A digital production operation management method for data quality assessment and analysis

Abstract

The present invention discloses a digital production operation management method for data quality assessment and analysis, including: step 1: constructing a multi-level distributed data acquisition system; step 2: designing a multimodal data fusion framework, combining graph neural network and time series data processing technology to realize data fusion and multidimensional analysis; step 3: constructing an intelligent scheduling engine to realize dynamic adaptive scheduling strategy to cope with real-time changes in the production environment; step 4: designing a decentralized self-organizing production network to realize flexible adjustment and optimization of the production process; step 5: realizing the coordinated optimization of product design and equipment maintenance through topology optimization, multi-objective optimization and machine learning algorithms; step 6: constructing a virtual production network to realize real-time monitoring, dynamic prediction and optimization decision support of the production system. Through the present invention, enterprises can break the limitations of traditional production models, improve production efficiency, flexibly respond to changes in market demand, and optimize resource allocation.

Description

Technical Field

[0001] The present invention relates to the technical field of resource management, and in particular to a digital production operation management method for data quality assessment and analysis. Background Art

[0002] With the advent of Industry 4.0, digital technology and intelligent management have gradually become the core driving force of modern production and operation management. Traditional production methods can no longer adapt to the increasingly complex market demand and production environment. Enterprises are facing growing production efficiency requirements and higher customization needs. In this case, how to improve the efficiency of data utilization in the production process and how to achieve flexibility and efficiency of the production process through precise analysis and management have become the key to enterprises gaining an advantage in global competition. The quality, accuracy and real-time nature of data play a vital role in production management, and how to extract valuable information from massive data and conduct scientific evaluation and analysis has become a difficult problem to improve productivity and corporate competitiveness.

[0003] Under the traditional production management model, production plans often rely on manual experience and historical data, and cannot quickly adapt to the rapid changes in market demand and fluctuations in the external environment. In addition, production systems often operate in an isolated manner, with insufficient coordination and optimization between various links, resulting in low production efficiency, serious waste of resources, and inability to meet flexible and changing production needs. As the manufacturing industry transforms to customized, small-batch, and flexible production, traditional manual intervention and quantitative production scheduling methods have become difficult to cope with complex and rapidly changing production scenarios.

[0004] Therefore, enterprises urgently need to use modern technology, especially data analysis and intelligent scheduling technology, to optimize each link in the production process and improve the accuracy and flexibility of resource allocation. Advanced data quality assessment and analysis methods can provide a more reliable basis for production decisions, ensuring that every decision in the production process is based on real-time and accurate data support. With the maturity of technologies such as big data, cloud computing and artificial intelligence, the intelligent management of production data has become more feasible, especially when the production process involves multiple systems and complex processes. The combination of intelligent scheduling and data analysis can significantly improve production efficiency and reduce operating costs. In this context, it is particularly important to create a production network that can self-organize and automatically adapt to changes in production needs through algorithms and intelligent scheduling. Summary of the invention

[0005] In order to solve the above problems, the present invention provides a digital production operation management method for data quality assessment and analysis.

[0006] To achieve the above purpose, the technical solution adopted by the present invention is as follows: A digital production operation management method for data quality assessment and analysis, comprising: Step 1: Build a multi-level distributed data collection system to collect multi-dimensional production data through sensors, cameras, and embedded devices, and combine edge computing technology for real-time transmission and preliminary processing to ensure high-quality data input; Step 2: Design a multimodal data fusion framework, combine graph neural networks with time series data processing technology, realize the fusion and multidimensional analysis of heterogeneous data such as sensor data, video images, and device logs, and mine potential data associations; Step 3: Based on the microservice architecture and event-driven mechanism, build an intelligent scheduling engine, combine optimization algorithms and reinforcement learning technology to implement dynamic adaptive scheduling strategies to cope with real-time changes in the production environment; Step 4: Use distributed autonomous systems and multi-agent technology to design a decentralized self-organizing production network, and achieve flexible adjustment and optimization of the production process through dynamic task allocation, resource coordination and evolution mechanism; Step 5: Combine generative design technology with predictive maintenance models to achieve collaborative optimization of product design and equipment maintenance through topology optimization, multi-objective optimization, and machine learning algorithms to improve the reliability and efficiency of the production system. Step 6: Build a virtual production network and combine it with digital twin technology to achieve real-time monitoring, dynamic prediction and optimization decision support of the production system, and guide the adjustment and optimization of the physical production system through a closed-loop feedback mechanism.

[0007] Further: Step 1 comprises: Through a multi-level distributed data acquisition system, sensors, cameras and embedded devices are used to collect multi-dimensional data from the production site. The sensor layer collects equipment operating status and environmental change data, the video monitoring layer obtains quality inspection visual data, and the embedded data acquisition layer transmits equipment operating data in real time. The data stream is transmitted to the edge computing node in real time through the transmission protocol. On the edge node, the raw data is preprocessed by noise filtering, formatting and standardization, and the self-supervised learning algorithm is used to identify and correct outliers and missing values, reducing the transmission burden and improving data real-time performance. Through data quality assessment and automatic correction mechanism, the collected data is dynamically scored and the data is automatically repaired.

[0008] Further: Step 2 comprises: Construct a multimodal data fusion framework, use graph neural network as the core model, abstract the equipment, production line, and material elements in the production process into graph nodes, and use graph convolution operations to propagate and fuse information on each data node; To solve the problem of integrating time series data and non-time series data in production environments, a spatiotemporal attention mechanism is introduced. Time series data is captured with long short-term memory networks or gated recurrent units, and features of non-time series data are extracted with convolutional neural networks. A multi-perspective joint modeling strategy is used to process heterogeneous data. Multiple data perspectives are defined according to different stages of the production process. The data from each perspective is integrated in the latent space through a deep generative model to extract common features and enhance the model's predictive ability. Perform multi-dimensional analysis and mining on the fused data, use frequent item set mining algorithm to discover association rules in the production process, perform trend prediction and anomaly detection through deep learning model, and set up real-time alarm mechanism.

[0009] Further: Step 3 includes: The scheduling engine is designed with a microservice architecture, which splits complex production scheduling tasks into multiple independent service modules. Each microservice exchanges data through message queues to ensure the flexibility and scalability of the system. At the same time, an event-driven mechanism is introduced so that changes in equipment status, material arrival, and order change events in the production process can trigger dynamic adjustments to scheduling strategies in real time. Build a scheduling decision system based on rule engine and machine learning model, combine business rules with scheduling strategy through rule engine, and realize rule-driven scheduling decision; A multi-objective optimization model is constructed using mixed integer linear programming and heuristic optimization algorithms; Design dynamic adaptive scheduling strategies, combine reinforcement learning and model predictive control, so that the scheduling system can continuously optimize the scheduling strategy based on real-time feedback.

[0010] Further: Step 4 includes: The distributed autonomous node design is adopted, and the equipment, production lines and warehouse units in the production network are regarded as autonomous nodes. Each node has independent decision-making and information processing capabilities. Through local perception and decision-making, the nodes can autonomously adjust their behavior to cope with environmental changes, thereby improving the flexibility and adaptability of the system. Introducing a multi-agent system model, taking nodes in the production network as intelligent agents, and achieving collaboration and coordination through local perception and information sharing; Design dynamic task allocation and resource coordination mechanism, and realize dynamic optimization allocation of tasks and resources through task allocation based on market mechanism and adaptive scheduling algorithm; Construct an adaptive evolution mechanism for self-organizing networks. Through evolutionary game theory and feedback mechanisms, the production network can self-adjust and evolve according to long-term changes and emergencies. Nodes participate in the game by optimizing their own strategies, and ultimately reach Nash equilibrium, achieving system stability and optimization.

[0011] Further: Step 5 comprises: Introducing generative design technology, using topology optimization, multi-objective optimization and artificial intelligence-assisted design systems to automatically generate efficient and manufacturable product design solutions; Establish a predictive maintenance model based on IoT technology and big data analysis, combined with machine learning algorithms to monitor equipment status in real time and predict faults. By analyzing the equipment's historical fault records and real-time operating data, the predictive maintenance model can identify potential faults in advance and reduce equipment downtime; During the design phase, the maintainability requirements of the equipment are embedded in the product design to optimize the maintainability and reliability of key components. At the same time, through real-time collection of equipment operation data and maintenance history, this information is fed back into the generative design system to achieve closed-loop optimization.

[0012] Further: Step 6 comprises: Build a virtual production network corresponding to the actual production system through physical simulation, system modeling and digital twin technology; Establish a data fusion and real-time monitoring system; use the Internet of Things technology to collect equipment status and production data information in the physical production system in real time, and integrate it with the virtual production network; use data fusion algorithms to pre-process multi-source data, eliminate noise and uncertainty, and ensure data accuracy and consistency; By utilizing the fusion results of virtual production networks and real-time data, dynamic prediction of production status and optimization decisions can be achieved. Through machine learning algorithms and optimization algorithms, equipment failures and demand fluctuations in the production process can be predicted, and production scheduling and resource allocation can be optimized.

[0013] Compared with the prior art, the present invention has the following technical advances: In the traditional production model, production decisions often rely on manual experience and historical data, which is highly subjective and has information lags. However, the digital management method of data quality assessment and analysis can ensure that high-quality and accurate information can be extracted from massive production data.

[0014] Modern manufacturing is facing rapidly changing market demands and production environments, which requires the production system to have the ability to respond quickly. By introducing intelligent scheduling algorithms and self-organizing production networks, the present invention can adapt to demand fluctuations in real time during the production process. This automated and intelligent adjustment capability enables the production system to quickly adapt to external changes, avoiding waste of production resources or idle equipment due to market fluctuations, thereby enhancing the flexibility of the production system.

[0015] Digital production operation management methods can accurately evaluate equipment operation status, production bottlenecks, and resource allocation efficiency through real-time monitoring and analysis of production data. Through intelligent scheduling and optimization algorithms, production schedules and equipment arrangements can be dynamically adjusted to eliminate bottlenecks in the production process and maximize the utilization of equipment and resources. Not only does it improve production efficiency, it also effectively reduces downtime, inventory backlogs, and production costs. In the traditional model, there may be incoordination between the various links on the production line, but through intelligent scheduling and optimization, the collaborative work of all links can achieve the best results.

[0016] In traditional production processes, decisions often rely on manual judgment and are easily affected by operator experience, subjective bias, or work pressure. In digital management methods, by introducing an automated intelligent scheduling system, human intervention and operational errors can be reduced, ensuring that every link in the production process can respond optimally based on real-time data. With the support of digital twin models, the production network can accurately predict potential problems and issue early warnings, thereby avoiding system failures or production accidents, thereby effectively reducing operational risks in the production process.

[0017] With the increasing demand for customized production, traditional production models are unable to efficiently cope with small-batch, multi-variety orders. Digital production operation management methods can not only optimize large-scale conventional production through the combination of data analysis and intelligent scheduling, but also flexibly adjust production plans and resource allocation according to different production needs to achieve customized production scheduling. For example, based on the analysis of the virtual production network, the production system can quickly respond to personalized orders, adjust the production line configuration, and achieve efficient customized production.

[0018] Through the combination of intelligent scheduling and algorithm models, the production system can not only make adjustments based on real-time data, but also form the ability to self-organize and automatically adapt throughout the production process. The production network will dynamically adjust its own operation mode and resource allocation mode based on real-time data flow, equipment status, production progress and other information to ensure that production activities can maintain efficient and stable operation in any environment. This adaptive ability not only improves production efficiency, but also greatly reduces the need for human intervention, making the production process more efficient and intelligent.

[0019] Ultimately, this approach enables companies to respond to rapid market changes with lower costs, higher efficiency and more flexible production methods through intelligent management, big data analysis and automated scheduling. Companies can quickly adjust production strategies, meet customers' personalized needs and improve market response speed while ensuring product quality and production efficiency, thereby occupying a favorable position in the fiercely competitive market.

[0020] Therefore, through the present invention, enterprises can break the limitations of traditional production models, improve production efficiency, flexibly respond to changes in market demand, and optimize resource allocation. Through the combination of intelligent scheduling and self-organizing capabilities, the production network can achieve automatic adjustment and adaptive optimization, maximize resource utilization, reduce operating costs, and ensure efficient and stable operation in a dynamic production environment, thereby enhancing the overall competitiveness of the enterprise. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0022] In the attached picture: Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0023] The following specific embodiments may be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments. The embodiments of the present invention will be described below in conjunction with the accompanying drawings.

[0024] like Figure 1 As shown, the present invention discloses a digital production operation management method for data quality assessment and analysis, comprising: Specifically, the specific implementation process of step 1 includes: 1.1 Construction of multi-level distributed data acquisition system Its purpose is to ensure that the required production data is collected in real time and accurately at different levels and links of the production network, forming multi-dimensional data input.

[0025] The implementation process includes: deploying a variety of data acquisition devices including sensors, cameras, embedded devices, etc., connecting and integrating them based on the IoT platform to form a real-time data stream. The specific steps are as follows: Sensor layer: Install sensors (such as temperature, pressure, vibration, humidity, load, etc.) on key equipment at the production site (such as robotic arms, conveyor belts, robots) to collect real-time data on equipment operating status and environmental changes during the process.

[0026] Video monitoring layer: Use industrial-grade cameras to collect visual data on the production line, especially images and videos of quality inspection links, to form data related to product appearance and processing accuracy.

[0027] Equipment embedded data acquisition layer: embed intelligent hardware (such as embedded computers, intelligent controllers, etc.) in production equipment, collect equipment operation data through built-in sensors, and connect to the data processing center in real time through wireless networks.

[0028] Through the above-mentioned multi-level collection system, the diversity and comprehensiveness of data sources are ensured, providing sufficient information for subsequent processing.

[0029] 1.2 Real-time transmission of data streams and preliminary processing of edge computing The purpose is to ensure that data collected from field devices can be quickly and seamlessly transmitted to the data processing system and initially processed through edge computing technology to reduce latency and improve response speed.

[0030] The implementation process includes: using edge computing nodes and data stream processing platforms to reduce the transmission cost of large amounts of data through near-end processing and provide real-time feedback. The specific steps are as follows: Selection and optimization of data transmission protocols: Data collected by sensors and video devices are transmitted to edge computing nodes in the form of real-time streams through wireless sensor networks or industrial Ethernet, and efficient transmission protocols (such as MQTT and CoAP) are used to ensure the stability and timeliness of data streams.

[0031] Data preprocessing of edge computing nodes: On edge nodes, the collected raw data is first preprocessed by noise filtering, formatting, and standardization. Edge computing nodes can apply basic data cleaning techniques (such as mean interpolation and outlier identification) to specific data types (such as temperature, pressure, etc.) and perform preliminary statistical analysis, thereby reducing unnecessary transmission burden and ensuring real-time performance.

[0032] The addition of the edge computing layer greatly improves the processing efficiency of real-time data, while reducing the delay and bandwidth pressure caused by large-scale data transmission.

[0033] 1.3 Data quality assessment and automatic correction Its purpose is to ensure that the quality of the collected data meets the requirements of production management, eliminate data that does not meet the quality standards, and automatically correct missing or abnormal data, thereby providing accurate input for subsequent data analysis and decision-making.

[0034] The implementation process includes: using data quality assessment algorithms, combining machine learning and self-supervised learning models to dynamically assess data quality, and using adaptive algorithms to repair data. The specific steps are as follows: Data quality scoring system: Based on the data quality assessment framework (such as ISO8000 standard or data quality scoring model), a comprehensive "data quality scoring system" is designed to evaluate the collected data in real time. This system will score the data based on the accuracy, completeness, consistency, timeliness and other dimensions of the data. If the score is lower than a certain threshold, it will be regarded as low-quality data.

[0035] Anomaly detection and correction based on self-supervised learning: Use anomaly detection algorithms based on self-supervised learning (such as autoencoders, variational autoencoders, etc.) to automatically identify abnormal values ​​and missing values ​​in sensor data, and fill and repair them through intelligent algorithms (such as K nearest neighbors, regression models, etc.). For example, for missing temperature sensor data, the system will automatically infer the missing value based on historical data and other states of the device to ensure data integrity.

[0036] Data quality feedback mechanism: Once the system identifies a data quality problem, it automatically triggers the feedback mechanism to feed back abnormal data to the edge computing node or on-site technicians for timely equipment calibration or replacement to prevent the abnormal data from further affecting the production process.

[0037] This step ensures that the collected data is of high quality and can meet the requirements of subsequent data analysis and decision-making, thus providing a reliable foundation for intelligent scheduling and production optimization.

[0038] 1.4 Data Integration and Data Warehouse Construction Its purpose is to integrate and archive high-quality data that has been pre-processed and corrected, and to provide historical data support for subsequent analysis and decision-making.

[0039] The implementation process includes: using a data integration platform and a distributed database system to integrate the multimodal data collected at each layer and store them in an efficient database system. The specific steps are as follows: Multi-source data integration platform: Use modern data integration technologies (such as ETL (Extract-Transform-Load), data virtualization, etc.) to uniformly process data from different sources (sensors, equipment, video surveillance, etc.). Use streaming processing technologies (such as Apache Kafka or Apache Flink) to batch import and synchronize data in real time to ensure that data can be updated in real time and accurately reflected in the data warehouse.

[0040] Distributed data warehouse construction: Deploy distributed databases (such as Hadoop, Apache Hive, Click House, etc.) on cloud platforms or locally to provide efficient query and processing capabilities for data storage. Optimize data storage efficiency and access speed through data partitioning, indexing and compression technologies, and provide convenient support for subsequent data analysis through multidimensional analysis model (OLAP) design.

[0041] Data integration and data warehouse construction will centralize and structure data storage, providing data basis for subsequent multi-dimensional analysis, predictive maintenance and intelligent scheduling.

[0042] 1.5 Continuous monitoring and data quality tracking Its purpose is to ensure continuous monitoring of the data collection process, timely detect and correct any problems in data collection, and ensure that the data quality is always at an acceptable level.

[0043] The implementation process includes: using real-time monitoring system and data quality tracking technology to establish a dynamic data quality tracking and alarm mechanism. The specific steps are as follows: Data monitoring platform: Build a real-time monitoring system based on a big data platform (such as Apache Spark, Prometheus, etc.) to monitor abnormal situations in the data collection, processing and transmission process in real time, such as data transmission delays, packet loss, quality degradation, etc.

[0044] Data quality tracking and feedback mechanism: By establishing a data quality tracking system, the quality changes of each data collection link are automatically recorded, and the "data lineage" technology is used to track the source, change and correction records of each piece of data, so as to quickly trace the problem and make timely intervention. Through continuous monitoring and data quality tracking, it is ensured that the data collected in the production process meets the requirements at each link, potential problems are discovered and adjusted in time, and subsequent decisions are ensured to be based on high-quality data.

[0045] Through the integration and innovation of these technical means, the entire process of data collection and preprocessing can be carried out efficiently and accurately, laying a solid foundation for subsequent data analysis, intelligent scheduling and production optimization. The collected multi-level and multi-modal data will undergo strict quality assessment and correction to ensure the integrity, accuracy and timeliness of the data, and achieve efficient storage and management through data integration platforms and data warehouses, ultimately providing strong data support for the intelligent scheduling and self-optimization of digital production networks.

[0046] Specifically, the specific implementation process of step 2 includes: 2.1 Multimodal Data Fusion Framework Design The goal is to design a fusion framework that can process data from different data sources (such as sensor data, device logs, video images, external environment information, etc.), and enable different forms of data to complement each other while maintaining their unique characteristics to form a unified data representation.

[0047] The implementation process includes: using graph neural network as the core model of data fusion, modeling multi-source data through graph structure, and using graph convolution operation to propagate and fuse information on each data node. The specific steps are as follows: Construct a production network diagram: abstract the various elements in the production process (such as equipment, production lines, materials, environment, etc.) into nodes of the diagram, and model the relationships between them (such as material flow, equipment interaction, environmental impact, etc.) as edges to form a "graph model" of the production process. Nodes can include data sources of different modalities, such as sensor data, video surveillance data, etc.

[0048] Graph convolutional networks propagate and fuse node information: The relationship between nodes is modeled through graph convolutional layers, integrating feature information from different data sources. For example, the temperature, humidity and other data collected from sensors by graph neural networks are combined with the appearance information in video surveillance data for multimodal fusion. The representation of each node is dynamically updated based on the status of its neighboring nodes, thereby obtaining a node feature that comprehensively considers various types of information.

[0049] Multimodal graph neural network extension: For complex multimodal data, a multimodal graph neural network can be further introduced. This model not only transfers information between nodes on the graph structure, but also encodes data of different modalities independently and then combines them. In this way, different modalities (such as sensor data and image data) can effectively interact and transfer information through a shared graph structure.

[0050] 2.2 Fusion processing of time series data and non-time series data The purpose is that in a production environment, a lot of data is time-series (such as equipment sensor data, production process monitoring data, etc.), while some data is non-time-series (such as product quality inspection images, environmental information, etc.). How to effectively fuse these time-series and non-time-series data is a key issue.

[0051] The implementation process includes: using the spatiotemporal attention mechanism for intelligent selection and weighted fusion of data. The specific steps are as follows: Time series data processing: For time series data, such as sensor temperature and pressure, long short-term memory networks or gated recurrent units are used to capture long-term dependencies in time series. These models can learn the changing trends and periodic fluctuations of data in the time dimension.

[0052] Non-time series data processing: For non-time series data, such as images, device status information, etc., convolutional neural networks can be used to extract their deep features. The features of image data can be extracted through pre-trained ResNet or EfficientNet networks, and device status data can be classified and processed through traditional machine learning models (such as support vector machines SVM).

[0053] Spatiotemporal attention mechanism: The spatiotemporal attention mechanism is introduced to weightedly fuse the features of time series data and non-time series data. For time series data, the spatiotemporal attention mechanism focuses on important time periods by dynamically adjusting the attention weights, while for non-time series data, the attention layer of the spatial dimension is used to determine which features are most critical to the current production status. This mechanism can flexibly adjust the contribution of different data sources to achieve accurate fusion.

[0054] 2.3 Joint Modeling of Heterogeneous Data Its purpose is to effectively process heterogeneous data from different production environments, equipment types, and sensor types, ensuring that different data dimensions can work together to ultimately form a unified knowledge representation.

[0055] The implementation process includes: adopting a multi-perspective joint modeling strategy to model heterogeneous data from different perspectives and integrate them in the model. The specific steps are as follows: Perspective definition and data preprocessing: According to different stages of the production process, multiple data perspectives are defined (such as production equipment perspective, production line perspective, quality control perspective, etc.), and feature extraction and normalization are performed on the data of each perspective. For example, the data in the production equipment perspective may include temperature, vibration, load, etc., while the data in the quality control perspective includes product images, quality scores, etc.

[0056] Multi-view fusion method: Based on deep generative models (such as variational autoencoders (VAEs), generative adversarial networks (GANs), etc.), a multi-view joint learning framework is designed to fuse data from each view in the latent space. This method can integrate the common features of different types of data in the model's latent space, thereby enhancing the model's predictive ability.

[0057] For example, using a variational autoencoder to map data from different views into a shared latent space:

[0058]

[0059] in, is a hidden variable, is the input data (including data from different perspectives), Represents hidden variables The prior distribution of The distribution of Represents the variational posterior distribution, which means that given the input data Under the condition of The distribution of are the mean and variance extracted from each view, is the parameter of the latent space learned by the generative model. By optimizing the latent space, the joint modeling and fusion of heterogeneous data can be achieved.

[0060] 2.4 Multidimensional Analysis and Mining of Fusion Data The purpose is to conduct in-depth analysis of the fused data, discover the potential correlations and patterns between the data, and provide insights for production decisions.

[0061] The implementation process includes: using multidimensional data analysis and deep learning models to perform complex pattern recognition, trend prediction, anomaly detection and other tasks after processing and fusion data. The specific steps are as follows: Association rule mining: Mining association rules in fused data based on frequent item set mining algorithms (such as Apriori or FP-Growth) to help discover potential correlation factors in the production process (for example, increased temperature may lead to an increased risk of equipment failure).

[0062] Deep learning prediction model: For trend prediction and pattern recognition in data, deep neural networks, convolutional neural networks or recurrent neural networks can be used for modeling. These models can extract complex nonlinear relationships between data when processing high-dimensional data, thereby providing accurate predictions for production scheduling, quality control, etc.

[0063] Anomaly detection and automatic alarm: Based on the fusion results of multimodal data, self-supervised learning or clustering-based anomaly detection methods (such as DBSCAN, Isolation Forest) are used to identify abnormal patterns in the production process, and a real-time automatic alarm mechanism is set up to ensure a quick response when problems arise in the production process.

[0064] Through the above steps, data fusion and multimodal analysis not only effectively solve the problem of integrating multi-source heterogeneous data, but also through innovative spatiotemporal attention mechanisms and multi-perspective joint modeling methods, the deep-level associations between data can be fully mined and converted into valuable information, further promoting the realization of intelligent decision-making. Through this method, data of different modalities can form a more comprehensive and accurate data model while maintaining their respective characteristics, providing strong data support for the optimization of production processes and production scheduling.

[0065] Specifically, the specific implementation process of step 3 includes: 3.1 Intelligent Scheduling Engine Architecture Design The goal is to design an engine architecture that can process multiple data inputs in the production environment in real time and dynamically adjust production plans and scheduling strategies based on multiple factors such as production demand, equipment status, order priority, etc.

[0066] The implementation process includes: using a scheduling engine framework based on microservice architecture and event-driven mechanism, combining rule engine and machine learning model to optimize the scheduling process. The specific steps are as follows: Microservice architecture design: The scheduling engine should use a microservice architecture to split complex production scheduling tasks into multiple independent services for flexible expansion and maintenance. Each microservice module is responsible for different tasks, such as production demand forecasting, equipment status monitoring, order management, etc. It runs independently and exchanges data through message queues (such as Kafka).

[0067] Event-driven mechanism: Through the event-driven mechanism, each node in the production process (such as equipment status change, material arrival, order change, etc.) will trigger related events, and the scheduling engine will dynamically adjust the scheduling strategy according to the type of event. For example, when the equipment fails, the system will trigger the "equipment failure" event, and the scheduling engine will automatically adjust the priority of the production task.

[0068] Rule engine and decision logic: Combined with rule engines (such as Drools), business rules and scheduling strategies are combined to achieve rule-driven scheduling decisions. Specifically, a series of scheduling rules can be set, such as "prioritize urgent orders" and "reduce workload when equipment load exceeds 80%", etc., to automatically select the appropriate production plan based on real-time data.

[0069] This microservice and event-driven architecture design enables the intelligent scheduling engine to respond quickly to external changes, and has high scalability and flexibility to adapt to the needs of dynamic production environments.

[0070] 3.2 Construction of Scheduling Decision Model Based on Optimization Algorithm Its purpose is to use optimization algorithms to provide optimal scheduling decisions for the production process under multi-objective conditions, so as to improve production efficiency, reduce costs, shorten production cycles and ensure product quality.

[0071] The implementation process includes: using mixed integer linear programming (MILP) and heuristic optimization algorithms (such as genetic algorithms and simulated annealing algorithms) to deal with complex production scheduling problems, especially when facing multiple optimization objectives and constraints. The specific steps are as follows: Multi-objective optimization model: In production scheduling, multiple optimization objectives usually need to be considered, such as minimizing production costs, maximizing resource utilization, minimizing delivery time, etc. This embodiment designs the following multi-objective optimization model:

[0072] in, is the vector of objective functions, indicating multiple objectives that need to be optimized simultaneously. , ,..., are different objective functions, X is the scheduling decision variable (such as task allocation, equipment scheduling, etc.), , ,..., is the weight of each objective, and the optimization problem seeks to find the optimal scheduling solution after comprehensively considering all objectives.

[0073] Introduction of constraints: Production scheduling problems should not only focus on the optimization of the objective function, but also consider various constraints, such as production capacity, material supply, equipment availability, etc. Constraints can be expressed by linear or nonlinear equations:

[0074] in, Represents constraints, such as material supply constraints, equipment capacity constraints, etc. The system needs to meet these constraints at the same time and perform optimal scheduling on this basis.

[0075] Application of heuristic algorithms: For large-scale and complex production scheduling problems, traditional optimization methods may be too computationally intensive. Therefore, heuristic algorithms such as genetic algorithms, particle swarm optimization, or simulated annealing can be used. These algorithms can quickly search in a multi-constraint, multi-objective environment to find a scheduling solution that is close to the optimal solution.

[0076] The basic idea of ​​genetic algorithm is to simulate natural selection and genetics principles, construct the initial population, and then continuously generate a new generation of population through operations such as selection, crossover, and mutation, and finally obtain a better production scheduling plan.

[0077] The particle swarm optimization algorithm simulates the foraging process of a flock of birds. It continuously adjusts the position of particles by moving them in the search space and finally converges to the global optimal solution.

[0078] By introducing these optimization algorithms, it is possible to handle complex constraints in various production scheduling and ensure the efficiency and feasibility of scheduling results.

[0079] 3.3 Dynamic Adaptive Scheduling Strategy The purpose is to enable the intelligent scheduling system to automatically adjust the scheduling plan according to the real-time changing production environment (such as equipment failure, production bottlenecks, sudden order demand, etc.), thereby improving the flexibility and adaptability of the production network.

[0080] The implementation process includes: combining reinforcement learning with adaptive scheduling algorithms to build an intelligent decision-making system that can continuously optimize scheduling strategies based on real-time feedback. The specific steps are as follows: Reinforcement learning model design: Design a scheduling optimization framework based on reinforcement learning, in which the scheduling engine is an intelligent agent whose task is to select the optimal action (i.e., scheduling decision) based on the state of the environment (such as production progress, equipment status, order priority, etc.). The intelligent agent continuously obtains reward signals (such as production efficiency, delivery time, quality control, etc.) through interaction with the environment, and optimizes the strategy based on these feedbacks.

[0081] The core goal of reinforcement learning is to maximize the cumulative reward. The model can be expressed as:

[0082] in, For the status Next action The value of For instant rewards, is the discount factor, is the new state to which the action is transferred after execution. Select the next action.

[0083] Adaptive scheduling mechanism: In combination with environmental changes (such as equipment failures, fluctuations in production demand, etc.), a scheduling mechanism based on model predictive control (MPC) is adopted. By acquiring production data in real time and predicting future production status, the scheduling strategy is adjusted to cope with sudden changes. For example, when the system predicts that equipment may fail, the scheduling engine will automatically adjust the task schedule and prioritize the tasks to production lines with better equipment status.

[0084] 3.4 Real-time feedback and optimization of scheduling results Its purpose is to dynamically evaluate the scheduling effect and further optimize the scheduling decision through real-time monitoring and feedback mechanism of scheduling results.

[0085] The implementation process includes: using closed-loop control mechanisms and data-driven optimization adjustments to enable the scheduling engine to continuously optimize the scheduling plan based on feedback from the production process. The specific steps are as follows: Real-time feedback system: Through the real-time data monitoring system (such as SCADA system), the scheduling results are continuously tracked, and the actual production situation (such as equipment operating status, material supply status, production progress, etc.) is input into the scheduling engine through the feedback mechanism for dynamic adjustment.

[0086] Self-learning and model updating: Self-adjust the scheduling strategy through machine learning algorithms (such as online learning and incremental learning). As production data continues to accumulate, the scheduling engine can gradually improve the decision model and improve scheduling accuracy.

[0087] By building a highly dynamic and intelligent scheduling engine, combined with advanced optimization algorithms and reinforcement learning mechanisms, the production scheduling system can respond to changes in production demand in real time and make optimal decisions. At the same time, the system has self-learning and adaptive adjustment capabilities, and can continuously optimize scheduling results in the face of complex and dynamic production environments, providing strong support for the efficient operation of the production network.

[0088] Specifically, the specific implementation process of step 4 includes: 4.1 Design of Self-Organizing Structure of Production Network The goal is to design a decentralized, flexible production network architecture that can self-adjust according to actual needs, and optimize the overall goal through local interactions between nodes.

[0089] The implementation process includes: using the technical framework of distributed autonomous systems and multi-agent systems to simulate the collaboration and adaptive adjustment between various nodes in the production network (such as equipment, production lines, warehouses, etc.). The specific steps are as follows: Distributed autonomous node design: In the production network, each production unit (such as equipment, process, warehouse, etc.) is designed as an autonomous node, and each node has independent decision-making and information processing capabilities. Through local data perception and decision-making capabilities, each node can independently perceive environmental changes and adjust its behavior based on local information.

[0090] Multi-agent system model: Each node in the production network is used as an intelligent agent. Each agent collaborates and shares information with other agents through local perception and information exchange. Each agent can independently determine the production scheduling strategy and coordinate with other agents based on information such as production tasks, equipment status, and material flow. For example, when a device fails, its agent will automatically report the failure status to the system, and other agents can adjust the priority of production tasks or reallocate resources based on this information.

[0091] 4.2 Dynamic Task Allocation and Scheduling in Self-Organizing Networks The purpose is to design a self-organizing mechanism that can dynamically perform task allocation and production scheduling, so that the production network can self-adjust the arrangement of production tasks according to real-time factors such as demand fluctuations, production bottlenecks and equipment status.

[0092] The implementation process includes: using a task allocation method based on market mechanism (such as auction mechanism) combined with an adaptive scheduling algorithm (such as adaptive genetic algorithm) to achieve dynamic optimization allocation of tasks and resources. The specific steps are as follows: Market mechanism for task allocation: The bidding and allocation process of tasks is simulated through the market mechanism. Production tasks are regarded as auction commodities. Each autonomous node in the production network (such as equipment, work center) acts as a bidder and bids according to its own resource status (such as load, production capacity, scheduling, etc.). Tasks will be allocated according to price (i.e. task processing capacity, response time, etc.). The task allocation model can be expressed as:

[0093] in, For Node The bidding price, is the load condition of the node, is the production capacity of the node, Predict the task processing time of the node. Through this mechanism, the task will be automatically assigned to the most suitable node to achieve optimal allocation of resources.

[0094] Adaptive Scheduling Algorithm: Combining the adaptive genetic algorithm and particle swarm optimization algorithm, the task allocation scheme is adjusted according to real-time data (such as equipment failure, production bottleneck, etc.). The genetic algorithm explores the task allocation space through selection, crossover, and mutation operations, while the particle swarm optimization fine-tunes the scheduling scheme through global search.

[0095] 4.3 Resource Coordination and Optimization in Self-Organizing Networks Its purpose is to enable the production network to achieve optimal coordination and allocation of resources, eliminate bottlenecks in the production process through self-organizing mechanisms and collaborative strategies, and improve overall production efficiency.

[0096] The implementation process includes: using collaborative filtering algorithm and collaborative optimization model to achieve resource coordination and optimization. The specific steps are as follows: Collaborative filtering and resource matching: Each node in the production network uses collaborative filtering algorithms to predict resource demand and supply and match resources based on historical production data, resource usage and other information. For example, during the production process, when the workload of a production line is too heavy, collaborative filtering algorithms can be used to find available resources (such as idle equipment, available processes, etc.), redistribute tasks and alleviate bottlenecks.

[0097] The basic idea of ​​the collaborative filtering algorithm is to predict future resource requirements based on similar historical situations by calculating the similarity between production nodes. The basic formula is:

[0098] in, For prediction nodes The resource requirements of the node The score of (i.e. resource usage), For Node With Node The similarity between For Node In Resources The actual needs.

[0099] Application of collaborative optimization algorithm: In order to eliminate bottlenecks in production, a collaborative optimization algorithm based on the global optimal solution is designed, so that each node in the production network can achieve the global optimal resource allocation through information sharing and collaborative decision-making under the condition of limited resources. For example, node A may not be able to complete a task alone, but through collaboration with nodes B and C, the overall task can be completed efficiently. This process can be achieved through the constraint optimization algorithm: Subordinate to

[0100] in, is the resource cost, is assigned to the node The amount of resources, is the resource allocation matrix, is the total demand for resources. Through this collaborative optimization strategy, each node in the production network can share resources and coordinate scheduling to maximize the production efficiency of the overall system.

[0101] 4.4 Adaptive Evolution Mechanism of Self-Organizing Production Networks Its purpose is to enable the production network to self-adjust and evolve through self-organizing mechanisms in the face of long-term changes and emergencies, so as to adapt to the dynamic changes in the external environment.

[0102] The implementation process includes: using evolutionary game theory and adaptive feedback mechanism to realize the evolution and adaptive adjustment of self-organizing production network. The specific steps are as follows: Evolutionary game model construction: Each node in the production network is regarded as a game participant. Each node improves its utility by optimizing its own strategy (such as resource allocation, production scheduling, etc.). The game strategy between nodes will evolve over time. By adjusting the strategy through repeated games, the Nash equilibrium point is finally found to achieve system stability and optimization. The game model can be expressed as:

[0103] in, For Node In Strategy The utility function under For Node The self-utility, For Node With Node The cost of collaboration between For collaborative decision making.

[0104] Adaptive feedback and evolution mechanism: Combining real-time feedback mechanism with evolutionary game theory, the production network can self-adjust and evolve according to actual production demand, resource usage, cooperation results between nodes, etc. For example, when the production efficiency of a certain node is low, its game strategy will be adjusted to improve the efficiency of the node by increasing resource allocation or optimizing production scheduling, and ultimately promote the evolution of the entire system.

[0105] By designing a decentralized, adaptive, self-organizing production network model, combined with distributed autonomous systems, multi-agent systems and evolutionary game theory, the production network can automatically adjust resource allocation, task scheduling and production strategies according to real-time environmental changes, thereby improving production efficiency, reducing costs, reducing resource waste, and enhancing the overall adaptability and flexibility of the system.

[0106] Specifically, the specific implementation process of step 5 includes: 5.1 Introduction and Optimization of Generative Design Its purpose is to use generative design technology to automatically generate optimized product design solutions through intelligent algorithms, thereby reducing design costs, improving design efficiency, and meeting customization needs in the production process.

[0107] The implementation process includes: introducing topology optimization algorithm, multi-objective optimization algorithm and artificial intelligence design assistance system, fully considering product structure, material properties, manufacturing process and production constraints in the generative design process, and ensuring that the generated design scheme has efficient manufacturability and operability in actual production. The specific steps are as follows: Topology optimization and generative design: Using topology optimization algorithms, given the initial framework and constraints of a product design, a structural design that meets performance requirements is automatically generated. This algorithm looks for the best way to distribute materials in the design space to achieve specific physical performance goals (such as stiffness, strength, weight, etc.). For example, by optimizing the structural design, reducing unnecessary material use, and improving the efficiency and lightweight of the structure, the mathematical expression of the topology optimization problem is:

[0108] in, is the displacement vector of the structure, is the material stiffness matrix, For the design space, the goal is to minimize the energy consumption of the structure by optimizing the material distribution of the structure while satisfying constraints such as strength and stiffness.

[0109] Multi-objective optimization design: Multi-objective optimization algorithms are used to comprehensively consider different design goals (such as cost, weight, functionality, etc.) and constraints (such as manufacturing difficulty, material selection, etc.) to generate multiple optimal design solutions. Heuristic algorithms such as genetic algorithms or particle swarm optimization can be used to find the global optimal solution in an iterative manner, helping the design team explore and select the most appropriate product design solution.

[0110] AI-assisted design: Combining deep learning and generative adversarial networks (GANs) to automatically generate product design solutions, especially when faced with complex and changing design requirements, AI can quickly come up with innovative designs. By training generative models, the system can automatically generate feasible design solutions based on historical data and design requirements. For example, in customized product design, AI models are used to learn different design parameters and provide real-time, personalized design solutions.

[0111] 5.2 Establishment and implementation of predictive maintenance model The purpose is to identify potential equipment failures in advance through predictive maintenance technology, reduce equipment downtime, improve production stability, and thus enhance the overall efficiency of the production network.

[0112] The implementation process includes: based on IoT technology and big data analysis, combined with machine learning algorithms (such as random forest, support vector machine, long short-term memory network LSTM, etc.), to achieve real-time monitoring of equipment status and fault prediction. The specific steps are as follows: Equipment data collection and sensor network construction: By deploying smart sensors (such as temperature sensors, vibration sensors, pressure sensors, etc.), real-time data on the equipment's operating status is collected. These data are transmitted to the cloud through the IoT platform for centralized processing. The sensor data collection system can include real-time monitoring of the equipment's operating status, load changes during operation, temperature fluctuations, etc., thereby providing the necessary input data for predictive maintenance.

[0113] Feature extraction and data preprocessing: Extract and preprocess the collected sensor data, select key features (such as vibration frequency, temperature change rate, working cycle, etc.), and perform denoising and standardization on the data to improve the accuracy of the prediction model. Commonly used data processing techniques include wavelet transform and principal component analysis, which help extract key features that affect the health status of the equipment.

[0114] Establishment of fault prediction model: Use machine learning algorithms to combine the historical fault records of the equipment with the current operating data to establish a device fault prediction model. For example, LSTM (Long Short-Term Memory Network) is used to process time series data to predict the probability of equipment failure in the future. By learning from a large amount of historical data, the model can identify potential failure modes in advance and predict the remaining useful life (RUL) of the equipment. Its basic prediction model can be expressed as:

[0115] in, is the input feature of time series data (such as sensor data), is the LSTM model obtained through training, and the predicted value is the remaining useful life (RUL) of the equipment.

[0116] Real-time fault diagnosis and early warning system: Based on the equipment health data output by the predictive maintenance model, a real-time fault diagnosis and early warning system is established. When the risk of equipment failure is predicted, the system will automatically issue a warning and provide maintenance suggestions (such as immediate shutdown, load adjustment, etc.). In addition, by combining equipment maintenance history with production plans, intelligent maintenance scheduling can be achieved, thereby optimizing equipment maintenance costs and downtime.

[0117] 5.3 Collaborative Optimization of Generative Design and Predictive Maintenance The goal is to combine generative design with predictive maintenance to optimize equipment design and maintenance strategies in production networks to improve equipment availability, reliability and production efficiency.

[0118] The implementation process includes: combining product structure optimization in generative design with equipment status monitoring in predictive maintenance to form a collaborative optimization closed-loop system. The specific steps are as follows: Design phase is synchronized with maintenance requirements: In the generative design process, the maintainability and maintainability of the equipment are considered, and the predictive maintenance requirements of the equipment are embedded in the product design. For example, the failure frequency of key components of the equipment is taken into account during the design phase, and components that are easy to maintain and replace are designed to reduce downtime when the equipment fails.

[0119] Design feedback based on health data: By collecting the operation data and maintenance history data of the equipment in real time, the data is fed back to the generative design system for closed-loop optimization. The design system automatically adjusts the design plan according to the actual operation status and maintenance history of the equipment to optimize the long-term stability and maintenance cost of the product. For example, when the failure rate of a certain type of equipment is high, the design system will analyze the cause through machine learning algorithms, and automatically modify the component design in subsequent designs to improve the reliability and maintenance efficiency of the components.

[0120] Adaptive maintenance strategy and design iteration: The collaborative optimization of generative design and predictive maintenance can not only optimize the design of existing products, but also provide real-time feedback to the design of future products to form an adaptive design and maintenance strategy. For example, by real-time monitoring of equipment status and obtaining equipment operation data, the design system can adjust product design based on this information to achieve dynamic optimization.

[0121] By deeply integrating generative design with predictive maintenance technology, we can not only improve the innovation and efficiency of product design, but also improve the reliability of equipment operation and the stability of production systems. Generative design enables products to achieve optimal configuration from structure to material selection, while predictive maintenance ensures the health and efficient operation of production equipment in a data-driven way. The coordinated optimization of the two will greatly enhance the adaptive and sustainable operation capabilities of the production network, and promote the entire production system towards intelligence and efficiency.

[0122] Specifically, the detailed implementation process of step 6 includes: 6.1 Construction and Integration of Virtual Production Network Its purpose is to create a virtual production network model to fully simulate the behavior and status of the physical production system, thereby providing support for production decision-making, production planning, resource allocation and optimization, and realizing global monitoring and scheduling of the system.

[0123] The implementation process includes: using physical simulation, system modeling and digital twin technology to digitize information such as production equipment, production lines, process flows, resource allocation, etc., so as to build a virtual production network model corresponding to actual production. The following are the specific steps: System modeling and simulation: Use system dynamics and discrete event simulation (DES) technology to build mathematical models of production systems to simulate the behavior of production lines, equipment operation, resource consumption, and process flows. These models should include elements such as production units, equipment performance, production plans, and demand fluctuations to ensure that the virtual production network can truly reflect the dynamic changes in actual production.

[0124] System dynamics model: This model simulates the dynamic behavior of the production process by establishing feedback loops between factors such as inventory, productivity, and demand fluctuations. Its mathematical expression usually contains multiple differential equations that represent the changing relationship between various parts of the production system at different time points. For example, the dynamic equation for inventory changes can be expressed as:

[0125] in, Indicates time The inventory level at the moment, is the production rate, is the required rate.

[0126] Discrete event simulation: Through the simulation model, different events in the production process (such as machine failure, production task completion, resource arrival, etc.) can be simulated on the timeline. This method is suitable for simulating complex, discrete production processes, and can help determine production bottlenecks, evaluate resource allocation strategies, and optimize production plans.

[0127] Construction of digital twins: By combining the physical production system with the virtual system, a digital twin model is created, so that the virtual model can reflect the status of the physical production system in real time. Digital twin technology uses sensors, IoT devices and real-time data acquisition systems to ensure the synchronization of the virtual production network with the actual production system, and can predict potential bottlenecks or problems by simulating and analyzing different production scenarios.

[0128] 6.2 Data Fusion and Real-time Monitoring System Its purpose is to collect data from the physical production system in real time and integrate it with the virtual production network, so as to achieve real-time monitoring, diagnosis and optimization of the production process, and ensure that the production system can respond quickly and accurately in different situations.

[0129] The implementation process includes: using IoT technology and big data analysis, combining real-time data streams with historical data, and building a comprehensive, multi-dimensional production monitoring platform. The specific steps are as follows: Data collection and real-time monitoring platform: By deploying sensors, cameras and IoT devices on production equipment, production lines and resource nodes, various data in the production process (such as equipment temperature, load, voltage, vibration, production quantity, etc.) are collected in real time. These data are sent to the data center in real time through the transmission network for centralized processing and analysis. Based on the real-time data stream, the status of the production line can be monitored in real time and reflected in the virtual production network in real time to ensure the synchronization of the virtual network and the actual production system.

[0130] Data fusion and preprocessing: In the process of data transmission, there may be problems such as data quality differences, inconsistent formats, and noise interference from different sources. By applying data fusion algorithms (such as Kalman filtering, particle filtering, etc.) and multi-sensor data fusion technology, data from different sensors are preprocessed and fused to eliminate data uncertainty and noise, and a standardized data model is formed to ensure data accuracy and consistency.

[0131] Kalman filter: used to estimate and optimize time series data, effectively filter out noise, and predict future data trends. 6.3 Forecasting and Optimization Decision Support Its purpose is to utilize the fusion results of virtual production networks and real-time data, through data analysis and machine learning models, to achieve dynamic prediction, optimal scheduling and resource allocation of production networks, and improve overall production efficiency and flexibility.

[0132] The implementation process includes: using machine learning (such as reinforcement learning, regression analysis) and optimization algorithms (such as genetic algorithms, particle swarm optimization, etc.) to dynamically predict and optimize the virtual production network. The specific steps are as follows: Production status prediction and trend analysis: Use machine learning algorithms to train historical production data to predict future production status, equipment failures, demand fluctuations, etc., and provide a basis for decision-making. Through regression analysis, a relationship model between demand and production rate can be established to predict future demand fluctuations. Optimize production scheduling and resource allocation: Use optimization algorithms, such as genetic algorithms or particle swarm optimization, to optimize the scheduling and allocation of production resources based on data in the virtual production network. By simulating the effects of different scheduling strategies, the optimal resource allocation plan is selected to reduce production bottlenecks, improve equipment utilization, and minimize production costs.

[0133] Particle swarm optimization algorithm: Particle swarm optimization finds the optimal solution by simulating the foraging process of bird flocks. The formula of the optimization process is:

[0134] in, For the The particle in The speed of the step, is the best historical position of the particle, is the global best position, , is the acceleration constant, , is a random number, is the inertia weight, which determines the search range and convergence speed, and generally decreases over time. is the position of the particle The particle in The current position of the generation and the position of the particle are the current production scheduling plan, and the goal is to find the optimal solution.

[0135] 6.4 Real-time Application and Feedback Mechanism of Digital Twins The purpose is to achieve real-time production system optimization and feedback through seamless connection of digital twins and physical production networks, ensuring that dynamic adjustments and optimization strategies in the production process can be immediately applied and verified.

[0136] The implementation process includes: establishing a closed-loop feedback mechanism so that the data flow between the virtual production network and the actual production system can be transmitted bidirectionally. The virtual model can not only reflect the state of the real system, but also guide the adjustment of the physical system according to the real-time optimization results. The specific steps are as follows: Real-time synchronization and feedback control: Through real-time monitoring and data transmission, the optimization decisions of the virtual production network can be fed back to the physical production system. If the virtual production network finds production bottlenecks or potential problems, the control system will immediately issue instructions to adjust production plans, resource allocation or equipment scheduling.

[0137] Real-time optimization application and adjustment: Based on the optimization results of the virtual production network, immediate adjustments are implemented. For example, when the virtual network predicts that a device will fail, the system can schedule backup equipment in advance and adjust the production plan to avoid stagnation or loss of efficiency in the physical production system.

[0138] By building a virtual production network and applying digital twin technology, not only can the production system be simulated and optimized in real time, but also dynamic and flexible production decisions and adjustments can be made. The combination of virtual production network and digital twin enables every link in the production process to be tested and optimized in a virtual environment, thereby maximizing production efficiency, reducing downtime, reducing costs, and enhancing the adaptability and stability of the production system.

[0139] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein with equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the scope of protection of the claims of the present invention.

Claims

1. A digital production operation management method for data quality assessment and analysis, characterized in that: include: Step 1: Build a multi-level distributed data collection system to collect multi-dimensional production data through sensors, cameras, and embedded devices, and combine edge computing technology for real-time transmission and preliminary processing to ensure high-quality data input; Step 2: Design a multimodal data fusion framework, combine graph neural networks with time series data processing technology, realize the fusion and multidimensional analysis of heterogeneous data such as sensor data, video images, and device logs, and mine potential data associations; Step 3: Based on the microservice architecture and event-driven mechanism, build an intelligent scheduling engine, combine optimization algorithms and reinforcement learning technology to implement dynamic adaptive scheduling strategies to cope with real-time changes in the production environment; Step 4: Use distributed autonomous systems and multi-agent technology to design a decentralized self-organizing production network, and achieve flexible adjustment and optimization of the production process through dynamic task allocation, resource coordination and evolution mechanism; Step 5: Combine generative design technology with predictive maintenance models to achieve collaborative optimization of product design and equipment maintenance through topology optimization, multi-objective optimization, and machine learning algorithms to improve the reliability and efficiency of the production system. Step 6: Build a virtual production network and combine it with digital twin technology to achieve real-time monitoring, dynamic prediction and optimization decision support of the production system, and guide the adjustment and optimization of the physical production system through a closed-loop feedback mechanism.

2. A digital production operation management method for data quality assessment and analysis according to claim 1, characterized in that: The step 1 comprises: Through a multi-level distributed data acquisition system, sensors, cameras and embedded devices are used to collect multi-dimensional data from the production site. The sensor layer collects equipment operating status and environmental change data, the video monitoring layer obtains quality inspection visual data, and the embedded data acquisition layer transmits equipment operating data in real time. The data stream is transmitted to the edge computing node in real time through the transmission protocol. On the edge node, the raw data is preprocessed by noise filtering, formatting and standardization, and the self-supervised learning algorithm is used to identify and correct outliers and missing values, reducing the transmission burden and improving data real-time performance. Through data quality assessment and automatic correction mechanism, the collected data is dynamically scored and the data is automatically repaired.

3. A digital production operation management method for data quality assessment and analysis according to claim 2, characterized in that: The step 2 comprises: Construct a multimodal data fusion framework, use graph neural network as the core model, abstract the equipment, production line, and material elements in the production process into graph nodes, and use graph convolution operations to propagate and fuse information on each data node; To solve the problem of integrating time series data and non-time series data in production environments, a spatiotemporal attention mechanism is introduced. Time series data is captured with long short-term memory networks or gated recurrent units, and features of non-time series data are extracted with convolutional neural networks. A multi-perspective joint modeling strategy is used to process heterogeneous data. Multiple data perspectives are defined according to different stages of the production process. The data from each perspective is integrated in the latent space through a deep generative model to extract common features and enhance the model's predictive ability. Perform multi-dimensional analysis and mining on the fused data, use frequent item set mining algorithm to discover association rules in the production process, perform trend prediction and anomaly detection through deep learning model, and set up real-time alarm mechanism.

4. A digital production operation management method for data quality assessment and analysis according to claim 3, characterized in that: The step 3 comprises: The scheduling engine is designed with a microservice architecture, which splits complex production scheduling tasks into multiple independent service modules. Each microservice exchanges data through message queues to ensure the flexibility and scalability of the system. At the same time, an event-driven mechanism is introduced so that changes in equipment status, material arrival, and order change events in the production process can trigger dynamic adjustments to scheduling strategies in real time. Build a scheduling decision system based on rule engine and machine learning model, combine business rules with scheduling strategy through rule engine, and realize rule-driven scheduling decision; A multi-objective optimization model is constructed using mixed integer linear programming and heuristic optimization algorithms; Design dynamic adaptive scheduling strategies, combine reinforcement learning and model predictive control, so that the scheduling system can continuously optimize the scheduling strategy based on real-time feedback.

5. A digital production operation management method for data quality assessment and analysis according to claim 4, characterized in that: The step 4 comprises: The distributed autonomous node design is adopted, and the equipment, production lines and warehouse units in the production network are regarded as autonomous nodes. Each node has independent decision-making and information processing capabilities. Through local perception and decision-making, the nodes can autonomously adjust their behavior to cope with environmental changes, thereby improving the flexibility and adaptability of the system. Introducing a multi-agent system model, taking nodes in the production network as intelligent agents, and achieving collaboration and coordination through local perception and information sharing; Design dynamic task allocation and resource coordination mechanism, and realize dynamic optimization allocation of tasks and resources through task allocation based on market mechanism and adaptive scheduling algorithm; Construct an adaptive evolution mechanism for self-organizing networks. Through evolutionary game theory and feedback mechanisms, the production network can self-adjust and evolve according to long-term changes and emergencies. Nodes participate in the game by optimizing their own strategies, and ultimately reach Nash equilibrium, achieving system stability and optimization.

6. A digital production operation management method for data quality assessment and analysis according to claim 5, characterized in that: The step 5 comprises: Introducing generative design technology, using topology optimization, multi-objective optimization and artificial intelligence-assisted design systems to automatically generate efficient and manufacturable product design solutions; Establish a predictive maintenance model based on IoT technology and big data analysis, combined with machine learning algorithms to monitor equipment status in real time and predict faults. By analyzing the equipment's historical fault records and real-time operating data, the predictive maintenance model can identify potential faults in advance and reduce equipment downtime; During the design phase, the maintainability requirements of the equipment are embedded in the product design to optimize the maintainability and reliability of key components. At the same time, through real-time collection of equipment operation data and maintenance history, this information is fed back into the generative design system to achieve closed-loop optimization.

7. A digital production operation management method for data quality assessment and analysis according to claim 6, characterized in that: The step 6 comprises: Build a virtual production network corresponding to the actual production system through physical simulation, system modeling and digital twin technology; Establish a data fusion and real-time monitoring system; use the Internet of Things technology to collect equipment status and production data information in the physical production system in real time, and integrate it with the virtual production network; use data fusion algorithms to pre-process multi-source data, eliminate noise and uncertainty, and ensure data accuracy and consistency; By utilizing the fusion results of virtual production networks and real-time data, dynamic prediction of production status and optimization decisions can be achieved. Through machine learning algorithms and optimization algorithms, equipment failures and demand fluctuations in the production process can be predicted, and production scheduling and resource allocation can be optimized.

Images