Aircraft product quality report verification method and system based on pre-trained language model

Abstract

The application provides an aviation product quality report verification method and system based on a pre-trained language model, relates to the technical field of aviation product quality report key data extraction and verification, and comprises the following steps: S1, acquiring an aviation product quality zero report text, extracting a feature length, and performing merging and blocking operations; S2, extracting key information feature text of the aviation product quality zero report text, and generating a test prompt; S3, performing context positioning on paragraphs of the aviation product quality zero report text; and S4, verifying the aviation product quality zero report text by using a large-scale language model. The application analyzes the quality problem zero report by using a large-scale pre-trained language model, improves the extraction accuracy of key information, realizes automatic verification of content, ensures the integrity and logical consistency of the content, can find common problems in quality control and optimize improvement schemes, and provides reliable data support for the improvement and standardization of the quality zero report.

Description

Technical Field

[0001] This invention relates to the field of key data extraction and verification technology for aviation product quality reports, specifically to a method and system for verifying aviation product quality reports based on a pre-trained language model. Background Technology

[0002] Zero-based reports play a crucial role in quality control, and their review and verification process is a key step in ensuring that product quality issues are effectively resolved. Zero-based reports typically contain a wealth of technical details, involving complex domain terminology and multi-layered logical relationships. They must clearly describe the time, location, phenomena, and environmental conditions of the quality problem, and provide a detailed analysis of the problem's nature, product structure, design history, and root causes. Therefore, the review and verification process of these reports demands a high level of completeness, consistency, and logical rigor in the information provided.

[0003] In practical applications, the verification and validation of zeroing reports are often time-consuming and require a high level of professional knowledge and experience from personnel. Common problems encountered during the verification process include incomplete information extraction, inconsistent content, and logical errors in complex scenarios, all of which can affect the effectiveness of zeroing quality issues. This invention aims to improve the accuracy of key information extraction from zeroing reports through automation, ensuring the comprehensiveness and logical consistency of information, thereby automating the report validation process and improving the efficiency and consistency of the quality control process.

[0004] With the development of large-scale pre-trained language models (LLMs), text analysis techniques based on natural language processing (NLP) are increasingly being applied in quality control and zero-reporting. Current technologies widely use Encoder-Only, Decoder-Only, and Encoder-Decoder architectures for information extraction and text generation. However, these models still suffer from incomplete information extraction and inconsistent generated content when processing zero-reporting analysis.

[0005] Currently, typical applications of large-scale language models include encoder-only architecture models such as BERT (Bidirectional Encoder Representations from Transformers), primarily used for text classification and information extraction; and decoder-only architecture models such as GPT (Generative Pre-trained Transformer), used for long text generation. While these models demonstrate excellent performance, they present numerous challenges in specific quality control applications, such as content verification and automated generation of zero-point reports. This invention aims to improve the accuracy of key data extraction in zero-point reports using large-scale language models, achieving automated verification and report generation. Summary of the Invention

[0006] To address the shortcomings of the existing technologies, the present invention aims to provide a method and system for verifying aviation product quality reports based on pre-trained language models. This method utilizes large-scale pre-trained language models to analyze quality problem zeroing reports, improving the accuracy of key information extraction and enabling automatic verification of report content to ensure its integrity and logical consistency. The invention can identify common problems in quality control and optimize improvement solutions, providing reliable data support for the improvement and standardization of quality zeroing reports.

[0007] Specifically, on the one hand, the present invention provides a method for verifying aviation product quality reports based on a pre-trained language model, which includes the following steps:

[0008] S1: Obtain the quality zeroing report text and use a language model to calculate P for each text segment. i The characteristic length len(P) i ), determine adjacent text paragraphs P i ,P i+1 If the merging conditions are met, merge to obtain the quality zeroing report text block set B;

[0009] S2: Parse the verification rules of the quality zeroing report text, perform component syntactic analysis, and obtain the syntactic structure tree S of the verification rules. i Identify noun phrases (NP) and verb phrases (VP) within the text, concatenate them to form key information feature text, construct a key information feature text set I, and then apply the test rule set R = {R1, R2, ..., R...} n Enter the prompt template T sequentially to obtain the inspection prompt set P;

[0010] S3: Obtain the quality zeroing report text block set B from step S1 and the key information feature text set I from step S2, input them into the ranking model based on the Transformer architecture, and use the similarity function Sim(B) j,I i ) Calculate the feature text I of each key information i With each quality zeroing report text block B j The fit is determined; the text blocks are sorted according to the fit to obtain the sorted quality zeroing report text block set; the top k text blocks are selected. As candidate context paragraphs, text merging is performed to obtain the context text of the zero-quality report text, which is C. j ;

[0011] S4: Verify rule R from step S2. j The corresponding inspection prompt set P and the context text C of the quality zeroing report text obtained in step S3. j The large-scale language model is concatenated and input, and the JSON format is set as the output format for the verification results to obtain the quality zeroing report verification results, which are used to improve the quality zeroing report.

[0012] Preferably, step S2 specifically includes:

[0013] S21: Obtain the inspection rule set R = {R1, ..., R...} of the quality zeroing report text. i ,…,R n The text of the zero-quality report is analyzed, and its verification rules are analyzed. Constituent syntactic analysis is performed to extract noun phrases, verb phrases, and key information features from the rules, resulting in the syntactic structure tree S of the verification rules. i ;

[0014] S22: Identify key information features in the quality zeroing report text; identify S in step S21. i The noun phrases (NP) and verb phrases (VP) are concatenated to form the key information feature text, thus constructing the key information feature text set I = {I1, ..., I...} i ,…,I n};

[0015] S23: Test the rule set R = {R1, R2, ..., R...} n Enter the prompt template T sequentially to obtain the inspection prompt set P.

[0016] Preferably, step S3 specifically includes:

[0017] S31: Obtain the quality zeroing report text block set B from step S1 and the key information feature text set I from step S2, and use the similarity function Sim(B) j ,I i To calculate key information feature text I i With Quality Zeroing Report Text Block B j The degree of compatibility;

[0018] S32: Based on the fit obtained in step S31, sort the text blocks according to the fit from largest to smallest, to obtain the sorted quality zeroing report text block set ranked(I i ,B);

[0019] S33: Obtain the sorted set of quality zeroing report text blocks from step S31, and select the first k text blocks. As candidate context paragraphs, text merging is performed to locate the context text C of the aviation product quality zeroing report text. j .

[0020] Preferably, the prompt template T in step S23 is: a standard prompt file in JSON format that determines whether the content meets the requirements based on the above content.

[0021] Preferably, the sorting of text blocks in step S32 is based on a sorting model with a Transformer architecture, which captures the relationship between the query and the candidates through a self-attention mechanism, thereby outputting the similarity.

[0022] Preferably, step S1 specifically includes:

[0023] S11: Determine the quality zeroing report text and the maximum input text length; set the quality zeroing report text D to contain n text paragraphs {P1,…P…P…} i …,P n}, and the maximum input text length of the model is L. max ;

[0024] S12: Perform text segmentation and use a language model to calculate P for each text segment. i The characteristic length len(P) i );

[0025] S13: Determine adjacent text segments P based on the characteristic length len() of the text segment. i ,P i+1 Does the merger condition meet?

[0026] S14: For text paragraphs that meet the merging criteria, perform the merge operation P'. i =P i ∪P i+1 , using P' i To replace P i and P i+1 Then re-evaluate P' i and P i+2 Determine if the merging conditions are met, and then merge to obtain the quality zeroing report text block set B.

[0027] Preferably, the JSON format in step S4 specifically includes: inspection items, inspection results, and inspection basis; the inspection items are inspection rules R. j The test results are categorized into three types: compliant, non-compliant, and inapplicable; the basis for the test is the reasoning behind the judgment.

[0028] Preferably, in step S12, the text paragraph splitting uses the text character splitting rule function Tokenizer() to split Chinese text according to Chinese characters, and English text according to English words.

[0029] On the other hand, the present invention provides an aviation product quality report verification system based on a pre-trained language model, which includes: a zeroing report text segmentation module, a key information and verification prompt module, a text paragraph context positioning module, a zeroing report verification module, and a verification result integration and output module;

[0030] The zeroing report text segmentation module divides long documents into blocks based on paragraphs and logic, with each text block adapting to the input length limit of a large-scale language model;

[0031] The key information and verification prompt module sets the key information to be verified in the zeroing report and its compliance standards, and classifies each item to determine its applicability during the verification process;

[0032] The text paragraph context localization module uses a rearrangement model to sort the segmented text blocks to achieve optimal order and contextual coherence;

[0033] The zeroing report verification module takes the zeroing report context and verification prompts as input into a large-scale language model, checks the zeroing report, records non-compliant and inapplicable items, and provides the basis for judgment to realize the compliance verification of the zeroing report;

[0034] The verification result integration and output module summarizes the verification results of each zeroing report and forms a unified summary table file.

[0035] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0036] (1) This invention uses a large-scale pre-trained language model to analyze the zero-out report of quality problems of aviation products, improves the accuracy of extracting key information in the text of the report, and realizes automatic verification of the report content to ensure its integrity and logical consistency.

[0037] (2) This invention can identify common problems in the quality control of aviation products and optimize improvement schemes based on the quality zeroing report, providing reliable data support for the improvement, optimization and standardization of the quality zeroing report of aviation products.

[0038] (3) The present invention realizes the zeroing report text segmentation, key information and verification prompts, text paragraph context positioning, zeroing report verification and verification result integration output through the aviation product quality report verification system, which greatly improves the verification efficiency and accuracy of the quality zeroing report text. Attached Figure Description

[0039] Figure 1 The flowchart shows the aviation product quality report verification method based on a pre-trained language model according to the present invention.

[0040] Figure 2 This is a schematic diagram of the aviation product quality report verification system of the present invention;

[0041] Figure 3 This is a structural diagram of the Transformer model of the present invention;

[0042] Figure 4 This is a schematic diagram of the interface of the aviation product quality report verification system of the present invention. Detailed Implementation

[0043] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings.

[0044] This invention provides a method for verifying aviation product quality reports based on a pre-trained language model, such as... Figure 1 As shown, the process involves: acquiring the zero-quality report text of aviation products; extracting feature lengths; performing merging and segmentation operations; extracting key information feature text from the aviation product quality zero-quality report text and generating inspection prompts; performing contextual location of paragraphs in the aviation product quality zero-quality report text; and using a large-scale language model to verify the aviation product quality zero-quality report text. This includes:

[0045] Step S1: Obtain the zeroing report text of aviation product quality, extract the feature length, and perform merging and block operations.

[0046] In a preferred embodiment of the present invention, the zero-out report text for aviation product quality includes five modules: problem location, mechanism analysis, problem reproduction, measures taken, and lessons learned. Each module consists of several hundred to several thousand words, with the longest possibly reaching several thousand words. Moreover, in a specific application embodiment, for example, the zero-out report text for the technical zero-out report of hydraulic plunger pump pressure swing problem contains a total of 8,363 characters.

[0047] Step S11: Determine the quality zeroing report text and the maximum input text length; set the quality zeroing report text D to contain n text paragraphs {P1,…P… i …,P n}, and the maximum input text length of the model is L. maxLimit the text block size. In this example, the zeroing report text contains 190 paragraphs, and L is set to... max The value is 2000, which is greater than the length of all paragraphs.

[0048] Step S12: Calculate P for each text segment using a language model. i The characteristic length len(P) i )for:

[0049] len(P i = len(Tokenizer(P) i ));

[0050] Where len(P) i P is the i-th text segment. i Characteristic length; P i The i-th text segment in the quality zeroing report text; Tokenizer() is the text character splitting rule function; i is the text segment number, i∈1,…,n; n is the total number of text segments.

[0051] The text character splitting rule function splits Chinese text according to Chinese characters, while English text is split according to English words. In this example, the feature length of all paragraphs is calculated, with the longest being 256.

[0052] Step S13: Determine the adjacent text segments P based on the characteristic length len() of the text segment. i ,P i+1 Whether the merger conditions are met, specifically:

[0053]

[0054] Where SouldMerge is the function to determine if the merging condition is met; True indicates that the merging condition is met; False indicates that the merging condition is not met; L max This represents the maximum length of the input text.

[0055] For text paragraphs that meet the merging criteria, perform the merge operation P'. i =P i ∪P i+1 If SouldMerge(P) i ,P i+1 When ) = True, use P' i To replace P i and P i+1 Then re-evaluate P' i and P i+2 Does the merging condition meet? Merge the blocks to obtain the Quality Zero Report text block set:

[0056] B = {B1,…,B} j ,…,B m};

[0057] Where B is the set of text blocks for the quality zeroing report; B j Let j be the j-th quality zeroing report text block; j is the quality zeroing report text block number; m is the total number of quality zeroing report text blocks. In this embodiment, the zeroing report paragraph texts are merged using this method to form 5 text blocks, among which the one with the largest feature length is 1955.

[0058] Step S2: Extract key information feature text from the aviation product quality zeroing report text and generate inspection prompts.

[0059] Step S21: Obtain the inspection rule set R = {R1, ..., R2} of the quality zeroing report text. i ,…,R n In the embodiment, the verification rule set R is:

[0060]

[0061] Analyze the inspection rules of the quality zeroing report text, and for each inspection rule R i By performing constituent syntactic analysis, noun phrases and verb phrases, as well as key information feature text, can be extracted from the rules; and a syntactic structure tree S for verifying the rules can be obtained. i for:

[0062] S i =ConstituencyParse(R i i∈{1,2,…,n};

[0063] Among them, S i To test rule R i The syntax tree structure; ConstituencyParse is the constituent parsing function; R i Let be the i-th verification rule.

[0064] In the embodiment, the syntax tree structure S1 is the result of constituent syntactic analysis after testing rule R1, and the result is as follows:

[0065] (ROOT

[0066] (CP

[0067] (PP(P to)(NP(QP(DNP(NN quality problem)(DEG of))(NN description))))

[0068] (IP

[0069] (VP

[0070] (ADVP(AD whether))

[0071] (VP

[0072] (VP(ADVP(AD explicit))(VP(VV explanation)))

[0073] (NP

[0074] (NP(NN occurs))

[0075] (CP(IP(NP(PN))(NP(NN time)(PU、)(NN location)(PU、)(NN timing))))))))

[0076] (PU?)))

[0077] Perform the same operation on inspection rule R2 and inspection rule R3.

[0078] Step S22: Identify the key information features of the quality zeroing report text; analyze the syntactic structure tree S obtained in step S21. i Identify the noun phrases (NP) and verb phrases (VP) within the text, concatenate them into a key information feature text, and then apply this to the validation rule R. i Key information feature text I i For the syntax structure tree S i The concatenation of noun phrases (NP) and verb phrases (VP) constructs a key information feature text set I = {I1, ..., I...} i ,…,I n There are n information items in total.

[0079] In the embodiment, the key information item I1 obtained by verifying rule R1 and syntax tree structure S1 is "quality problem description, time, location, and timing". Similarly, I2 and I3 can be obtained as "problematic product, structure, function, and performance" and "fault tree analysis", respectively.

[0080] Step S23: Analyze the test rule set R = {R1, R2, ..., R...} n After entering the prompt template T sequentially, the following inspection prompt is obtained:

[0081]

[0082] in, To test rule R i The verification prompts are as follows: T is the prompt template, which determines whether the JSON format of the prompt file meets the requirements based on the content above; FillTemplate is the verification prompt function.

[0083] The set of verification prompts is constructed as follows:

[0084]

[0085] Where P is the test hint set.

[0086] Step S3: Contextualize paragraphs in the zero-quality report text for aviation products.

[0087] Step S31: Obtain the quality zeroing report text block set B from step S1 and the key information feature text set I from step S2, and use them as input sequences to process [CLS]I using a ranking model based on the Transformer architecture. j [SEP]B i [SEP]; such as Figure 3 The diagram shows the Transformer model structure of this invention. The ranking model of the Transformer architecture captures the relationship between the query and the candidates through a self-attention mechanism, thereby outputting the similarity. The similarity function Sim(B) is used. j ,I i To calculate the feature text I of each key information. i With each quality zeroing report text block B j The degree of compatibility.

[0088] Step S32: Sort the text blocks based on the fit obtained in step S31, sorting them from largest to smallest fit, to obtain the sorted set of text blocks for the quality zeroing report:

[0089]

[0090] Among them, ranked(I i B) is the sorted set of text blocks for the quality zeroing report; {j1,j2,…,j m Let {1,2,…,m} be a sequence such that satisfying This holds true for all p = 1, 2, ..., m-1.

[0091] Step S33: Output the context paragraph location of the zero-quality report text for aviation products; from the sorted quality zero-quality report text block set ranked(I) obtained in step S31 i B), Select the first k text blocks. As candidate context paragraphs, text merging is performed to locate the context text of the aviation product quality zeroing report:

[0092]

[0093] Among them, C jThe context text for the zeroing-out report of aviation product quality; combine is the text merging operation function; k is the text block value parameter, with a default value of 1 or 2.

[0094] Step S4: Use a large-scale language model to verify the zeroing report text for aviation product quality.

[0095] The test rule R in step S23 is respectively j Corresponding test prompt set The context text C of the zeroing report text of aviation product quality obtained in step S33 j The data is concatenated and input into a large-scale language model, with JSON format set as the output format for the test results. The output format is JSON and includes: test items, test results, and test criteria. The test items are the test rules R. j The test results are categorized into three types: compliant, non-compliant, and inapplicable; the test basis is the judgment reason. The prompt specifies that the large model should be output in JSON format in the following sub-format:

[0096] {"Validation Item":"R"} j ", / / Validation rule R j

[0097] "Validation Result": "Compliant" | "Disagree" | "Not Applicable", / / Validation Result

[0098] "Basis": "Reason for Judgment" / / Basis for Judgment

[0099] R1, R2, ..., R n and its corresponding By inputting the data into the large-scale language model in this manner, in this example, the final model output will be in the following format:

[0100] [{“Verification Item”: “Does the description of the quality problem clearly state the time, place, and timing of its occurrence?”, “Verification Result”: “Non-compliant”, “Basis”: “The zero-based report does not clearly state the time, place, timing, phenomenon (consequence), and environmental conditions of the occurrence. The report mainly focuses on the mechanism analysis, fault tree analysis, and corrective measures of the problem, but does not describe in detail the specific time and place of occurrence.”};

[0101] {“Verification Item”: “Does it briefly describe the structure, working principle, function, and performance of the problematic product?”, “Verification Result”: “Does not meet the requirements”, “Basis”: “The zeroing report mainly focuses on fault analysis, location, and mechanism analysis, and does not provide a detailed description of the structure, working principle, function, and performance of the problematic product.”};

[0102] {“Verification Item”: “Has a fault tree analysis been established?”, “Verification Result”: “Compliant”, “Basis”: “Based on the provided zeroing report, a fault tree analysis was indeed established. The report clearly mentions the fault tree with the rotor assembly fatigue crack as the top event, and lists each branch and factor in detail.”}

[0103] The output inspection results of the zero-tolerance reports for aviation product quality are integrated and summarized, and the judgment results of aviation product quality are displayed in tabular form. The verification results of the zero-tolerance reports are integrated and summarized so that management can effectively analyze the compliance status of each verification item, identify common problems, and identify areas for improvement. The output of this step is a unified summary table, showing the verification results of each report, as shown in Table 1, including the compliance of each verification item, whether it meets the requirements, and the basis for its judgment.

[0104] Table 1: Results of Quality Assessment for Aviation Products

[0105]

[0106] Next, the JSON format output from S4 is parsed, and the corresponding items are filled into a table. For different reports, this process is repeated, filling in the report number, verification items, verification results, and basis in sequence. Finally, the table is saved and output in Excel format, forming the system's final output.

[0107] The second aspect of this invention proposes an aviation product quality report verification system based on a pre-trained language model. Relying on a large language model, it achieves automated verification, recognition, and information extraction of zero-quality reports, significantly improving the verification efficiency and accuracy of zero-quality report text. The system includes: a zero-quality report text segmentation module, a key information and verification prompt module, a text paragraph context positioning module, a zero-quality report verification module, and a verification result integration and output module. The overall system flow is as follows: Figure 2 As shown.

[0108] The zeroing report text segmentation module divides long documents into blocks based on paragraphs and logic, ensuring that each block is logically complete and adapts to the input length limits of large-scale language models.

[0109] The key information and verification prompt module sets the key information that must be verified in the zeroing report and its compliance standards, and classifies each item to determine its applicability during the verification process.

[0110] The text paragraph context localization module uses a rearrangement model to reorder the segmented text blocks, ensuring the optimal order of task processing and contextual coherence.

[0111] The zeroing report verification module utilizes a large-scale language model. By inputting the report context and verification prompts into the large-scale language model, it checks the compliance of the zeroing report, records non-compliance and inapplicable items, and provides the basis for judgment to ensure the completeness and compliance of the report.

[0112] The verification result integration and output module summarizes the verification results of each zeroing report and designs a unified summary table to facilitate management in analyzing the compliance status of each verification item and identifying common problems and areas for improvement.

[0113] like Figure 4 The diagram shows the interface of the aviation product quality report verification system of the present invention. Users can select and upload report files in PDF / DOCX format using the "Browse" and "Upload" buttons. The system will display the document content in the "Content Preview" area. After clicking "Submit Verification," the system verifies the report content item by item according to preset standards and displays the verification results and basis in the table below. Users can also save the verification results as a document using the "Export Results" button. In this embodiment, the application of the aviation product quality report verification system verifies that the system obtained by the present invention has good practicality and improves the verification efficiency and accuracy of zero-quality report texts.

[0114] The beneficial effects of this invention are as follows: This invention provides a method for verifying aviation product quality reports based on pre-trained language models. It uses a large-scale pre-trained language model to analyze aviation product quality problem zeroing reports, improves the accuracy of extracting key information from the text in the reports, and realizes automatic verification of the report content to ensure its integrity and logical consistency. Based on the quality zeroing reports, it identifies common problems in aviation product quality control and optimizes improvement schemes, providing reliable data support for the improvement, optimization, and standardization of aviation product quality zeroing reports.

[0115] The embodiments described above are merely 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 verifying aviation product quality reports based on a pre-trained language model, characterized in that: S1: Obtain the quality zeroing report text and use a language model to calculate the quality of each text segment. Feature length Determine adjacent text paragraphs If the merging criteria are met, merge the data to obtain the quality zeroing report text block set. ; S2: Parse the verification rules of the quality zeroing report text, perform constituent syntactic analysis, and obtain the syntactic structure tree of the verification rules. Identify noun phrases (NPs) and verb phrases (VPs) within the text, concatenate them to form key information feature text, and construct a key information feature text set. , will verify the rule set Enter the prompt template in sequence , obtain the set of verification prompts ; Step S2 is as follows: S21: Obtain the inspection rule set for the quality zeroing report text The analysis process involves analyzing the validation rules of the quality zeroing report text, performing constituent syntactic analysis, extracting noun phrases and verb phrases, as well as key information feature text, to obtain the syntactic structure tree of the validation rules. ; S22: Identify key information features in the zero-quality report text; In step S21 of identification The noun phrases (NP) and verb phrases (VP) are concatenated to form the key information feature text, thus constructing a key information feature text set. ; S23: Verify the rule set Enter the prompt template in sequence , obtain the set of verification prompts ; S3: Obtain the quality zeroing report text block set from step S1. and the key information feature text set in step S2 Input a ranking model based on the Transformer architecture and use a similarity function. Calculate the feature text of each key information With each quality zeroing report text block The fit is determined; the text blocks are sorted according to the fit to obtain the sorted quality zeroing report text block set; the previous selection is performed. text blocks As candidate context paragraphs, text merging is performed to obtain the context text of the quality zeroing report text. ; S4: Verify the rules in step S2 Corresponding test prompt set The context text of the quality zeroing report text obtained in step S3 The large-scale language model is concatenated and input, and the JSON format is set as the output format for the verification results to obtain the quality zeroing report verification results, which are used to improve the quality zeroing report.

2. The method for verifying aviation product quality reports based on pre-trained language models according to claim 1, characterized in that: Step S3 is as follows: S31: Obtain the quality zeroing report text block set from step S1 and the key information feature text set in step S2 Using similarity functions To calculate key information feature text With the quality zeroing report text block The degree of compatibility; S32: Sort the text blocks based on the fit obtained in step S31, sorting them from largest to smallest fit, to obtain the sorted quality zeroing report text block set. ; S33: Obtain the sorted quality zeroing report text block set from step S31, and select the first... text blocks As candidate context paragraphs, text merging is performed to locate the context text of the aviation product quality zeroing report. .

3. The method for verifying aviation product quality reports based on pre-trained language models according to claim 1, characterized in that: Step S23 prompt template For: A standard JSON-formatted prompt file that determines whether the content above meets the requirements.

4. The method for verifying aviation product quality reports based on pre-trained language models according to claim 2, characterized in that: In step S32, the text blocks are sorted using a sorting model based on the Transformer architecture. The self-attention mechanism is used to capture the relationship between the query and the candidates, thereby outputting the similarity.

5. The method for verifying aviation product quality reports based on pre-trained language models according to claim 1, characterized in that: Step S1 is as follows: S11: Determine the quality zeroing report text and maximum input text length; Set the quality zeroing report text. Include A text paragraph And the maximum input text length of the model is ; S12: Perform text segmentation and use a language model to calculate the value of each text segment. Feature length ; S13: Based on the characteristic length of the text paragraph Determine adjacent text paragraphs Does the merger condition meet? S14: For text paragraphs that meet the merging criteria, perform the merge operation. ,use to replace and Then re-evaluate and Does the merging condition meet? Merge the data to obtain the quality zeroing report text block set. .

6. The method for verifying aviation product quality reports based on pre-trained language models according to claim 1, characterized in that: The JSON format in step S4 specifically includes: inspection items, inspection results, and inspection basis; the inspection items are the inspection rules. The test results are categorized into three types: compliant, non-compliant, and inapplicable; the basis for the test is the reasoning behind the judgment.

7. The method for verifying aviation product quality reports based on pre-trained language models according to claim 5, characterized in that: Step S12 involves splitting text paragraphs using a text character splitting rule function. For Chinese text, split the text according to Chinese characters; for English text, split it according to English words.

8. A verification system for aviation product quality reports based on a pre-trained language model as described in any one of claims 1 to 7, characterized in that, It includes: The module includes a zeroing report text segmentation module, a key information and verification prompt module, a text paragraph context positioning module, a zeroing report verification module, and a verification result integration and output module. The zeroing report text segmentation module divides long documents into blocks based on paragraphs and logic, with each text block adapting to the input length limit of a large-scale language model; The key information and verification prompt module sets the key information to be verified in the zeroing report and its compliance standards, and classifies each item to determine its applicability during the verification process; The text paragraph context localization module uses a rearrangement model to sort the segmented text blocks to achieve optimal order and contextual coherence; The zeroing report verification module takes the zeroing report context and verification prompts as input into a large-scale language model, checks the zeroing report, records non-compliant and inapplicable items, and provides the basis for judgment to realize the compliance verification of the zeroing report; The verification result integration and output module summarizes the verification results of each zeroing report and forms a unified summary table file.

Images