Method for processing evidence of quality management system audit based on minimum audit unit

Abstract

The application discloses a quality management system audit evidence processing method based on minimum audit units, and belongs to the technical fields of computer document processing, knowledge graph retrieval and quality management system auditing. The method comprises the following steps: taking a target product or project entity as a starting point, traversing a quality knowledge graph according to the relationships between entities, and obtaining a minimum audit unit set; receiving an audit execution instruction and an audit evidence file carrying a target minimum audit unit identifier, performing character recognition, extracting key information according to a rule input field list, and generating evidence feature data labeled with original evidence positions; loading a set of judgment rules bound to the unit, and generating rule-level judgment and state auxiliary information item by item; and outputting an audit result set recording rule calculation field values and original evidence positions. The application establishes data association of graph retrieval, evidence fields, judgment rules and original positions relying on minimum audit unit identifiers, reduces irrelevant rules participating in operation, and can distinguish between insufficient evidence and abnormal triggering situations.

Description

Technical Field

[0001] This invention relates to the fields of computer document processing, knowledge graph retrieval, and quality management system auditing, specifically to a method for processing evidence in quality management system audits based on the smallest audit unit. Background Technology

[0002] Quality management system audits are an important means of verifying whether an organization's quality management activities comply with the system specifications and identifying quality risk points. When conducting audits, auditors need to collect substantial audit evidence and, in conjunction with relevant quality incidents, management processes, and audit clauses for each product or project, make conformity assessments to complete the system audit evaluation.

[0003] In the current quality management system audit process, auditors mostly rely on manually sorting through various documents related to the audited entity and manually searching and matching the corresponding audit clauses and audit basis. For complex products or projects, there are numerous quality incidents and long management process chains involved. The relationships between audit evidence, quality incidents, management processes, and audit clauses are scattered and stored in different document and form systems, lacking a unified structured association.

[0004] Therefore, there is an urgent need for a quality management system audit evidence processing method based on the smallest unit that can automatically link and trace, and assist in the collection and processing of audit evidence. Summary of the Invention

[0005] To address the lack of definitive data correlation between audit evidence, quality incidents, management processes, and audit clauses, this invention provides a method for processing audit evidence in a quality management system based on the minimum audit unit: The method uses a type-restricted quality knowledge graph path to convert the target product or project identifier into a candidate minimum audit unit identifier; it determines the target minimum audit unit based on the audit execution instruction and converts the corresponding audit evidence file into evidence feature data with the original evidence location; it selects the corresponding decision rule set using the target minimum audit unit identifier; and it outputs a rule-by-rule audit result set containing rule-level decisions, status auxiliary information, and the original evidence locations of existing field values ​​involved in rule calculation.

[0006] Specifically, the present invention provides a method for processing audit evidence of a quality management system based on the smallest audit unit, including: Obtain the target product or project identifier for the audit task to be processed, and identify the product or project entity that matches the target product or project identifier in the pre-established quality knowledge graph; take the product or project entity as the starting point for traversal, and traverse the associated quality event entities, management process entities and minimum audit unit entities in sequence along the occurrence relationship, cause attribution relationship and audit basis relationship, and summarize the identifiers of the minimum audit unit entities obtained by traversal to form a candidate minimum audit unit set. The system receives audit execution instructions and audit evidence files. The audit execution instructions include the identifier of the target minimum audit unit selected from the candidate minimum audit unit set, and the audit evidence files correspond to the target minimum audit units. The system performs text recognition and key information extraction on the audit evidence files, and stores the extracted field names, field values, and the original evidence positions of the field values ​​in the audit evidence files in association to form evidence feature data. Based on the identifier of the target minimum review unit, load the set of judgment rules bound to the target minimum review unit from the dynamic review clause library, substitute the field values ​​in the evidence feature data that match the input fields of each judgment rule into the corresponding judgment rule, and generate rule-level judgment and status auxiliary information respectively. Generate and output an audit result set. Each rule-level result in the audit result set shall include at least the target minimum audit unit identifier, rule identifier, rule-level judgment, status auxiliary information, and the original evidence location of the existing field values ​​that participated in the calculation of the corresponding judgment rule.

[0007] Preferably, the smallest audit unit is the smallest knowledge unit in the audit standard that can independently make a compliance determination, defined as a triple: MAU = <ID, Desc, R MAU >; ; ; Where MAU is the smallest review unit, ID is the identifier of the smallest review unit; Desc is a natural language description of the content reviewed by the smallest review unit; R MAU For the set of decision rules associated with the smallest review unit, r i Let i be the i-th judgment rule, where Type is the rule type, Condition uniformly represents the violation trigger predicate, and Action represents the rule-level judgment output when the violation trigger predicate is true.

[0008] Preferably, the quality knowledge graph is defined as a directed graph: G=(E,R KG A); Where G is the quality knowledge graph, E is the entity set, and R is the entity set. KG Let A be the set of relations and A be the set of attributes.

[0009] Preferably, the entity set includes at least a product or project entity, a quality event entity, a management process entity, and a minimum audit unit entity.

[0010] Preferably, the set of relationships includes at least the occurrence relationship from the product or project entity to the quality event entity, the cause attribution relationship from the quality event entity to the management process entity, and the audit basis relationship from the management process entity to the smallest audit unit entity.

[0011] Preferably, the decision rule set includes sufficiency rules, temporal rules, correlation consistency rules, and quantitative comparison rules.

[0012] Preferably, the sufficiency rule calculates the field coverage using the following formula: C=|F K ∩K| / |K|; Where C represents field coverage, K represents the list of key information fields that should be included in the target minimum audit unit, and F represents the field coverage rate. K This refers to the set of provided fields extracted from the audit evidence documents and matched with the list of key information fields; When C is below the preset coverage threshold, the sufficiency rule's violation trigger predicate is established, outputting a rule-level judgment of insufficient evidence, and setting the set K\F... K The fields in the table are used as missing fields.

[0013] Preferably, the timing rule takes the design task issuance date, review date, and design drawing release date as input fields. When the sequential relationship that the design task issuance date is earlier than the review date and the review date is earlier than the design drawing release date is not met, the timing rule's violation trigger predicate is established and a rule-level judgment of timing abnormality is output.

[0014] Preferably, the association consistency rule is used to compare the same attribute value that has been identified as the same entity across files or records, without re-determining the entity correspondence in this rule; for non-numerical attributes, when the attribute values ​​are different, an inconsistent rule-level judgment is output; for numerical attributes, when the difference between attribute values ​​exceeds the preset tolerance range, an inconsistent rule-level judgment is output.

[0015] Preferably, the quantitative comparison rule is used to compare the measured values ​​in the audit evidence with the numerical range specified by the standard or document, and outputs a rule-level judgment of numerical anomaly when the measured value does not fall within the numerical range.

[0016] Furthermore, the present invention provides a quality management system audit evidence processing system based on the smallest audit unit, including a processor, a memory, a document input interface and an auditor terminal communication interface; the processor executes the computer program in the memory to complete the graph path query, evidence field extraction, rule-by-rule calculation and audit result set output.

[0017] Furthermore, the present invention provides a computer-readable storage medium on which a computer program stored is executed by a processor to implement a method for processing audit evidence in a quality management system.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention represents quality management system clauses as the smallest audit unit with a unique identifier, audit content description, and judgment rule set. It also establishes directed associations between products or projects, quality events, management processes, and the smallest audit unit through a quality knowledge graph, enabling computers to determine the scope of candidate audits based on the target product or project.

[0019] 2. This invention uses the smallest audit unit identifier as an index to connect the target audit object, evidence fields, judgment rules, and original evidence locations into a continuous data processing chain. It can distinguish three rule-by-rule states: insufficient evidence, abnormal triggering, and non-triggered abnormality, and supports the audit results to be retrieved back to the input document.

[0020] 3. Starting from the target product or project entity, this invention only traverses along the occurrence relationship, cause attribution relationship, and audit basis relationship to the smallest audit unit entity, so that the smallest audit unit that is unrelated to the target object does not enter the subsequent rule calculation.

[0021] 4. After the target MAU is selected in the review instruction of this invention, the identifier simultaneously determines the evidence to be processed, the fields to be extracted, and the rule set to be loaded, establishing a stable input dependency relationship. Using the same MAU identifier to uniformly constrain evidence and rules avoids mismatches between evidence, fields, and rules, simplifies task scheduling logic, ensures consistency between extraction and judgment logic, facilitates rule iteration and traceability of review issues, and improves the reliability of automated review.

[0022] 5. During extraction, this invention saves the "field name, field value, and original evidence location" in association; during output, it continues to retain the MAU identifier, rule identifier, judgment status, and the actual evidence location involved in the calculation, thus enabling direct retrieval of the original text from the audit conclusion. This results in rule-level results that allow for retrieval of original evidence, distinguishes between scenarios with missing fields, supports fine-grained post-audit review, meets the evidence archiving and traceability requirements of the quality management system, and facilitates the review and iteration of algorithms and rules.

[0023] 6. This invention records a missing marker when an input field is missing, without fabricating the original evidence location for a non-existent field, and without directly converting "insufficient evidence" into a system non-compliance conclusion. The auditor makes the final confirmation based on supplementary evidence. Attached Figure Description

[0024] Figure 1 This is a data flow diagram of the execution system of the quality management system audit evidence processing method based on the minimum audit unit of the present invention; Figure 2 This is the main flowchart of the quality management system audit evidence processing method based on the smallest audit unit of the present invention; Figure 3 This is a schematic diagram of entities and relationship paths in a quality knowledge graph according to an embodiment of the present invention. Detailed Implementation

[0025] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. It should be understood that the following embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.

[0026] like Figure 1 As shown, as one aspect of the present invention, the present invention provides a method for processing audit evidence of a quality management system based on the smallest audit unit, comprising: S1. Construct a quality knowledge graph that includes the smallest audit unit entity, specifically including: The quality knowledge graph is used to store directed relationships between target products or projects, historical quality events, management processes, and the smallest audit unit; the dynamic audit clause library is used to store the smallest audit unit identifier, audit content description, rule input fields, rule parameters, and the set of judgment rules bound to that unit.

[0027] The constructed quality knowledge graph is a semantic network, defined as a directed graph G: G=(E,R KG A); Where E represents a set of entities, which includes at least product or project entities, quality event entities, management process entities, and the smallest audit unit entity.

[0028] R KG It represents a set of graph relationships, including at least the occurrence relationship from product or project entities to quality event entities, the cause attribution relationship from quality event entities to management process entities, and the audit basis relationship from management process entities to the smallest audit unit entities.

[0029] A is a set of attributes that describe the characteristics of an entity or relationship, such as the severity of the problem, the version number of the clause, the date of the evidence, etc.

[0030] The smallest audit unit (MAU) is the smallest knowledge unit in the audit standard that can be independently determined to meet compliance requirements; it is defined as a triple. MAU = <ID, Desc, R MAU >; ; ; Wherein, ID is a unique identifier representing the smallest review unit; Desc represents the review content description capable of independent evidence verification; R MAU This represents the set of decision rules bound to this smallest review unit. Each decision rule r i Represented as r i=<Type,Condition,Action> Where Type represents the rule type, Condition represents the violation trigger predicate, and Action represents the rule-level judgment output when the violation trigger predicate is true.

[0031] S11. Construction and storage structure of quality knowledge graph.

[0032] By combining quality management system standards, historical quality incident records, and management process information, define product or project entities, quality incident entities, management process entities, and the smallest audit unit entities, and limit the start type, end type, and traversal order of occurrence relationships, cause attribution relationships, and audit basis relationships.

[0033] In one logical storage method, each entity record includes at least an entity type, an entity identifier, and an entity attribute, and each relation record includes at least a relation type, a starting entity identifier, an ending entity identifier, and a relation attribute; the starting entity and ending entity in the relation record point to the corresponding entity record respectively.

[0034] S12, Data extraction and knowledge construction in the quality domain.

[0035] Using quality management system standard documents, historical quality event records, audit records, and management process documents as data sources, product or project identifiers, quality events, responsible processes, and standard clause mappings are extracted to form corresponding entity records and relationship records. The entity and relationship paths of the quality knowledge graph, such as... Figure 3 As shown.

[0036] For structured data, existing fields are directly mapped to entities, attributes, and relationships; for semi-structured or unstructured files, chapters, entries, and text content are parsed first, and then candidate entities and relationships are extracted.

[0037] When extracting candidate entities or relations using natural language processing, the extraction results are confirmed by quality experts before being written into the formal quality knowledge graph; the confirmed entity identifiers and relation types serve as the data basis for subsequent path queries.

[0038] For semi-structured data such as quality manuals, procedure documents, and record templates, the header level, tables, key-value pairs, and clause numbers are parsed to convert product or project identifiers, quality event descriptions, responsible processes, and standard clause references into candidate entity attributes or candidate relationships.

[0039] For unstructured data such as scanned copies of audit reports, meeting minutes, and review reports, the scanned copies are first processed by optical character recognition to form text arranged by page number, and the electronic files are formed into text units by paragraph; then, candidate entities, attributes, and relationships are identified according to preset entity types and relationship types, and the source file and text position are preserved.

[0040] For example, from the statement "XX model radar experienced a solder joint failure during a high temperature and humidity test, and the responsible process was design input review," candidate records can be generated for the product or project entity "XX model radar," the quality event entity "solder joint failure during a high temperature and humidity test," and the management process entity "design input review process." After confirmation by quality experts, these records are written into the corresponding entities and relationships.

[0041] Evidence preprocessing and quality knowledge graph construction are independent of each other: the graph is used to determine the candidate minimum audit unit, while the evidence preprocessing unit takes the list of input fields declared by the binding rule of the target minimum audit unit as the extraction target and converts the corresponding audit evidence file into evidence feature data F.

[0042] Each field in the rule input field list must include at least the field name and the rule identifier that references that field; the field name is used to determine the category of information to be extracted, and the rule identifier is used to distribute the field record to the corresponding decision rule after extraction is completed.

[0043] During extraction, optical character recognition is first performed on the scanned document, retaining page numbers and recognized text ranges, or paragraph identifiers and character ranges are read from the electronic document; then, candidate field values ​​are generated according to field names, field aliases, or text fragments that match the semantics of the fields. Natural language processing models can be used for candidate fragment recognition, but there are no limitations on the specific model structure, training sample size, or confidence threshold.

[0044] For design input review records, the input fields declared by the target minimum review unit binding rule can include the reviewed object, document date, approval record, and reviewed input items; the evidence preprocessing unit only identifies candidate field values ​​for this input field list and does not directly send irrelevant paragraphs into subsequent rule calculations.

[0045] Each confirmed matching candidate field value forms a field record.<FieldName, FieldValue,SourceLocation> Where FieldName is the field name, FieldValue is the field value, and SourceLocation is the location of the field value in the original audit evidence file.

[0046] When the same field name appears multiple times in one or more audit evidence files, each appearance will form a separate field record and the source file and original evidence location will be saved separately, without the later appearance overwriting the previous appearance; when the rule is executed, only the field record that actually participates in the rule calculation will be selected.

[0047] For example, the field list K of the sufficiency rule includes functional performance requirements, environmental adaptability requirements, security requirements, regulatory requirements, interface requirements, and information on previous similar designs; the evidence feature data F stores the field values ​​and original evidence locations of each matched field, and unmatched fields are marked as missing only after the rule is calculated.

[0048] S13, Multi-source knowledge alignment and standardized storage.

[0049] Name unification and duplicate record merging are performed on entity records pointing to the same object from different data sources, and the confirmed entities, entity attributes and directed relationships are written into the graph database; the same smallest audit unit is associated with the corresponding record in the dynamic audit clause library with a unique identifier.

[0050] When different documents use the same product or project identifier, the same standard clause number, or the same minimum audit unit identifier, the corresponding records can be aligned according to the business identifier; candidate records that are only similar in name but lack a definite identifier should be confirmed by quality experts to determine whether they point to the same object.

[0051] The aligned entity records retain the standard name and source attribute, while the relationship records retain the relationship type, starting entity identifier, and ending entity identifier, enabling path queries to be reproduced by entity identifier and relationship type, without directly replacing formal relationships with text similarity.

[0052] Furthermore, it may also include: S14, controlled updates of the quality knowledge graph and the dynamic audit clause library.

[0053] When audit standards or confirmed quality event information change, entities, relationships, and judgment rules can be added or adjusted. Updated content must be confirmed by quality experts or rule maintenance personnel before being invoked by the audit task to avoid unconfirmed data automatically changing the graph path or rule judgment semantics.

[0054] The update process first registers new audit findings, standard changes, or rule maintenance comments as candidate update records, and indicates the entities, relationships, or smallest audit unit rules that are to be affected, without directly rewriting the currently effective data.

[0055] Candidate update records should at least distinguish between graph data updates and clause rule updates: the former is used to add or adjust entities and relationships, while the latter is used to adjust rule input fields, rule parameters, conditions, or actions.

[0056] For graph data updates, verify whether the starting entity type, relation type, and ending entity type of the newly added relation conform to the schema layer constraints, and verify whether the referenced entity identifier already exists.

[0057] For rule updates, the rule content before and after the update is saved separately, and it is verified whether all the input fields required by the rule are declared in the field list, whether the rule parameters have value descriptions, and whether the semantics of the violation trigger predicate and Action are consistent.

[0058] When updating content or rule sets involving the smallest audit unit, maintain consistency in the references between the smallest audit unit identifier, audit content description, rule identifier, and rule input fields; if consistency cannot be maintained, create a new rule record and prevent the old rule from being invoked by new tasks.

[0059] Candidate updates become effective records after being confirmed by quality experts or rule maintainers; unconfirmed auditor feedback serves only as a basis for candidate updates and does not automatically change relationship weights, rule thresholds, or rule judgment results.

[0060] Before the update takes effect, perform an integrity check on the relationship path from the product or project entity to the smallest audit unit entity, and verify with existing example evidence whether field extraction, rule loading, and rule-level status output can still form a closed data chain.

[0061] After the update takes effect, the new audit task reads the effective entity, relationship and rule records; historical audit results continue to retain the smallest audit unit identifier and rule identifier actually used for subsequent review.

[0062] The aforementioned controlled update method enables the quality knowledge graph and dynamic audit clause library to change with confirmed business information, while avoiding the direct conversion of single manual feedback into automatic learning parameters that would alter subsequent audit results.

[0063] S2. Construct a dynamic review clause library that includes the smallest review unit, specifically including: S21. The natural language review standard is broken down into several smallest review units.

[0064] The review criteria for natural language expressions are analyzed, background explanations are removed, and complex constraint clauses are broken down into multiple minimum review units (MAUs). Each MAU corresponds to a single judgment dimension, including the judgment object, constraints, and judgment rules. Each MAU can independently perform compliance judgment and output the judgment result. Specifically, this includes: S211, Split standard text entries.

[0065] The review criteria in natural language form are first roughly divided into clauses and sub-clauses, removing explanatory, introductory, and descriptive statements, and retaining only binding judgment statements.

[0066] S212, Single constraint stripping.

[0067] If a sentence contains multiple decision conditions ("satisfies A, B, and C simultaneously"), it is split into multiple sentences. Each MAU carries only one decision dimension to avoid a single unit outputting multiple decision results.

[0068] S213, Stripping away embellishments and background descriptions.

[0069] Retain independently verifiable audit content and form a Desc, assign a unique ID to this audit content, and bind one or more decision rules required to execute the audit content to R. MAU .

[0070] S214, Independence check.

[0071] Each decomposed MAU can be input into the review reasoning module separately, and the compliance judgment result can be output independently. If a decomposed unit must depend on another unit to make a judgment, it means that the decomposition is not in place and needs to be merged.

[0072] S215, Structured Package MAU.

[0073] Assign a unique identifier to each MAU, and store fields including: Unit ID, Decision Object, Constraints, Decision Rules, Allowed Values / Thresholds, and Failure Prompt.

[0074] The dynamic review clause library stores the smallest review unit identifier, review content description, required evidence type, rule input fields, and judgment rules in a linked manner. Taking the design input review involved in clause 8.3.4 of the GJB 9001C standard as an example, the following smallest review unit and rule records can be generated: Table 1: Example of the smallest audit unit (GJB 9001C 8.3.4 section)

[0075] S22. Store the smallest review unit instances in a structured manner and create a scene index. Record the ID, Desc, and R of each smallest review unit. MAU Rule input fields and rule parameters are stored together and indexed by ID. After the audit task determines the smallest audit unit, it uses this ID to read the corresponding list of input fields and the set of judgment rules. Rule updates take effect after confirmation, and the history retains the identifier of the rule it invoked.

[0076] The dynamic review terms library is essentially a collection of instantiated review units. The parsed collection of MAU instances, representing the smallest review unit, is stored in a structured manner. Each MAU is configured with a unique identifier, judgment rules, applicable scenario tags, version, and effective status. A scenario index is built to support the addition, editing, and decommissioning of MAUs, enabling online dynamic updates and resulting in the dynamic review terms library. During review task execution, a subset of valid MAUs is selected based on the business scenario for use in review inference.

[0077] S3. Obtain the target product or project identifier for the pending audit task, and identify the product or project entity matching the target product or project identifier in the pre-established quality knowledge graph; using this entity as the starting point, traverse related entities sequentially along the occurrence relationship, cause attribution relationship, and audit basis relationship, and summarize the smallest audit unit entity identifiers reached into a candidate smallest audit unit set. Specifically, this includes: S31. Obtain the target product or project identifier for the pending audit task.

[0078] The system receives the quality audit task to be executed and reads the unique business identifier bound to the task. This identifier is either a product model code or a project number, which is used to uniquely identify the audit object corresponding to this audit and distinguish the audit scenarios of different equipment products and different rectification projects.

[0079] For example, if the current audit task is a zero-fault audit of a certain type of aircraft engine, then the corresponding product model identifier is read and obtained.

[0080] S32. Locate product or project entities in the quality knowledge graph based on target identifiers.

[0081] Match the target product or project identifier with the entity identifier field in the product or project entity record. If no entity is found, output the status of "target entity not found" and stop the current path query; if an entity is found, determine it as the starting point for traversal; if there are multiple records with the same entity identifier, first deduplicate them by entity identifier, and transfer the records that cannot be deduplicated to the quality expert for confirmation.

[0082] For example, the model identifier can be used to match the aircraft engine product entity stored in the knowledge graph, which can then be used as the starting point for subsequent link retrieval.

[0083] S33. Starting from the matching product or project entity, perform branch traversal in the order of occurrence relationship, cause attribution relationship, and audit basis relationship.

[0084] First, by following the relationship between product / project entities and quality event entities, all historical faults, quality anomalies, and other quality event entities associated with the product / project are retrieved. Then, by following the cause attribution relationship corresponding to the quality event entities, the associated management process entities such as design, process, and operation and maintenance are traced. Finally, based on the audit basis relationship between management process entities and audit rules, all bound minimum audit units (MAU) entities are traversed.

[0085] S34. Summarize the set of candidate minimum review units.

[0086] The identifier of the smallest audit unit entity reached by each traversal branch is added to the candidate set; when different branches reach the same smallest audit unit entity, only one set member is retained according to the smallest audit unit identifier. Once a branch reaches the smallest audit unit entity, the extension of that branch stops, without affecting the continuation of traversal by other branches.

[0087] As an embodiment of the present invention, taking the design input review of “XX model radar” as an example, in S1, the map path query unit starts from the product entity corresponding to “XX model radar”, reaches the quality event entity of “solder joint failure in high temperature and high humidity test” along the occurrence relationship, reaches the management process entity of “design input review process” along the cause attribution relationship, and then reaches the minimum review unit entity of MAU-GJB9001C-8.3.4-01 along the review basis relationship, and adds the minimum review unit identifier to the candidate minimum review unit set.

[0088] S4. Receive audit execution instructions and audit evidence documents. The audit execution instructions include the identifier of the target minimum audit unit selected from the candidate minimum audit unit set; perform text recognition and key information extraction on the corresponding audit evidence documents, and associate and store the field names, field values, and the original evidence locations of the field values ​​to form evidence feature data. Specifically, this includes: S41. Receive audit execution instructions and corresponding audit evidence documents.

[0089] The system parses the target minimum audit unit identifier carried in the audit execution instruction and receives the audit evidence file corresponding to the target minimum audit unit; the audit evidence file can be a scanned document or an electronic file with a document text layer.

[0090] For example, if the target minimum audit unit is "fault phenomenon description completeness audit", then the system will receive the equipment fault zeroing report document to be audited as the audit evidence document.

[0091] S42. Perform text recognition and key information extraction on the audit evidence documents.

[0092] When the input is a scanned document, optical character recognition is performed to generate recognized text arranged by page number; the evidence preprocessing unit extracts the corresponding field values ​​from the document text layer of the recognized text or electronic document according to the input field list declared in the target minimum review unit binding rule.

[0093] S43. Associate the stored field value with the location of the original evidence.

[0094] Each field record includes a field name, a field value, and the location of the original evidence. For scanned documents, the location of the original evidence consists of the page number and the text range of the field value within the optical character recognition (OCR) text; for electronic documents, the location of the original evidence consists of the paragraph identifier and the character range of the field value within the corresponding paragraph. When the same field name appears multiple times, separate field records are created for each occurrence, and subsequent occurrences do not overwrite previous occurrences.

[0095] In this embodiment, the audit execution instruction selects MAU-GJB9001C-8.3.4-01 as the target minimum audit unit from the candidate minimum audit unit set.

[0096] The evidence preprocessing unit reads the list of input fields for the sufficiency rule corresponding to the minimum audit unit of the target. The list of key information fields K to be included in the sufficiency rule includes functional performance requirements, environmental adaptability requirements, security requirements, regulatory requirements, interface requirements, and information on previous similar designs.

[0097] As an example, the field records generated from the corresponding audit evidence files are matched to four field names: functional performance requirements, environmental adaptability requirements, security requirements, and interface requirements. Each field record is associated with its source audit evidence file and the location of the original evidence.

[0098] When the evidence document being reviewed is a scanned document, optical character recognition is performed on the scanned document to obtain the recognized text arranged according to page numbers; key information extraction is performed on the recognized text to extract at least one field from the reviewed object, document date, approval record, and reviewed input items; the field name, field value, page number, and text position of each field are combined to form a field record, and the field records constitute the evidence feature data.

[0099] S5. Based on the target minimum audit unit identifier, load the bound judgment rule set from the dynamic audit clause library, substitute the field values ​​matching the input fields of each rule into the corresponding rule, and generate rule-level judgment and status auxiliary information for each rule; the status auxiliary information is one of the following: abnormal fields, missing fields, or no abnormal marker. Specifically, it includes: S51. Load the corresponding judgment rule set R based on the target minimum review unit identifier. MAU .

[0100] The system reads the corresponding record in the dynamic audit clause library using the target minimum audit unit identifier as an index, and obtains the judgment rule set, each rule input field and rule parameter bound to the record; if the corresponding record is not found or the record is in an inactive state, the system outputs the rule set as unavailable and does not use rules from other minimum audit units to replace it.

[0101] For example, when the target minimum audit unit is the sufficiency audit of design input review records, the sufficiency rules bound to that unit and its list of key information fields K to be included are loaded from the dynamic audit clause library.

[0102] R MAU This is the set of decision rules associated with this smallest review unit. Each decision rule in the set of decision rules is represented as follows: r i =<Type,Condition,Action> ; Wherein, Type represents the rule type, Condition represents the violation trigger predicate, and Action represents the rule-level judgment output when the violation trigger predicate is true.

[0103] The decision rule set includes sufficiency rules, temporal rules, association consistency rules, and quantitative comparison rules, among which: Sufficiency rule: Calculate field coverage using the following formula: C=|F K ∩K| / |K|; Where C represents field coverage, K represents the list of key information fields that should be included in the target minimum audit unit, and F represents the field coverage rate. K This represents the set of provided fields extracted from the audit evidence documents and matched with the list of key information fields; when C is lower than the preset coverage threshold, the sufficiency rule violation trigger predicate is established, outputting a rule-level judgment of "insufficient evidence," and setting the set K\F... K The fields in the table are used as missing fields.

[0104] Sequential rules; For design input review records, the sequential rules take the design task assignment date, review date, and design drawing release date as input fields. When the sequential relationship of "the design task assignment date is earlier than the review date, and the review date is earlier than the design drawing release date" is not met, the sequential rule's violation trigger predicate is established and a rule-level judgment of sequential abnormality is output.

[0105] Association consistency rules are used to compare the same attribute value of the same entity across files or records. For non-numeric attributes, a rule-level judgment of inconsistency is output when the attribute values ​​are different. For numeric attributes, a rule-level judgment of inconsistency is output when the difference between attribute values ​​exceeds the preset tolerance range.

[0106] The quantitative comparison rule is used to compare the measured values ​​in the audit evidence with the numerical range specified by the standard or document. When the measured value does not fall within the numerical range, a rule-level judgment of numerical anomaly is output.

[0107] In this embodiment, the sufficiency rule is constructed according to the field name. K ={functional performance requirements, environmental adaptability requirements, security requirements, interface requirements}, construct the missing field set K\F K ={regulatory requirements, previous similar design information}, and calculate C=4 / 6. When the preset coverage threshold is 90%, C is lower than the threshold, the sufficiency rule outputs a rule-level judgment of "insufficient evidence", and records the regulatory requirements and previous similar design information as missing fields.

[0108] S52. Complete the matching of evidence fields with judgment rule input fields.

[0109] The system reads the input fields declared by each judgment rule and searches for the corresponding field record in the evidence feature data by field name. When there are multiple records for the same field name, all candidate records are retained, and the specific rule selects the record to actually participate in the calculation according to its comparison object. Fields not declared as input by the current rule do not participate in the calculation of that rule.

[0110] S53. Substitute the matching field value into the corresponding judgment rule and perform the calculation.

[0111] For each decision rule in the decision rule set, read the input fields declared by that rule; for sufficiency rules, first construct the set of provided fields and the set of missing fields and calculate the field coverage; for other rules, calculate the violation trigger predicate when all required input fields are complete.

[0112] For each decision rule, perform the following calculations: When the evidence feature data lacks the input fields required to calculate the judgment rule, a rule-level judgment of "insufficient evidence" is generated, and the missing input fields are recorded as missing fields.

[0113] When all the input fields required to calculate the judgment rule are complete and the violation trigger predicate is true, the Action is executed and the corresponding exception rule-level judgment is generated.

[0114] When all the input fields required to calculate the judgment rule are complete and the violation trigger predicate is not true, a rule-level judgment of "no exception triggered" is generated.

[0115] Each rule-level decision is written independently into the audit result set, and "insufficient evidence" is not directly converted into a conclusion that the auditee does not meet the requirements of the quality management system.

[0116] S54. Generate rule-level judgments and corresponding status auxiliary information.

[0117] After each judgment rule is calculated, the rule-level judgment of that rule is output independently; if the input field is missing, the missing field is recorded; if the violation trigger predicate is true, the abnormal field is recorded; if the input field is complete and the violation trigger predicate is false, no abnormal marker is recorded.

[0118] S6. Generate and output the audit result set; each rule-level result includes the target minimum audit unit identifier, rule identifier, rule-level judgment, status auxiliary information, and the original evidence location of the existing field values ​​that participated in the corresponding rule calculation; missing fields only record the missing marker, not the original evidence location. Specifically, this includes: S61. Initialize the data storage structure of the audit result set.

[0119] The system pre-builds an audit result set, which is used to collect the output content of all rules under the target minimum audit unit. Each rule-level result has multiple storage fields reserved in advance, which are used to store the target minimum audit unit identifier, rule identifier, rule-level judgment, status auxiliary information, field values ​​involved in the calculation, and the location of the original evidence.

[0120] For example, for the sufficiency review of design input review records, an independent result entry is created for each judgment rule.

[0121] S62, Traverse the rules, output and assemble the results of a single rule.

[0122] Read the rule-level judgment and status auxiliary information of each rule obtained in the previous step in sequence, and write the target minimum review unit identifier and the rule identifier of the current rule into the corresponding result entries. At the same time, obtain the existing field values ​​that participated in the calculation of this judgment rule, and store the field values ​​in the corresponding field positions in the result set.

[0123] S63. Process the original evidence location field based on the status auxiliary information.

[0124] When the status auxiliary information is an abnormal field or has no abnormal marker, record the existing field values ​​that actually participated in the rule calculation and their original evidence positions; when the status auxiliary information is a missing field, record the missing field with a missing marker but do not record the original evidence positions, while retaining the original evidence positions of other existing input field values.

[0125] For example, if the design input review record does not match a field required by regulations, the record for that field is marked as missing and the original evidence position is left blank; fields such as functional performance requirements that have already been matched still retain their original evidence positions.

[0126] S64. Summarize the results of each rule level to form a complete audit result set.

[0127] All processed rule-level results are collected into a unified audit result set. Audit reports can be generated based on this audit result set, and manual reviewers can quickly view the judgment conclusion, field information, and evidence source of each rule.

[0128] In this step, the result generation unit outputs the rule-level result corresponding to the sufficiency rule. The rule-level result includes the target minimum audit unit identifier, the rule identifier of the sufficiency rule in the dynamic audit clause library, the rule-level judgment of "insufficient evidence," the two missing fields mentioned above, and the original evidence positions of the four existing field values ​​actually involved in the coverage calculation. For the two missing fields, only the missing marker is recorded, not the original evidence position. The audit terminal displays the corresponding original text fragment based on the original evidence position, and the auditor confirms the missing status or supplements the audit evidence. The system associates and stores the confirmation information or supplementary evidence information entered by the auditor with the rule-level result, and the final audit finding record is formed after the auditor confirms it.

[0129] S7. Conduct audit confirmation, final audit findings record, and compile standardized audit evidence items.

[0130] S71. The system sends the audit result set and the location of the original evidence corresponding to each rule-level result to the auditor's terminal; the auditor's terminal displays the corresponding original text fragments according to the location of the original evidence.

[0131] S72. Auditors can input confirmation information, correction information, or supplementary evidence information for rule-level results. The system will associate and store the manual input with the corresponding rule-level results to form the final audit findings record.

[0132] S73. Machine-generated "insufficient evidence," abnormal triggering, or non-triggered abnormality are all rule-based statuses and are not directly used as the final non-compliance conclusion; the final compliance conclusion is confirmed by the auditor in combination with the original evidence and supplementary evidence.

[0133] S74. Collection of standardized audit evidence items after auditor confirmation.

[0134] S741. After the auditor completes the confirmation, the system reads each rule-level result and the corresponding manual confirmation information, maps the target minimum audit unit identifier and its associated standard clause identifier to the audit clause identifier, maps the conclusion after manual confirmation to the compliance conclusion, maps the status auxiliary information to the evidence issue type, and maps the original evidence location to the evidence tracing information.

[0135] S742. For existing fields, the evidence entry carries the field value that actually participated in the rule calculation and its original evidence location; for missing fields, only the missing marker is recorded, and no non-existent original evidence text or original evidence location is generated.

[0136] S743. The generated audit evidence items shall include at least the audit clause identifier, rule identifier, compliance conclusion after manual confirmation, evidence issue type, original text citation of existing evidence, location of original evidence, and missing mark, so that one rule-level result corresponds to one audit evidence item that can be reviewed.

[0137] S744. The system will collect and store the audit evidence items confirmed by the auditors into the quality management system audit evidence set, and maintain the association between the item and the original rule-level results, manual confirmation information and supplementary evidence information.

[0138] S745. The audit evidence set of the quality management system can be used for subsequent audit archiving and review; for machine rule-level results that have not yet been manually confirmed, they are only saved as records to be confirmed and do not directly form final non-conformities or rectification instructions.

[0139] S746. When an auditor provides supplementary evidence, the newly added field record saves its field value and the location of the original evidence, and associates it with the original rule-level result; whether to re-execute the corresponding judgment rule is determined by the review instruction issued by the auditor, and the original rule-level result is retained as a historical record.

[0140] For example, if a rule outputs "Insufficient Evidence" due to a missing "Time of Failure" field, the system first records the missing marker but does not record the original evidence location for that field. When the auditor confirms that supplementary evidence is indeed needed, the compliance conclusion is recorded as "Evidence to be Supplemented," rather than directly as "Non-compliant." After the supplementary document arrives, the system generates a new field record, and the auditor issues a review instruction.

[0141] Optional Implementation: Historical Quality Event Grouping and Statistics. The system can group quality event entities according to the responsible process field in the historical quality event records, count the number of quality events corresponding to each management process, and associate the grouping and statistics results with the candidate minimum audit unit identifiers obtained through the graph path for storage, for sorting and displaying candidate minimum audit units during audit planning. This optional implementation does not change the execution order of S1 to S7.

[0142] like Figure 2 As shown, as one aspect of this invention, the present invention provides a quality management system audit evidence processing system based on the smallest audit unit. Figure 1 As shown, the system includes at least a data input interface 101, a quality knowledge graph 102, a dynamic audit clause library 103, a graph path query unit 104, an evidence preprocessing unit 105, a rule execution unit 106, a result generation unit 107, and an auditor terminal 108; the graph path query unit 104, the evidence preprocessing unit 105, the rule execution unit 106, and the result generation unit 107 are implemented by the processor executing computer programs in the memory.

[0143] The data input interface 101 is used to receive the target product or project identifier, audit execution instructions, and audit evidence documents in the form of scanned documents or electronic files.

[0144] The quality knowledge graph 102 is used to store product or project entities, quality event entities, management process entities, and the smallest audit unit entities, and provides entity and relationship data to the graph path query unit.

[0145] like Figure 3 As shown, the quality knowledge graph is represented as follows: ; Where E is the entity set, RKG is the graph relation set, and A is the entity attribute and relation attribute set.

[0146] In one logical storage method, each entity record includes at least an entity type, an entity identifier, and an entity attribute; each relation record includes at least a relation type, a starting entity identifier, an ending entity identifier, and a relation attribute. Entity attributes and relation attributes belong to set A, and the starting and ending entities in the relation record point to the corresponding entity records in set E.

[0147] The main chain of the quality knowledge graph is unified as follows: Product or project entity -> Quality incident entity -> Management process entity -> Smallest audit unit entity; In the quality knowledge graph, the relationships between adjacent entities are sequentially: occurrence relationship, cause attribution relationship, and audit basis relationship. For example, starting with the product entity corresponding to "XX model radar", the occurrence relationship leads to the quality event entity "solder joint failure in high temperature and high humidity test", the cause attribution relationship leads to the management process entity "design input review process", and the audit basis relationship leads to the smallest audit unit entity MAU-GJB9001C-8.3.4-01.

[0148] Fault phenomena, product or project identifiers, responsible processes, and standard clause mappings in historical quality incident records can be annotated by quality experts and written into the map; when extracting entities using natural language processing, the extraction results are confirmed by quality experts before being written into the formal map.

[0149] The dynamic review clause library 103 stores the smallest review unit identifier, the review content description, and the set of judgment rules corresponding to the smallest review unit. Each judgment rule in the set of judgment rules... Represented as: ; Where Type is the rule type; Condition is the violation trigger predicate; and Action is the abnormal result generated when the violation trigger predicate is true.

[0150] The rule types include at least one of the following: sufficiency rule, compliance rule, temporal rule, correlation consistency rule, and quantitative comparison rule.

[0151] The dynamic review clause library also stores the input fields and rule parameters required for the rules. For example, the input fields for the design input review record sufficiency rule include functional performance requirements, environmental adaptability requirements, safety requirements, regulatory requirements, interface requirements, and information on previous similar designs; the input fields for the time-series rule include the design task assignment date, review date, and design drawing release date.

[0152] The map path query unit 104 is used to receive the target product or project identifier from the data input interface and output the candidate minimum review unit identifier.

[0153] When performing a path query, the graph path query unit starts from the target product or project entity and traverses only sequentially along the occurrence relationship, cause attribution relationship, and audit basis relationship. If the starting entity has multiple occurrence relationships, the quality event entity has multiple cause attribution relationships, or the management process entity has multiple audit basis relationships, the graph path query unit continues traversing each branch that satisfies the relationship order. When a branch reaches the smallest audit unit entity, it stops extending that branch without affecting the continued traversal of other branches. The graph path query unit adds the identifier of the smallest audit unit entity reached by each branch to the candidate smallest audit unit set. When different branches reach the same smallest audit unit entity, only one member is retained in the candidate smallest audit unit set according to the smallest audit unit identifier. The auditor issues an audit execution instruction based on the audit planning results. The audit execution instruction includes the target smallest audit unit identifier selected from the candidate smallest audit unit set.

[0154] The evidence preprocessing unit 105 is used to determine the target minimum audit unit from the candidate minimum audit units according to the target minimum audit unit identifier in the audit execution instruction, and to receive the corresponding audit evidence file and output evidence feature data.

[0155] like Figure 2 As shown, the evidence documents to be reviewed can be scanned documents or electronic documents. When the input is a scanned document, the evidence preprocessing unit performs optical character recognition (OCR) to generate recognized text arranged by page number. During recognition, the page number of the field value and its text range in the OCR text are retained for subsequent display of the original evidence.

[0156] The evidence preprocessing unit reads the input fields of the judgment rule declaration bound to the target minimum review unit based on the target minimum review unit identifier, drives the extraction of key information with the rule input field list, and extracts field values ​​corresponding to the input field list from the identified text of the scanned document or the document text layer of the electronic document. For example, for design input review records, at least one field is extracted from the reviewed object, document date, approval record, and reviewed input items. Each field record can be represented as: ; Wherein, FieldName is the field name, FieldValue is the field value, and SourceLocation is the location of the original evidence. For scanned documents, SourceLocation consists of the page number where the field value is located and the text range of that field value in the optical character recognition (OCR) text; for electronic documents, SourceLocation consists of the paragraph identifier where the field value is located and the character range of that field value in the corresponding paragraph. Each field record constitutes the evidence feature data F. When the same field name appears multiple times in the audited evidence document, a separate field record is created for each appearance, and the corresponding field value and original evidence location are saved separately, so that subsequent appearances do not overwrite previous appearances.

[0157] In one implementation, the evidence preprocessing unit uses a natural language processing model to identify the field names and values ​​in the list of rule input fields, and generates a SourceLocation according to the aforementioned location representation of scanned documents or electronic files.

[0158] The rule execution unit 106 is used to read the judgment rule set with the target minimum review unit identifier as the index and use the evidence feature data as the rule input.

[0159] The rule execution unit loads the corresponding judgment rule set based on the target minimum review unit identifier, reads the input fields of each rule declaration, and performs field matching in the evidence feature data.

[0160] Rule-by-rule execution and output status: The rule execution unit first distinguishes between sufficient rules and other rules besides sufficient rules.

[0161] For sufficiency rules, if a field does not match a state that is processed by the field coverage calculation, the general input field missing short-circuiting is not performed before the coverage calculation; instead, the provided field set FK and the missing field set K\FK are constructed first, and then the field coverage C is calculated and compared with the preset coverage threshold.

[0162] Sufficiency Rule Implementation Example: Let K be the list of key information fields that the target minimum audit unit should include. The sufficiency rule first obtains field records whose field names match K from the evidence feature data, constructs a set of provided fields FK according to the field names, and constructs a set of missing fields K\FK, then calculates the field coverage C according to the following formula. When there are multiple field records with the same field name, it is counted as a provided field in FK for coverage, but each field record is still retained separately: ; When C is lower than the preset coverage threshold, the violation trigger predicate is true, the rule outputs "insufficient evidence", and K\FK is taken as the missing field set; when C is not lower than the preset coverage threshold, the violation trigger predicate is false, and the rule outputs "no exception triggered".

[0163] For example, if K contains six fields, and four of the evidence matches, then C = 4 / 6. When 90% is configured as the coverage threshold for this rule, the rule outputs "Insufficient Evidence" and lists the unmatched fields.

[0164] For rules other than the sufficiency rule, the following state processing is performed: If any input field necessary to complete the calculation of this rule is missing, the rule is subjected to general input field missing short-circuiting, the output "Insufficient evidence" is displayed, the missing fields are listed, and the calculation of the violation triggering predicate of this rule is stopped.

[0165] If all input fields are complete, calculate the violation trigger predicate; if the predicate is true, output the abnormal result defined by the rule. If the input fields are complete and the predicate is false, output "No exception triggered".

[0166] The above statuses are rule-by-rule statuses. "Insufficient evidence" only indicates that the current input documents are insufficient to complete the calculation of this rule; it does not directly indicate that the auditee does not meet the quality management system requirements. The final conformity conclusion can be confirmed by the auditor based on supplementary evidence.

[0167] Example of a sequential rule: For design input review records, if the design task assignment date is t1, the review date is t2, and the design drawing release date is t3, the violation trigger predicate of the sequential rule can be expressed as: ; When Condition is true, output "Timing anomaly"; when Condition is false, output "Anomaly not triggered". If any field of t1, t2, or t3 is missing, output "Insufficient evidence" instead of "Facts do not match".

[0168] Other rule implementation examples: When forming evidence feature data, each field record is associated with its source audit evidence document; when there are multiple field records for the same field name, each field record and its respective original evidence location are retained separately.

[0169] The association consistency rule only compares field records that have already been identified as belonging to the same entity; it does not re-determine the entity correspondence within the rule. For the same non-numeric attribute of the same entity, if the attribute values ​​are different, the output is "Inconsistent"; for the same numeric attribute of the same entity, if the difference between attribute values ​​exceeds the preset tolerance range, the output is "Inconsistent". When multiple field records exist for the same field name, the rule-level result outputs the original evidence location of each field record actually participating in this comparison; field records with the same name that did not participate in this comparison remain in the evidence feature data but are not used as the original evidence location corresponding to this rule calculation.

[0170] The quantization comparison rule is used to compare the measured value with the range specified by the standard or document. When the measured value is less than the lower limit of the range or greater than the upper limit of the range, "numerical anomaly" is output.

[0171] The result generation unit 107 is used to combine the rule identifier, rule-level judgment, status auxiliary information, and the original evidence position of the existing field values ​​involved in the rule calculation into an audit result set.

[0172] The result generation unit generates one rule-level result for each rule. The rule-level result must include at least: The target minimum audit unit identifier; rule identifier; rule-level judgment; status auxiliary information, which is one of the following: abnormal field, missing field, or no abnormality marker; the original evidence location of the existing field values ​​involved in the calculation of this rule.

[0173] When all rule input fields are complete and the violation trigger predicate is true, the status auxiliary information records the abnormal fields and the original evidence positions of the abnormal fields and other existing field values ​​involved in the rule calculation. When a rule input field is missing, the status auxiliary information records the missing field, marks the missing field as missing but does not record the original evidence position, and records the original evidence positions of other existing input field values. When all rule input fields are complete and the violation trigger predicate is false, the status auxiliary information records no abnormality mark and records the original evidence positions of existing field values ​​involved in the rule calculation.

[0174] The auditor terminal 108 is used to receive and display the audit result set and the corresponding original text fragments, and to receive confirmation information, correction information or supplementary evidence information input by the auditor.

[0175] The system associates and stores the information entered by the auditor with the corresponding rule-level results to form the final audit findings record; the machine-generated rule-by-rule status is not directly used as the final non-compliance conclusion.

[0176] The above embodiments are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for processing audit evidence in a quality management system based on the smallest audit unit, executed by a computer, characterized in that: It includes: Obtain the target product or project identifier for the audit task to be processed, and identify the product or project entity that matches the target product or project identifier in the pre-established quality knowledge graph; take the product or project entity as the starting point for traversal, and traverse the associated quality event entities, management process entities and minimum audit unit entities in sequence along the occurrence relationship, cause attribution relationship and audit basis relationship, and summarize the identifiers of the minimum audit unit entities obtained by traversal to form a candidate minimum audit unit set. The system receives audit execution instructions and audit evidence files. The audit execution instructions include the identifier of the target minimum audit unit selected from the candidate minimum audit unit set, and the audit evidence files correspond to the target minimum audit units. The system performs text recognition and key information extraction on the audit evidence files, and stores the extracted field names, field values, and the original evidence positions of the field values ​​in the audit evidence files in association to form evidence feature data. Based on the identifier of the target minimum review unit, load the set of judgment rules bound to the target minimum review unit from the dynamic review clause library, substitute the field values ​​in the evidence feature data that match the input fields of each judgment rule into the corresponding judgment rule, and generate rule-level judgment and status auxiliary information respectively. Generate and output an audit result set. Each rule-level result in the audit result set shall include at least the target minimum audit unit identifier, rule identifier, rule-level judgment, status auxiliary information, and the original evidence location of the existing field values ​​that participated in the calculation of the corresponding judgment rule.

2. The method for processing audit evidence of a quality management system based on the smallest audit unit according to claim 1, characterized in that, The smallest unit of review is the smallest knowledge unit in the review standard that can independently make a compliance determination, defined as a triple: ALWAYS = <ID, Desc, R MAU >; ； ； Where MAU is the smallest review unit, ID is the identifier of the smallest review unit; Desc is a natural language description of the content reviewed by the smallest review unit; R MAU For the set of decision rules associated with the smallest review unit, r i Let i be the i-th judgment rule, where Type is the rule type, Condition uniformly represents the violation trigger predicate, and Action represents the rule-level judgment output when the violation trigger predicate is true.

3. The method for processing audit evidence of a quality management system based on the smallest audit unit according to claim 1, characterized in that, A quality knowledge graph is defined as a directed graph: G=(E,R KG ,A); Where G is the quality knowledge graph, E is the entity set, and R is the... KG Let A be the set of relations and A be the set of attributes.

4. The method for processing audit evidence of a quality management system based on the smallest audit unit according to claim 3, characterized in that, The entity set includes at least product or project entities, quality event entities, management process entities, and the smallest audit unit entity.

5. The method for processing audit evidence of a quality management system based on the smallest audit unit according to claim 4, characterized in that, The set of relationships includes at least the occurrence relationship from the product or project entity to the quality event entity, the cause attribution relationship from the quality event entity to the management process entity, and the audit basis relationship from the management process entity to the smallest audit unit entity.

6. The method for processing audit evidence of a quality management system based on the smallest audit unit according to claim 5, characterized in that, The decision rule set includes sufficiency rules, temporal rules, association consistency rules, and quantitative comparison rules.

7. The method for processing audit evidence of a quality management system based on the smallest audit unit according to claim 6, characterized in that, The sufficiency rule calculates field coverage using the following formula: C=|F K ∩K| / |K|; Where C represents field coverage, K represents the list of key information fields that should be included in the target minimum audit unit, and F represents the field coverage rate. K This refers to the set of provided fields extracted from the audit evidence documents and matched with the list of key information fields; When C is below the preset coverage threshold, the sufficiency rule's violation trigger predicate is established, outputting a rule-level judgment of insufficient evidence, and setting the set K\F... K The fields in the table are used as missing fields.

8. The method for processing audit evidence of a quality management system based on the smallest audit unit according to claim 6, characterized in that, The timing rule takes the design task assignment date, review date, and design drawing release date as input fields. When the sequential relationship that the design task assignment date is earlier than the review date and the review date is earlier than the design drawing release date is not met, the timing rule's violation trigger predicate is established and a timing exception rule-level judgment is output.

9. The method for processing audit evidence of a quality management system based on the smallest audit unit according to claim 6, characterized in that, Association consistency rules are used to compare the same attribute value of the same entity across files or records; for non-numeric attributes, when the attribute values ​​are different, the rule-level judgment of inconsistency is output. For numerical attributes, when the difference between attribute values ​​exceeds the preset tolerance range, a rule-level judgment of inconsistency is output; the quantitative comparison rule is used to compare the measured values ​​in the audit evidence with the numerical range specified by the standard or document, and when the measured values ​​do not fall into the numerical range, a rule-level judgment of numerical abnormality is output.

Images