CN-122018468-A - Large-model-based aluminum veneer production control method and system

CN122018468ACN 122018468 ACN122018468 ACN 122018468ACN-122018468-A

Abstract

The invention relates to the technical field of aluminum veneer production control, in particular to a large-model-based aluminum veneer production control method and a large-model-based aluminum veneer production control system, which comprise the steps of constructing an aluminum veneer production full-process knowledge graph, and receiving natural language production task requirements including customer descriptions of aluminum veneer surface coating colors, plate thicknesses and special-shaped cutting contours; the method comprises the steps of carrying out semantic analysis and entity extraction on the requirement by using a production instruction analysis engine, converting the requirement into a normalized technological parameter query request capable of matching with a knowledge graph node, traversing a historical successful production batch with highest correlation degree of the knowledge graph retrieval, extracting a complete technological parameter chain covering a whole working procedure, calling a parameter optimization model based on an attention mechanism, and carrying out dynamic fine adjustment on the parameter chain by combining real-time equipment state data of a production line to generate an optimal production control instruction set. The method improves the accuracy and adaptability of production control and adapts to the personalized demands of customers.

Inventors

WANG SHIPENG
LU FEI
WEI SHUAI
JIAO XIAOBING
LU GUANGCHAO
ZHU CHENGLONG
ZHANG YONG
SHENG JUNFENG
XUE ZENGMING
LIANG ZHIGANG
JIAO PENGHAO
ZHANG YINGJIE
ZHANG HAO
Mao Shuaijun

Assignees

河南鑫瑞佳新材料有限公司
河南凯美特科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260313

Claims (10)

1. The aluminum veneer production control method based on the large model is characterized by comprising the following steps of: Constructing a full-process knowledge graph of aluminum veneer production, and receiving natural language production task requirements input by a user, wherein the natural language production task requirements comprise descriptions of coating colors, plate thicknesses and special-shaped cutting outlines on the surface of the aluminum veneer by a client; Carrying out semantic analysis and entity extraction on the natural language production task demands by using a production instruction analysis engine, and converting the surface coating color, the plate thickness and the special-shaped cutting profile information of the aluminum veneer described by a customer into a normalized process parameter query request capable of matching corresponding nodes in the full-process knowledge graph; Traversing the full-process knowledge graph based on the normalized process parameter query request, searching a historical successful production batch with highest degree of association with each requirement in the normalized process parameter query request, and extracting a complete process parameter chain corresponding to the historical successful production batch, wherein the complete process parameter chain covers a blanking process parameter, a molding process parameter, a welding process parameter, a polishing process parameter and a spraying process parameter; And calling a parameter optimization model based on an attention mechanism, and carrying out dynamic fine adjustment and adaptive correction on each parameter in the extracted complete process parameter chain by combining with real-time equipment state data of the current production line to generate an optimal production control instruction set suitable for the current production environment.
2. The method for controlling aluminum veneer production based on large model as claimed in claim 1, wherein the semantic parsing and entity extraction of the natural language production task requirement are performed by using a production instruction parsing engine, comprising: Deploying a pre-trained large language model as a production instruction analysis engine; Inputting the natural language production task requirement into the production instruction analysis engine, wherein the production instruction analysis engine utilizes a word vector matrix arranged in the production instruction analysis engine to segment and embed words into an input text; Performing context semantic coding on the word embedding sequence through a multi-layer transducer encoder structure of the production instruction analysis engine to generate a feature vector sequence containing global semantic information; On the characteristic vector sequence, a classifier trained for the aluminum veneer production field is applied to identify key semantic segments belonging to the process requirement category in the text; carrying out named entity identification operation on the identified key semantic segments, and specifically identifying color card coding entities corresponding to the color of the aluminum veneer surface coating, millimeter value entities corresponding to the thickness of the plate and two-dimensional contour point set description entities corresponding to the special-shaped cutting contour; And packaging the identified color card coding entity, millimeter value entity and two-dimensional contour point set description entity according to a preset process parameter data structure to form the normalized process parameter query request.
3. The method of claim 2, wherein traversing the full-process knowledge graph based on the normalized process parameter query request retrieves a historical successful production lot with highest association with each requirement in the normalized process parameter query request, comprising: The full-process knowledge graph comprises raw material attribute data, equipment operation parameter data, historical process formula data and finished product quality detection data, and the association relation among different types of data is structurally stored in the full-process knowledge graph in a node and edge mode; taking the color card coding entity in the normalized process parameter query request as an initial query node, performing graph traversal in a coating material attribute subgraph of the full-process knowledge graph, and positioning all historical production batch nodes using color card coding coatings; Screening all the located historical production batch nodes by taking the millimeter value entity as a constraint condition, and eliminating the historical production batch nodes with unmatched plate thickness parameters; Further calculating the geometrical similarity between the associated profile description data and the two-dimensional profile point set description entity of the remaining historical production batch nodes, and screening a historical production batch node set with the geometrical similarity higher than a preset threshold value; calculating comprehensive scores of the historical finished product quality detection data corresponding to each node in the historical production batch node set; And selecting the historical production batch node with the highest comprehensive score as the historical successful production batch with the highest association degree, and extracting all upstream process data nodes taking the historical production batch node as an end point from the full-process knowledge graph to form the complete process parameter chain.
4. The method for controlling aluminum veneer production based on large model according to claim 3, wherein the calling the attention mechanism-based parameter optimization model, combining the real-time equipment state data of the current production line, dynamically fine-tuning and adaptively correcting each parameter in the extracted complete process parameter chain comprises the following steps: acquiring real-time equipment state data of the current production line from a production monitoring system, wherein the real-time equipment state data comprises a cutter abrasion coefficient of a numerical control punch, an atomization pressure value of a spray gun of a spraying robot and current furnace temperature distribution data of a baking furnace; Inputting the original blanking process parameters, the original forming process parameters, the original welding process parameters, the original polishing process parameters and the original spraying process parameters in the complete process parameter chain and the real-time equipment state data into the parameter optimization model based on the attention mechanism; the parameter optimization model based on the attention mechanism firstly calculates the correlation weight between each parameter in the real-time equipment state data and each original procedure parameter in the complete process parameter chain; According to the calculated correlation weight, carrying out weighted correction on the original process parameters in the complete process parameter chain, wherein the original process parameters with high correlation with the equipment state abnormal item obtain a correction amplitude with a larger degree; After the weighting correction, the parameter optimization model based on the attention mechanism outputs the adjusted adaptive blanking process parameter, the adaptive forming process parameter, the adaptive welding process parameter, the adaptive polishing process parameter and the adaptive spraying process parameter, and the optimal production control instruction set is formed by the adaptive parameters together.
5. The aluminum veneer production control method based on large model as recited in claim 4, further comprising the steps of executing the optimal production control instruction set and collecting production feedback data: Issuing adaptive blanking process parameters in the optimal production control instruction set to a numerical control cutting machine controller of a blanking workshop, and driving a numerical control cutting machine to perform aluminum plate blanking according to the specified contour and size; Issuing the adaptive forming process parameters in the optimal production control instruction set to a numerical control bending machine controller of a forming workshop, and driving a numerical control bending machine to form the plate according to a specified angle and a bending sequence; During the blanking and forming processes, collecting multi-dimensional feedback data in the actual production process through a sensor array arranged at a key node of a production line, wherein the multi-dimensional feedback data comprises a smoothness measured value of a cutting edge of an aluminum plate, an actual angle deviation value of a bending angle and a timestamp record transmitted among the processes; and comparing the collected multidimensional feedback data with the corresponding expected parameters in the optimal production control instruction set in real time to generate a process-level production state deviation report.
6. The aluminum veneer production control method based on the large model as claimed in claim 5, further comprising the step of iteratively updating the full-process knowledge graph based on production feedback data: After a complete production batch is finished, summarizing the multidimensional feedback data collected by the subsequent procedures of welding, polishing and spraying to form complete production execution track data of the complete production batch; Correlating a final quality detection report of a complete production batch with the production execution track data, and marking the production batch as a successful batch or a failed batch; using a graph structure updating algorithm, and adding the marked production execution track data as a new production example node into the full-process knowledge graph; Establishing a correlation edge between the new production example node and the existing raw material attribute node, equipment parameter node and process parameter node in the full-process knowledge graph, wherein the weight of the correlation edge carries out initialization assignment according to the quality degree of the production batch production result; And periodically recalculating and updating the weights of all the associated edges in the full-process knowledge graph, wherein the basis of weight updating is the statistical information of newly added successful production examples and failed production examples.
7. The method of claim 6, further comprising processing a production anomaly event using the production command parsing engine: When an online detection device on a production line identifies that the product quality is abnormal, natural language alarm information containing abnormal feature description is automatically generated; inputting the natural language alarm information into the production instruction analysis engine; the production instruction analysis engine analyzes the natural language alarm information and identifies an abnormal type, an abnormal occurrence procedure and an abnormal severity level; Based on the identified abnormal type and the abnormal occurrence procedure, the production instruction analysis engine performs reverse traceability query in the full-procedure knowledge graph, and searches production records with similar abnormal conditions in history and correction measures taken by the production records; And generating a targeted abnormal treatment proposal instruction according to the inquired correction measure record and the current real-time equipment state data, wherein the abnormal treatment proposal instruction comprises equipment parameter adjustment quantity, process step pause or material replacement scheme.
8. The method of claim 7, wherein calculating the geometric similarity between the associated profile data and the two-dimensional profile point set description entity comprises: Extracting profile description data associated with the historical production batch nodes from the knowledge graph, and analyzing the profile description data into a historical profile point set; Resolving the two-dimensional profile point set description entity into a target profile point set; Carrying out coordinate normalization processing on the historical contour point set and the target contour point set to enable the historical contour point set and the target contour point set to be in the same size and coordinate system; Applying an iterative nearest point algorithm to calculate a minimum root mean square error required to align the historical contour point set to the target contour point set; And performing reciprocal operation on the calculated minimum root mean square error, normalizing the minimum root mean square error to a range from zero to one, and obtaining a numerical value which is the geometric similarity.
9. The method for controlling aluminum veneer production based on a large model according to claim 8, wherein the periodically re-calculating and updating weights of all associated sides in the full-process knowledge graph comprises: Setting a fixed weight updating period, and counting all newly added successful production examples and failed production examples in the period when each weight updating period is finished; For each associated side in the full-process knowledge graph, searching all historical production instance nodes passing through the associated side, and screening out newly added instances belonging to the current updating period; counting the number of successful production examples and the number of failed production examples in the screened newly added examples; according to the number of successful production examples and the number of failed production examples, recalculating the current confidence weight of the associated edge according to a preset formula; And replacing the original weight value of the associated edge with the calculated current confidence coefficient weight to complete dynamic update of the knowledge association strength in the full-process knowledge graph.
10. A large model based aluminum veneer production control system comprising a memory, a processor and a computer program stored in the memory and running on the processor, characterized in that the processor, when executing the computer program, realizes the steps of a large model based aluminum veneer production control method according to any one of the preceding claims 1 to 9.

Description

Large-model-based aluminum veneer production control method and system Technical Field The invention relates to the technical field of aluminum veneer production control, in particular to an aluminum veneer production control method and system based on a large model. Background The aluminum veneer production comprises a plurality of closely related working procedures such as blanking, forming, welding, polishing, spraying and the like, the rationality of technological parameters of each working procedure directly determines the quality of products, and the production control process needs to consider the individual demands of customers and the actual conditions of production sites. In the existing aluminum veneer production control mode, the production requirements proposed by customers are presented in a natural language form and comprise key information such as color of a coating on the surface of an aluminum veneer, thickness of a plate, special-shaped cutting outline and the like, the requirements are manually analyzed and converted into executable production process parameters, and then historical production parameters are selected as references by depending on experience of operators to finally formulate production control instructions. The manual analysis of the natural language requirement is easy to generate understanding deviation, so that the process parameter setting is error, the prior art lacks the system integration of the full process knowledge of the aluminum veneer production, the historical successful production batch and the complete process parameter chain which are highly matched with the current customer requirement can not be quickly and accurately searched, and the parameter selection efficiency is low and the subjectivity is strong. In addition, once the technological parameters in the existing production control are determined, the technological parameters can be kept fixed, the dynamic adjustment can not be carried out by combining the real-time equipment state of a production line, the equipment operation fluctuation under different production environments is difficult to adapt, the production quality is unstable, the personalized and high-precision requirements of customers on aluminum veneer products can not be met, and a production control mode capable of accurately analyzing the requirements, efficiently matching the parameters and realizing the dynamic adjustment is required. Disclosure of Invention The invention aims to solve the defects in the prior art, and provides a large-model-based aluminum veneer production control method and a large-model-based aluminum veneer production control system. In order to achieve the purpose, the invention adopts the following technical scheme that the aluminum veneer production control method based on the large model comprises the following steps: Constructing a full-process knowledge graph of aluminum veneer production, and receiving natural language production task requirements input by a user, wherein the natural language production task requirements comprise descriptions of coating colors, plate thicknesses and special-shaped cutting outlines on the surface of the aluminum veneer by a client; Carrying out semantic analysis and entity extraction on the natural language production task demands by using a production instruction analysis engine, and converting the surface coating color, the plate thickness and the special-shaped cutting profile information of the aluminum veneer described by a customer into a normalized process parameter query request capable of matching corresponding nodes in the full-process knowledge graph; Traversing the full-process knowledge graph based on the normalized process parameter query request, searching a historical successful production batch with highest degree of association with each requirement in the normalized process parameter query request, and extracting a complete process parameter chain corresponding to the historical successful production batch, wherein the complete process parameter chain covers a blanking process parameter, a molding process parameter, a welding process parameter, a polishing process parameter and a spraying process parameter; And calling a parameter optimization model based on an attention mechanism, and carrying out dynamic fine adjustment and adaptive correction on each parameter in the extracted complete process parameter chain by combining with real-time equipment state data of the current production line to generate an optimal production control instruction set suitable for the current production environment. As a further aspect of the present invention, the semantic parsing and entity extraction of the natural language production task requirement using a production instruction parsing engine includes: Deploying a pre-trained large language model as a production instruction analysis engine; Inputting the natural language production task requirement into the production