Onboard embedded software-oriented demand semantic tracking method and system

Abstract

The invention provides a demand semantic tracking method and system for airborne embedded software, and relates to the field of software engineering automation and quality assurance, and the method comprises the following steps: adaptive demand entry extraction: obtaining an extraction seed provided by a user, and automatically generating an enhanced regular expression set through a rule enhancement algorithm; extracting a demand entry; verifying a hardware perception tracking relationship: identifying and extracting functional module entities and hardware peripheral entities of the demand entries by using a pre-constructed domain-specific hardware ontology knowledge base, and calculating a tracking similarity score between the demand entries; and on the basis of the semantic fingerprints of the demand items, establishing a demand main item, establishing a tracking relation set on the basis of the corresponding tracking similarity scores, and generating and exporting a structured tracking matrix report for displaying a multi-tracking relation aggregation result. According to the method, through combination of adaptive extraction, hardware perception verification and structured optimization, the accuracy, robustness and auditing efficiency of airborne embedded software demand tracking are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of software engineering automation and quality assurance technology, specifically to a method and system for requirement semantic tracking of airborne embedded software. Background Technology

[0002] In the development of aerospace and high-reliability airborne embedded software (such as following the GJB5000B standard), requirements engineering exhibits significant multi-stage and multi-level requirements document transmission, necessitating cross-level requirements correlation. A typical requirements tracing chain usually runs from top to bottom: System / Subsystem Specification (SSS), Software Development Task Specification (USER), Software Requirements Specification (SRS), and Software Design Specification (SDD). Constructing and maintaining an accurate Requirements Traceability Matrix (RTM) throughout the entire lifecycle is crucial for verifying requirements coverage, preventing requirements drift, and passing airworthiness certification or quality audits. Existing methods have the following shortcomings: Defect 1: Rigidity and low adaptability of relying on a single regular expression. Existing tools mostly rely on fixed regular expressions to extract identifiers within a single document. However, engineering document formats are often heterogeneous, and relying solely on simple regular expression logic or keyword retrieval is difficult to adapt to the stylistic differences of different writers and the complexity of unstructured text. This results in a high false positive and false negative rate when dealing with non-standardized descriptions, and it is extremely dependent on the operator's regular expression writing skills, lacking practical engineering applicability.

[0003] Deficiency 2. Existing technologies lack hardware context-aware semantic pseudo-similarity misjudgment, a pain point unique to airborne embedded software. Existing technologies mostly use general natural language processing (NLP) to calculate text similarity to assist in tracking, but lack sensitivity to hardware context.

[0004] For example, "XXX module PCIe controller initialization" and "YYY module PCIe controller initialization" have extremely high text similarity, but belong to completely different functional modules; "XXX module QSPI controller initialization" and "XXX module SPI controller initialization" differ by only one character, but they correspond to completely different hardware peripherals and driver logic in embedded systems. Existing technologies can usually effectively distinguish between these literally similar but logically unrelated or functionally orthogonal hardware requirements, making it impossible to deeply verify whether lower-level requirements truly support and implement higher-level requirements, and making it difficult to discover substantial logical breaks.

[0005] Deficiency 3. Lack of aggregation and deduplication mechanisms in many-to-many mappings. In complex airborne software, requirement relationships often present a complex network structure, with situations where "a high-level task is decomposed into multiple low-level requirements" or "multiple low-level designs reuse the same high-level requirement." Existing solutions lack effective aggregation and deduplication algorithms and cannot correctly handle duplicate references or multiple mappings. This leads to redundant and chaotic generated traceability matrices that cannot intuitively reflect the completeness of requirement coverage and are difficult to support efficient quality auditing. Summary of the Invention

[0006] In view of this, embodiments of this application provide a method and system for requirement semantic tracking for airborne embedded software, aiming to solve the problems of requirement tracking gaps and semantic pseudo-similarity misjudgments caused by heterogeneous document formats and hardware-software coupling in the development of airborne embedded software; by combining adaptive extraction, hardware-aware verification and structured optimization, the accuracy, robustness and auditing efficiency of requirement tracking for airborne embedded software are significantly improved.

[0007] This application provides the following technical solution: a requirement semantic tracking method for airborne embedded software, comprising: Adaptive requirement item extraction: Obtain requirement item identifiers provided by the user for the target document type as extraction seeds; based on the extraction seeds, automatically generate an enhanced regular expression set adapted to multiple document formats through a rule enhancement algorithm; use the enhanced regular expression set to extract requirement items from heterogeneous engineering documents and perform text normalization processing; Hardware-aware tracking relationship verification: Using a pre-built domain-specific hardware ontology knowledge base, feature entity recognition is performed on the extracted requirement items to extract the corresponding functional module entities and hardware peripheral entities; based on the identified functional module entities and hardware peripheral entities, the hardware context consistency score between any two requirement items to be associated is calculated; and the hardware context consistency score is weighted and fused with the traditional text similarity score based on text content to obtain the final tracking similarity score. Tracking matrix construction and optimization: For the content of the standardized requirement items, a unique semantic fingerprint is generated; based on the semantic fingerprint, redundant items with semantic similarity exceeding a set threshold within the same document are identified and excluded, and a main item representing the semantic requirement is established; among the main items, a set of tracking relationships is established based on the corresponding tracking similarity scores; and based on the main items and the set of tracking relationships, a structured tracking matrix report displaying the aggregation results of multiple tracking relationships is generated and exported.

[0008] According to one embodiment of this application, the calculation process of the hardware context consistency score includes: If the functional module entity of the first requirement item does not match the functional module entity of the second requirement item between two requirement items to be associated, a first penalty is applied to the hardware context consistency score. If the hardware peripheral entity of the first requirement item and the hardware peripheral entity of the second requirement item are marked as functionally orthogonal in the hardware ontology knowledge base, a second penalty is applied to the hardware context consistency score, and the strength of the second penalty is greater than or equal to the strength of the first penalty.

[0009] According to one embodiment of this application, the tracking similarity score The calculation formula is:

[0010] in, The hardware context consistency score is given. The weights for the hardware context consistency score, The traditional text similarity score is given. The weights for the traditional text similarity scores are defined, and the following settings are given: > .

[0011] According to one embodiment of this application, a unique semantic fingerprint is generated for the normalized requirement item content, including: For the standardized content of the requirement item, a unique semantic vector is generated using a pre-trained word embedding model or semantic encoder, which serves as the semantic fingerprint of the item.

[0012] According to one embodiment of this application, based on the semantic fingerprint, identifying and excluding redundant entries within the same document whose semantic similarity exceeds a set threshold includes: Calculate the cosine similarity between the semantic fingerprints of different requirement entries within the same document; When the cosine similarity exceeds a set threshold, the different requirement entries are determined to be semantically redundant entries, and one of them is established as the main entry.

[0013] According to one embodiment of this application, while generating the semantic fingerprint, the precise hash value of the normalized requirement item content is also calculated; identifying the redundant items further includes: identifying requirement items with the same item identifier and the same precise hash value, then determining them as semantically duplicated redundant items, and establishing one of them as the main item.

[0014] According to one embodiment of this application, the automatic generation operation performed by the rule enhancement algorithm includes at least one of the following: The separators in the extracted seed are blurred to generate a matching pattern with higher tolerance. Automatically inject synonyms and case variations into the keywords extracted from the seeds; Replace the number sequence in the extracted seed with dynamic number wildcards.

[0015] According to one embodiment of this application, the hardware ontology knowledge base includes at least: A functional module dictionary, used to store the names of functional modules involved in airborne software; A hardware peripheral dictionary is used to store the orthogonality relationship between hardware interfaces and controller types and their corresponding functions. An operation behavior dictionary is used to store verbs or state words that describe operations.

[0016] This application also provides a requirement semantic tracking system for airborne embedded software, including a rule enhancement module for receiving extraction seeds input by the user and automatically synthesizing an enhanced set of regular expressions through a rule enhancement algorithm; The document parsing and extraction module is used to load heterogeneous project documents, extract requirement items from the heterogeneous project documents using the enhanced regular expression set, and perform text normalization processing. A hardware knowledge ontology base is used to store hardware entities, functional module entities, and corresponding functional orthogonality definitions in the airborne domain. The feature entity extraction module is used to call the hardware knowledge ontology library to identify and extract functional module entities and hardware peripheral entities from the extracted requirement entries; The similarity calculation and verification module is used to calculate the hardware context consistency score between any two required items to be associated; and to perform a weighted fusion of the hardware context consistency score and the traditional text similarity score based on text content to obtain the final tracking similarity score. The aggregation, deduplication, and matrix generation module is used to generate a unique semantic fingerprint for the normalized requirement item content, and based on the semantic fingerprint, identify and exclude redundant items in the same document whose semantic similarity exceeds a set threshold, and establish the main item representing the semantic requirement; establish a set of tracking relationships between the main items based on the corresponding tracking similarity scores; and generate and export a structured tracking matrix report that displays the aggregation results of multiple tracking relationships based on the main items and the set of tracking relationships.

[0017] This invention addresses three major challenges faced by airborne embedded software engineering: heterogeneous document formats, semantic pseudo-similarity in hardware-software coupling, and redundancy in tracking matrices. It achieves the following beneficial effects: To address the rigidity and low adaptability of single regular expression matching in documents, this invention proposes and implements a paradigm-guided rule adaptive enhancement mechanism that automatically synthesizes a highly robust set of enhanced regular expressions, significantly reducing the difficulty of rule configuration and parsing error rate, and solving the robustness challenge of extraction.

[0018] To address the semantic pseudo-similarity problem caused by hardware-software coupling, this invention proposes a hardware ontology-aware and semantic dimension-weighted model. This model extracts functional modules and hardware peripheral entities, and performs mandatory functional orthogonality verification on the tracking relationships, effectively solving the accuracy challenge of traditional semantic models in distinguishing between entities such as "XXX module PCIe" and "YYY module PCIe".

[0019] For aggregation and deduplication mechanisms under many-to-many mapping, this invention proposes a document-internal deduplication and multiple relation M:N aggregation strategy based on semantic fingerprints, which ensures that the final tracking matrix rows and columns consist of only unique master entries, greatly improving the readability and auditing efficiency of the matrix.

[0020] In summary, this embodiment achieves high-precision extraction of cross-document requirement items by combining adaptive extraction, hardware-aware verification, and structured optimization. It effectively avoids semantic pseudo-similarity misjudgment through hardware context awareness, supports structured aggregation and deduplication of complex tracking relationships, and completely solves the accuracy and robustness challenges faced by existing semi-automated tools in the field of airborne embedded software. It significantly improves the accuracy, robustness, and auditing efficiency of airborne embedded software requirement tracking. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a schematic diagram of the demand semantic tracking system structure according to an embodiment of the present invention; Figure 2 This is a flowchart of the requirement semantic tracking relationship verification process in the requirement semantic tracking method of this invention. Figure 3 This is an example of a tracking matrix output in an embodiment of the present invention. Detailed Implementation

[0023] The embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0024] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. This application can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0025] like Figure 1 As shown, the requirement semantic tracking system for airborne embedded software described in this invention can be deployed on a server or local terminal, and mainly includes the following functional modules: The rule enhancement module is used to receive the extraction seed input by the user and automatically synthesize an enhanced set of regular expressions through the rule enhancement algorithm; The document parsing and extraction module is used to load heterogeneous project documents, extract requirement items from the heterogeneous project documents using the enhanced regular expression set, and perform text normalization processing. A hardware knowledge ontology base is used to store hardware entities, functional module entities, and corresponding functional orthogonality definitions in the airborne domain. The feature entity extraction module is used to call the hardware knowledge ontology library to identify and extract functional module entities and hardware peripheral entities from the extracted requirement entries; The similarity calculation and verification module is used to calculate the hardware context consistency score between any two required items to be associated; and to perform a weighted fusion of the hardware context consistency score and the traditional text similarity score based on text content to obtain the final tracking similarity score. The aggregation, deduplication, and matrix generation module generates unique semantic fingerprints for the normalized requirement entries. Based on these fingerprints, it identifies and excludes redundant entries within the same document whose semantic similarity exceeds a set threshold, establishing a main entry representing the semantic requirement. Among these main entries, a set of tracking relationships is established based on their corresponding tracking similarity scores. Finally, based on the main entries and the set of tracking relationships, a structured tracking matrix report displaying the aggregation results of multiple tracking relationships is generated and exported. The present invention provides a requirement semantic tracking method for airborne embedded software, comprising: Adaptive requirement item extraction: Obtain requirement item identifiers provided by the user for the target document type as extraction seeds; based on the extraction seeds, automatically generate an enhanced regular expression set adapted to multiple document formats through a rule enhancement algorithm; use the enhanced regular expression set to extract requirement items from heterogeneous engineering documents and perform text normalization processing; Hardware-aware tracking relationship verification: Using a pre-built domain-specific hardware ontology knowledge base, feature entity recognition is performed on the extracted requirement items to extract the corresponding functional module entities and hardware peripheral entities; based on the identified functional module entities and hardware peripheral entities, the hardware context consistency score between any two requirement items to be associated is calculated; and the hardware context consistency score is weighted and fused with the traditional text similarity score based on text content to obtain the final tracking similarity score. Tracking matrix construction and optimization: For the content of the standardized requirement items, a unique semantic fingerprint is generated; based on the semantic fingerprint, redundant items with semantic similarity exceeding a set threshold within the same document are identified and excluded, and a main item representing the semantic requirement is established; among the main items, a set of tracking relationships is established based on the corresponding tracking similarity scores; and based on the main items and the set of tracking relationships, a structured tracking matrix report displaying the aggregation results of multiple tracking relationships is generated and exported.

[0026] According to some embodiments of the present invention, the calculation process of the hardware context consistency score includes: If the functional module entity of the first requirement item does not match the functional module entity of the second requirement item between two requirement items to be associated, a first penalty is applied to the hardware context consistency score. If the hardware peripheral entity of the first requirement item and the hardware peripheral entity of the second requirement item are marked as functionally orthogonal in the hardware ontology knowledge base, a second penalty is applied to the hardware context consistency score, and the strength of the second penalty is greater than or equal to the strength of the first penalty.

[0027] According to some embodiments of the present invention, the tracking similarity score The calculation formula is:

[0028] in, The hardware context consistency score is given. The weights for the hardware context consistency score, The traditional text similarity score is given. The weights for the traditional text similarity scores are defined, and the following settings are given: > .

[0029] According to some embodiments of the present invention, a unique semantic fingerprint is generated for the content of the normalized requirement item, including: For the standardized content of the requirement item, a unique semantic vector is generated using a pre-trained word embedding model or semantic encoder, which serves as the semantic fingerprint of the item.

[0030] According to some embodiments of the present invention, based on the semantic fingerprint, identifying and excluding redundant entries within the same document whose semantic similarity exceeds a set threshold includes: Calculate the cosine similarity between the semantic fingerprints of different requirement entries within the same document; When the cosine similarity exceeds a set threshold, the different requirement entries are determined to be semantically redundant entries, and one of them is established as the main entry.

[0031] According to some embodiments of the present invention, while generating the semantic fingerprint, the precise hash value of the normalized requirement entry content is also calculated; identifying the redundant entries further includes: identifying requirement entries with the same entry identifier and the same precise hash value, then determining them as semantically duplicated redundant entries, and establishing one of them as the main entry.

[0032] According to some embodiments of the present invention, the automatic generation operation performed by the rule enhancement algorithm includes at least one of the following: The separators in the extracted seed are blurred to generate a matching pattern with higher tolerance. Automatically inject synonyms and case variations into the keywords extracted from the seeds; Replace the number sequence in the extracted seed with dynamic number wildcards.

[0033] According to some embodiments of the present invention, the hardware ontology knowledge base includes at least: A functional module dictionary, used to store the names of functional modules involved in airborne software; A hardware peripheral dictionary is used to store the orthogonality relationship between hardware interfaces and controller types and their corresponding functions. An operation behavior dictionary is used to store verbs or state words that describe operations.

[0034] The preferred embodiment of the method of the present invention includes the following three key stages, which describe in detail how to solve the three types of defects mentioned in the background art: To address the first shortcoming and the problems of high difficulty in writing regular expressions and poor adaptability to unstructured documents in existing technologies, this invention proposes a requirement item extraction method based on example guidance and rule adaptive enhancement. This method allows users to input only simplified feature words or example text, and the system automatically synthesizes highly robust matching patterns through feature generalization and boundary expansion algorithms.

[0035] To address the second defect, this invention proposes a tracking relationship verification method based on a hardware ontology dictionary and semantic dimension weighting. This method decomposes requirement items into three dimensions: functional entities, hardware peripherals, and operational behaviors, and performs mandatory consistency verification on these core hardware coupling features, thereby effectively distinguishing requirements with high textual similarity but unrelated functions.

[0036] To address the third deficiency—the lack of aggregation and deduplication for duplicate entries or multiple references, and the inability to correctly handle situations where a low-level requirement traces multiple high-level requirements—this invention proposes a tracking matrix optimization method based on feature fingerprints and multi-relationship aggregation. This method aims to eliminate redundant information in the source document while accurately representing complex many-to-many (M:N) tracking relationships.

[0037] In practice, the detailed implementation process of the requirement semantic tracking method for airborne embedded software of the present invention is as follows: Step S1: Example-guided adaptive requirement item extraction (solving robustness issues); Step S2: Hardware-aware tracking relationship verification (solving accuracy issues); Step S3: Semantic fingerprint-based tracking matrix optimization (addressing redundancy and complexity issues).

[0038] Step S1 can be further refined as follows: Step S11: Construct a configurable metadata extraction template. The system provides a visual configuration interface to obtain the extraction seeds defined by the user for specific document levels (such as SRS, SDD). The extraction seeds include: 1) Identifier characteristics: Typical example of a user-input request number (e.g., “SRS_Flight_001” or “R-1.2”).

[0039] 2) Anchor keywords: Characteristic words that mark the beginning of a requirement (e.g., "requirement", "function", "input", "output").

[0040] 3) Boundary constraints: Define the termination conditions of requirement items (e.g., encountering the next heading, encountering a blank line, or encountering a specific terminator).

[0041] Step S12: The Automatic Augmentation system for regular expressions based on feature analysis receives the extraction seeds in Step S1 and automatically generates an extended set of regular expressions through the rule enhancement engine. The specific enhancement logic includes: 1) Delimiter fuzzification: The system automatically identifies delimiters (such as "_", "-", ".") in the user input example and replaces them with a more tolerant regular expression pattern (such as [\s_\-\.]+) to accommodate possible format inconsistencies in the document (for example, automatically compatible with "SRS-Flight-001" and "SRS_Flight_001").

[0042] 2) Keyword synonym and fault tolerance extension: For the anchor keywords input by the user, the system automatically injects common case variants and thesaurus (for example, when inputting "Req", the system automatically generates (Req|REQ|Requirement|需求)\s*[::]?) and automatically adds support for variable-length matching of Chinese and English punctuation and whitespace characters.

[0043] 3) Digital sequence generalization: Automatically identify the digital part in the example and replace it with a dynamic counter wildcard (such as \d+ or [0-9a-fA-F]+) to construct a general number indexing rule.

[0044] Step S13: Multi-dimensional document structure parsing and candidate entry location Use the enhanced regular set generated in Step S12, combined with the Document Object Model (DOM) parsing technology, to scan the target document: 1) Style-aware extraction: Not only based on the text content, but also read the style attributes (such as Heading1, bold, font size) of Word / PDF documents. If the text matches the enhanced regular expression and has a specific style (such as "title style"), a higher extraction confidence is given.

[0045] 2) Hierarchical context binding: When extracting low-level requirements (such as SDD), automatically trace back to the nearest chapter title upward and bind the chapter number as an implicit attribute to the requirement entry to solve the problem that it is impossible to distinguish requirements with the same name only by the main text.

[0046] Step S14: Rule correction and solidification based on interactive feedback The system displays a preview of the extraction results in the preprocessing stage: 1) Abnormal highlighting: Automatically mark areas with irregular formats or suspected omissions.

[0047] 2) Incremental learning: If the user manually corrects the extraction range of a certain item, the system reverse analyzes the corrected text features, dynamically updates the enhanced rule base in step S2, and solidifies the updated rules into a dedicated parsing template for this type of document for reuse in subsequent batch processing.

[0048] Step S2 can be further refined as follows: Step S21: Build and initialize the hardware ontology knowledge base.

[0049] The system first establishes a domain-specific knowledge base, which is used to define and store key hardware / software entities and their hierarchical relationships: 1) Functional Entity / Module Dictionary : Includes the names of all core functional modules involved in the airborne software (e.g., flight control module, communication module, power management module, storage module, etc.).

[0050] 2) Hardware peripherals / interface dictionary It includes all key hardware interfaces and controller types (e.g., PCIe, SPI, QSPI, Ethernet, CAN, ADC, WDT).

[0051] 3) Operation Behavior Dictionary It includes common verbs and status words (e.g., initialize, send, receive, read, configure, fault).

[0052] The knowledge base supports user-defined maintenance and can define the functional orthogonality of peripherals through a hierarchical structure (for example, although QSPI and SPI are both serial interfaces, they are defined as two independent and non-interchangeable entities in airborne systems).

[0053] Step S22: Identification and extraction of feature entities for requirement items; such as Figure 2 As shown.

[0054] For high-level needs to be tracked (e.g., from SRS) and lower-level demands (For example, from SDD), using the knowledge base established in step S21 and the customized Named Entity Recognition (NER) model, both are parsed and features are extracted to obtain a set of triples:

[0055] This requirement description covers the functional module entities involved.

[0056] The hardware peripheral entities involved in the requirements description.

[0057] This requirement describes the core operational behavior.

[0058] Step S23: Calculate the hardware context consistency score Based on the feature entities extracted in step S22, calculate and Hardware context consistency score This score primarily focuses on the forced matching of key hardware entities: 1) Functional module consistency verification: If and If not matched, then Assigning penalty weights (Close to 0). This mechanism is used to distinguish between "initialization of module XXX" and "initialization of module YYY".

[0059] 2) Critical peripheral consistency verification: If and Although the texts are similar, in In the knowledge base, if something is marked as functionally orthogonal (e.g., QSPI is not equivalent to SPI), then... Assign strong penalty weights (Set directly to 0). This mechanism is used to distinguish between misjudgments of "QSPI" and "SPI".

[0060] 3) Soft matching of operational behaviors: and Soft matching based on ontology dictionary is performed between them. For example, "initialization" can be traced to "configuration" but not to "destruction".

[0061] final, The calculation logic ensures that the score will not be reset to zero only when two requirement items are highly consistent or compatible in the dimensions of core functional entities and key hardware peripherals.

[0062] Step S24: Integrate multi-dimensional tracking similarity aggregation to obtain the final tracking similarity score. It is determined by the hardware context consistency score. Similarity score with traditional text (Currently, the weighted fusion is obtained using cosine similarity calculated based on BERT embedding)

[0063] in, and These are the weighting coefficients, and .

[0064] In this invention, Set to a higher weight (e.g.) = 0.7, = 0.3), to ensure that hardware context consistency dominates in tracking relationship judgment.

[0065] Examples are given below: Assumption Description: "The flight control module should be configured with a PCIe interface..."; Assumption Description: "The communication module should be configured with a PCIe bus..."; Extract Entities For "flight control", For "communication". Due to entity inconsistency, even text similarity... Very high, ultimately Also because The penalty is applied if the value falls below the threshold, and the tracking relationship is blocked, thus solving the problem of mistracking across modules.

[0066] Step S25: Establish a tracking relationship, only when Greater than the preset tracking threshold Only then did the system... and An effective requirement tracing relationship is automatically established between them and recorded in the tracing matrix.

[0067] S3 can be further refined as follows: Step S31: Requirement item normalization and semantic fingerprint generation.

[0068] The system processes each requirement item extracted in steps S13 / S22. Preprocessing: 1) Text normalization: Remove non-functional noise from the content of entries, including but not limited to: timestamps, version numbers, tabs, extra spaces, and minor punctuation differences that do not affect semantics.

[0069] 2) Semantic fingerprint generation: Based on the normalized entry content, a pre-trained word embedding model or semantic encoder is used (e.g., in computation...). The model used at the time generates a unique semantic vector. This vector serves as the semantic fingerprint of the entry, used for subsequent similarity deduplication. Simultaneously, the entry ID is extracted as the identifier key. .

[0070] 3) Precise fingerprint generation: In addition, a traditional, irreversible, precise content hash value is calculated for the normalized text content. It is used for the most stringent deduplication and integrity checks.

[0071] Step S32: Redundancy identification and main entry establishment within the document.

[0072] The system scans the interior of each source document (such as SRS or SDD) to identify and process redundant entries: Strict deduplication (exact citation): Identification Same and Identical entries. These are determined to be completely duplicate references. The system retains the first instance as the Master Entry, and the rest are marked as redundant references and excluded.

[0073] For example: The input document contains... (ID:SDD-1.1, Content: "PCIe Interface Initialization") and (ID:SDD-1.1, Content: "PCIe Interface Initialization"). Because... and The system will be exactly the same. Mark as redundant references and exclude.

[0074] 2) Semantic deduplication (fuzzy repetition): Identification Different but with semantic fingerprints The cosine similarity between them is extremely high (e.g. ) entries. These entries were determined to be duplicate entries that had been renamed or slightly modified.

[0075] The system marks this type of information and includes it in the deduplication log for auditors to manually verify; or, depending on the configuration, it automatically merges the most recent semantic entries into a single main entry.

[0076] Step S33: Establishment and identification of multiple tracking relationships. After performing deduplication within the document (retaining only the main entry), the system uses the data obtained in step S24. As a result, a set of all valid tracking relationships was established. :

[0077] The system then identifies and labels multiple tracking relationships, and an example of the output tracking matrix is ​​shown below. Figure 3 As shown.

[0078] This invention receives user examples and automatically generates a highly robust set of enhanced regular expressions to accurately extract cross-document requirement items. The core of this invention lies in introducing a hardware ontology knowledge base to identify the functional modules and hardware peripheral entities of the requirement items, and then calculating a tracking similarity score using a hardware-sensitive weighted algorithm. . Assigning hardware context consistency scores Higher weighting effectively distinguishes requirements with similar text but orthogonal functions (e.g., identical interface initializations in different modules). Simultaneously, the system performs semantic fingerprint-based deduplication and M:N relationship aggregation, automatically generating structured traceability matrix reports, significantly improving the efficiency and accuracy of requirements quality auditing for airborne software engineering.

[0079] This invention is particularly applicable to the development of airborne embedded software under the GJB5000 system, which involves complex hardware constraints and multi-level requirement decomposition, and enables the automatic extraction of cross-document requirement items, semantic consistency verification, and automated construction of the tracking matrix.

[0080] 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 requirement semantic tracking method for airborne embedded software, characterized in that, include: Adaptive requirement item extraction: Obtain the requirement item identifiers provided by the user for the target document type and extract the seed; Based on the extracted seeds, an enhanced set of regular expressions adapted to multiple document formats is automatically generated using a rule enhancement algorithm; The enhanced regular expression set is used to extract requirement items from heterogeneous project documents and perform text normalization processing; Hardware-aware tracking relationship verification: Using a pre-built domain-specific hardware ontology knowledge base, feature entity recognition is performed on the extracted requirement items to extract the corresponding functional module entities and hardware peripheral entities; Based on the identified functional module entities and hardware peripheral entities, calculate the hardware context consistency score between any two requirement items to be associated; and then weight and fuse the hardware context consistency score with the traditional text similarity score based on text content to obtain the final tracking similarity score. Tracking matrix construction and optimization: For the content of the standardized requirement items, a unique semantic fingerprint is generated; based on the semantic fingerprint, redundant items with semantic similarity exceeding a set threshold within the same document are identified and excluded, and a main item representing the semantic requirement is established; among the main items, a set of tracking relationships is established based on the corresponding tracking similarity scores; and based on the main items and the set of tracking relationships, a structured tracking matrix report displaying the aggregation results of multiple tracking relationships is generated and exported.

2. The requirement semantic tracking method for airborne embedded software according to claim 1, characterized in that, The calculation process for the hardware context consistency score includes: If the functional module entity of the first requirement item does not match the functional module entity of the second requirement item between two requirement items to be associated, a first penalty is applied to the hardware context consistency score. If the hardware peripheral entity of the first requirement item and the hardware peripheral entity of the second requirement item are marked as functionally orthogonal in the hardware ontology knowledge base, a second penalty is applied to the hardware context consistency score, and the strength of the second penalty is greater than or equal to the strength of the first penalty.

3. The requirement semantic tracking method for airborne embedded software according to claim 1, characterized in that, The tracking similarity score The calculation formula is: in, The hardware context consistency score is given. The weights for the hardware context consistency score, The traditional text similarity score is given. The weights for the traditional text similarity scores are defined, and the following settings are given: > .

4. The requirement semantic tracking method for airborne embedded software according to claim 1, characterized in that, For the standardized requirement item content, a unique semantic fingerprint is generated, including: For the standardized content of the requirement item, a unique semantic vector is generated using a pre-trained word embedding model or semantic encoder, which serves as the semantic fingerprint of the item.

5. The requirement semantic tracking method for airborne embedded software according to claim 1, characterized in that, Based on the semantic fingerprint, redundant entries within the same document whose semantic similarity exceeds a set threshold are identified and excluded, including: Calculate the cosine similarity between the semantic fingerprints of different requirement entries within the same document; When the cosine similarity exceeds a set threshold, the different requirement entries are determined to be semantically redundant entries, and one of them is established as the main entry.

6. The requirement semantic tracking method for airborne embedded software according to claim 5, characterized in that, While generating the semantic fingerprint, the precise hash value of the normalized requirement item content is also calculated. Identifying the redundant entries further includes: identifying requirement entries with the same entry identifier and the same exact hash value, then determining them as semantically redundant entries, and establishing one of them as the main entry.

7. The requirement semantic tracking method for airborne embedded software according to claim 1, characterized in that, The automatic generation operation performed by the rule enhancement algorithm includes at least one of the following: The separators in the extracted seed are blurred to generate a matching pattern with higher tolerance. Automatically inject synonyms and case variations into the keywords extracted from the seeds; Replace the number sequence in the extracted seed with dynamic number wildcards.

8. The requirement semantic tracking method for airborne embedded software according to claim 1, characterized in that, The hardware ontology knowledge base includes at least: A functional module dictionary, used to store the names of functional modules involved in airborne software; A hardware peripheral dictionary is used to store the orthogonality relationship between hardware interfaces and controller types and their corresponding functions. An operation behavior dictionary is used to store verbs or state words that describe operations.

9. A requirement semantic tracking system for airborne embedded software, characterized in that, The rule enhancement module is used to receive the extraction seed input by the user and automatically synthesize an enhanced set of regular expressions through the rule enhancement algorithm; The document parsing and extraction module is used to load heterogeneous project documents, extract requirement items from the heterogeneous project documents using the enhanced regular expression set, and perform text normalization processing. A hardware knowledge ontology base is used to store hardware entities, functional module entities, and corresponding functional orthogonality definitions in the airborne domain. The feature entity extraction module is used to call the hardware knowledge ontology library to identify and extract functional module entities and hardware peripheral entities from the extracted requirement entries; The similarity calculation and verification module is used to calculate the hardware context consistency score between any two required items to be associated; and to perform a weighted fusion of the hardware context consistency score and the traditional text similarity score based on text content to obtain the final tracking similarity score. The aggregation, deduplication, and matrix generation module is used to generate a unique semantic fingerprint for the normalized requirement item content, and based on the semantic fingerprint, identify and exclude redundant items in the same document whose semantic similarity exceeds a set threshold, and establish the main item representing the semantic requirement; establish a set of tracking relationships between the main items based on the corresponding tracking similarity scores; and generate and export a structured tracking matrix report that displays the aggregation results of multiple tracking relationships based on the main items and the set of tracking relationships.

Images