build a closed loop around chapter task packages, evidence constraint maps, intelligent review, and the propagation of the impact of changes, in order to solve the problems of scattered compilation basis, coarse review granularity, weak version linkage, and insufficient interpretability of search results in the existing technology. Summary of the Invention

[0012] To address the aforementioned issues, this application provides an intelligent technical data compilation system and method based on evidence constraints. This system aims to resolve problems such as the fragmented basis for compiling technical data for existing equipment users, difficulty in ensuring consistency between chapters, difficulty in quickly locating affected chapters after changes in technical status, and the lack of version basis and evidentiary explanation in search results.

[0013] The technical solution of this application is: a system and method for intelligent compilation of technical data based on evidence constraints, including a data governance module, a constraint graph construction module, a large model training and optimization module, an automated generation module, an intelligent review module, an intelligent retrieval module, a change impact analysis module, and an interface integration module.

[0014] The system comprises the following modules: a data governance module to process multi-source heterogeneous data into objectified data suitable for compilation and review; a compilation constraint graph construction module to establish structured relationships between chapter content and technical basis; a large model training and optimization module to form an auxiliary compilation large model adapted to the compilation scenarios of equipment user technical data; an automated generation module to output initial drafts with evidence identifiers based on chapter task packages; an intelligent review module to conduct rule review, graph consistency review, and semantic review of the initial drafts; an intelligent retrieval module to return data fragments and evidence chains consistent with the current valid version based on natural language intent; a change impact analysis module to locate affected chapters based on technical status changes and trigger targeted re-editing; and an interface integration module to connect the data flow of the design, configuration, maintenance, and delivery systems.

[0015] In a preferred embodiment, the data governance module does not simply store the original document, but splits paragraphs, steps, table entries or warning blocks with relatively complete semantics and logic in the technical information into chapter unit objects, and splits the corresponding basis into evidence objects; each chapter unit object establishes a mapping relationship with multiple evidence objects, thereby forming a unified data foundation for subsequent generation, review, retrieval and updating.

[0016] In a preferred embodiment, the constraint map construction module establishes a heterogeneous map around equipment objects, systems / components, maintenance actions, parameter thresholds, safety warnings, failure modes, chapter units, evidence sources, and version baselines, enabling the system to provide a structured expression of "which parameters and warnings should be referenced for a certain maintenance action under a certain configuration, which preceding chapters should be consistent with, and which valid evidence the current basis comes from."

[0017] In a preferred embodiment, the automated generation module does not directly input open-ended prompts into the large model. Instead, the system first generates chapter task packages. These chapter task packages include at least a target chapter identifier, a template skeleton, a set of required entities, a set of required parameters, a set of terminology constraints, a set of evidence citations, a set of prohibited expressions, and output format constraints. This assists in the compilation of the large model by generating initial drafts of data according to the chapter task packages, ensuring that the generation process is driven by structured constraints rather than relying entirely on free text reasoning.

[0018] In a preferred embodiment, the intelligent review module employs a combination of rule review, graph consistency review, and semantic review. Rule review identifies issues such as incorrect chapter formatting, missing templates, abnormal unit usage, and inconsistent terminology. Graph consistency review identifies issues such as parameter and object mismatches, incorrect step order, missing cross-references, and conflicting applicable configurations. Semantic review identifies ambiguous statements, logical contradictions, omitted warnings, and insufficient evidence. The combined results of these three types of reviews output a problem list and revision suggestions more suitable for the equipment user's technical documentation scenario.

[0019] In a preferred embodiment, after receiving design change orders, technical status notifications, service announcements, fault feedback, and standard and specification update information, the change impact analysis module first maps the changed content to object nodes, parameter threshold nodes, or safety warning nodes in the constraint diagram. Then, it identifies the affected chapter unit objects along the version association, reference relationship, and change propagation relationship, thereby only rewriting and reviewing high-impact chapters, reducing repeated revisions of the entire document.

[0020] In a preferred embodiment, the intelligent retrieval module parses the user's natural language request into four categories of retrieval elements: object, action, condition, and state. Then, it combines the graph path, chapter unit object vector representation, and keyword precise matching results for fusion and sorting. This allows maintenance personnel to not only see relevant paragraphs but also simultaneously obtain the applicable configuration, the current valid version, and source evidence, thereby improving the credibility and interpretability of the retrieval results.

[0021] In a preferred embodiment, the large model training and optimization module uses structured training samples corresponding to the equipment user technical data compilation task for domain adaptation training. The training samples include at least task instruction fields, chapter task package fields, evidence object fields, historical chapter fields, review issue fields, and revision result fields, enabling the model to learn the mapping relationship of "input constraints, citing valid evidence, avoiding existing review issues, and outputting content that conforms to chapter specifications."

[0022] In a preferred embodiment, the system includes a release control unit. Candidate release versions are only marked as valid versions when the structural integrity score, evidence coverage score, and consistency score all reach preset thresholds. If errors are found in the new version after release, the system restores to the previous valid version through version baselines and chapter-level change chains to prevent the erroneous content from spreading further.

[0023] This invention also provides an intelligent auxiliary compilation method for equipment user technical data, including multi-source data governance, compilation constraint graph construction, chapter task package generation, initial draft generation, intelligent review, natural language retrieval, and identification and targeted re-compilation of affected chapters after changes in technical status.

[0024] The intelligent compilation system and method for technical data based on evidence constraints in this application have the following advantages:

[0025] 1. By breaking down technical documentation into chapter-based units and binding evidence objects and version baselines to each chapter-based unit, the generated documentation has a clear basis, facilitating review, tracking, and accountability;

[0026] 2. By using chapter task packages to drive generation and joint review of rules / graphs / semantics, we can improve the consistency of parameters across chapters, the integrity of processes, and the integrity of security warnings, and reduce the risk of logical conflicts caused by simply relying on free generation;

[0027] 3. By establishing a propagation chain of "technical status change - affected nodes - affected chapters" through the change impact analysis module, the relevance and timeliness of technical data updates can be improved;

[0028] 4. By parsing the intent of the object-action-condition-state and displaying the chain of evidence, the accuracy of operation and maintenance retrieval and the interpretability of the results can be improved, making it easier for users to obtain operational basis consistent with the current effective configuration. Attached Figure Description

[0029] Figure 1 This is a flowchart illustrating an intelligent auxiliary compilation method for equipment user technical data according to one embodiment of the present invention.

[0030] Figure 2 This is a schematic diagram of the overall architecture of a system and method for intelligent compilation of technical data based on evidence constraints, according to one embodiment of the present invention. Detailed Implementation

[0031] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of this application will be described in more detail below with reference to the accompanying drawings. In the drawings, the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The described embodiments are only some, not all, of the embodiments of this application. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application. The embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0032] The first aspect of this application provides a system and method for intelligent compilation of technical data based on evidence constraints, such as... Figure 1As shown, the intelligent auxiliary compilation method for equipment user technical data in this embodiment includes steps S1 to S6.

[0033] Step S1: Multi-source data access and governance. The system acquires raw data from the design system, configuration management system, standard specification library, historical technical data library, fault and maintenance record library, manual input terminal, and technical status reporting system. For structured data, the system maps fields and verifies codes through interfaces; for semi-structured and unstructured data, the system performs format parsing, text extraction, terminology normalization, chapter segmentation, and source location. After processing, the system forms source record objects, chapter unit objects, and evidence objects. Preferably, a chapter unit object can be a complete step, a technical description, an alarm block, a parameter table entry, or a troubleshooting logic description.

[0034] In this implementation, the chapter unit object includes at least the following fields: chapter unit identifier, title, scope of application, content, document type, parent chapter identifier, list of source evidence, current version number, status identifier, and last verification time. The evidence object includes at least the following fields: evidence identifier, evidence type, source document, source location, applicable configuration, effective date, expiration date, credibility level, and summary description. Through this objectification process, subsequent generation, review, retrieval, and updating can share the same set of objects, rather than processing separate, fragmented original documents.

[0035] Step S2 involves constructing a constraint diagram for compiling equipment user technical data. The system extracts equipment model, system, component, maintenance action, parameter threshold, safety warning, failure mode, chapter unit, evidence source, and version baseline from objectified data, and establishes relationships between nodes. For example, when a chapter describes "replacing a certain type of fuel filter," the system associates that chapter unit with the corresponding equipment configuration, filter part number, torque parameters, required tools, safety warnings, and source standard clauses. When the chapter needs to reference disassembly steps in a preceding chapter or functional check steps in a subsequent chapter, the system also establishes process dependencies.

[0036] Preferably, the relationships include at least: applicability relationships, indicating which configurations a chapter or piece of evidence is applicable to; dependency relationships, indicating the order of operation steps; constraint relationships, indicating parameters, tools, or security conditions that an action needs to meet; reference relationships, indicating references between chapters and between chapters and evidence; version association relationships, indicating the version baseline corresponding to the current chapter; and change propagation relationships, indicating the downstream chapter nodes that may be affected when a node changes. Through the above graph, the system can answer questions such as "which parameters must a chapter contain?" and "which chapters will be affected by a configuration change?"

[0037] Step S3: Generate chapter task packages and output initial drafts of the data. Upon receiving a data compilation task, the system extracts the corresponding template skeleton, mandatory entity set, mandatory parameter set, terminology constraint set, evidence citation set, and output format constraints from the graph based on the target data type and target chapter, forming a chapter task package. After reading the chapter task package, the system assists in compiling the large model first generates paragraph structures based on the template skeleton, then fills in the main text according to the mandatory entities and parameters, and finally writes evidence identifiers, applicable configuration identifiers, and version identifiers for each paragraph or step.

[0038] In a specific example, when the task is "Generate a chapter on pressure regulating valve inspection in the maintenance manual for a certain type of equipment's hydraulic system," the chapter task package should include at least: the chapter title "Pressure Regulating Valve Inspection"; the set of applicable configurations; related component objects; a set of required parameters, such as pressure thresholds, temperature ranges, and allowable leakage values; a set of required warnings, such as pressure release warnings and power failure warnings; references to preceding steps, such as "power failure" and "pressure relief"; references to subsequent steps, such as "reset check" and "functional verification"; and a set of evidence objects, such as design specification clauses, test report summaries, and previously published versions. The text output by the large model based on this task package is no longer a generalized description, but a chapter draft with clear constraints and traceable identifiers.

[0039] Step S4: Perform joint intelligent review and generate a candidate release version. The system sequentially performs rule review, graph consistency review, and semantic review on the initial draft of the data. The rule review mainly checks the chapter hierarchy, numbering format, unit standardization, terminology consistency, template completeness, and prohibited expressions; the graph consistency review mainly checks whether the component objects and configurations match, whether the parameter values ​​match the corresponding actions, whether the warning items cover dangerous actions, whether the relationship between preceding and following steps is complete, and whether cross-chapter references are invalid; the semantic review mainly identifies ambiguities in statements, logical contradictions, missing conditions, and "conclusions without basis" content.

[0040] Preferably, the system binds each review issue to a specific evidence object and a chapter unit object. For example, when a chapter contains a torque value inconsistent with the standard clause, the system not only marks the location of the issue but also provides the source of the conflict, the location of the standard clause, and the suggested revision value. When a chapter's "disassembly and assembly steps" omit necessary safety warnings, the system automatically prompts the corresponding alarm content based on the safety warning nodes and dangerous action nodes in the diagram. After revision, the system re-executes the review until the release threshold is reached, forming a candidate release version.

[0041] Step S5: Perform intelligent retrieval for operation and maintenance scenarios. After the user inputs their natural language requirements, the system decomposes these requirements into four categories: object, action, condition, and state. For example, for the requirement "How to check for abnormal outlet pressure of a hydraulic pump under high temperature conditions," the system can identify the object as "hydraulic pump," the action as "check," the condition as "high temperature conditions," and the state as "abnormal outlet pressure." Subsequently, the system performs keyword precise matching, chapter / unit vector similarity retrieval, and graph path retrieval to obtain a set of candidate results, which are then sorted based on version validity, configuration fit, evidence credibility, and user scenario matching.

[0042] The final search results are not limited to isolated paragraphs, but can include operation step sequences, relevant parameter entries, necessary warnings, applicable configurations, recent change descriptions, and source evidence. This allows maintenance personnel to clearly understand "why this content is recommended," "whether this content applies to the current equipment configuration," and "what the basis for this content is" when reviewing the results.

[0043] Step S6: Perform change impact analysis and targeted rewriting. When the system receives a design change order, configuration replacement notification, technical status report, fault feedback, or standard specification update, the change impact analysis module first maps the change to object nodes, parameter threshold nodes, safety warning nodes, or fault mode nodes in the graph. Then, it calculates the chapter impact score based on version association, reference relationship, and change propagation relationship. For chapter unit objects with an impact score higher than the threshold, the system marks them as pending rewriting and triggers the automatic generation module to generate a new chapter task package only for the corresponding chapter. The intelligent review module then reviews the new version.

[0044] Preferably, the system retains records linking the previous and new versions, the corresponding evidence, and the review conclusions to form a chapter-level traceable change chain. When a user subsequently retrieves that chapter, the system can also display the reason for the most recent change and its effective date, thus preventing the user from continuing to use an outdated version.

[0045] Implementation Method 2: Data Governance and Object Modeling

[0046] In this implementation, the data governance module operates in a manner that follows the structure of "raw record layer - object extraction layer - relationship building layer". The raw record layer stores the original documents and interface data; the object extraction layer generates source record objects, chapter unit objects, and evidence objects; and the relationship building layer is responsible for establishing mapping relationships between chapter units, between chapter units and evidence, and between chapter units and equipment configurations.

[0047] For historical technical data, the system preferably segments it by chapter titles, list items, table rows, and alarm blocks; for maintenance records and fault cases, the system preferably extracts them in a structured manner according to "fault phenomenon - diagnostic conditions - troubleshooting steps - handling conclusion"; for standards and specifications, the system preferably extracts the clause number, constraint content, scope of application, and effective date. Through the above rules, the data governance module can uniformly transform content originally scattered in different files into a set of calculable, referable, and traceable objects.

[0048] Implementation Method 3: Chapter Task Pack Generation Strategy

[0049] In this implementation, the chapter task package can be considered as a structured control load before the large model is called. The template skeleton in the chapter task package is used to constrain the paragraph order, the required entity set is used to constrain the systems, components, tools, fixtures, consumables, and other objects that must be involved in this chapter, the required parameter set is used to constrain numerical items, the terminology constraint set is used to constrain the standard terminology, the evidence citation set is used to limit the scope of usable evidence, the prohibited expression set is used to eliminate promotional or uncertain expressions that do not conform to the specifications, and the output format constraint is used to limit the chapter style, numbering method, and citation method.

[0050] Preferably, the automated generation module first generates chapter outlines, then generates paragraphs for each section, and verifies whether any required entities, parameters, or warnings are missing during the generation process. If any omissions are found, the system prioritizes completing them during the generation stage, rather than relying entirely on subsequent review stages, thereby reducing the cost of repeated revisions.

[0051] Implementation Method Four: Joint Review Strategy

[0052] In this implementation, the rule review unit establishes an executable rule base based on template specifications, numbering specifications, terminology specifications, and unit of measurement specifications; the graph consistency review unit focuses on compiling constraint graphs and verifies object relationships, parameter constraints, reference chains, and applicable configurations; and the semantic review unit uses the trained auxiliary model to identify potential semantic risks.

[0053] For example, for requirements such as "confirm that residual pressure has been released before disassembly", if the text contains disassembly steps but the corresponding action node in the diagram is marked as a high-risk action and is not associated with any warning node, the diagram consistency review unit can identify the problem of missing safety warnings; if different chapters give different values ​​for the torque value of the same component, and the credibility level and effective time of the evidence object show that one of them has failed, the system can automatically give the source of conflict and suggest the effective value to be adopted; if the semantic review unit identifies vague expressions such as "appropriate", "when necessary", "depending on the situation" without accompanying triggering conditions, the system can mark them as content to be clarified.

[0054] Implementation Method 5: Linked Retrieval and Update

[0055] In this implementation, the retrieval logs are not only used to statistically analyze user interests, but also to optimize the graph structure and training samples. If certain chapters are frequently searched but users repeatedly add filtering criteria, it indicates that the granularity of the original chapter unit objects or the search tags are insufficient. The system can write this phenomenon back to the data resource library, where the data governance module can add object tags or split chapter units. If certain revision suggestions are continuously adopted after manual confirmation, the system can add them as positive samples to the training sample library to enhance subsequent generation and review capabilities.

[0056] Implementation Method Six: Training Sample Construction and Distribution Control

[0057] In this implementation, a single training sample constructed by the large model training and optimization module consists of a task instruction field, a chapter task package field, an evidence object field, a historical chapter field, a review issue field, and a revision result field. The task instruction field characterizes the document type, target chapter, and applicable configuration of this compilation task; the chapter task package field characterizes the template skeleton, mandatory entity set, mandatory parameter set, terminology constraint set, and prohibited expression set; the evidence object field characterizes the valid evidence that can be cited in this generation; the historical chapter field characterizes existing published versions of the same chapter; the review issue field characterizes the types of issues that have occurred in the historical compilation of this chapter; and the revision result field characterizes the final revised version that was manually adopted.

[0058] Preferably, the system uses “chapter task package + valid evidence object + historical problem feedback” as input constraints and “approved chapter text + paragraph-level evidence identifier + version identifier” as output supervision signals for training, so that the model learns a controlled generation mode for equipment data scenarios, rather than just learning a general text continuation mode.

[0059] In this implementation, the release control unit performs gating checks on candidate release versions. If any of the structural integrity score, evidence coverage score, or consistency score falls below the corresponding threshold, the candidate release version remains in a pending review state; if all scores meet the threshold, it is marked as a valid version and written to the release repository. The system also records the version identifier, chapter mapping, and evidence snapshot of the previous valid version, forming a rollback entry. When a problem is subsequently discovered in a new version after release, the system restores the previous valid version based on the rollback entry and re-triggers the joint review of the affected chapters.

[0060] Implementation Method 7: System Deployment

[0061] In this embodiment, the system is deployed in a computing environment including a processor, memory, graph database, object storage, vector database, and message bus. The data governance module runs on the object extraction service node, the constraint graph construction module runs on the graph construction service node, and the automated generation module, intelligent review module, and intelligent retrieval module call the same auxiliary large model construction service; the change impact analysis module receives design change orders, technical status notifications, and standard update events through the message bus; the interface integration module interacts with the design system, configuration management system, and data system through standardized interface protocols.

[0062] Preferably, chapter unit objects and evidence objects are stored in an object store or relational database, inter-chapter relationships, evidence relationships, and version propagation relationships are stored in a graph database, and chapter text vectors and retrieval intent vectors are stored in a vector database. Using this deployment method, the system can perform differentiated processing on object queries, relational reasoning, and semantic retrieval, thereby supporting chapter-level targeted rewriting, evidence chain tracing, and valid version retrieval.

[0063] Through the above implementation methods, the present invention forms a closed-loop technical solution of "data governance - graph constraint - chapter task package generation - joint review - intelligent retrieval - change propagation", which is not only applicable to the technical data of aviation equipment users, but can also be extended to the compilation scenarios of user data, maintenance data and support data of other complex equipment.

[0064] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A system and method for intelligent compilation of technical data based on evidence constraints, characterized in that, It includes a data governance module, a constraint graph construction module, a large model training and optimization module, an automated generation module, an intelligent review module, an intelligent retrieval module, a change impact analysis module, and an interface integration module; The data governance module is used to collect structured, semi-structured, and unstructured data generated during the equipment research, development, manufacturing, testing, delivery, and maintenance phases. It performs format normalization, version alignment, terminology standardization, content deduplication, and semantic indexing on the collected data to generate a data resource library. The data resource library includes at least source record objects, chapter unit objects, and evidence objects. The constraint map construction module is used to extract equipment object nodes, system / component nodes, operation and maintenance action nodes, parameter threshold nodes, safety warning nodes, fault mode nodes, chapter unit nodes, evidence source nodes, and version baseline nodes based on the data resource library, and establish applicable relationships, dependency relationships, constraint relationships, reference relationships, version association relationships, and change propagation relationships to form a constraint map for compiling equipment user technical data. The large model training and optimization module is used to compile constraint graphs based on the data resource library and the equipment user technical data to construct domain training samples, perform domain adaptation training and compilation task optimization training on the basic model, and obtain an auxiliary large model for compilation. The automated generation module is used to receive data compilation tasks, and extract the corresponding chapter template skeleton, mandatory entity set, mandatory parameter set, evidence reference set and output constraint rules from the equipment user technical data compilation constraint diagram according to the data type, applicable configuration, target chapter and compilation purpose, generate chapter task package, input the chapter task package into the auxiliary compilation large model, and output the first draft of the data with paragraph-level evidence identifier and scope of application identifier. The intelligent review module is used to perform structural integrity review, parameter consistency review, process dependency review, terminology uniformity review, security warning integrity review, and evidence coverage review on the initial draft of the materials, generate a list of issues, revision suggestions, and review scores, and feed the review results back to the automated generation module for targeted revision; The intelligent retrieval module is used to receive user natural language requests, parse the natural language requests into object-action-condition-state retrieval intents, and combine the equipment user technical data to compile constraint diagrams, the data resource library, and the approved document versions to perform mixed retrieval of chapter unit objects and evidence objects, and output the document paragraphs, operation steps, applicable configurations, and evidence chains corresponding to the retrieval intents. The change impact analysis module is used to receive design change orders, technical status notifications, fault feedback, service notices or standard specification update information, identify affected chapter unit objects based on the version baseline node and the change propagation relationship, and trigger the automated generation module and the intelligent review module to perform targeted rewriting and review of the affected chapters; The interface integration module is used to interact with design systems, configuration management systems, quality management systems, fault diagnosis systems, standard and specification libraries, data systems, or technical status management systems, and to send back the approved technical document versions, search logs, and user revision feedback to external systems.

2. The intelligent compilation system and method for technical data based on evidence constraints as described in claim 1, characterized in that, The data governance module archives the collected data hierarchically according to "equipment model - applicable configuration - document type - chapter number - knowledge point number", and establishes a unique identifier, title attribute, scope of application attribute, content attribute, source attribute, version attribute and status attribute for each chapter unit object; wherein, the content attribute is used to record the chapter text, and the source attribute is used to record the evidence object identifier corresponding to the chapter text.

3. The intelligent compilation system and method for technical data based on evidence constraints as described in claim 1, characterized in that, The evidence objects established by the constraint map construction module include at least design basis, standard provisions, test conclusions, failure cases, maintenance records, and manual confirmation conclusions. For each evidence object, the source document, source location, applicable configuration, effective time, failure time, and credibility level are recorded. At least one of the following relationships is established between the chapter unit node and the evidence source node: "direct basis", "supplementary basis", or "verification basis".

4. The intelligent compilation system and method for technical data based on evidence constraints as described in claim 1, characterized in that, The chapter task package includes a target chapter identifier, a template skeleton, a set of required entities, a set of required parameters, a set of terminology constraints, a set of evidence citations, a set of prohibited expressions, and output format constraints. When generating the initial draft of the document, the automated generation module first determines the chapter structure based on the template skeleton, then generates the main text based on the set of required entities and the set of required parameters, and finally writes evidence identifiers, scope of application identifiers, and version identifiers for the paragraphs or steps in the main text.

5. The intelligent compilation system and method for technical data based on evidence constraints as described in claim 1, characterized in that, The intelligent review module includes a rule review unit, a graph consistency review unit, and a semantic review unit. The rule review unit verifies chapter hierarchy, numbering format, units of measurement, terminology, and template integrity. The graph consistency review unit verifies object relationships, parameter values, process dependencies, and cross-references. The semantic review unit identifies ambiguous statements, logical conflicts, omissions in warnings, and insufficient evidence. When conflicts exist among the three review results, the final review conclusion is determined based on the credibility level of the evidence object and the validity of the version.

6. The intelligent compilation system and method for technical data based on evidence constraints as described in claim 1, characterized in that, The change impact analysis module calculates the chapter impact score based on the object nodes, parameter threshold nodes, security warning nodes, and fault mode nodes involved in the change. The chapter impact score is determined based on at least two of the following: relationship propagation depth, number of references, security level, applicable configuration coverage, and change type. When the chapter impact score is higher than a preset threshold, the corresponding chapter unit object is marked as pending recompilation.

7. The intelligent compilation system and method for technical data based on evidence constraints as described in claim 1, characterized in that, After parsing the user's natural language requirements, the intelligent retrieval module generates object vectors, action vectors, condition vectors, and state vectors, and performs a fusion sorting of keyword precise matching, vector similarity retrieval, and graph path retrieval. The retrieval results include at least chapter titles, text fragments, operation order, applicable configurations, source evidence, and version validity prompts. The interface integration module writes back the manually confirmed revisions, review conclusions, and retrieval feedback to the data resource library. The adopted revisions are written as positive samples into the training sample library, and the rejected revisions and their reasons are written as constraint samples into the training sample library to drive the large model training and optimization module to perform continuous optimization.

8. The intelligent compilation system and method for technical data based on evidence constraints as described in claim 1, characterized in that, The domain training samples constructed by the large model training and optimization module include at least the task instruction field, chapter task package field, evidence object field, historical chapter field, review question field, and revision result field; wherein, the task instruction field is used to characterize the data type, target chapter, and applicable configuration, and the revision result field is used to characterize the revision content adopted by the human or the reason for the revision rejected by the human. The system also includes a release control unit, which marks a candidate release version as a valid version when the candidate release version meets the structural integrity threshold, evidence coverage threshold, and consistency threshold, and retains the rollback entry corresponding to the previous valid version; when subsequent review or manual verification finds that there are problems with the release version, the rollback entry is called to restore the previous valid version.

9. A method for intelligently assisting in the compilation of equipment user technical data, characterized in that, The system and method for intelligent compilation of technical data based on evidence constraints as described in any one of claims 1 to 8 includes the following steps: S1. Collect multi-source heterogeneous data generated throughout the entire life cycle of the equipment, and perform format normalization, terminology standardization, semantic indexing and version alignment on the multi-source heterogeneous data to generate source record objects, chapter unit objects and evidence objects. S2. Construct a constraint diagram for compiling equipment user technical data based on the source record object, chapter unit object, and evidence object. The constraint diagram for compiling equipment user technical data describes the relationship between equipment objects, maintenance actions, parameter thresholds, safety warnings, chapter units, evidence sources, and version baselines. S3. Based on the data compilation task and the equipment user technical data, compile constraint diagrams to generate chapter task packages, and based on the chapter task packages, call the auxiliary compilation large model to output the data draft with evidence identification. S4. Perform structural integrity review, parameter consistency review, process dependency review, terminology uniformity review, and evidence coverage review on the initial draft of the document, and make targeted revisions to the initial draft of the document based on the review results to obtain a candidate release version; S5. Respond to the user's natural language search request, parse the user's needs into object-action-condition-state search intent, perform mixed search on the chapter unit objects and evidence objects corresponding to the candidate release version or the approved version, and output data fragments, operation steps and evidence chains. S6. In response to design change orders, technical status notifications, fault feedback, or standard specification updates, identify the affected chapter / unit objects and perform targeted rewriting, review, and version release on the affected chapter / unit objects.

10. The intelligent assisted compilation method for equipment user technical data as described in claim 9, characterized in that, In step S6, for the affected chapter unit objects, the system retains the associated records of the version before the change, the version after the change, the corresponding evidence objects, and the review conclusions, forming a chapter-level traceable change chain, and displays the current valid version and the reason for the most recent change to the user when the search output is displayed.

Images