Industrial measurement data intelligent analysis report generation method and system based on large language model

Abstract

The invention discloses an industrial measurement data intelligent analysis report generation method and system based on a large language model, and relates to the technical field of industrial measurement, and the method comprises a multi-source data and abstraction module which carries out the structural processing of original data DAT output by measurement analysis software; and the large language model service and interface module is used for submitting the PCT to a selected large language model LLM through a multi-model adaptation interface to execute semantic reasoning and output a structured text TXT. Structured analysis and semantic association modeling are carried out on original measurement data through multi-source data and an abstract module, when overall alignment deviation or local feature anomaly exists in the original data, a hierarchical dependency relationship between the data can be established through a semantic graph structure, a deviation transmission path is revealed, and the accuracy of the measurement data is improved. The large language model is analyzed in a unified data context, and the utilization depth of measurement data can be effectively improved, so that the accuracy and integrity of report analysis are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial measurement, and specifically provides an intelligent analysis report generation method and system for industrial measurement data based on a large language model. Background Art

[0002] Industrial measurement is a key quality control link in the industrial manufacturing process, and its core lies in accurately obtaining and evaluating the geometric dimensions, form and position tolerances, and assembly relationships of products. The actual measurement data of products is obtained through coordinate measuring equipment, optical scanning equipment, or laser point cloud acquisition equipment, and alignment processing, feature extraction processing, and deviation calculation processing are performed with the theoretical model to determine whether the product meets the design tolerance requirements. Industrial measurement is not only used for the qualification determination of single-piece products, but also widely used in process verification, equipment accuracy evaluation, and batch quality stability analysis, and is an important technical foundation for realizing precision manufacturing and intelligent manufacturing. With the continuous improvement of manufacturing precision requirements, industrial measurement data presents multi-dimensional, high-precision, and large-scale characteristics, posing higher requirements for data processing capabilities and analysis intelligence levels.

[0003] Existing industrial measurement methods mostly generate reports by filling fixed templates, only listing and simply judging measurement values, lacking the ability to perform hierarchical evaluation based on data distribution characteristics, unable to classify and analyze deviation patterns, and also unable to infer potential causes based on measurement results. Summary of the Invention

[0004] Aiming at the deficiencies of the prior art, the present invention provides an intelligent analysis report generation method and system for industrial measurement data based on a large language model.

[0005] To achieve the above objectives, the present invention is realized through the following technical solutions: An intelligent analysis report generation method and system for industrial measurement data based on a large language model, including: A multi-source data and abstraction module that structurally processes the original data DAT output by measurement analysis software, including: extracting fields, performing type conversion, and reconstructing the structure on the original data DAT through a templated data parser to generate a data object JON, and performing node extraction and relationship construction on the data object JON through a measurement semantic graph construction unit to generate a semantic graph structure SUB; A domain knowledge management and prompt engineering module that establishes a prompt template library, including role identity templates, analysis process templates, domain rule templates, and output structure templates, and filters and combines the prompt templates, data object JON, and semantic graph structure SUB according to the user-configured report to generate a prompt context PCT; The large language model service and interface module submits the prompt context PCT to the selected large language model LLM through the multi-model adaptation interface to perform semantic reasoning and output structured text TXT. The output parsing post-processor performs consistency verification on the structured text TXT. The multimodal content automatic synthesis and typesetting module is used to perform layout mapping on the structured text TXT and data table TAB to generate report documents DOC; The human-computer interaction module and feedback optimization module are used to provide an interactive report editing interface and a feedback learning interface.

[0006] Preferably, the multi-source data and abstraction module includes a templated data parser and a measurement semantic graph construction unit. The templated data parser performs data format determination on the original data DAT, extracts fields from the alignment matrix MAT, feature deviation data DEV, and point cloud statistical data PCD in the original data DAT according to a preset field mapping rule MAP; performs data type conversion, unit unification, and precision normalization on the extracted field data, reconstructs the field hierarchy according to the structure transformation rule TRN; and encapsulates the processed field data according to a unified data structure to generate a data object JON. The measurement semantic graph construction unit performs node extraction processing on the field data in the data object JON to generate a node set NOD; it performs relation abstraction processing on the alignment relation data, benchmark reference data and deviation data in the data object JON to generate a relation edge set REL; and it establishes a hierarchical dependency structure based on the node set NOD and the relation edge set REL to generate a semantic graph structure SUB.

[0007] Preferably, the domain knowledge management and prompting engineering module includes a prompt template management unit and a context dynamic assembler; The prompt template management unit is used to build an extensible prompt word library, including: Role constraint templates are used to restrict the output identity of large language models, including: injecting preset identity constraint text into the prompt context to limit the boundaries of the model's analytical perspective, the scope of use of professional terms, and the criteria for judging problems; The reasoning process template is used to construct a fixed sequence of analysis steps, set phased reasoning order control rules for input data, and generate overall evaluation results, anomaly location results, statistical analysis results, cause inference results, and improvement suggestions in sequence according to the preset step sequence. Domain rule templates are used to constrain the knowledge boundaries of the content generated by the model, including: establishing a set of rule mappings between quality problems and potential causes and a set of industry terminology mappings; The structure control template is used to format the chapter structure and presentation of the generated text, including preset chapter title sets, paragraph hierarchy rules, keyword emphasis rules, and tag output rules;

[0008] The prompt template management unit establishes a template category index and an applicable scenario index; it also establishes parent-child reference relationships and calling relationships between templates according to hierarchical nesting rules, forming a template structure set. The context dynamic assembler parses the user configuration report to generate a set of report control parameters; performs rule matching and priority sorting on the template structure set according to the set of report control parameters to determine the target template combination sequence; performs condition filtering and field screening on the data object JON and semantic graph structure SUB to generate a data subset; performs similarity matching and content extraction on the historical case library to generate a supplementary text set; and arranges the target template combination sequence, data subset, and supplementary text set in order according to preset splicing rules to generate a prompt context.

[0009] Preferably, the large language model service module includes a multi-model adaptation interface and an output parsing post-processor; The multi-model adaptation interface will submit the context prompt to the selected large language model; the multi-model adaptation interface will perform role identity constraint processing and domain framework constraint processing; The role identity constraint processing includes: establishing a role identity database, configuring identity identifier parameters, professional ability tags, and terminology usage scope parameters for different roles; identifying and generating roles based on the current task type after receiving the prompt context; and extracting the corresponding identity definition text from the role identity database and inserting it into the control instruction section of the prompt context. The domain framework constraint processing includes: establishing a set of analysis step rules; generating structured guidance text from the set of analysis step rules in a preset order; and inserting the structured guidance text into a specified logical segment after identity enhancement.

[0010] The multi-model adaptation interface performs structural identifier parsing on the prompt context, identifies the analysis step identifier information therein, and generates a stage index set; based on the stage index set, the prompt context is divided into data interpretation stage prompt content, cause inference stage prompt content, and action suggestion stage prompt content, and corresponding stage prompt sub-contexts are generated respectively; For each stage's prompt sub-context, the corresponding inference control parameter set is loaded and a stage call message is constructed. The inference control parameter set includes randomness control parameters, candidate range control parameters, and repetition suppression parameters. The stage call messages are sent to the large language model in a preset stage call order to obtain the output results of each stage. The output results of each stage are concatenated in the order of the stages to generate a structured text TXT.

[0011] Preferably, the output parsing post-processor performs consistency verification processing on the structured text TXT, including: performing field recognition and structure extraction processing on the structured text TXT to generate a set of structured fields; comparing the set of structured fields with the original measurement data item by item, performing field integrity verification, numerical range verification, unit consistency verification, and logical relationship verification according to preset verification rule parameters, and generating field parsing results; judging the field parsing results according to a confidence threshold parameter, performing a regular verification process when the field confidence is greater than or equal to the threshold, and directly marking the field as an anomaly when the field confidence is lower than the threshold.

[0012] Preferably, the multimodal content automatic synthesis and typesetting module includes: an intelligent typesetting engine and a multi-format rendering and export unit; The intelligent typesetting engine performs syntactic structure parsing on the structured text TXT, generates a chapter tree structure according to preset chapter hierarchy rules, performs field filtering and data reorganization processing on the measurement result fields in the standardized data object, converts the numerical fields into tabular data structures, and performs attribute marking processing on abnormal data fields according to preset threshold rules. Load the set of typesetting template rules and layout constraint rules, perform position matching and sequence arrangement processing on the chapter tree structure and table data structure, and generate a report document DOC; The multi-format rendering and export unit is used to perform target format encapsulation processing on the report document DOC, specifically including: The target format identifier is determined based on the output type specified by the user; the corresponding format rendering engine is called to perform format mapping and style conversion processing on the report document DOC, and then further encapsulated into the target format file FIL.

[0013] This invention also provides a method for generating intelligent analysis reports of industrial measurement data based on a large language model, including: S1. Obtain the raw data DAT output by the measurement software, convert the raw data DAT into a data structure JON through a templated data parser, and extract nodes and construct relationships for the data object JON through a measurement semantic graph construction unit to generate a semantic graph structure SUB. S2. Based on the user configuration report set by the user, filter and combine the prompt template, the data object JON, and the semantic graph structure SUB to generate the prompt context PCT; S3. The prompt context PCT is submitted to the large language model LLM through the role preset injection module ROLN to perform semantic reasoning and output the structured text TXT; S4. Perform consistency verification on the structured text TXT through the output parsing post-processor. S5. Perform layout mapping on the structured text TXT and data table TAB to generate a report document DOC, and further encapsulate it into a target format file FIL.

[0014] Compared with the prior art, the present invention has the following advantages: This invention performs structured analysis and semantic association modeling of raw measurement data through multi-source data and abstract modules. When the raw data has overall alignment deviation or local feature anomalies, it can establish hierarchical dependencies between data through semantic graph structure and reveal the deviation propagation path. This allows the large language model to be analyzed in a unified data context, which can effectively improve the depth of measurement data utilization, thereby improving the accuracy and completeness of report analysis and enhancing the support capability for production decision-making.

[0015] This invention uses a domain knowledge management and prompting engineering module to control the model input through role and process constraints. Combined with a phased reasoning and consistency verification mechanism, it verifies the generated text. When there are conflicts or logical deviations in the model output, the output parsing post-processor performs item-by-item comparison and anomaly marking, thereby ensuring that the generated report content is consistent with the original measurement data. The reasoning control and verification closed-loop mechanism can effectively improve the professionalism and reliability of the report and reduce the cost of manual review. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the system architecture of the present invention; Figure 2 This is a schematic diagram of the method flow of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Please see Figure 1 This application provides a method and system for generating intelligent analysis reports of industrial measurement data based on a large language model, including: The multi-source data and abstraction module performs structured processing on the raw data DAT output by the measurement and analysis software, including: performing field extraction, type conversion and structure reconstruction on the raw data DAT through a templated data parser to generate a data object JON; and performing node extraction and relationship construction on the data object JON through a measurement semantic graph construction unit to generate a semantic graph structure SUB. The domain knowledge management and prompting engineering module establishes a prompt template library, including role identity templates, analysis process templates, domain rule templates, and output structure templates. Based on the user configuration report, it filters and combines prompt templates, data objects JON, and semantic graph structures SUB to generate prompt contexts PCT. The large language model service and interface module submits the prompt context PCT to the selected large language model LLM through the multi-model adaptation interface to perform semantic reasoning and output structured text TXT. The output parsing post-processor performs consistency verification on the structured text TXT. The multimodal content automatic synthesis and typesetting module is used to perform layout mapping on the structured text TXT and data table TAB to generate report documents DOC; The human-computer interaction module and feedback optimization module are used to provide an interactive report editing interface and a feedback learning interface.

[0019] Specifically, users can evaluate or correct the report content through a pop-up feedback panel. The feedback panel provides a problem type selection control, allowing users to choose a corresponding tag from a preset problem type set, including categories such as incorrect cause inference, suggestions that do not conform to actual production conditions, and inaccurate data interpretation. Simultaneously, the feedback panel provides a text input area where users can enter corrections to the current report content, describing correct cause judgments, reasonable improvement suggestions, or correct data interpretation methods.

[0020] After a user submits feedback, the system listens for and captures the user's actions, and performs structured parsing processing on the feedback, converting the user's feedback content into a structured feedback instance. The structured feedback instance includes a user-corrected content field and a problem type tag field. The user-corrected content field records the user's input of corrective text, and the problem type tag field identifies the problem category to which the feedback information belongs.

[0021] The system performs text pattern analysis on the feedback instances, identifying semantic patterns, rule expressions, or causal relationship descriptions in the user's corrections to extract reusable knowledge rules or example expressions and generate a set of rule candidates. This set of rule candidates is then submitted to the prompting engineering module for review and integration, used to update existing domain rule templates.

[0022] The multi-source data and abstraction module includes a templated data parser and a measurement semantic graph construction unit. The templated data parser performs data format determination on the original data DAT, and extracts fields from the alignment matrix MAT, feature deviation data DEV, and point cloud statistical data PCD in the original data DAT according to a preset field mapping rule MAP. It then performs data type conversion, unit unification, and precision normalization on the extracted field data, reconstructs the field hierarchy according to the structure transformation rule TRN, and encapsulates the processed field data according to a unified data structure to generate a data object JON.

[0023] The measurement semantic graph construction unit performs node extraction processing on the field data in the data object JON to generate a node set NOD; it performs relation abstraction processing on the alignment relation data, benchmark reference data and deviation data in the data object JON to generate a relation edge set REL; and it establishes a hierarchical dependency structure based on the node set NOD and the relation edge set REL to generate a semantic graph structure SUB.

[0024] Specifically, after the templated data parsing unit converts the raw data into data objects, the measurement semantic graph construction unit performs node extraction processing on the field data in the data objects. The node extraction process first identifies the types of the fields in the data objects. Based on the measurement element categories represented by the fields, the alignment matrix data, feature measurement data, and deviation data are mapped to different types of data nodes, and a unique node identifier is assigned to each node, thereby generating a node set.

[0025] Subsequently, the measurement semantic graph construction unit performs relational abstraction processing on the relational information in the data object. Specifically, this includes: establishing alignment relationships between reference coordinates based on alignment matrix data, establishing reference relationships between features and references based on reference reference fields, establishing deviation correlation relationships between measurement results and theoretical models based on deviation data, and abstracting the above relationships into a relational edge structure to generate a relational edge set.

[0026] After completing the construction of the node set and the relation edge set, the system organizes the nodes and relation edges according to the preset hierarchical dependency rules to establish a hierarchical dependency structure between the measurement data.

[0027] In a preferred embodiment of the present invention, the domain knowledge management and prompting engineering module includes a prompt template management unit and a context dynamic assembler; The prompt template management unit is used to build an extensible prompt word library, including: Role constraint templates are used to restrict the output identity of large language models, including: injecting preset identity constraint text into the prompt context to limit the boundaries of the model's analytical perspective, the scope of use of professional terms, and the criteria for judging problems; The reasoning process template is used to construct a fixed sequence of analysis steps, set phased reasoning order control rules for input data, and generate overall evaluation results, anomaly location results, statistical analysis results, cause inference results, and improvement suggestions in sequence according to the preset step sequence. Domain rule templates are used to constrain the knowledge boundaries of the content generated by the model, including: establishing a set of rule mappings between quality problems and potential causes and a set of industry terminology mappings; The structure control template is used to format the chapter structure and presentation of the generated text, including preset chapter title sets, paragraph hierarchy rules, keyword emphasis rules, and tag output rules.

[0028] Specifically, here's an example of a prompt word template: system_prompt = """ You are an experienced quality engineer in the aerospace manufacturing industry, skilled at identifying process problems, speculating on root causes, and proposing feasible recommendations from measurement data. Please strictly adhere to the following requirements when analyzing industrial measurement data and output the results in Chinese: 1. Analytical logic: start with the whole and then move to the parts, start with the data and then move to the reasons, make inferences based on the deviation distribution pattern and domain knowledge, and avoid unfounded guesses; 2. Output format: Must use Markdown, and must contain only the following sections; the order of the sections cannot be changed: (1). Overview of the testing task (Briefly describe the object being tested, the test items, the alignment method, and whether the alignment error is acceptable.) (2). Overall conformity conclusion (Clearly define the overall compliance status, and mark the number of warning / non-compliance features and key deviations) (3). Key findings and detailed analysis (Sorted by importance, analyze the distribution pattern of deviations, the degree of exceeding the standard, and the correlation of data) (4). Inference of potential causes (if possible) (Based on domain knowledge, speculate on the root cause of the bias and explain the basis for the inference.) (5). Recommendations and follow-up measures (Propose specific and actionable suggestions, listed in points) (6) Professional requirements: The terminology should be accurate (such as “profile degree”, “RMS error”, “tolerance range”), and the expression should be rigorous, avoiding vague expressions (such as “probably”).

[0029] [Measurement data begins] Project Name: {project_name} Alignment method: {alignment_method} Alignment RMS error: {rms_error} mm (Requirement: {error_requirement}) The feature measurement results are as follows (unit: mm): {features_table_in_markdown} Point cloud overall deviation statistics: maximum value = {max_dev} mm, minimum value = {min_dev} mm, standard deviation = {std_dev} mm, and the proportion of points exceeding the standard deviation = {over_tolerance_ratio}.

[0030] [End of measurement data] Domain knowledge reference: {domain_knowledge} (e.g., "a slight bulge in the middle of the blade back may be related to deformation of the clamp support") Please complete the analysis based on the above information, paying particular attention to the deviation distribution pattern and warning / non-compliance characteristics.

[0031] In a preferred embodiment of the present invention, the prompt template management unit establishes a template category index and an applicable scenario index; and establishes parent-child reference relationships and calling relationships between templates according to hierarchical nesting rules to form a template structure set; The context dynamic assembler parses the user configuration report to generate a set of report control parameters; performs rule matching and priority sorting on the template structure set according to the set of report control parameters to determine the target template combination sequence; performs condition filtering and field screening on the data object JON and semantic graph structure SUB to generate a data subset; performs similarity matching and content extraction on the historical case library to generate a supplementary text set; and arranges the target template combination sequence, data subset, and supplementary text set in order according to preset splicing rules to generate a prompt context.

[0032] Specifically, the large language model service module includes a multi-model adaptation interface and an output parsing post-processor; The multi-model adaptation interface will submit the context prompt to the selected large language model; the multi-model adaptation interface will perform role identity constraint processing and domain framework constraint processing; The role identity constraint processing includes: establishing a role identity database, configuring identity identifier parameters, professional ability tags, and terminology usage scope parameters for different roles; identifying and generating roles based on the current task type after receiving the prompt context; and extracting the corresponding identity definition text from the role identity database and inserting it into the control instruction section of the prompt context. The domain framework constraint processing includes: establishing a set of analysis step rules; generating structured guidance text from the set of analysis step rules in a preset order; and inserting the structured guidance text into a specified logical segment after identity enhancement.

[0033] Specifically, the multi-model adaptation interface performs structural identifier parsing on the prompt context, identifies the analysis step identifier information therein, and generates a stage index set; based on the stage index set, the prompt context is divided into data interpretation stage prompt content, cause inference stage prompt content, and action suggestion stage prompt content, and corresponding stage prompt sub-contexts are generated respectively; For each stage's prompt sub-context, the corresponding inference control parameter set is loaded and a stage call message is constructed. The inference control parameter set includes randomness control parameters, candidate range control parameters, and repetition suppression parameters. The stage call messages are sent to the large language model in a preset stage call order to obtain the output results of each stage.

[0034] The output results of each stage are concatenated in the order of the stages to generate a structured text TXT.

[0035] In a preferred embodiment of the present invention, the output parsing post-processor performs consistency verification processing on the structured text TXT, including: performing field recognition and structure extraction processing on the structured text TXT to generate a set of structured fields; comparing the set of structured fields with the original measurement data item by item, performing field integrity verification, numerical range verification, unit consistency verification, and logical relationship verification according to preset verification rule parameters, and generating field parsing results; judging the field parsing results according to a confidence threshold parameter, performing a regular verification process when the field confidence is greater than or equal to the threshold, and directly marking the field as an anomaly when the field confidence is lower than the threshold.

[0036] Specifically, the multimodal content automatic synthesis and typesetting module includes an intelligent typesetting engine and a multi-format rendering and export unit; The intelligent typesetting engine performs syntactic structure parsing on the structured text TXT, generates a chapter tree structure according to preset chapter hierarchy rules, performs field filtering and data reorganization processing on the measurement result fields in the standardized data object, converts the numerical fields into tabular data structures, and performs attribute marking processing on abnormal data fields according to preset threshold rules. Specifically, the intelligent typesetting engine follows the preset "Quality Inspection Report Format Specification" to achieve dynamic layout planning (automatic pagination based on content length), automatic element arrangement (inserting text, data tables, statistical charts, and screenshots in logical order), and unified style management (automatically applying title styles, fonts, colors, headers, and footers).

[0037] Load the set of typesetting template rules and layout constraint rules, perform position matching and sequence arrangement processing on the chapter tree structure and table data structure, and generate a report document DOC; The multi-format rendering and export unit is used to perform target format encapsulation processing on the report document DOC, including: The target format identifier is determined based on the output type specified by the user; the corresponding format rendering engine is called to perform format mapping and style conversion processing on the report document DOC, and then further encapsulated into the target format file FIL.

[0038] Specifically, the multi-format rendering and export unit drives different renderers to generate files in specified formats based on the complete report document object in memory. It supports Word / PowerPoint (editable), PDF (fixed format), Excel / CSV (structured data) and DMO / XML (industrial data exchange format) to meet the needs of different application scenarios.

[0039] like Figure 2 As shown, the present invention provides a method for generating intelligent analysis reports of industrial measurement data based on a large language model, the steps of which include: S1. Obtain the raw data DAT output by the measurement software, convert the raw data DAT into a data structure JON through a templated data parser, and extract nodes and construct relationships for the data object JON through a measurement semantic graph construction unit to generate a semantic graph structure SUB. S2. Based on the user configuration report set by the user, filter and combine the prompt template, the data object JON, and the semantic graph structure SUB to generate the prompt context PCT; S3. The prompt context PCT is submitted to the large language model LLM through the role preset injection module ROLN to perform semantic reasoning and output the structured text TXT; S4. Perform consistency verification on the structured text TXT through the output parsing post-processor. S5. Perform layout mapping on the structured text TXT and data table TAB to generate a report document DOC, and further encapsulate it into a target format file FIL.

[0040] This embodiment takes the aero-engine blade profile inspection scenario as an example to illustrate the specific implementation process of the present invention. Those skilled in the art can adjust the relevant parameters and configurations according to the actual application scenario to achieve the technical effects of the present invention.

[0041] Specifically, the scenario description is as follows: For the inspection of aero-engine blade profiles, it is necessary to generate an intelligent report based on the comparison results between point cloud data and CAD digital models, which includes deviation analysis, cause speculation and process suggestions, to guide subsequent production and quality control.

[0042] Specifically, the implementation process: Data Acquisition and Standardization: The blade point cloud data is acquired through a 3D scanning device, and the best fit and alignment with the CAD model is completed using measurement software to calculate the profile deviation. The system calls the "point cloud + feature" parsing adapter to convert the alignment error, feature deviation, and point cloud statistical data into the above-mentioned standardized JSON format, and automatically extracts the deviation chromatogram and key section line diagram.

[0043] Task configuration and context preparation: The user selects the "detailed analysis" report depth and the "PDF+Word" output format; the context dynamic assembler filters the "blade profile analysis" exclusive prompt word template, extracts the feature deviation and point cloud statistical subset from the JSON data, and combines it with the relevant summary of "clamp deformation causing blade back deviation" in historical cases to assemble the prompt context.

[0044] Large-scale model-driven content generation: The prompt context is submitted to the language large-scale model, which analyzes it according to the chained prompt logic to generate the following structured text: 1. Overview of the Testing Task The object of this inspection is an aero-engine blade, and the inspection items are the profile of the blade back and blade basin. The best fit alignment method is adopted, and the alignment RMS error is 0.032mm (meeting the requirement of <0.05mm).

[0045] 2. Overall Conformity Conclusion The blade profile generally meets the design requirements and has no out-of-range features. However, there is a local slight convex deviation (+0.08mm) in the middle of the blade back, which is close to the upper limit of tolerance (0.10mm) and requires close attention.

[0046] 3. Key Findings and Detailed Analysis The deviation value of the middle part of the leaf back (feature ID: F001) is +0.08mm, which is within the warning range. This deviation is locally concentrated and not a systemic shift. The deviation of the leaf basin surface (feature ID: F002) is +0.02mm, which is evenly distributed and meets the requirements. The standard deviation of the point cloud is 0.025mm, which indicates that the deviation dispersion is small and the production process is stable.

[0047] 4. Inference of Potential Causes Based on domain knowledge, the deviation pattern of the localized slight convexity in the middle of the blade back is highly consistent with the typical characteristics of the slight support deformation of the No. 3 clamp, suggesting that the deviation may be caused by wear of the No. 3 clamp or displacement of the support point position.

[0048] 5. Recommendations and Follow-up Measures 1) Prioritize retesting all blades processed using fixture No. 3 in this batch to check for any batch deviation risks; 2) Conduct a comprehensive inspection of the support structure of fixture No. 3, measure the wear of the support points, and calibrate or replace it if necessary; 3) In subsequent production, increase the frequency of regular inspections of fixture No. 3 (it is recommended to adjust from once a month to once every two weeks) to avoid deviation risks in advance.

[0049] Content parsing and post-processing: The system extracts structured fields such as "overall conclusions", "key deviations" and "recommendations", verifies that the deviation values ​​are consistent with the original JSON data, and calibrates the accuracy of professional terms such as "profile" and "tolerance limit".

[0050] Automatic multimodal document synthesis: The intelligent typesetting engine lays out the report title, test task overview, data summary table (including feature ID, deviation value, judgment result), deviation chromatogram, key cross-sectional line graph, and detailed analysis text in sequence according to the standard layout. It automatically applies the company's unified font style and header and footer (including company logo, report number, test time).

[0051] Output and Delivery: Generate reports in two formats: PDF (fixed format, for quality review) and Word (editable version, for process adjustment records). Users can download them directly or push them to the enterprise QMS system for archiving via the interface.

[0052] The output format must be Markdown and include the following sections: 1. Overview of the detection task; 2. Overall compliance conclusions; 3. Key findings and detailed analysis; 4. Potential cause inferences (if possible); 5. Recommendations and follow-up measures.

[0053] Safety and cost control measures: Data anonymization: Before data is submitted to the large model, confidential information about the workpiece (such as model number, production batch and other sensitive fields) is automatically removed, and only measurement-related data is retained; Caching optimization: Cache the analysis results of frequently occurring detection scenarios (such as routine detection of the same type of workpiece), and directly call the cached results for repeated scenarios to reduce API call costs.

[0054] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via a wired or wireless network. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0055] Some of the data in the above formulas are numerical calculations with dimensions removed, and the contents not described in detail in this specification are all prior art known to those skilled in the art.

[0056] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A system for generating intelligent analysis reports of industrial measurement data based on a large language model, characterized in that, include: The multi-source data and abstraction module performs structured processing on the raw data DAT output by the measurement and analysis software, including: performing field extraction, type conversion and structure reconstruction on the raw data DAT through a templated data parser to generate a data object JON; and performing node extraction and relationship construction on the data object JON through a measurement semantic graph construction unit to generate a semantic graph structure SUB. The domain knowledge management and prompting engineering module establishes a prompt template library, including role identity templates, analysis process templates, domain rule templates, and output structure templates. Based on the user configuration report, it filters and combines prompt templates, data objects JON, and semantic graph structures SUB to generate prompt contexts PCT. The large language model service and interface module submits the prompt context PCT to the selected large language model LLM through the multi-model adaptation interface to perform semantic reasoning and output structured text TXT. The output parsing post-processor performs consistency verification on the structured text TXT. The multimodal content automatic synthesis and typesetting module is used to perform layout mapping on the structured text TXT and data table TAB to generate report documents DOC; The human-computer interaction module and feedback optimization module are used to provide an interactive report editing interface and a feedback learning interface.

2. The intelligent analysis report generation system for industrial measurement data based on a large language model according to claim 1, characterized in that, The multi-source data and abstraction module includes a templated data parser and a measurement semantic graph construction unit. The templated data parser performs data format determination on the original data DAT, and extracts fields from the alignment matrix MAT, feature deviation data DEV, and point cloud statistical data PCD in the original data DAT according to a preset field mapping rule MAP. It then performs data type conversion, unit unification, and precision normalization on the extracted field data, reconstructs the field hierarchy according to the structure transformation rule TRN, and encapsulates the processed field data according to a unified data structure to generate a data object JON. The measurement semantic graph construction unit performs node extraction processing on the field data in the data object JON to generate a node set NOD; it performs relation abstraction processing on the alignment relation data, benchmark reference data and deviation data in the data object JON to generate a relation edge set REL; and it establishes a hierarchical dependency structure based on the node set NOD and the relation edge set REL to generate a semantic graph structure SUB.

3. The intelligent analysis report generation system for industrial measurement data based on a large language model according to claim 1, characterized in that, The domain knowledge management and prompting engineering module includes a prompt template management unit and a context dynamic assembler; The prompt template management unit is used to build an extensible prompt word library, including: Role constraint templates are used to restrict the output identity of large language models, including: injecting preset identity constraint text into the prompt context to limit the boundaries of the model's analytical perspective, the scope of use of professional terms, and the criteria for judging problems; The reasoning process template is used to construct a fixed sequence of analysis steps, set phased reasoning order control rules for input data, and generate overall evaluation results, anomaly location results, statistical analysis results, cause inference results, and improvement suggestions in sequence according to the preset step sequence. Domain rule templates are used to constrain the knowledge boundaries of the content generated by the model, including: establishing a set of rule mappings between quality problems and potential causes and a set of industry terminology mappings; The structure control template is used to format the chapter structure and presentation of the generated text, including: preset chapter title set, paragraph hierarchy rules, keyword emphasis rules, and tag output rules; The prompt template management unit establishes a template category index and an applicable scenario index; it also establishes parent-child reference relationships and calling relationships between templates according to hierarchical nesting rules, forming a template structure set. The context dynamic assembler parses the user configuration report to generate a set of report control parameters; performs rule matching and priority sorting on the template structure set according to the set of report control parameters to determine the target template combination sequence; performs condition filtering and field screening on the data object JON and semantic graph structure SUB to generate a data subset; performs similarity matching and content extraction on the historical case library to generate a supplementary text set; and arranges the target template combination sequence, data subset, and supplementary text set in order according to preset splicing rules to generate a prompt context.

4. The intelligent analysis report generation system for industrial measurement data based on a large language model according to claim 1, characterized in that, The large language model service module includes a multi-model adaptation interface and an output parsing post-processor; The multi-model adaptation interface will submit the context prompt to the selected large language model; the multi-model adaptation interface will perform role identity constraint processing and domain framework constraint processing; The role identity constraint processing includes: establishing a role identity database, configuring identity identifier parameters, professional ability tags, and terminology usage scope parameters for different roles; identifying and generating roles based on the current task type after receiving the prompt context; and extracting the corresponding identity definition text from the role identity database and inserting it into the control instruction section of the prompt context. The domain framework constraint processing includes: establishing a set of analysis step rules; generating structured guidance text from the set of analysis step rules in a preset order; and inserting the structured guidance text into a specified logical segment after identity enhancement.

5. The intelligent analysis report generation system for industrial measurement data based on a large language model according to claim 4, characterized in that, The multi-model adaptation interface performs structural identifier parsing on the prompt context, identifies the analysis step identifier information therein, and generates a stage index set; based on the stage index set, the prompt context is divided into data interpretation stage prompt content, cause inference stage prompt content, and action suggestion stage prompt content, and corresponding stage prompt sub-contexts are generated respectively; For each stage's prompt sub-context, the corresponding inference control parameter set is loaded and a stage call message is constructed. The inference control parameter set includes randomness control parameters, candidate range control parameters, and repetition suppression parameters. The stage call messages are sent to the large language model in a preset stage call order to obtain the output results of each stage. The output results of each stage are concatenated in the order of the stages to generate a structured text TXT.

6. The intelligent analysis report generation system for industrial measurement data based on a large language model according to claim 5, characterized in that, The output parsing post-processor performs consistency verification on the structured text TXT, including: performing field recognition and structure extraction on the structured text TXT to generate a set of structured fields; comparing the set of structured fields with the original measurement data item by item, and performing field integrity verification, numerical range verification, unit consistency verification, and logical relationship verification according to preset verification rule parameters to generate field parsing results; judging the field parsing results according to the confidence threshold parameter, and performing a regular verification process when the field confidence is greater than or equal to the threshold, and directly marking the field as an anomaly when the field confidence is lower than the threshold.

7. The intelligent analysis report generation system for industrial measurement data based on a large language model according to claim 1, characterized in that, The multimodal content automatic synthesis and typesetting module includes: an intelligent typesetting engine and a multi-format rendering and export unit; The intelligent typesetting engine performs syntactic structure parsing on the structured text TXT, generates a chapter tree structure according to preset chapter hierarchy rules, performs field filtering and data reorganization processing on the measurement result fields in the standardized data object, converts the numerical fields into tabular data structures, and performs attribute marking processing on abnormal data fields according to preset threshold rules. Load the set of typesetting template rules and layout constraint rules, perform position matching and sequence arrangement processing on the chapter tree structure and table data structure, and generate a report document DOC; The multi-format rendering and export unit is used to perform target format encapsulation processing on the report document DOC, including: The target format identifier is determined based on the output type specified by the user; the corresponding format rendering engine is called to perform format mapping and style conversion processing on the report document DOC, and then further encapsulated into the target format file FIL.

8. A method for generating intelligent analysis reports of industrial measurement data based on a large language model, characterized in that, include: S1. Obtain the raw data DAT output by the measurement software, convert the raw data DAT into a data structure JON through a templated data parser, and extract nodes and construct relationships for the data object JON through a measurement semantic graph construction unit to generate a semantic graph structure SUB. S2. Based on the user configuration report set by the user, filter and combine the prompt template, the data object JON, and the semantic graph structure SUB to generate the prompt context PCT; S3. The prompt context PCT is submitted to the large language model LLM through the role preset injection module ROLN to perform semantic reasoning and output the structured text TXT; S4. Perform consistency verification on the structured text TXT through the output parsing post-processor. S5. Perform layout mapping on the structured text TXT and data table TAB to generate a report document DOC, and further encapsulate it into a target format file FIL.

Images