Large-model-based manufacturing enterprise data medium system and method

Abstract

The invention discloses a manufacturing enterprise data center system and method based on a large model, and the system comprises a data collection layer which is used for constructing a multi-source heterogeneous data collection system; the data storage layer is used for constructing a data warehouse system comprising multiple layers and realizing data lake storage; the data processing layer and the service layer are used for deploying a model and integrating an anomaly detection module, a multi-modal fusion engine and a prediction model; the service layer is used for an interface group, integrating a Drive / ClickHouse query engine and a Redis cache, and supporting multi-condition combined query and millisecond response; deploying a visual service module and an RBAC / ABAC authority control system to realize data grading sharing; and the application layer is used for developing a production optimization application, a supply chain management application and a CRM application. Through a uniform data acquisition interface and an efficient data acquisition tool, mass data inside and outside an enterprise can be integrated, data islands are broken, centralized management and sharing of the data are realized, and availability and value of the data are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of enterprise informatization management, and particularly to a data middle platform system and method for manufacturing enterprises based on large models. Background Art

[0002] In manufacturing enterprises, with the advancement of digital transformation, a vast amount of data has been generated and accumulated within the enterprises. This data has a wide range of sources, including various business systems such as production equipment, supply chain management, customer relationship management, and financial management. However, manufacturing enterprises currently face many challenges in data management.

[0003] On the one hand, the data is scattered and stored in different systems, forming data islands. The data formats and standards are not unified, resulting in difficulties in data sharing and circulation, and the data value cannot be fully utilized. On the other hand, traditional data processing and analysis methods are inefficient in the face of large-scale, multi-source heterogeneous data, and it is difficult to extract valuable information and knowledge from it, unable to provide timely and accurate support for enterprise decision-making.

[0004] In addition, the business requirements of manufacturing enterprises are complex and diverse, and they evolve rapidly with market changes. Traditional data architectures and technologies are difficult to quickly respond to business changes, with insufficient flexibility and scalability. Summary of the Invention

[0005] The purpose of the present invention is to overcome the deficiencies of the prior art. To achieve the above purpose, a data middle platform system and method for manufacturing enterprises based on large models are adopted to solve the problems raised in the above background art.

[0006] A data middle platform system for manufacturing enterprises based on large models includes a data collection layer, a data storage layer, a data processing layer, a service layer, and an application layer;

[0007] The data collection layer is used to build a multi-source heterogeneous data collection system. Among them, the data collection system includes an industrial Ethernet interface and field bus protocol adaptation module, a highly compatible database connection module, and an intelligent file parsing engine;

[0008] The data storage layer is used to adopt a collaborative architecture of a distributed file system and a distributed database to build a multi-layer data warehouse system and implement data lake storage in Parquet / ORC columnar format;

[0009] The data processing layer is used to deploy a pre-trained AI large model, integrate a deep anomaly detection module, a multi-modal fusion engine, and a prediction model; configure a transfer learning framework to achieve model fine-tuning, and associate device parameters, process data, and quality inspection results to form a decision knowledge base;

[0010] The service layer is used to provide a RESTful API interface group, integrate the Druid / ClickHouse query engine and Redis cache, support multi-condition combined queries and millisecond-level responses; deploy a visualization service module and an RBAC / ABAC permission control system to achieve hierarchical data sharing;

[0011] The application layer is used to develop production optimization applications, supply chain management applications, and CRM applications, and dock with the ERP / MES system through a microservices architecture and an API gateway to form a closed loop for functions such as equipment failure prediction, process parameter optimization, and customer segmentation recommendation.

[0012] As a further solution of the present invention: The data acquisition layer includes a lightweight acquisition agent, a real-time data pipeline, and a log audit module;

[0013] The lightweight acquisition agent is used to be deployed at the device edge, integrate ZigBee / Modbus protocol conversion modules, and support low-power transmission of sensor data;

[0014] The real-time data pipeline is built based on Apache Kafka, uses the Snappy compression algorithm and AES-256 end-to-end encryption for industrial data stream transmission with millisecond-level latency;

[0015] The log audit module is used to record acquisition task IDs, data volume statistics, and CRC check information, and support breakpoint resumption and traceability of abnormal tasks.

[0016] As a further solution of the present invention: The anomaly detection module in the data processing layer includes a vibration signal time series prediction model and an anomaly root cause analysis module;

[0017] The vibration signal time series prediction model is built based on a GRU network, sets dynamic threshold alarm rules, and uses a GAN network to generate synthetic anomaly samples;

[0018] The anomaly root cause analysis module correlates the SPC control chart with process parameter fluctuations for multi-dimensional traceability.

[0019] As a further solution of the present invention: The production optimization application includes a digital twin modeling module, a genetic algorithm optimization engine, and a predictive maintenance dashboard;

[0020] The digital twin modeling module is used to update model parameters based on ANSYS simulation data and real-time sensor feedback;

[0021] The genetic algorithm optimization engine is used to solve the optimal process parameter combination with the minimum energy consumption as the objective function;

[0022] The predictive maintenance dashboard is used to display the predicted remaining useful life (RUL) value and the spare parts inventory linkage information.

[0023] As a further solution of the present invention: the supply chain management application includes a supplier evaluation index system, a dynamic procurement strategy module, and a logistics path optimization module;

[0024] The supplier evaluation index system is used to obtain the on-time delivery rate, the PPM value of quality defects, and the price fluctuation coefficient;

[0025] The dynamic procurement strategy module is used to obtain a strategy based on the combinatorial auction algorithm and support the solution of multi-objective constraints;

[0026] The logistics path optimization module is used to integrate the map API to calculate the optimal delivery route in real time.

[0027] As a further solution of the present invention: the CRM application includes a customer lifetime value prediction model, a work order semantic analysis module, and a personalized recommendation engine;

[0028] The customer lifetime value prediction model is constructed using the XGBoost integrated learning framework;

[0029] The work order semantic analysis module is constructed based on Transformer and automatically classifies customer complaint types;

[0030] The personalized recommendation engine is used to generate differentiated marketing strategies according to the RFM score.

[0031] The technical solution of the second aspect: an application method applied to a data middle platform system of a manufacturing enterprise based on a large model as described in any one of the above, including the following steps:

[0032] Step S1, through the interfaces and tools of the data acquisition layer, according to the predetermined acquisition strategy, collect data in real time or at regular intervals, obtain data from various data sources, and transmit the data to the data storage layer;

[0033] Step S2, the data storage layer receives the collected data, stores it in the data lake or data warehouse of the corresponding data storage layer, and performs preliminary preprocessing;

[0034] Step S3, use the large model of the data processing layer to process and analyze the stored data, which includes data cleaning, fusion, classification, and prediction, and generate data insights and knowledge;

[0035] Step S4, the service layer provides the processed data and analysis results to the application layer and external systems through service interfaces, and at the same time realizes data sharing;

[0036] Step S5: The application layer develops applications based on the services of the data middle platform and integrates them with the business system, applying the value of the data middle platform to actual business operations.

[0037] Compared with the prior art, the present invention has the following technical effects:

[0038] By adopting the above technical solution, through a unified data collection interface and efficient data collection tools, it is possible to integrate the massive data inside and outside the enterprise, break data islands, realize centralized management and sharing of data, and improve the availability and value of data.

[0039] Based on the powerful natural language processing ability of the large model, it is possible to simplify the operation process of the data middle platform to the greatest extent. Users can obtain the required data in the form of voice commands, and at the same time, it can be presented to users in various ways.

[0040] Based on the data processing ability of the large model, it can automatically process and analyze large-scale, multi-source heterogeneous data, mine deep information and potential laws in the data, and provide more accurate and forward-looking decision-making support for the enterprise. For example, it can predict equipment failures in advance and reduce the risk of production interruptions; accurately predict market demand and optimize production plans and inventory management.

[0041] The service layer and application layer provided by the data middle platform provide data and analysis capabilities to various internal departments and external partners of the enterprise in a service-oriented form, reducing the threshold of data application, promoting collaborative innovation and business expansion within the enterprise, and at the same time improving the enterprise's response speed and competitiveness to market changes.

[0042] The deep integration with the enterprise's existing business system enables the data middle platform to integrate into the enterprise's business processes, realize real-time interaction and feedback of business data, optimize business processes, and improve the overall operation efficiency and management level of the enterprise. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The following will describe in detail the specific embodiments of the present invention in conjunction with the drawings:

[0044] Figure 1 It is a schematic diagram of the enterprise data middle platform architecture of the disclosed embodiment of the present application;

[0045] Figure 2 It is a step flow block diagram of the application method of the enterprise data middle platform of the disclosed embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0046] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0047] Please refer to Figure 1 , in the embodiment of the present invention, a data middle platform system for manufacturing enterprises based on a large model includes a data acquisition layer, a data storage layer, a data processing layer, a service layer, and an application layer;

[0048] The data acquisition layer is used to construct a multi-source heterogeneous data acquisition system. Among them, the data acquisition system includes an industrial Ethernet interface and a field bus protocol adaptation module, a highly compatible database connection module, and an intelligent file parsing engine;

[0049] In this embodiment, the data acquisition layer includes a lightweight acquisition agent, a real-time data pipeline, and a log audit module;

[0050] The lightweight acquisition agent is used to be deployed at the device edge end, integrate a ZigBee / Modbus protocol conversion module, and support low-power transmission of sensor data;

[0051] The real-time data pipeline is built based on Apache Kafka, and uses the Snappy compression algorithm and AES-256 end-to-end encryption for industrial data stream transmission with millisecond-level latency;

[0052] The log audit module is used to record the acquisition task ID, data volume statistics, and CRC check information, and support breakpoint resumption and traceability of abnormal tasks.

[0053] In the specific implementation, a data acquisition interface system highly adapted to the complex data sources in the manufacturing industry is designed, covering various industrial communication protocols and database connection methods. For the data of production equipment, the industrial Ethernet interface is combined with various field bus protocols (such as Profibus, Profinet, etc.) to ensure accurate and high-speed acquisition of high-precision data from various sensors, including but not limited to physical quantities such as temperature, pressure, humidity, flow, displacement, etc., as well as the operation status information of the equipment (such as start / stop, fault codes, etc.), meeting the requirements for real-time monitoring and refined management of the production process. For the business system databases within the enterprise, database connection interfaces with high compatibility are developed, supporting seamless docking with mainstream relational databases (such as Oracle, MySQL, SQL Server, PostgreSQL, etc.) and some NoSQL databases (such as MongoDB, Redis, etc.). Through optimized SQL query statements and database driver programs, structured data in systems such as ERP, MES, PLM, CRM, SCM, etc., such as core business data like bill of materials (BOM), production orders, inventory records, customer information, financial accounts, etc., are efficiently extracted to ensure data integrity and consistency. When processing external file data, various file transfer protocols such as FTP, SFTP, HTTP, HTTPS, etc. are supported, and an intelligent file parsing engine is integrated, which can automatically identify and process common file formats such as CSV, XML, JSON, Excel, PDF, etc., realizing effective acquisition of external data such as market research reports, industry standard documents, design drawings, etc., broadening the breadth and depth of the enterprise's data resources.

[0054] Comprehensively utilize data collection tools and technical frameworks to build an efficient and reliable collection process. For ETL tools, an open-source framework with distributed processing capabilities and a rich plugin ecosystem is selected (such as Apache Sqoop, Talend, Pentaho DataIntegration, etc.), combined with custom data conversion scripts (based on programming languages such as Python and Java), to achieve batch extraction, cleaning, transformation, and loading operations of large-scale data. By configuring optimized connection parameters, data partitioning strategies, and incremental update mechanisms, ensure the efficient completion of data migration tasks without affecting the performance of the source system, and improve the throughput and stability of data collection. For data collection scenarios with extremely high real-time requirements, a high-performance message queue system based on the publish-subscribe model is introduced (such as Apache Kafka, RabbitMQ, ActiveMQ, etc.), combined with lightweight data collection agent programs (deployed at the data source end), to achieve low-latency, high-concurrency data collection and transmission. These agent programs have adaptive data compression and encryption functions, and can dynamically adjust the transmission strategy according to the network conditions to ensure the fast and secure transmission of data in complex network environments. At the same time, to ensure the reliability and traceability of data collection, a complete logging and monitoring system is integrated, which details key information such as each step, data volume, and transmission status during the collection process, facilitating the timely discovery and solution of problems, and providing a strong basis for data quality auditing.

[0055] The data storage layer is used to build a multi-layer data warehouse system by collaborating the distributed file system and the distributed database, and implement data lake storage in Parquet / ORC columnar format;

[0056] In the specific implementation, a hybrid storage architecture combining a distributed file storage system (such as Ceph, GlusterFS, MinIO, etc.) and a distributed database (such as HBase, Cassandra, ScyllaDB, etc.) is constructed to meet the storage requirements of massive, multi-source heterogeneous data in manufacturing enterprises. The distributed file storage system utilizes its distributed metadata management, data redundancy backup, and high scalability to be responsible for storing large-scale unstructured data, such as raw log files generated by device operations, high-definition image data, video surveillance materials, complex design documents (such as CAD, SolidWorks files), and text files in various formats. By adopting technologies such as distributed hash tables (DHT), etc., to achieve fast data location and retrieval, and combining data redundancy strategies such as erasure codes to ensure high availability and fault tolerance of data, even when some nodes fail, the integrity and accessibility of data can still be guaranteed. The distributed database, relying on its powerful distributed transaction processing capabilities, efficient index structures (such as B+ trees, LSM trees and their variants), and column storage optimization technologies, focuses on storing structured data, such as enterprise business transaction data, master data, metadata, etc. During the storage process, by reasonably designing the table structure, partitioning strategies (such as partitioning by time, region, business type, etc.) and establishing appropriate indexes (including primary key indexes, secondary indexes, composite indexes, etc.), fast write, query, and update operations on structured data are realized, ensuring data consistency and integrity, and providing solid data support for the daily business operations and decision-making analysis of enterprises.

[0057] A data storage platform based on the concept of data lake is built, which saves the original data in high-performance column storage formats such as Parquet and ORC, and makes full use of its high compression ratio, efficient query performance, and good Schema evolution characteristics to achieve low-cost and efficient storage of massive original data. On top of the data lake, a multi-level data warehouse architecture is constructed, including the original data layer (ODS), the detailed data layer (DWD), the aggregated data layer (DWS), and the application data layer (ADS). The ODS layer stores the original data collected from various data sources, maintaining the original appearance and integrity of the data, and serving as the basic data source for subsequent data processing. The DWD layer cleans, transforms, and preliminarily integrates the original data through the ETL process and data modeling tools (such as Apache Hive, Spark SQL, Flink SQL, etc.), organizes the data according to business subject domains (such as production, sales, procurement, finance, R & D, etc.), and constructs star or snowflake data models to form the core layer of the enterprise's data assets. The DWS layer further summarizes, aggregates, and performs light calculations on the data based on the DWD layer to generate wide-table data for analysis topics, providing an efficient data query interface for data analysis and decision support. The ADS layer extracts data from the DWS layer according to the specific needs of different business departments and application scenarios, and performs customized processing to generate data marts for specific business applications (such as production optimization data mart, supply chain analysis data mart, customer relationship management data mart, etc.), providing accurate and efficient data services for various business applications of the enterprise, and realizing the full-process management and application of data from original collection to value release.

[0058] The data processing layer is used to deploy a pre-trained large AI model, integrate a deep anomaly detection module, a multi-modal fusion engine, and a prediction model; configure a transfer learning framework to implement model fine-tuning, and associate device parameters, process data, and quality inspection results to form a decision knowledge base;

[0059] In this embodiment, the anomaly detection module of the data processing layer includes a vibration signal time series prediction model and an anomaly root cause analysis module;

[0060] The vibration signal time series prediction model is constructed based on the GRU network, sets dynamic threshold alarm rules, and uses the GAN network to generate synthetic anomaly samples;

[0061] The anomaly root cause analysis module associates the SPC control chart with process parameter fluctuations for multi-dimensional traceability.

[0062] In the specific implementation, a large AI model is introduced as the core engine for data processing and analysis to cope with the complex and changeable data processing tasks and decision support needs of manufacturing enterprises. In the pre-training stage of the model, large-scale public data sets in the manufacturing field (such as manufacturing-related data sets in the UCI machine learning database, industrial data competition data sets on the Kaggle platform, etc.) and massive historical data accumulated over the years within the enterprise are fully integrated, covering multiple dimensions such as production process parameters, equipment life cycle data, market sales data, customer behavior data, and supply chain data. The model is efficiently trained to enable the model to learn the common characteristics, potential patterns, and complex correlations of manufacturing data. For example, by learning from massive equipment operation data, the model can grasp the parameter change laws, characteristic patterns, and correlations between different types of fault types of equipment in normal operation and fault states; through the analysis of market sales data, it can gain insight into seasonal fluctuations in market demand, changes in product life cycles, and consumer preferences and purchasing trends in different regions and customer groups. During the fine-tuning stage, based on the company's specific business needs and data characteristics, we carefully collected and organized high-precision labeled data in professional fields, such as equipment failure cases, product quality defect samples, customer segmentation label data, and key points of market trend changes that have been strictly labeled by senior engineers and industry experts. We use transfer learning technology to transfer the general knowledge of the pre-trained model to the company's specific business scenarios, and optimize and adjust the model in a targeted manner, significantly improving the model's prediction accuracy, recall rate, and generalization ability in the company's actual business, so that it can better serve the company's production decisions, marketing, customer service and other key business links.

[0063] In the specific process of data processing, the large model fully exerts its powerful intelligent analysis and decision-making support capabilities. In the data cleaning process, deep learning-based anomaly detection algorithms (such as unsupervised anomaly detection based on autoencoders, anomaly detection based on generative adversarial networks, and time series anomaly detection based on recurrent neural networks such as LSTM and GRU) are used, combined with traditional statistical analysis methods (such as the 3σ rule, box plot analysis, outlier detection based on clustering, etc.), to intelligently identify and accurately process error values, outliers, and duplicate data in the data. For example, when processing device temperature data, the model can accurately determine whether the temperature value at a certain moment is abnormal based on the statistical distribution characteristics of historical data, the normal operating range of the device, and the time series correlation of the data, and reasonably correct the abnormal data through various data repair strategies (such as weighted interpolation method based on neighboring data, prediction value repair method based on time series models, outlier replacement method based on machine learning classifiers, etc.), ensuring the quality and reliability of the data and providing a solid data foundation for subsequent data analysis and decision-making. In the data fusion stage, semantic understanding technologies in natural language processing (such as word vector representation, semantic parsing, text classification and clustering, etc.), feature extraction and image recognition technologies in computer vision (such as convolutional neural networks, object detection and recognition algorithms, etc.), and multi-modal data fusion algorithms (such as feature fusion based on attention mechanisms, probability fusion based on Bayesian networks, etc.) are used to deeply fuse and perform correlation analysis on data from different data sources and different formats. For example, when analyzing product quality problems, it is possible to organically integrate the operation parameter data of production equipment, the inspection report data of raw materials (which may include text descriptions and numerical indicators), the execution data of production processes (such as processing paths, process parameter settings, etc.), and the image or video detection data of products, and accurately identify the key factors and potential problems affecting product quality through semantic understanding and feature association, providing strong support for enterprises to improve production processes, optimize raw material procurement, and strengthen quality control.In terms of data classification and prediction, based on deep learning neural network models (such as multi-layer perceptrons, convolutional neural networks, recurrent neural networks, and their combined architectures, such as CNN-LSTM, Transformer-Encoder, etc.), historical data is deeply mined and learned to build multiple targeted prediction models, including deep survival analysis models for equipment fault prediction, LSTM fault prediction models based on attention mechanisms, time series decomposition and deep learning fusion models for market demand prediction (such as ARIMA-LSTM, Prophet-GRU, etc.), Bayesian deep learning models for product quality prediction, and multi-model fusion prediction systems based on ensemble learning. These models can accurately predict the probability distribution of equipment faults, the changing trends of market demands, the fluctuations of product quality, etc., providing forward-looking decision-making basis for enterprises' production plan arrangements, equipment maintenance, formulation of marketing strategies, inventory management optimization, etc., and significantly improving the operation efficiency, management level, and market competitiveness of enterprises.

[0064] The service layer is used to provide a group of RESTful API interfaces, integrate the Druid / ClickHouse query engine and Redis cache, support multi-condition combined queries and millisecond-level responses; deploy a visualization service module and an RBAC / ABAC permission control system to achieve hierarchical data sharing;

[0065] In this embodiment, a data service interface suite with rich functions and excellent performance is provided to meet the diverse data requirements of different departments within the enterprise and external partners. The data query service is designed based on the mature RESTful architecture, deeply integrating high-performance query optimization engines (such as Apache Druid, ClickHouse, Presto, etc.) and efficient caching mechanisms (such as Redis, Memcached, Caffeine, etc.). It can support complex combinations of query conditions, including but not limited to multi-field fuzzy queries, precise spatio-temporal range queries, multi-table association queries, full-text search queries, and complex condition filtering based on function expressions. Through functions such as paging queries, sorting queries, and aggregation queries, it realizes the efficient retrieval and flexible display of large-scale data. When processing query requests, the service layer can intelligently select the optimal data storage nodes, query paths, and index strategies according to the query conditions, and combine cache hit technology and query result prefetch mechanism to quickly return the query results to users within milliseconds, ensuring the efficiency, timeliness, and accuracy of data queries. Even if business personnel do not have a deep programming foundation, they can easily carry out data analysis work, deeply explore the value behind the data, greatly reducing the threshold and cost of data analysis, and improving the overall data application ability and innovation vitality of the enterprise through simple and intuitive drag-and-drop operations, parameter configuration, and SQL-style query statement writing. The data visualization service integrates professional visualization toolkits such as Echarts, Tableau, PowerBI, Plotly, etc. It can automatically select the most suitable visualization chart types (such as line charts, bar charts, pie charts, scatter plots, heat maps, Sankey diagrams, box plots, radar charts, word clouds, maps, etc.) and report templates (such as Excel reports, PDF reports, HTML dynamic reports, PPT presentations, etc.) according to the type, characteristics of the data, and user needs, and present the data to users in an intuitive, beautiful, and easy-to-understand form. At the same time, it supports multi-dimensional dynamic interaction functions such as data drilling, linked analysis, filtering, zooming and panning, and hover tips, enabling users to explore the information behind the data more deeply, discover potential laws and trends in the data, provide more intuitive and powerful support for enterprise decision-making, and promote the implementation of data-driven decision-making culture within the enterprise.

[0066] A secure, reliable, flexible, and manageable data sharing platform is built, adopting industry-leading identity authentication and authorization mechanisms such as OAuth 2.0 and JWT, combined with role-based access control (RBAC), attribute-based access control (ABAC), and token policies to strictly authenticate access requests from different internal departments and external partners of the enterprise and perform fine-grained permission management. For example, according to the internal organizational structure, department responsibilities, and job roles within the enterprise, different data access permissions are assigned to different users or user groups through an access control list (ACL). For instance, front-line operators in the production department can access equipment operation data and process parameter data related to production tasks, but only have read-only permissions for financial data, customer sensitive information, etc.; while professional staff in the finance department can perform read and write operations on financial data and access some production data and sales data with restrictions according to business needs, but have no access to the core technical materials of the R & D department. For external partners, corresponding data access permissions are dynamically granted according to the scope of cooperation projects, business needs, and confidentiality agreements. Through technical means such as token validity period control, access frequency limitation, and data desensitization processing, it is ensured that data can be efficiently shared and circulated on the premise of being secure and controllable, promoting the collaborative innovation and development of the enterprise ecosystem, enhancing the competitiveness and influence of the enterprise in the industrial chain, and maximizing the release of data value.

[0067] The application layer is used to develop production optimization applications, supply chain management applications, and CRM applications, and is docked with the ERP / MES system through a microservices architecture and an API gateway to form a closed-loop for functions such as equipment fault prediction, process parameter optimization, and customer segmentation recommendation.

[0068] In this embodiment, the production optimization application includes a digital twin modeling module, a genetic algorithm optimization engine, and a predictive maintenance dashboard;

[0069] The digital twin modeling module is used to update model parameters based on ANSYS simulation data and real-time sensor feedback;

[0070] The genetic algorithm optimization engine is used to solve the optimal process parameter combination with the minimum energy consumption as the objective function;

[0071] The predictive maintenance dashboard is used to display the remaining useful life (RUL) prediction value and the spare parts inventory linkage information.

[0072] In this embodiment, the supply chain management application includes a supplier evaluation index system, a dynamic procurement strategy module, and a logistics path optimization module;

[0073] The supplier evaluation index system is used to obtain the on-time delivery rate, the PPM value of quality defects, and the price fluctuation coefficient;

[0074] The dynamic procurement strategy module is used to obtain strategies based on the combinatorial auction algorithm and support the solution of multi-objective constraints;

[0075] The logistics path optimization module is used to integrate the map API to calculate the optimal delivery route in real time.

[0076] In this embodiment, the CRM application includes a customer life cycle value prediction model, a work order semantic analysis module, and a personalized recommendation engine;

[0077] The customer life cycle value prediction model is constructed using the XGBoost ensemble learning framework;

[0078] The work order semantic analysis module is constructed based on Transformer to automatically classify customer complaint types;

[0079] The personalized recommendation engine is used to generate differentiated marketing strategies according to the RFM score.

[0080] In the specific implementation, based on the powerful data service and analysis capabilities of the data center, a series of targeted and practical enterprise applications have been developed, covering multiple key business areas such as production, supply chain, and customer relationship management, providing strong support for the digital transformation and innovative development of enterprises. The production optimization application uses big data analysis technology and machine learning algorithms to collect, deeply analyze, and intelligently mine the massive data generated during the production process in real time. By building a digital twin model of the production process, it realizes real-time monitoring of production equipment, fault prediction, performance optimization, and dynamic adjustment of production plans. For example, by deploying a large number of sensors on production equipment, the operating parameters of the equipment (such as temperature, pressure, vibration, current, etc.), processing technology parameters (such as cutting speed, feed rate, machining depth, etc.), and product quality data (such as dimensional accuracy, surface roughness, physical performance indicators, etc.) are collected in real time. Using machine learning algorithms (such as support vector machines, decision trees, neural networks, etc.) to construct an equipment fault prediction model, it can predict possible equipment failures hours or even days in advance and send early warning notifications in a timely manner, providing sufficient preparation time for equipment maintenance and effectively reducing equipment downtime and improving production efficiency. At the same time, through real-time analysis and optimization of production process data, using intelligent optimization algorithms such as genetic algorithms and particle swarm optimization algorithms, production parameters are adjusted, production processes are optimized, product quality and production efficiency are improved, production costs are reduced, and the competitiveness of enterprises in the market is enhanced. The supply chain management application integrates various data of upstream and downstream enterprises in the supply chain, including supplier information, raw material inventory data, logistics distribution information, production plan data, and market demand forecast data, etc. Through the use of predictive analysis models and intelligent optimization algorithms, it realizes functions such as visualization management of the supply chain, inventory optimization, supplier evaluation, and collaborative procurement. For example, by analyzing historical sales data, market trend data, and macroeconomic data, using time series analysis and machine learning algorithms (such as ARIMA, LSTM, Prophet, etc.) to construct a demand prediction model, it can accurately predict the demand quantity, time node, and price fluctuation trend of raw materials, providing a scientific basis for the enterprise's procurement plan and inventory management, realizing the dynamic balance of inventory, and reducing inventory costs. At the same time, by comprehensively evaluating data such as the delivery period, product quality, price, and service level of suppliers and establishing a supplier evaluation model (such as based on the analytic hierarchy process, fuzzy comprehensive evaluation method, TOPSIS method, etc.), it can select high-quality suppliers for cooperation, optimize procurement strategies, improve the stability and reliability of the supply chain, and enhance the overall operating efficiency of the enterprise. The customer relationship management application, relying on the data center's in-depth analysis capabilities of customer data, uses customer segmentation models, customer lifetime value analysis models, and personalized recommendation algorithms to achieve precise customer segmentation, precise push of marketing activities, and personalized customization of customer services, improving customer satisfaction and loyalty, promoting the enhancement of customer value and the growth of the enterprise's sales performance.For example, by analyzing data such as customers' basic information, purchase behavior, preferences, location, and consumption frequency, and using clustering analysis, decision tree algorithms, RFM models, etc., customers are segmented into different groups. Based on the characteristics and needs of each segment, personalized marketing plans and service strategies are developed, such as precisely pushing product information, providing exclusive promotional activities, and customizing personalized services, to improve customers' response rate and purchase conversion rate, enhance the stickiness between customers and the enterprise, and achieve win-win development for both the enterprise and customers.

[0081] During the application development process, emphasis is placed on deep integration and fusion with the enterprise's existing business systems. The microservices architecture and API gateway technology are adopted to seamlessly embed the capabilities of the data middle platform into the enterprise's business processes in the form of services. For example, in the enterprise's ERP system, when a sales order changes, the demand forecasting and production plan optimization services of the data middle platform are triggered through API calls. The data middle platform conducts real-time analysis and calculation based on the latest sales order data, inventory data, production capacity data, equipment status data, and supply chain data, etc., quickly generates an optimized production plan, and feeds it back to the ERP system to achieve automated and intelligent-driven business processes, improving the enterprise's operation efficiency and response speed. In the MES system, the data middle platform can push the optimized suggestions for equipment operation parameters and process adjustment instructions to the terminal devices at the production site in real time to guide the operators to perform precise operations and ensure the efficient and stable operation of the production process. In the CRM system, customers' behavior data can flow into the data middle platform in real time.

[0082] Example 2:

[0083] Data Collection and Storage

[0084] For example, in an automobile manufacturing enterprise, the data collection layer connects various sensors on the production line through the sensor data interface to collect data such as torque, temperature, and vibration during the assembly process of automobile engines in real time, with a sampling frequency of 10 times per second; connects to the enterprise's ERP system database through the database connection interface to obtain order information (including order quantity, vehicle model, delivery date, etc.), inventory information (raw material inventory, spare part inventory, etc.), and financial information (cost data, revenue data, etc.), and updates the data regularly every day; receives the design documents of new automobile models (CAD drawings, technical specification manuals, etc.) from the design department through the file transfer interface.

[0085] The collected data is transmitted to the data storage layer. The real-time data during the engine assembly process is stored in specific partitions in the data lake in its original format, and partition management is carried out according to date and production line; structured data such as orders and inventory is stored in the distributed database HBase, and unstructured data such as design documents is stored in the distributed file storage system HDFS. At the same time, a data warehouse is built on top of the data lake to integrate and model data according to subject areas such as production, sales, and finance. For example, in the production subject area, the engine assembly data is associated with the production order data to form a production fact table containing fields such as order number, engine model, assembly time, and quality inspection results.

[0086] Data Processing and Analysis

[0087] The large model in the data processing layer uses a pre-trained model based on the Transformer architecture, which is pre-trained on large-scale datasets such as the enterprise's historical production data, sales data, and equipment operation data, enabling it to learn the basic characteristics and business logic of manufacturing data. Then, for the specific needs of the automotive manufacturing enterprise, the model is fine-tuned using the enterprise's internal labeled data (such as equipment failure records, product quality defect data, etc.).

[0088] During the data processing, the large model cleans the engine assembly data, automatically identifies and corrects outliers caused by sensor failures or interference, such as correcting data that significantly deviates from the normal torque range; performs data fusion, associates the engine assembly data with production order data and equipment maintenance data, and analyzes the relationship between engine assembly quality and equipment maintenance cycle under different order batches; performs data classification and prediction, predicts the probability of engine failure within a certain period in the future based on historical data, and the change trend of the order volume of different vehicle models in the next quarter.

[0089] Service Provision and Application Development

[0090] The service layer develops a data query service through which production managers can query the current operating status data of the engine assembly production line in real time (such as the real-time output of each workstation, equipment utilization rate, etc.); the data analysis service provides algorithm tools such as clustering analysis and regression analysis, and quality engineers can use these tools to analyze engine quality data and identify key factors affecting quality; the data visualization service displays production data in the form of a dashboard, including production trend charts, equipment failure rate bar charts, etc., facilitating management to intuitively understand the production situation.

[0091] The application layer develops production optimization applications based on the data middle platform. According to the analysis results of the large model on the engine assembly data, it automatically adjusts the assembly process parameters (such as torque tightening values, assembly sequences, etc.), optimizes the production process, improves the assembly efficiency and product quality; develops supply chain management applications, uses the data middle platform to integrate supplier data, inventory data, and production plan data, realizes accurate prediction of raw material procurement and dynamic optimization of inventory, and reduces supply chain costs; develops customer relationship management applications, segments customers by analyzing customer order data and market feedback data, formulates personalized marketing plans, and improves customer satisfaction and loyalty.

[0092] System Integration and Business Applications

[0093] Deeply integrate the large model-based data middle platform with the enterprise's existing ERP system, MES system, and CRM system. In the ERP system, when the sales department enters a new car order, the data middle platform immediately obtains the order information, conducts demand forecasting and production plan optimization, and feeds back the optimized production plan to the ERP system; in the MES system, the data middle platform pushes the optimization suggestions for the engine assembly process parameters to the operators at the production site in real time to guide the production process; in the CRM system, according to the analysis results of the customer relationship management application, accurate marketing is carried out for customers to improve the sales conversion rate. Through this integration, seamless connection between the data middle platform and business systems is achieved, and the overall digital operation ability of the enterprise is improved.

[0094] The technical solution of the second aspect: As Figure 2 shown, an application method applied to a data middle platform system for manufacturing enterprises based on a large model as described in any one of the above, includes the following steps:

[0095] Step S1: Through the interfaces and tools of the data collection layer, according to the predetermined collection strategy, collect data in real time or at regular intervals from various data sources, and transmit the data to the data storage layer;

[0096] Step S2: The data storage layer receives the collected data, stores it in the data lake or data warehouse of the corresponding data storage layer, and conducts preliminary preprocessing;

[0097] Step S3: Use the large model of the data processing layer to process and analyze the stored data, which includes data cleaning, fusion, classification, and prediction, and generate data insights and knowledge;

[0098] Step S4: The service layer provides the processed data and analysis results to the application layer and external systems through service interfaces, and at the same time realizes data sharing;

[0099] Step S5: The application layer develops applications based on the services of the data middle platform, integrates with the business system, and applies the value of the data middle platform to actual business operations.

[0100] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents, and all should be included within the protection scope of the present invention.

Claims

1. A data middle platform system for manufacturing enterprises based on large models, characterized in that, It includes a data acquisition layer, a data storage layer, a data processing layer, a service layer, and an application layer; The data acquisition layer is used to construct a multi-source heterogeneous data acquisition system. Among them, the data acquisition system includes an industrial Ethernet interface and a fieldbus protocol adaptation module, a highly compatible database connection module, and an intelligent file parsing engine; The data storage layer is used to build a multi-layer data warehouse system by using a collaborative architecture of a distributed file system and a distributed database, and realize data lake storage in Parquet / ORC columnar format; The data processing layer is used to deploy a pre-trained AI large model, integrate a deep anomaly detection module, a multi-modal fusion engine, and a prediction model; configure a transfer learning framework to realize model fine-tuning, and associate device parameters, process data, and quality inspection results to form a decision knowledge base; The service layer is used to provide a RESTful API interface group, integrate Druid / ClickHouse query engines and Redis caches, support multi-condition combined queries and millisecond-level responses; deploy a visualization service module and an RBAC / ABAC permission control system to realize hierarchical data sharing; The application layer is used to develop production optimization applications, supply chain management applications, and CRM applications, and dock with the ERP / MES system through a microservices architecture and an API gateway to form a closed loop of functions such as equipment failure prediction, process parameter optimization, and customer segmentation recommendation.

2. The data middle platform system for manufacturing enterprises based on a large model according to claim 1, wherein The data acquisition layer includes a lightweight acquisition agent, a real-time data pipeline, and a log audit module; The lightweight acquisition agent is used to be deployed at the device edge, integrate a ZigBee / Modbus protocol conversion module, and support low-power transmission of sensor data; The real-time data pipeline is built based on Apache Kafka, uses the Snappy compression algorithm and AES-256 end-to-end encryption for industrial data stream transmission with millisecond-level latency; The log audit module is used to record the acquisition task ID, data volume statistics, and CRC check information, and support breakpoint resumption and traceability of abnormal tasks.

3. According to the large model-based manufacturing enterprise data middle platform system of claim 1, it is characterized by: The anomaly detection module of the data processing layer includes a vibration signal time series prediction model and an anomaly root cause analysis module; The vibration signal time series prediction model is built based on a GRU network, sets dynamic threshold alarm rules, and uses a GAN network to generate synthetic anomaly samples; The anomaly root cause analysis module conducts multi-dimensional traceability by associating SPC control charts with process parameter fluctuations.

4. The data middle platform system for manufacturing enterprises based on large models according to claim 1, wherein The production optimization application includes a digital twin modeling module, a genetic algorithm optimization engine, and a predictive maintenance dashboard; The digital twin modeling module is used to update model parameters based on ANSYS simulation data and real-time sensor feedback; The genetic algorithm optimization engine is used to solve the optimal process parameter combination with the minimum energy consumption as the objective function; The predictive maintenance dashboard is used to display the predicted remaining useful life (RUL) value and spare parts inventory linkage information.

5. The data middle platform system for manufacturing enterprises based on large models according to claim 1, characterized in that The supply chain management application includes a supplier evaluation index system, a dynamic procurement strategy module, and a logistics path optimization module; The supplier evaluation index system is used to obtain the on-time delivery rate, the PPM value of quality defects, and the price fluctuation coefficient; The dynamic procurement strategy module is used to obtain strategies based on the combinatorial auction algorithm and support multi-objective constraint solving; The logistics path optimization module is used to integrate the map API to calculate the optimal distribution route in real time.

6. The data middle platform system for manufacturing enterprises based on large models according to claim 1, wherein The CRM application includes a customer lifetime value prediction model, a work order semantic analysis module, and a personalized recommendation engine; The customer lifetime value prediction model is constructed using the XGBoost integrated learning framework; The work order semantic analysis module is constructed based on Transformer and automatically classifies customer complaint types; The personalized recommendation engine is used to generate differentiated marketing strategies according to the RFM score.

7. A method of application for a data middle platform system of manufacturing enterprises based on large models as described in any one of claims 1 to 6, characterized in that, It includes the following steps: Step S1: Through the interfaces and tools of the data collection layer, according to the predetermined collection strategy, collect data in real time or at regular intervals, obtain data from various data sources, and transmit the data to the data storage layer; Step S2: The data storage layer receives the collected data, stores it in the data lake or data warehouse of the corresponding data storage layer, and performs preliminary preprocessing; Step S3: Use the large model of the data processing layer to process and analyze the stored data, which includes data cleaning, fusion, classification, and prediction, and generate data insights and knowledge; Step S4: The service layer provides the processed data and analysis results to the application layer and external systems through service interfaces, and at the same time realizes data sharing; Step S5: The application layer develops applications based on the services of the data middle platform, integrates with the business system, and applies the value of the data middle platform to the actual business.

Images