Digital file management system for product research and development

Abstract

The invention provides a digital file management system for product research and development, and the system comprises an interactive service layer which is configured to receive a multi-modal technical document and external system data inputted by a user, and provide a visual interface and a service calling interface; the core processing layer is in communication connection with the interactive service layer, and the core processing layer comprises a large language model LLM processing module, a formalization and protocol converter and a model verification processor; and the data support layer is in communication connection with the core processing layer and is used for storing and managing system operation data. The logic elements of the natural language technical document are extracted through the large language model, the standard formalized model is generated, a semantic gap between the natural language and machine executable logic is bridged, and the purpose of improving research and development quality and efficiency is achieved.

Description

Technical Field

[0001] This invention belongs to the field of digital R&D management technology, and in particular relates to a digital document management system for product R&D. Background Technology

[0002] In the development of complex systems (such as spacecraft and nuclear power plants), system design and behavior are essentially defined by logical propositions contained in a series of technical documents (such as requirements and design reports). The correctness, consistency, and completeness of these documents directly affect the safety and cost of the product.

[0003] However, existing technologies face two challenges: 1. The problem of "loose logic" in natural language documents: Mainstream practices rely on natural language documents that suffer from structural deficiencies and semantic ambiguity, resulting in "loosely logical" expressions. This leads to: Human interpretation bias: can easily lead to design errors and project delays in team collaboration; Machine processing barrier: Its loose structure is difficult for computers to automatically parse, and it cannot be directly connected to automated verification tools.

[0004] 2. The application threshold of formal methods: Although formal methods (such as model checking) can provide rigorous mathematical guarantees for system verification and have been adopted by international safety standards such as DO-178C and EN50128, their engineering application faces extremely high barriers, as detailed below: Highly dependent on experts: Natural language documents must be manually parsed and formally transcribed by professionals in a tedious and error-prone manner; Inefficient process: This manual process severely restricts the large-scale promotion of formal methods and their practical engineering benefits.

[0005] Therefore, there is a seemingly insurmountable gap between the ease of writing natural language and the rigor of formal verification in the current state of technology.

[0006] To overcome the above-mentioned difficulties, existing technologies have provided a variety of partial solutions, but all of them have fundamental limitations. The existing technologies are as follows: 1. A document management system without semantic awareness: Traditional systems like Git and SVN are limited to managing file versions and permissions, completely unable to understand the semantics and logic of document content. This makes them unable to identify and prevent the aforementioned "loose logic" problems, such as conflicts in technical specifications across documents and inconsistencies in design descriptions.

[0007] 2. Limitations of rule checking tools: Existing rule-based compliance inspection tools can only perform superficial checks such as terminology and formatting. Their static rule systems cannot understand complex engineering logic and are powerless to address key quality issues such as deeper logical consistency and design completeness.

[0008] 3. Verification platforms that rely on "ideal input": While standalone MBSE and formal verification tools (such as TLA+ and Coq) possess powerful analytical capabilities, their operation presupposes a perfectly formalized system model as input. This technology places the highest technical hurdle—the conversion from natural language to a formal model—entirely on engineers, making it difficult to popularize in engineering practice.

[0009] In summary, existing technologies lack intelligent document management systems that can automatically understand unstructured technical documents and proactively build and maintain the formal logical models behind them.

[0010] The aforementioned deficiencies in existing technologies have resulted in a severe lack of support from the document management system for the two key management systems: For Quality Management Systems (QMS): Traditional systems cannot provide an automated, real-time, and unambiguous chain of evidence for evidence-based decision-making. The quality inspection of document content relies entirely on manual post-review, making it difficult to effectively implement the "risk prevention" and "process approach" of QMS.

[0011] For Project Management Systems (PMS): Project management relies heavily on manual reporting and milestone reviews, making it impossible to automatically and objectively identify technical risks and the impact of changes from design documents, leading to delayed decision-making and uncontrolled costs.

[0012] For the reasons mentioned above, the current development of complex products relies heavily on manual labor, which severely restricts the quality and efficiency of research and development. Summary of the Invention

[0013] In view of the problems existing in the prior art, the present invention provides a digital document management system for product development.

[0014] This disclosure provides a digital document management system for product development, comprising: The interaction service layer is configured to receive multimodal technical documents and external system data input by users, and provide a visual interface and service call interface; The core processing layer is communicatively connected to the interaction service layer, and the core processing layer includes: The Large Language Model (LLM) processing module is used to perform deep semantic parsing on the input natural language technical documents, extract the logical elements, and generate a structured intermediate representation containing logical semantics. A formal reduction converter, connected to the LLM processing module, is used to automatically convert the structured intermediate representation into at least one formal model and corresponding system property reduction. A model verification processor, connected to the formal reduction transformer, is used to automatically verify the formal model and output verification results and analysis reports. The data support layer, which communicates with the core processing layer, is used to store and manage system operation data.

[0015] Optionally, the large language model LLM processing module includes: Domain-adaptive and fine-tuning units optimize the basic large language model using relevant domain technical literature and standard specifications, thereby improving the accuracy of the large language model in understanding engineering semantics; The deep semantic parsing unit is used to extract propositions that define the properties of system components and the relationships between components from technical documents; The context-aware and fusion unit is used to integrate information from different documents and drawings, build a globally unified system view, and perform consistency verification. The interaction and evolution unit is used to output clarified inputs, optimized model parameters, and updated parsing strategies based on real-time detected semantic ambiguity and logical conflict signals, as well as negative example data from the model validation processor, through an active clarification mechanism and a negative example-driven negative sample learning algorithm.

[0016] Optionally, the consistency check includes: During the document writing process, the current input content is compared with the information already existing in the knowledge graph in real time, and potential logical conflicts or deviations from the norms are indicated based on the comparison results.

[0017] Optionally, when the consistency verification result indicates that the technical document has changed, the change impact analysis is automatically triggered. That is, the model verification processor performs regression verification on the affected formal model to determine the chain reaction and consistency risk caused by the change.

[0018] Optionally, the formal reduction transformer includes: The mapping rule base and template processing unit are used to store rules that map natural language patterns into formal logic components; A multi-paradigm model synthesis unit is used to automatically generate at least one of state machines, sequential logic formulas, and process algebra models based on system characteristics. Attribute specification generator is used to automatically derive and generate formalized system attribute specifications based on design intent or user input.

[0019] Optionally, the model validation processor includes: A multi-paradigm validator integration framework for unified scheduling and management of various formal verification tools; The verification executor is used to distribute verification tasks and computing resources based on the multi-paradigm verifier integration framework. It realizes the verification of formal models and the discovery of counterexamples through task scheduling, state space exploration and attribute determination algorithms. The counterexample analyzer is used to convert the counterexample path data generated when verification fails into a structured report containing natural language descriptions and visual time series diagrams; The results synthesizer is used to automatically generate structured verification proof documents based on the verification pass conclusions output by the verification executor, through report generation and compliance document templates. The performance optimizer is designed to improve verification efficiency and expand the scale of problems that can be handled by abstracting and refining, symbolic model detection, and parallel distributed computing, based on the computational complexity challenges faced by verification tasks. It outputs optimized verification tasks and resource allocation schemes.

[0020] Optionally, the multi-paradigm verifier integration framework includes a model detector, a theorem prover, or an SMT solver.

[0021] Optionally, the data support layer stores a unified knowledge graph, which is based on entities, attributes and relationships extracted from all technical documents, and constructs and maintains a global semantic association network of the system through graph database technology and ontology reasoning mechanism.

[0022] Optionally, the interaction service layer further includes a system visualization and monitoring center, which is configured as follows: Based on the verification results, generate and display at least one quantitative indicator among the entity's verification status, defect convergence trend, and project health. The quantitative indicators are pushed to the external system in real time through a standardized interface.

[0023] Optionally, the system is deployed as a microservices architecture; The large language model LLM processing module is built on an open-source large language model that has undergone domain-adaptive pre-training and fine-tuning.

[0024] The digital document management system for product development provided by this invention extracts the logical elements of natural language technical documents through a large language model and generates a standard formal model. This achieves automatic conversion from unstructured natural language technical documents to formal models, thereby bridging the semantic gap between natural language and machine-executable logic. It solves the problems of human comprehension bias and computer processing obstacles caused by the loose logic of natural language documents in the prior art, thus achieving the goal of improving R&D quality and efficiency. Attached Figure Description

[0025] The above and other objects, features and advantages of this disclosure will become more apparent from the more detailed description of exemplary embodiments of this disclosure taken in conjunction with the accompanying drawings, in which the same reference numerals generally represent the same components.

[0026] Figure 1 A schematic diagram of a digital document management system for product development provided in this embodiment of the disclosure. Detailed Implementation

[0027] The embodiments of this disclosure will now be described in detail with reference to the accompanying drawings.

[0028] The term "documents" as used herein includes all written records related to the product, such as design requirements, design specifications / reports, and test reports; the term "system" refers to the complete product or document management system. Counterexample Analysis Report: This is a structured document automatically generated by the system's model verification processor after it discovers that a formal model violates a certain property specification, using its counterexample analyzer component. The core purpose of this report is to transform the machine-readable raw counterexamples output by the verification tool into an engineering analysis document that engineers can directly understand and use to guide debugging. MBSE: Model-Based System Engineering; LLM: Large Language Model; QMS: Quality Management System; LTL: refers to Linear Sequential Logic; CTL: Refers to Computation Tree Logic; ECSS: European Cooperation for Space Standardization; DO-178C: This refers to the standard jointly developed by RTCA (Radio Technical Committee on Aeronautics) and EUROCAE (European Organization for Civil Aviation Equipment). Its full name is DO-178C: Software Considerations in Airborne Systems and Equipment Certification. ISO 26262: refers to Road Vehicles Functional Safety, which is a set of international standards for automotive electronic and electrical systems. It provides a complete set of processes, methods and management systems to avoid hazards caused by the malfunction of electronic and electrical systems, thereby ensuring the functional safety of automobiles. Digital model: refers to a hybrid model that combines the structured features of MBSE with the mathematical semantics of formal methods; Structured representation refers to an information representation paradigm whose core lies in forcibly constraining the internal logic and organization of information within a predefined, machine-parseable framework while retaining the vocabulary and sentence structure of human-readable natural language. It does not convert natural language into another programming or mathematical language, but rather standardizes, normalizes, and encapsulates it, transforming it from a "loosely logical" state to a "rigorously logical" one. Properties are logical propositions that describe the inherent characteristics and behavioral rules of a system or its components (entities). They define what an entity "is" and the constraints that dictate its "how it must behave," representing invariants that the system must always satisfy or behavioral rules that must be triggered under certain conditions. In other words, properties define the behavioral rules and constraint objectives of the system. Relationships refer to the organizational structure and interaction paths of a system; they are propositions describing the logical or physical connections between different components (entities) of the system. They define "how entities connect" and "how they interact," constructing the organizational network and behavioral interaction map of the entire system. Structured intermediate representation: refers to a machine-readable, logically semantically rich transitional data structure built between natural language and formal specifications; Alloy: refers to structural modeling languages ​​and their accompanying analysis tools. Its core concept is "to discover design flaws before coding through lightweight formal methods." B-method refers to a formal approach based on refinement and mathematical proof, used to specify, design, and implement highly reliable and secure software systems. CCS: Calculus of Communicating Systems, is a theory used to describe concurrent systems, but it focuses more on viewing concurrent systems as mathematical models that can be constructed from basic components through specific combination rules. It is more like a simplified, formalized "calculus" system. Coq: refers to the interactive theorem prover; CSP: Communicating Sequential Processes, a modeling language / theory for describing concurrent systems. Its core concept is that a concurrent system consists of several processes running in parallel, communicating synchronously with each other through channels. It emphasizes the communication itself, rather than the internal state of the processes. Isabelle / HOL: Isabelle is an interactive theorem prover. It provides an environment that allows users to define formal systems (such as various logic systems and programming languages), formally express definitions, lemmas, and theorems within these systems, and then complete the proof step by step under the rigorous checks of the system. HOL is an abbreviation for Higher-Order Logic. This is the most mature and commonly used "instance" of the Isabelle framework. JSON: A lightweight, text-based data format that organizes data using key-value pairs. It is easy for humans to read and write, and also easy for machines to parse and generate. SMT: Satisfiability Modulo Theories, whose solvers include Z3 and CVC5. A solver is an algorithm engine or computer program that performs automated logical reasoning and ultimately provides a solution. TLA+: Refers to a formal language and toolset for modeling, reducing, and verifying concurrent and distributed systems. Its core idea is that "all systems can be described as the changes of their states over time." XML stands for Extensible Markup Language, which uses tags to define data and data structures. It is more verbose than JSON, but very powerful, especially in document markup and complex data domains.

[0029] Formal methods enable comprehensive and effective automated analysis and verification by transforming sets of propositions from natural language into precise, unambiguous mathematical expressions.

[0030] Its core value lies in its ability to systematically answer several key questions about this set of propositions: Consistency: Are there any contradictions among these propositions? Completeness: Are any key scenarios missing from the description? Correctness: Do these propositions ensure that the system will function as expected? In this way, formal methods elevate system design from relying on personal experience and sampling tests to a new level based on mathematical and comprehensive logical guarantees.

[0031] Current complex product development heavily relies on natural language processing (NLP) technical documents. The inherent "loose logic" of NLP, combined with the "semantic ignorance" of existing document management tools and the "high barrier to entry" of advanced verification methods, leads to the following challenges. Under current technological conditions, these problems are primarily solved manually, severely restricting development quality and efficiency. This embodiment aims to systematically address the following technical challenges arising from this: ① The contradiction between the high barrier to entry for verification technology and its widespread applicability in engineering: How can we overcome the heavy reliance of advanced verification techniques such as formal methods on professional knowledge, significantly lower the application threshold, and enable engineers without a formal background to easily obtain rigorous mathematical-level verification guarantees, thereby resolving the contradiction between their "theoretically powerful" and "practically difficult to use" nature?

[0032] ② The contradiction between superficial document management and the need for a deep management system: For QMS (Quality Management System): How to break through the limitations of traditional document management systems that can only manage "document format" (version, permissions) and achieve automated verification of the logical correctness and consistency of "document content", thereby providing automated and real-time in-depth support for the basic principles of QMS such as "evidence-based decision-making", "process approach" and "risk prevention".

[0033] For PMS (Project Management System): How to address the problem that traditional systems cannot automatically and objectively gain insight into project technical risks and status from technical documentation? Current project management heavily relies on manual reporting and milestone reviews, resulting in poor visibility of technical status, delayed risk assessment, and long decision-making chains. This implementation aims to provide PMS with real-time, quantitative data based on the completeness and consistency of document logic, enabling accurate insight and proactive early warning of project technical progress and risks.

[0034] ③ Static defects in technical documentation are difficult to detect, and dynamic changes are difficult to control: Static correctness and completeness: How to systematically and automatically detect and ensure the integrity of the technical documentation system. Logical correctness: Eliminate logical contradictions and descriptive conflicts that exist within a document and across documents; Content completeness: Identify and warn of missing coverage and definition omissions in key scenarios to ensure complete system-level descriptions; Dynamic coordination: How to automatically and accurately analyze and warn of the chain reaction and consistency failure risks caused by changes in documents, between documents, and even in the global system model during the design change process, so as to prevent global logical conflicts caused by local modifications.

[0035] ④ Bridging the semantic gap between natural language and formal models.

[0036] This embodiment integrates large language models and formal methods to construct a complete toolchain from unstructured input to formal verification. The core lies in significantly lowering the application threshold of formal methods, transforming them from a highly specialized expert skill into a readily available daily design tool for a wide range of engineers. This transformation enables the early, automatic, and more comprehensive application of formal methods in the design and verification of complex products (such as satellites and spacecraft), achieving rigorous mathematical-level verification of system designs. This not only fundamentally identifies and eliminates deep-seated logical flaws, compensating for the shortcomings of traditional testing, but also elevates Quality Management System (QMS) and Project Management System (PMS) to a new level based on objective logical data, thereby systematically improving product correctness, reliability, and R&D efficiency.

[0037] To address the aforementioned problems, this embodiment discloses a digital document management system for product development, comprising: The interaction service layer is configured to receive multimodal technical documents and external system data input by users, and provide a visual interface and service call interface; The core processing layer is communicatively connected to the interaction service layer, and the core processing layer includes: The Large Language Model (LLM) processing module is used to perform deep semantic parsing on the input natural language technical documents, extract the logical elements, and generate a structured intermediate representation containing logical semantics. A formal reduction converter, connected to the LLM processing module, is used to automatically convert the structured intermediate representation into at least one formal model and corresponding system property reduction. A model verification processor, connected to the formal reduction transformer, is used to automatically verify the formal model and output verification results and analysis reports. The data support layer, which communicates with the core processing layer, is used to store and manage system operation data.

[0038] Interaction service layer: This layer serves as the information exchange hub between the system and its internal and external environments. Based on user commands, external system data, and internal service requests, it achieves the technical goal of aggregating internal needs and providing external capabilities through unified interface specifications and business service encapsulation. It outputs a visual interface, structured data, and service call results, thus connecting users, external systems, and the core processing layer to ensure efficient, accurate, and bidirectional information flow. This includes: User interaction interface: The natural language input interface supports multimodal interaction via voice and text. Visual modeling tools enable graphical system architecture design; A real-time collaborative editing environment that supports concurrent work by multiple roles; The intelligent document workbench provides writing assistance and real-time consistency checks.

[0039] External system integration: Engineering toolchain: data import and integration with management systems for CAD / CAE design tools and simulation software; QMS (Quality Management System): Transforms quality requirements into inspectable technical rules; PMS (Project Management System): Enables the linkage between technical progress and project milestones; PLM / ERP systems: Correlation analysis between technical parameters and business resources; Integration of verification tools: Task distribution and result collection for external formal verification tools.

[0040] Core business services: Document compliance check service: Automatically performs basic compliance checks; Semantic parsing service: cross-document association analysis and consistency check; Formal transformation service: Generate formal models of different levels on demand; Verification Analysis Service: Automated verification task scheduling and impact analysis.

[0041] System Visualization and Monitoring Center: Based on the verification results generated by the core processing layer and the historical information stored in the data support layer, visualization technologies such as dynamic dashboards, trend charts, and traceability views are used to achieve the technical goal of aggregating the underlying state of the system and data into advanced management insights. This outputs key quantitative indicators such as the verification status of entities (e.g., pass rate), defect convergence trends, project health, and quality status, providing project managers and quality engineers with real-time and objective decision-making basis, and realizing the fundamental purpose of data-driven R&D management.

[0042] Core processing layer: This layer serves as the technical processing hub of the system. Based on the multi-source heterogeneous data and verification requests gathered from the interactive service layer, it achieves the technical goals of deep analysis, semantic understanding, logical mapping, and mathematical verification of raw data through the collaborative pipeline operation of four core components: multimodal input adapter, large language model LLM processing module, formal reduction transformer, and model verification processor. It outputs verification conclusions for decision-making, counterexample analysis reports, and a continuously updated unified knowledge graph, thereby overcoming the "semantic gap" problem, laying the foundation for quality, and providing automated verification capabilities for the document management system.

[0043] Multimodal input adapter: Function: Unified reception and preprocessing of various types of input data, converting them into an internal standard format.

[0044] Composition: Includes processing modules such as text parsing, image recognition, and data standardization.

[0045] Text parsing module: Connects to the large language model LLM processing module to perform preliminary processing on natural language technology documents; Image recognition module: Based on computer vision technology, it extracts component symbols and connection relationships from engineering drawings and architecture diagrams; Data standardization module: Converts data from tools such as CAD / CAE / simulation into a unified intermediate representation.

[0046] The LLM processing module (semantic understanding core) is a domain-specific customized large language model. Its function is to deeply understand the engineering semantics of technical documents and transform them into logically rich, structured knowledge. Unlike general-purpose large language models, it achieves accurate parsing of complex engineering language through the following four collaborative functional units: Domain Adaptive and Fine-Tuning Unit: Based on technical documents, standards, and historical project documents from high-reliability industries, it achieves the technical goal of accurately understanding engineering semantics through domain adaptive pre-training and task-specific fine-tuning, and optimizes instructions using paired data (natural language, formal logic). It outputs a language model with domain expertise and conforms to formal specifications, ultimately providing a reliable semantic parsing foundation for the document management system.

[0047] Deep Semantic Parsing Unit: Based on multi-source heterogeneous input data composed of natural language, formal language and specific computer languages, it uses deep semantic parsing technology and complex logical relationship extraction algorithms to achieve the technical goal of accurately extracting logical elements from technical documents, outputting structured intermediate representations and associated contextual information, and thus providing the core data foundation for static consistency verification of the document management system.

[0048] Context-aware and fusion unit: Based on multimodal information from different documents and drawings, it achieves the technical goal of building a unified system view through cross-document reference resolution and multimodal information alignment technology. It outputs globally unique knowledge entities and cross-modal consistency verification results, thereby achieving the management purpose of maintaining the global knowledge consistency and integrity of the document management system.

[0049] Interaction and Evolution Unit: Based on real-time detected semantic ambiguity and logical conflict signals, as well as negative example data from the verification core, it achieves the technical goal of continuously improving the accuracy of the system through an active clarification mechanism and a negative example-driven negative sample learning algorithm. It outputs clarified accurate input, optimized model parameters, and updated parsing strategies, ultimately enhancing the adaptive capabilities of the file management system and the user experience.

[0050] Formal reduction transformer (logic mapping core): This component serves as a crucial bridge connecting "understanding" and "verification," acting as the logical mapping center of the system. Based on the structured intermediate representation output by the Large Language Model (LLM) processing module, it achieves the technical goal of unambiguously mapping engineering semantics to a precise mathematical logic model through the collaborative work of multiple dedicated components within it. It outputs a verifiable formal model and attribute specification for the core system, laying a solid foundation for the automated verification of the document management system and fundamentally bridging the gap between natural language and machine-processable logic.

[0051] Mapping rule base and template processing unit: Based on the domain-standard formal logic paradigm and engineering expression habits, this paper achieves the technical goal of automatically matching and converting common natural language patterns (such as "when..., it must be...") into formal components through an internalized domain knowledge, scalable mapping rule base and template system. It outputs preliminary structured logical expressions, thereby improving conversion efficiency and accuracy and reducing sensitivity to the original input expression.

[0052] Multi-paradigm model synthesis unit: Based on the preliminary structured logical expressions and the characteristics of the described system, this paper uses various algorithms such as state machine synthesis, temporal logic formula generation, and process algebra modeling to achieve the technical goal of automatically generating the most suitable formal model on demand. It outputs multi-paradigm formal models of the system or components (such as state machines, LTL / CTL formulas, etc.), so as to enable the document management system to flexibly adapt to the verification needs of different complex systems and provide the optimal verification strategy.

[0053] Interface Contracts and System Composition Framework: Positioned as the core support architecture for managing the complexity of component interactions in a file management system, this system, based on formal contract theory and combinatorial reasoning, aims to provide a unified methodology for modular design, large-scale verification, and precise change impact analysis of the file management system by defining standardized interface contract specifications and system composition rules. It outputs a component interface contract library and combinatorial verification strategies, ultimately providing scalable and reasonable integrity assurance for the development of complex systems.

[0054] Attribute reduction generator: Based on the verification requirements automatically derived from the design intent and the user-defined verification requirements, the system uses an attribute derivation algorithm and a guided interactive interface that guides users to define attributes step by step. This automatically converts informal user requirements into formal attributes and guides users to supplement and confirm the technical goals of the attribute definitions in natural language. The system outputs a complete and accurate set of system attribute specifications, ensuring that the verification activities of the document management system can cover implicit design intent and respond to explicit natural language instructions from users, thereby fully supporting the fundamental purpose of customized verification tasks.

[0055] Model Validation Processor (Mathematical Validation Core): This component serves as the core of the system's mathematical verification. Based on the system model and attribute reduction output by the formal reduction converter, it achieves the technical goal of automating and exhaustively verifying key system attributes through its internal multi-paradigm verifier integration framework. It outputs verification conclusions accompanied by natural language explanations and visualized time series diagrams, thus providing mathematical-level correctness guarantees for the document management system. Its core architecture consists of the following components: Multi-paradigm validator integration framework: Positioned as a unified support architecture for coordinating and managing various verification tools within the model verification processor, it leverages the complementary capabilities and performance of verification paradigms such as model checking, theorem proving, and SMT solving. Through scalable verifier interface specifications and intelligent scheduling algorithms, it achieves the technical goal of unified encapsulation and on-demand scheduling of heterogeneous verification tools, outputting optimal verifier selection decisions and task distribution instructions. This ultimately establishes a core technical mechanism for automatically selecting the optimal verification path based on the characteristics of the verification task.

[0056] Verify the executor: Based on the multi-paradigm validator integration framework, the distributed verification tasks and computing resources, through an automated verification pipeline consisting of task scheduling, state space exploration and attribute determination algorithms, achieve the technical goals of rigorous verification and counterexample discovery of formal models. It outputs the basic conclusion of "verification passed" or "verification failed" and the original counterexample path data that led to attribute violations, thereby driving the file management system to automatically execute the deep verification process.

[0057] Counterexample Analyzer: Based on the raw counterexample path data output by the verification executor, natural language generation and visualization technologies are used to transform machine-guided counterexample information into technical objectives of fault scenarios that engineers can understand. The output is a structured counterexample report containing natural language descriptions (e.g., "at t=15ms, ... causes a 2ms pass-through short circuit") and a visualized timing diagram. This transforms complex verification results into intuitive debugging information, thereby significantly reducing the barrier to entry for using the file management system.

[0058] Result synthesizer: Based on the "verification passed" conclusion output by the verification executor, the technical goal of automatically generating structured verification documents is achieved through report generation and compliance document templates. This results in the output of standardized verification reports that meet the requirements of the quality management system, thereby automatically generating audit evidence for the document management system and satisfying industry compliance requirements.

[0059] Performance optimizer: To address the computational complexity challenges posed by verification tasks, such as state space explosion, this paper employs abstraction and refinement, symbolic model detection, and parallel distributed computing techniques to improve verification efficiency and expand the scale of problems that can be handled. The output is an optimized verification task and resource allocation scheme, thereby ensuring the feasibility of the document management system for efficiently verifying complex industrial systems.

[0060] The data support layer is used for unified knowledge storage. This layer serves as the unified data hub of the system. Based on the various structured knowledge and historical data generated by the core processing layer, it achieves the technical goal of providing efficient and reliable data support for upper-layer services through domain-oriented professional data storage and management strategies. It outputs unified and standardized data access results, thereby ensuring the fundamental purpose of ensuring the consistency, integrity, and traceability of data in the document management system.

[0061] Its internal structure consists of the following core data warehouses: Unified Knowledge Graph: Based on entities, attributes, and relationships extracted from all technical documents, this system utilizes graph database technology and ontology reasoning mechanisms to achieve the technical goal of constructing and maintaining a global semantic relationship network. It outputs relational query results and consistency verification feedback, thus fulfilling its core objective of serving as a "single source of fact" for the document management system and ensuring all functions operate based on a unified and accurate context. The graph database serves as the storage engine and efficient query engine for the unified knowledge graph, while the ontology reasoning mechanism acts as its brain, upgrading the knowledge graph from a static relational map to an expert system with logical judgment capabilities.

[0062] Formal model library: Based on the models and verification history generated by the formal specification converter, and through versioned storage and structured data management, the goal of full lifecycle management of formal specifications, interface contracts and verification evidence is achieved. Specific model versions, contracts and verification history are output, thus providing the core data foundation for traceability analysis, impact assessment and incremental verification of the document management system.

[0063] Standards and Specifications Library: Based on industry standards, corporate specifications, and quality management system requirements, this system stores structured and configurable rule templates to achieve the technical goal of transforming text standards into machine-readable and executable rules. It outputs specific inspection rules and verification attribute templates, thereby providing authoritative evidence for the automated compliance inspection and quality verification of the document management system.

[0064] Operation logs and audit database: Based on all user operations, document changes, verification task executions, and other key events within the system, high-throughput logging and storage technologies are used to achieve the technical goal of complete and tamper-proof recording of all system activities. This results in the output of operation history records, compliance audit reports, and system operation indicators, thereby meeting the mandatory requirements of the Quality Management System (QMS) for "process traceability" and "evidence-based decision-making," and providing data support for system operation and maintenance and fault diagnosis.

[0065] Permissions and configuration library: Based on the enterprise's organizational structure, role definition, and project management strategy, this system achieves centralized and unified management of users, roles, access permissions, and system operating parameters through a fine-grained permission model and a configurable strategy engine. It outputs permission verification decisions, personalized user views, and system configuration parameters, thereby ensuring the security of intellectual property and sensitive data in the file management system and supporting complex collaboration scenarios involving multiple projects and teams.

[0066] The inputs to this system are designed to interface with multi-source heterogeneous data within the R&D system, specifically including: Informal input: This type of input requires the system to perform deep semantic parsing and structural transformation; Natural Language Technology Documentation: Product requirements, design specifications, interface control documents, test cases and reports, etc.; Graphical design documents: system architecture diagrams, engineering drawings, logic flowcharts, etc.; Collaborative process assets: design review comments, meeting minutes, experience feedback, etc.

[0067] Structured and formal input: This type of input can be understood and utilized by the system directly or after adaptation; Engineering data and models: Design data and simulation results from CAD / CAE / EDA tools; Validating assets: specifications, models, and validation results from existing formal validation tools; Standards and Specifications: Digital texts of industry standards, corporate specifications, and quality management system processes; Management data: Process and status data from the Project Management System (PMS) and Quality Management System (QMS).

[0068] System output: The output of this system provides directly usable valuable information to users in different roles and downstream systems, specifically including: Structured output: Specifications and Models: Formal specification documents (LTL / CTL formulas, state machines, process algebras), standardized data models and interface definitions; Machine-readable knowledge: structured natural language descriptions (JSON / XML format), unified knowledge graph datasets; Reusable assets: formal model library, verification attribute templates, compliance rule set.

[0069] Visual output: Architecture and Traceability View: System architecture overview and real-time requirement traceability chain view; Status monitoring dashboard: Project health dashboard, real-time verification status dashboard, quality indicator trend chart; Visualization of analysis results: defect distribution chart, change impact propagation diagram, and counterexample time series diagram.

[0070] Analysis Report: Verification and Analysis Reports: Automated verification reports (including pass / fail attributes and counterexamples with natural language interpretation), multi-dimensional compliance check reports; Impact Analysis Report: Change Impact Analysis Report (precisely listing the affected documents, components, and validation items), Design Consistency Audit Report; Quality and Management Reports: Quality Status Assessment Report, Project Technical Risk Early Warning Report.

[0071] This system systematically solved all the identified technical problems through the collaborative work of functions at all levels: Regarding "lowering the application threshold of high-assurance verification technology": Through a natural language interface at the interface layer and an LLM processor at the core processing layer, automatic conversion from natural language familiar to engineers to formal specifications is achieved. Users do not need to master professional formal languages ​​to initiate and obtain rigorous verification results, thereby democratizing verification capabilities.

[0072] Regarding "making up for the lack of in-depth support for QMS and PMS in the document management system": Support for QMS: Through services such as document compliance checks and semantic parsing at the service layer, the "evidence-based decision-making," "process approach," and "risk prevention" principles of QMS are transformed into automatically executable check rules. The unified knowledge graph at the data layer ensures that all decisions are based on a single, consistent source of facts. Support for PMS: Through the application-layer verification status dashboard and quality trend analysis, as well as the service-layer verification analysis service, real-time, quantitative technical status data (such as verification pass rate, number of defects, and consistency health) is provided to the PMS. This data is integrated with the project management system (PMS) through the interface layer, enabling project managers to objectively understand technical risks and assess the actual progress of tasks, realizing a transformation from a "report-dependent" to a "data-driven" management model.

[0073] Regarding "systematically resolving the issues of static correctness, completeness, and dynamic consistency in the technical documentation system": Static correctness and completeness: Deep semantic parsing and logical understanding are performed through the LLM processor in the core processing layer, combined with semantic parsing services in the service layer, automatically detecting logical contradictions and missing content within and across documents. The model validation processor then performs a completeness check on the generated formal model. Dynamic coordination: Through the service layer's verification and analysis service, when a change occurs, the core processing layer's model verification processor is automatically invoked to perform impact scope analysis and regression verification, ensuring that local modifications do not disrupt global consistency and effectively preventing "chain reactions".

[0074] Regarding "bridging the semantic gap between natural language and formal models": The core processing layer's "LLM processor" and "formal reduction transformer" together form the technological bridge to solve this core problem. The LLM processor is responsible for understanding natural language, while the reduction transformer is responsible for performing precise mappings. Working together, they realize an automated pipeline from unstructured information to accurate mathematical models.

[0075] This embodiment can also be extended and optimized through the following implementation methods and modifications: Enhancement and specialization of core methodologies: Construction of Domain-Specific Systems: Based on the core architecture of this invention, MBSE collaborative platforms can be built for specific domains (such as aerospace and automotive electronics). This variation enhances the system's understanding and inspection of domain-specific specifications (such as DO-178C and ISO 26262) by deepening the domain ontology library and pre-trained corpus.

[0076] Formal Integration of Quality Management: This involves formally modeling the processes and review criteria within the Quality Management System (QMS) and deeply integrating them with the document verification process. This transformation enables fully automated compliance verification across the entire chain, from "quality requirements" to "design implementation" and then to "verification evidence," significantly improving audit efficiency.

[0077] Expansion of application scope: Dynamic Generation of Intelligent Document Templates and Specifications: By leveraging the formal model library and knowledge graph of this system, reverse operations can be performed to automatically generate standardized document templates, writing outlines, and content specifications. This transformation can solidify excellent design patterns into enterprise standards, directly enhancing the starting quality and efficiency of document compilation.

[0078] Knowledgeization of Standards and Specifications: Using the large language model (LLM) processing module of this system, in-depth parsing and structured transformation of vast natural language standard and specification documents are carried out to form a "digital standard library" that can be understood and reasoned by machines. This transformation can ensure that design activities are always based on the latest and most accurate specifications and can automatically identify potential deviations between designs and standards.

[0079] Diversification and Intelligence of Implementation Means: Development and Interaction of Natural Language-Driven Formal Tools: Extend the application of LLM from the front-end of "specification generation" to "the use and development assistance of formal tools themselves". For example, through natural language instructions, let LLM automatically generate scripts for specific verification scenarios, configure complex verification parameters, or explain deep error messages in the tool chain, thereby further reducing the usage threshold of the entire formal technology stack.

[0080] Deepening of Multimodal Interaction: Beyond text, develop the ability to deeply understand and perform two-way conversion for graphics such as engineering drawings and system architecture diagrams. The system can not only extract information from graphics but also reverse-label verification results (such as counterexamples) back to the original drawings, enabling the linkage traceability and visual debugging of "text-graphics-model".

[0081] Expansion of System Architecture: Distributed and Cloud-Native Architecture: Deploy the core verification service as a microservice cluster in the cloud. Through distributed computing technology, share the computing load brought by large-scale system verification, and through containerization technology, achieve elastic scaling of resources to support the concurrent use of large R & D teams and the verification tasks of ultra-large-scale systems.

[0082] Lightweight and Incremental Deployment: This invention can also be integrated into existing PLM or ALM systems in the form of lightweight components and run as a high-level intelligent plugin. Support incremental model verification and knowledge graph update, avoid triggering full-system re-verification due to minor changes, and improve practicality and ease of use.

[0083] As Figure 1 shown, the system data input of this embodiment: The system receives multi-source inputs through three entrances in the interaction service layer: User Interaction Interface: Receive natural language documents, drawings, review opinions, etc.; External system integration: interfacing with CAD / CAE models, simulation software models and results, standards and specifications, etc. Core business services: handling management data, verifying requests, etc.

[0084] The core processing layer constructs a cascaded information refinement and enhancement mechanism through the sequential processing and collaboration of four functional components: Multimodal input adapter → standardizes various input data; Large Language Model (LLM) processing module → deep semantic parsing and knowledge extraction; Formal reduction converter → logical mapping and model generation; Model verification processor → mathematical verification and counterexample analysis.

[0085] Data Support System: The data support layer provides comprehensive support for the processing flow, including a unified knowledge graph that stores semantic parsing results, a formal model library that manages models and validation history, a standards and specifications library that provides compliance basis, an operation audit library that records all system operations, and a permission configuration library that controls access permissions at each layer.

[0086] Output Stage: All outputs are aggregated and presented uniformly through the system visualization and monitoring center, including: Decision support reports: verification reports, compliance documents, analysis reports, etc.; Visual outputs include: status dashboards, traceability views, quality trend charts, etc. Structured data output: machine-readable data assets such as formal models and knowledge graph datasets; Integration with PMS (Project Management System): Key quantitative indicators such as entity verification status, defect convergence trend, and project health are pushed to the project management system in real time and proactively through standardized interfaces, enabling accurate insight and proactive early warning of technical progress and risks.

[0087] In a specific scenario, this system is implemented as a web platform based on a microservices architecture. The front end provides users with an interactive interface accessible through a browser, while the core back end services are developed using Python and Java, and are encapsulated and deployed using containerization technology.

[0088] Interaction Service Layer: The user interface is built using the React framework and equipped with a standard RESTful API gateway to receive call requests from external systems (such as PLM and QMS).

[0089] Core processing layer: The large language model LLM processing module is based on the open-source Llama 3 model architecture and uses hundreds of thousands of technical documents and industry standards (such as ECSS and DO-178C) accumulated within the enterprise for domain-adaptive pre-training and fine-tuning.

[0090] The formal reduction converter and model verification processor integrate open-source formal tools such as NuSMV (model checking) and Z3 (SMT solver), and are uniformly scheduled through a custom wrapper.

[0091] Data support layer: Graph database technology is used to store a unified knowledge graph, and PostgreSQL relational database is used to store a formal model library and a standard specification library, ensuring data persistence and the ability to perform complex relational queries.

[0092] The specific workflow example is as follows: Take, for example, a systems engineer handling a "satellite power controller" design document: Document Creation and Intelligent Assistance: Engineers log into the system and create a new "Power Controller Interface Design Specification" in the "Intelligent Document Workbench." When they type "output voltage accuracy," the system automatically suggests the applicable enterprise standard "Q / ABC-2023: Output voltage accuracy should be better than ±2%." The engineer can then adopt this suggestion with a single click, ensuring consistency between terminology and specifications.

[0093] Real-time consistency check: When the engineer writes "overvoltage protection threshold is 32V" later in the document, the system background service immediately starts the semantic parsing service and performs a logical comparison with the "rated output voltage 28V" already defined in the knowledge graph within milliseconds. The system then triggers a prompt in the document sidebar: "The difference (4V) between the current setting (32V) and the rated value (28V) does not reach the usual redundancy design margin (recommended >10%), please confirm." This demonstrates the system's reliance on design experience and its real-time intervention.

[0094] One-click verification and feedback: After the document is completed, engineers can directly select the relevant paragraphs and click the "Generate Specification and Verify" button without leaving the platform. The system automatically converts the natural language into a formal specification (such as generating the LTL formula G(voltage<35)) in the background and calls the verification engine. The verification result is presented in the form of a dashboard card: "Verification passed: 3 attributes". At the same time, the system automatically generates a digitally signed, tamper-proof "Verified" status mark for the document in the document management library and establishes a traceable link with the relevant requirement items.

[0095] System Integration Example: Once the "Power Controller Interface Design Specification" is validated, its structured data (such as interface type, voltage, and power consumption) is automatically synchronized to the PLM (Product Lifecycle Management) system via the enterprise application integration interface, generating or updating the corresponding bill of materials items. The entire history of the document's review, modification, and validation is fully recorded in the operation log and audit database. When a quality engineer initiates an audit in the QMS, a compliance documentation package compliant with standards can be generated with a single click, greatly simplifying the audit process.

[0096] Compared with the prior art, this embodiment has the following significant advantages: 1. Significantly lowers the application threshold for high-assurance verification technology: By using natural language interaction and automated formal specification generation, engineering technicians can obtain mathematical-level verification guarantees without being proficient in formal methods. This breaks the dilemma of advanced verification techniques being "too sophisticated for most people" and greatly promotes the popularization of formal methods in engineering practice.

[0097] 2. Achieved in-depth governance of the quality of technical document content: The system can automatically and in real time detect and warn of logical contradictions, descriptive conflicts, and missing content within and across documents, eliminating design defects caused by "loose logic" at the source and significantly shifting the quality control point from "late testing" to "early design".

[0098] 3. It provides objective and quantifiable decision-making basis for project and quality management: For Project Management System (PMS): Through quantitative indicators such as entity verification status and defect convergence trends, it provides a real and transparent view of the technical status of project management, enabling schedule assessment and risk decision-making to shift from relying on subjective reports to being driven by objective data.

[0099] For Quality Management System (QMS): It transforms abstract QMS principles into automatically executable technical rules and provides structured, traceable, and objective evidence for management review, realizing a shift in quality management from post-audit to pre-prevention and process control.

[0100] 4. It provides a systematic solution to the dynamic coordination challenges of change management in complex systems: Through automated impact analysis and regression verification, it is possible to accurately predict and warn of the "chain reaction" triggered by local changes in the global system, effectively preventing systemic risks that "affect the whole body from a single change," and solving the core pain points of traditional methods, such as difficulty, lag, and easy omission in change impact assessment.

[0101] 5. It broke down the key barriers between natural language and machine logic: By deeply integrating LLM with formal methods, an intelligent bridge connecting human abstract thinking with precise machine execution has been successfully built, achieving a leap from "human understanding" to "machine understanding," and laying the core technological foundation for the intelligent upgrade of the entire R&D system.

[0102] 6. A core set of enterprise knowledge that can be retained and reused has been constructed: During the operation of the system, the continuously constructed and enriched unified knowledge graph and formal model library transform scattered and implicit expert experience into explicit assets that can be queried, verified, and reused, forming a "digital brain" that supports the continuous innovation and knowledge inheritance of the enterprise.

[0103] This embodiment systematically improves the R & D quality, efficiency, and reliability of complex products, marking a fundamental transformation of document management from the traditional "document warehousing" mode to the intelligent "quality and knowledge engine" mode.

[0104] A formal model is an abstract computational model that describes the logical behavior of a system using precise mathematical language.

[0105] Suppose there is a simple satellite power control unit, and its partial natural language requirements are as follows: The system has two modes: normal and safe.

[0106] In the normal mode, the scientific payload can be turned on.

[0107] If the battery voltage is lower than the threshold V_min, the system must automatically switch to the safe mode.

[0108] In the safe mode, the scientific payload must be turned off to save power.

[0109] After the scientific payload is turned off, it needs to wait at least T seconds before it can be turned on again.

[0110] Now, build this system into a formal model. The following are several common model paradigms: Example 1: State machine model: This is the most intuitive model, which describes the possible states of the system, as well as the events and conditions that cause state transitions.

[0111] States: normal mode, safe mode.

[0112] Transition conditions: Voltage < V_min causes a transition from the normal mode to the safe mode.

[0113] Voltage >= V_min causes a transition from the safe mode to the normal mode.

[0114] Variable: payload status (on / off).

[0115] This model clearly defines the logic of mode switching.

[0116] Example 2: Temporal logic formula (used to describe property specifications): State machines describe "what the system can do," while sequential logic uses mathematical formulas to strictly define "what the system must / absolutely cannot do." These formulas are the attribute specifications that need to be verified.

[0117] To address the above requirements, the following formal properties can be defined: Security attribute: Specifies that "some bad things will never happen".

[0118] Requirement 4: "In safe mode, scientific payloads must be turned off."

[0119] Formal Specification (LTL): G(System Mode = Security Mode -> Load State = Off); G stands for "always globally". The formal specification means that at any point in time, if the system is in safe mode, then the load state must be off.

[0120] Active attribute: It stipulates that "some good things will eventually happen".

[0121] Implicit requirement: The system must not deadlock and should be able to respond to commands.

[0122] Formal specification (LTL): G(load state = off -> F(load state = on)); F stands for "finally". The formal specification means that if the load is turned off at any time, it will eventually be turned on again. (This is a weaker liveness; a stronger specification would combine mode and time constraints.)

[0123] Temporal attribute: Specifies time-related behaviors.

[0124] Requirement 5: "After the scientific payload is turned off, it needs to wait at least T seconds before it can be turned on again."

[0125] Formal specification: G((load state changes from on to off)->(load state = off U time > T)); U is the "until" operator, which in a formal specification means that once the load is turned off, it must remain off until more than T seconds have elapsed.

Claims

1. A digital document management system for product development, characterized in that, include: The interaction service layer is configured to receive multimodal technical documents and external system data input by users, and provide a visual interface and service call interface; The core processing layer is communicatively connected to the interaction service layer, and the core processing layer includes: The Large Language Model (LLM) processing module is used to perform deep semantic parsing on the input natural language technical documents, extract the logical elements, and generate a structured intermediate representation containing logical semantics. A formal reduction converter, connected to the LLM processing module, is used to automatically convert the structured intermediate representation into at least one formal model and corresponding system property reduction. A model verification processor, connected to the formal reduction transformer, is used to automatically verify the formal model and output verification results and analysis reports. Digital document management systems also include: The data support layer, which communicates with the core processing layer, is used to store and manage system operation data.

2. The digital document management system for product development according to claim 1, characterized in that, The large language model LLM processing module includes: Domain-adaptive and fine-tuning units optimize the basic large language model using relevant domain technical literature and standard specifications, thereby improving the accuracy of the large language model in understanding engineering semantics; The deep semantic parsing unit is used to extract propositions that define the properties of system components and the relationships between components from technical documents; The context-aware and fusion unit is used to integrate information from different documents and drawings, build a globally unified system view, and perform consistency verification. The interaction and evolution unit is used to output clarified inputs, optimized model parameters, and updated parsing strategies based on real-time detected semantic ambiguity and logical conflict signals, as well as negative example data from the model validation processor, through an active clarification mechanism and a negative example-driven negative sample learning algorithm.

3. The digital document management system for product development according to claim 2, characterized in that, The consistency check includes: During the document writing process, the current input content is compared with the information already existing in the knowledge graph in real time, and potential logical conflicts or deviations from the norms are indicated based on the comparison results.

4. The digital document management system for product development according to claim 3, characterized in that, When the consistency verification result indicates that the technical document has changed, the change impact analysis is automatically triggered. That is, the model verification processor performs regression verification on the affected formal model to determine the chain reaction and consistency risk caused by the change.

5. The digital document management system for product development according to claim 1, characterized in that, The formal reduction transformer includes: The mapping rule base and template processing unit are used to store rules that map natural language patterns into formal logic components; A multi-paradigm model synthesis unit is used to automatically generate at least one of state machines, sequential logic formulas, and process algebra models based on system characteristics. Attribute specification generator is used to automatically derive and generate formalized system attribute specifications based on design intent or user input.

6. The digital document management system for product development according to claim 1, characterized in that, The model validation processor includes: A multi-paradigm validator integration framework for unified scheduling and management of various formal verification tools; The verification executor is used to distribute verification tasks and computing resources based on the multi-paradigm verifier integration framework. It realizes the verification of formal models and the discovery of counterexamples through task scheduling, state space exploration and attribute determination algorithms. The counterexample analyzer is used to convert the counterexample path data generated when verification fails into a structured report containing natural language descriptions and visual time series diagrams; The results synthesizer is used to automatically generate structured verification proof documents based on the verification pass conclusions output by the verification executor, through report generation and compliance document templates. The performance optimizer is designed to improve verification efficiency and expand the scale of problems that can be handled by abstracting and refining, symbolic model detection, and parallel distributed computing, based on the computational complexity challenges faced by verification tasks. It outputs optimized verification tasks and resource allocation schemes.

7. The digital document management system for product development according to claim 6, characterized in that, The multi-paradigm verifier integration framework includes a model detector, a theorem prover, or an SMT solver.

8. The digital document management system for product development according to claim 1, characterized in that, The data support layer stores a unified knowledge graph, which is based on entities, attributes and relationships extracted from all technical documents. Through graph database technology and ontology reasoning mechanism, it constructs and maintains a global semantic association network for the system.

9. The digital document management system for product development according to claim 1, characterized in that, The interactive service layer also includes a system visualization and monitoring center, which is configured as follows: Based on the verification results, generate and display at least one quantitative indicator among the entity's verification status, defect convergence trend, and project health. The quantitative indicators are pushed to the external system in real time through a standardized interface.

10. The digital document management system for product development according to claim 1, characterized in that, The system is deployed in a microservices architecture; The large language model LLM processing module is built on an open-source large language model that has undergone domain-adaptive pre-training and fine-tuning.

Images