5G factory full-process visual monitoring platform based on digital twinning

Abstract

The invention relates to the technical field of factory control, in particular to a 5G factory full-process visual monitoring platform based on digital twinning, which comprises the steps of arranging edge computing nodes in a factory, constructing a space-time coupling control network according to space-time relevance of a production process and a material flow path, and dynamically adjusting a topological structure according to a real-time production state. Equipment operation parameters, environment data and product quality tracing data are collected, and a quality-process parameter association database is established; constructing a multi-physics field digital twinborn model based on the quality-process parameter association database, and establishing an association model from a quality result to a process parameter by the multi-physics field digital twinborn model through a backward reasoning algorithm; the self-learning control unit takes a quality index as a reward function training control strategy based on the correlation model, and adjusts PLC control parameters and formula parameters; the active intervention control unit analyzes the multi-dimensional time sequence data to predict the equipment trend; and superposing the equipment trend on the three-dimensional model in a thermodynamic diagram mode for fusion display.

Description

Technical Field

[0001] This invention relates to the field of factory control technology, specifically to a 5G-based full-process visualization monitoring platform for factories using digital twins. Background Technology

[0002] Real-time monitoring and visual management of factory production processes have become key technologies for improving production efficiency and product quality. Existing factory monitoring technologies mainly suffer from the following technical problems: First, traditional monitoring systems are mostly based on wired network transmission, which is difficult to wire in complex factory environments, has high network latency, and cannot meet the high real-time requirements of control scenarios. Second, existing monitoring platforms lack effective 3D visualization methods, making it difficult for operators to understand the operational status of the entire production process, affecting the efficiency of fault diagnosis and decision-making. Third, their capabilities in data fusion and analysis are limited, failing to fully utilize multi-source heterogeneous data for intelligent analysis and predictive control.

[0003] Furthermore, although digital twin technology offers new solutions for factory monitoring, existing digital twin applications are mostly limited to local processes and lack a comprehensive monitoring platform that covers the entire process.

[0004] To address this, a 5G-based full-process visualization monitoring platform for factories is proposed. Summary of the Invention

[0005] The purpose of this invention is to provide a 5G-based factory end-to-end visualization monitoring platform that combines 5G network and digital twin technology to ensure the continuity of factory operation.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A 5G-based factory end-to-end visualization and monitoring platform based on digital twins includes: The network construction module deploys 5G edge computing nodes in the factory's four-level architecture. Based on the 5G edge computing nodes, a spatiotemporal coupling control network is constructed by combining the spatiotemporal correlation of the production process and the material flow path. The spatiotemporal coupling control network dynamically adjusts its topology according to the real-time production status. The data acquisition and processing module collects equipment operating parameters, environmental data, and product quality traceability data based on a spatiotemporal coupled control network, and establishes a quality-process parameter correlation database. The digital twin model construction module constructs a multi-physics digital twin model based on the quality-process parameter correlation database, and uses the multi-physics digital twin model to establish a correlation model from quality results to process parameters through a reverse reasoning algorithm. The control decision execution module includes a self-learning control unit and an active intervention control unit. The self-learning control unit trains the control strategy based on the correlation model and uses quality indicators as the reward function, and adjusts the PLC control parameters and formula parameters. The active intervention control unit analyzes multi-dimensional time series data to predict equipment trends. The visualization module overlays device trends onto the 3D model in the form of a heat map for integrated display.

[0007] Preferably, the spatiotemporal coupling control network construction process includes: Factory environmental parameters are collected based on a spatiotemporal coupled control network. These parameters include temperature field distribution data, humidity gradient data, airflow velocity vector, dust concentration distribution, light intensity distribution, and noise spectrum data. Based on the material flow path analysis algorithm, the spatiotemporal correlation model of the production process is performed, and the process coupling coefficient between equipment is extracted. The process weight matrix between 5G edge computing nodes is calculated based on the process coupling coefficient, and a spatiotemporal coupling control network topology in the form of a weighted directed graph is constructed.

[0008] Preferably, the topology adjustment process includes: The system monitors changes in production tasks and equipment operating status. When a work order switch and / or equipment status change is detected, a topology reconstruction algorithm is triggered. The minimum spanning tree algorithm in graph theory is used to recalculate the optimal production path. At the same time, the process weight matrix is ​​updated according to the current production load and network latency conditions to achieve dynamic reconstruction of the control network.

[0009] Preferably, the process of constructing the quality-process parameter association database includes: Collect equipment operating parameters, including motor speed, torque value, power consumption, bearing temperature, vibration acceleration, and displacement deviation; Obtain factory environmental parameters; collect product quality traceability data, including product dimensional accuracy, surface roughness, hardness value, component content, and defect type identifier; synchronize and associate the above data according to timestamps to establish a multi-dimensional data matrix-based associated database.

[0010] Preferably, the process of constructing the multiphysics digital twin model includes: A structural mechanical model of the equipment is constructed based on the finite element analysis method, and the stress distribution and deformation are calculated. A fluid flow model was established using computational fluid dynamics to simulate the flow and heat transfer processes of coolant. An electromagnetic induction model was constructed using electromagnetic field simulation technology to analyze the magnetic induction intensity and electromagnetic force distribution. By integrating the structural mechanics model, fluid flow model, and electromagnetic induction model through a multiphysics coupling algorithm, a unified multiphysics digital twin simulation model is formed.

[0011] Preferably, the structural mechanics model construction process includes: acquiring the equipment's geometric parameters and material properties, and performing mesh generation; defining boundary conditions and load conditions, and calculating nodal displacement deformation and stress distribution; the fluid flow model construction process includes: establishing a fluid domain geometric model, setting inlet velocity, outlet pressure, and wall boundary conditions; using the Navier-Stokes equations to describe fluid motion, using a turbulence model to handle turbulence effects, and simulating coolant flow and heat transfer processes; the electromagnetic induction model construction process includes: establishing an electromagnetic field distribution model based on Maxwell's equations, defining coil current excitation and magnetic material properties, and calculating magnetic induction intensity and electromagnetic force distribution.

[0012] Preferably, the association model construction process includes: Extract historical quality inspection results and corresponding process parameter data from the quality-process parameter association database; A Bayesian network algorithm is used to establish the conditional probability relationship between quality defect types and process parameter deviations, and the network parameters are determined by maximum likelihood estimation. A feature selection algorithm is used to identify key process parameters that have a significant impact on product quality, and a parameter importance ranking table is established. A mathematical mapping function between quality indicators and key process parameters is established based on multiple regression analysis, and the parameter sensitivity coefficients and confidence intervals are calculated to form a complete correlation model between quality and process parameters.

[0013] Preferably, the self-learning control unit obtains parameter sensitivity coefficients based on the correlation model, uses a deep Q-network algorithm with a comprehensive quality index composed of product qualification rate, production efficiency, and energy consumption as the reward function, determines the parameter adjustment range according to the quality impact predicted by the correlation model, updates the neural network weights through an experience playback mechanism, and outputs the PLC control parameter adjustment amount and formula parameter correction value; the active intervention control unit analyzes multi-dimensional time-series data including equipment vibration spectrum characteristics, temperature change trends, current waveform distortion rate, and sound spectrum characteristics, uses a long short-term memory network to extract time-series features and predict the probability of equipment abnormality, and issues an early warning signal and recommends intervention measures when the probability of abnormality exceeds a set threshold.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention deploys 5G edge computing nodes within a four-tier factory architecture to construct a spatiotemporally coupled control network based on the spatiotemporal correlation of production processes, enabling dynamic adjustment of the control topology. This solution collects environmental parameters such as temperature, humidity, airflow, and dust, and combines process coupling analysis with edge computing capabilities to adaptively reconstruct the control path. When work order switching or equipment status changes are detected, the optimal control path is reconstructed using graph theory algorithms, and the process weight matrix is ​​updated based on real-time load and network latency, giving the platform high flexibility and rapid response capabilities. This improves the robustness and resource scheduling efficiency of the control network, offering significant advantages for flexible manufacturing and variable task execution in complex production environments.

[0015] 2. This invention proposes a quality-process parameter correlation database and a multi-physics digital twin model, realizing the correlation of data throughout the entire process from equipment operating status and environmental factors to product quality. By collecting various equipment parameters and combining environmental information and quality traceability data, a multi-dimensional data matrix is ​​established, and a reverse inference model between quality and process parameters is constructed using Bayesian network algorithms and multiple regression models. This model can not only predict the impact of key parameters on quality results but also output parameter sensitivity and confidence intervals, providing a quantitative basis for subsequent control optimization, and exhibiting higher prediction accuracy and parameter interpretation capabilities.

[0016] 3. This invention combines deep reinforcement learning algorithms with temporal prediction networks to construct a control decision-making system with self-learning capabilities and proactive intervention mechanisms, thereby improving the intelligence level of equipment operation. It possesses continuous learning capabilities, enabling it to continuously optimize control effects during production. Simultaneously, the proactive intervention control unit incorporates a long short-term memory network to analyze multi-dimensional time-series data, predict abnormal trends in equipment, and automatically issue warnings and intervention suggestions when risks exceed thresholds. This not only enhances predictive maintenance capabilities but also significantly reduces equipment failure rates and unplanned downtime, ensuring the continuity and safety of factory operations and driving the transformation from responsive management to predictive intelligent operation and maintenance. Attached Figure Description

[0017] Figure 1 A schematic diagram of the structure of a 5G factory full-process visualization monitoring platform based on digital twin provided by the present invention; Figure 2 This invention provides a schematic diagram of the entire process of visual monitoring. Figure 3 This is a schematic diagram of the multiphysics digital twin model process provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the device status adjustment process provided in an embodiment of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Example 1:

[0020] This invention provides a 5G factory end-to-end visualization monitoring platform based on digital twins, referring to... Figure 1 It includes a network construction module, a data acquisition and processing module, a digital twin model construction module, a control decision execution module, and a visualization display module.

[0021] For details, please refer to [link / reference]. Figure 2 The technical solution is as follows: 5G edge computing nodes are deployed in the four-level architecture of the factory (factory-workshop-production line-equipment). Based on the 5G edge computing nodes, a spatiotemporal coupling control network is constructed by combining the spatiotemporal correlation of the production process and the material flow path. The spatiotemporal coupling control network dynamically adjusts its topology according to the real-time production status. Furthermore, the construction process of the spatiotemporal coupling control network includes: Collect factory environmental parameters, including temperature field distribution data, humidity gradient data, airflow velocity vector, dust concentration distribution, light intensity distribution, and noise spectrum data; Based on the material flow path analysis algorithm, the spatiotemporal correlation model of the production process is performed, and the process coupling coefficient between equipment is extracted. The process weight matrix between 5G edge computing nodes is calculated based on the process coupling coefficient, and a spatiotemporal coupling control network topology in the form of a weighted directed graph is constructed.

[0022] Specifically, in a 5G factory environment, the system first comprehensively collects multi-dimensional environmental parameters of the factory, including the spatial distribution of the temperature field, the gradient changes of humidity in different areas, the speed and direction of airflow in the workshop, the concentration distribution of dust in the space, the light intensity level of each area, and the spectral characteristics of environmental noise. Based on this environmental data and the actual flow path of materials on the production line, the system uses specialized analysis algorithms to model the spatiotemporal correlation of the entire production process. By analyzing the interdependencies between various devices in the process, the system quantifies and extracts the process coupling coefficients between devices. Finally, the system calculates the process weight matrix between 5G edge computing nodes based on these coupling coefficients and constructs a spatiotemporal coupling control network topology reflecting the process correlations between devices in the form of a weighted directed graph.

[0023] In this embodiment, a spatiotemporal coupled control network based on 5G edge computing is constructed through multi-dimensional environmental parameter acquisition and material flow path analysis, improving the adaptability and real-time response capability of the factory control system to complex production environments. By modeling process coupling and constructing a process weight matrix, the relationships between equipment are quantified, forming a dynamically adjustable weighted directed graph topology. This enhances the flexibility and robustness of network control, improving production efficiency and intelligence.

[0024] The specific process by which the material flow path analysis algorithm performs spatiotemporal correlation modeling of the production process includes: A production line topology diagram is constructed, modeling each process node and material transport path as a directed graph structure. Data on material flow time, dwell time, and transport distance between processes are collected. Based on the temporal characteristics of material flow, the temporal dependence and spatial adjacency between processes are calculated. Spatiotemporal feature vectors of process nodes are extracted using a graph convolutional neural network, and key process nodes and bottlenecks are identified through an attention mechanism. A spatiotemporal correlation matrix is ​​constructed to quantify the coupling strength between processes in the time and spatial dimensions, forming a mathematical model reflecting the spatiotemporal correlation of the production process. By constructing a production line topology diagram and introducing graph convolutional neural networks and attention mechanisms, in-depth mining and modeling of spatiotemporal correlations in the process flow are achieved, enabling the identification of key nodes and bottlenecks, and improving the intelligence level of production optimization and scheduling decisions.

[0025] Furthermore, the topology adjustment process includes: The system monitors changes in production tasks and equipment operating status. When a work order switch and / or equipment status change is detected, a topology reconstruction algorithm is triggered. The minimum spanning tree algorithm in graph theory is used to recalculate the optimal production path. At the same time, the process weight matrix is ​​updated according to the current production load and network latency conditions to achieve dynamic reconstruction of the control network.

[0026] Specifically, the data on changes in production tasks includes key information such as work order number, product model and specifications, production quantity, delivery time, process flow path, quality requirements standards, raw material specifications, production priority, batch information, and special process requirements. In the 5G factory digital twin monitoring platform, when production plans are adjusted, the system needs to capture the product specification changes of the new work orders in real time, such as changes in dimensional parameters, differences in material composition requirements, and adjustments to processing accuracy standards when switching from producing product model A to product model B. Simultaneously, changes in production quantity directly affect the load distribution and equipment usage strategies of the production line, requiring the system to monitor the match between planned output and actual capacity. Changes in the process flow path are another important monitoring target. Different products may require different processing steps or different processing parameters, and these changes will trigger the reconstruction of the control network topology.

[0027] The specific data on equipment operating status encompasses multi-dimensional information such as equipment power-on / off status, operating mode switching, fault alarm information, maintenance status, changes in equipment performance parameters, and network connection status. In the digital twin monitoring platform, equipment status changes include transitions from standby mode to operating mode, from normal operation to fault shutdown, and from production mode to maintenance mode. Real-time changes in equipment performance parameters, such as decreased processing accuracy, fluctuations in operating speed, and abnormal increases in energy consumption, all fall under the category of equipment status data that needs to be monitored. Furthermore, network-level status information, such as the communication connection status between 5G edge computing nodes and equipment, data transmission quality, and changes in network latency, is also crucial for system monitoring. When a status change such as equipment failure, communication interruption, or performance degradation is detected, the system immediately triggers a topology reconstruction algorithm. By recalculating the optimal production path and updating the process weight matrix, it adapts to new equipment configurations and operating conditions, ensuring that the entire control network maintains optimal control performance and production efficiency even when equipment status changes.

[0028] In this embodiment, changes in production tasks and the real-time operating status of each device are continuously monitored. When a change in production work order or equipment operating status is detected, a topology reconstruction algorithm is triggered. Combined with the minimum spanning tree algorithm and dynamic updates of the process weight matrix, adaptive reconstruction of the control network topology is achieved. In cases of work order switching, product specification adjustments, or equipment status changes, the optimal production path is quickly adjusted to ensure network communication efficiency and the continuous effectiveness of control logic. This enhances the rapid response capability and robustness to changing production environments, ensuring the continuity, reliability, and optimal resource allocation of the production process.

[0029] The specific process of updating the process weight matrix based on the optimal production path, current production load, and network latency conditions includes: Real-time monitoring of CPU utilization, memory usage, and network bandwidth usage of each 5G edge computing node is performed to calculate node load factors. Network probing technology is used to measure communication latency, packet loss rate, and jitter parameters between nodes, establishing a network performance evaluation model. Based on the optimal production path determined by the minimum spanning tree algorithm, combined with node load factors and network performance parameters, a dynamic weight adjustment algorithm is used to recalculate the values ​​of each element in the process weight matrix. A load balancing strategy is introduced: when a node's load is too high, its weight value in the weight matrix is ​​reduced; when network latency increases, the weight coefficients of relevant paths are adjusted accordingly, ensuring that the process weight matrix can reflect the current network status and production conditions in real time. By introducing node load factors and the network performance evaluation model, combined with the dynamic weight adjustment algorithm and load balancing strategy, real-time optimization of the process weight matrix is ​​achieved. Path weights are dynamically adjusted based on CPU usage, communication latency, and other statuses, ensuring efficient and stable production path selection even under network fluctuations or changes in node load, improving the real-time performance, reliability, and resource utilization efficiency of the control network.

[0030] Collect equipment operating parameters, environmental data, and product quality traceability data to establish a quality-process parameter correlation database; Furthermore, the process of constructing the quality-process parameter association database includes: Collect equipment operating parameters, including motor speed, torque value, power consumption, bearing temperature, vibration acceleration, and displacement deviation; The system acquires factory environmental parameters and collects product quality traceability data, including dimensional accuracy, surface roughness, hardness, component content, and defect type identifiers. This data is then synchronized and correlated according to timestamps to establish a multi-dimensional data matrix-based relational database. Specifically, it comprehensively collects key equipment operating parameters, including motor speed, output torque, power consumption, bearing temperature, equipment vibration acceleration, and axial displacement deviations. Simultaneously, it collects workshop environmental data, covering temperature changes, relative humidity levels, atmospheric pressure, and air cleanliness levels. Furthermore, it collects comprehensive product quality traceability data, including dimensional accuracy, surface roughness, material hardness, component content analysis results, and defect type identifiers. The system precisely synchronizes and correlates all these multi-source heterogeneous data according to timestamps to establish a multi-dimensional data matrix-based quality-process parameter relational database.

[0031] In this embodiment, by synchronously collecting equipment operating parameters, environmental data, and product quality traceability data, a multi-dimensional data matrix-based database of quality-process parameters is constructed. This achieves a precise mapping between the production process and quality results, enhancing the systematic nature and traceability of quality analysis. Timestamp association ensures data synchronization, providing a solid data foundation for subsequent modeling, control optimization, and quality prediction, thereby strengthening the factory's intelligent analysis and process control capabilities.

[0032] A multiphysics digital twin model is constructed based on the aforementioned quality-process parameter correlation database. (See details...) Figure 3 The multiphysics digital twin model is used to establish a correlation model from quality results to process parameters through a reverse reasoning algorithm; Furthermore, the process of constructing the multiphysics digital twin model includes: A structural mechanical model of the equipment is constructed based on the finite element analysis method, and the stress distribution and deformation are calculated. A fluid flow model was established using computational fluid dynamics to simulate the flow and heat transfer processes of coolant. An electromagnetic induction model was constructed using electromagnetic field simulation technology to analyze the magnetic induction intensity and electromagnetic force distribution. By integrating the structural mechanics model, fluid flow model, and electromagnetic induction model through a multiphysics coupling algorithm, a unified digital twin simulation model is formed.

[0033] Specifically, the system employs the finite element method to construct a structural mechanics simulation model of the equipment, and uses numerical calculation methods to analyze the stress distribution and structural deformation of the equipment under working loads. Computational fluid dynamics (CFD) technology is used to establish a fluid flow simulation model to simulate the flow state and heat transfer process of the coolant inside the equipment. Electromagnetic field simulation technology is used to construct an electromagnetic induction model, and to deeply analyze the magnetic field distribution characteristics and potential electromagnetic interference effects during motor operation. Finally, the system organically integrates the structural mechanics model, fluid flow model, and electromagnetic induction model through an advanced multiphysics coupling algorithm, forming a unified digital twin simulation model that comprehensively reflects the multiphysics characteristics of the equipment.

[0034] In this embodiment, three simulation models—structural mechanics, fluid flow, and electromagnetic induction—are constructed and integrated using a multiphysics coupling algorithm to form a digital twin simulation model. This model simulates the multiphysics interaction characteristics of equipment during actual operation. It enables collaborative analysis of key factors such as stress distribution, heat conduction, and electromagnetic interference, providing simulation basis for tracing the root causes of quality problems and optimizing process parameters. This enhances the predictability of equipment operation and the realism of digital modeling.

[0035] Furthermore, the structural mechanics model construction process includes: acquiring the equipment's geometric parameters and material properties, and performing mesh generation; defining boundary conditions and load conditions, and calculating nodal displacement deformation and stress distribution; the fluid flow model construction process includes: establishing a fluid domain geometric model, setting inlet velocity, outlet pressure, and wall boundary conditions; using the Navier-Stokes equations to describe fluid motion, using a turbulence model to handle turbulence effects, and simulating coolant flow and heat transfer processes; the electromagnetic induction model construction process includes: establishing an electromagnetic field distribution model based on Maxwell's equations, defining coil current excitation and magnetic material properties, and calculating magnetic induction intensity and electromagnetic force distribution.

[0036] The fluid flow region is extracted from the 3D CAD model of the equipment, including the internal spatial structure such as coolant circulation pipes, lubricating oil channels, and gas flow channels. The boundary definition of the fluid domain needs to consider the interface between the solid wall and the fluid. Boolean operations are used to subtract the solid portion from the overall geometric model of the equipment to obtain the pure fluid flow space. During geometric simplification, it is necessary to retain geometric features that significantly affect flow characteristics, such as pipe bends, valves, and filters, while simplifying or ignoring minor structures that have a smaller impact on the overall flow, in order to balance computational accuracy and efficiency.

[0037] The fluid domain meshing employs a hybrid meshing strategy. Boundary layer meshes are used near the wall to capture wall shear flow characteristics, while tetrahedral or hexahedral meshes are used in the core flow region to ensure computational stability. Mesh density distribution follows the flow gradient principle, with meshes finer in regions of rapid velocity change, such as inlets, outlets, and bends, and appropriately coarser in relatively stable straight pipe sections. Boundary conditions include velocity distribution or mass flow rate specifications at the inlet boundary, pressure conditions at the outlet boundary, no-slip conditions at the wall boundaries, and symmetry constraints at symmetric boundaries. For flow problems involving heat transfer, isothermal or constant heat flux density boundary conditions are also required at the wall to accurately simulate the heat exchange process during equipment operation.

[0038] The electromagnetic field distribution model solution process uses the finite element method to discretize the Maxwell equations in the spatial domain. For static magnetic field problems, the magnetic vector potential equation is solved. For time-varying electromagnetic field problems, the frequency domain or time domain method is used to solve the coupled electric and magnetic field distributions. Finally, key physical quantities such as magnetic induction intensity, electric field intensity, and electromagnetic force density at various points in space are obtained.

[0039] In the structural mechanics model construction, the geometric parameters of the equipment and the physical properties of the materials are first obtained. The model is then rationally meshed, and corresponding boundary and load conditions are defined. Numerical calculations are then used to obtain the displacement deformation and stress distribution at each node. In the fluid flow model construction, the system establishes a geometric model of the fluid domain, sets velocity boundary conditions at the fluid inlet, pressure boundary conditions at the outlet, and boundary conditions on the walls. The Navier-Stokes equations are used to describe the fluid motion, and a turbulence model is used to handle complex turbulence effects, achieving accurate simulation of coolant flow and heat transfer processes. In the electromagnetic induction model construction, the system establishes an electromagnetic field distribution model based on Maxwell's equations, defines the current excitation parameters of the coil and the properties of the magnetic material, and obtains the spatial distribution of magnetic induction intensity and electromagnetic force through numerical calculations.

[0040] The reverse reasoning algorithm is a diagnostic reasoning method based on Bayesian network theory. Its core idea is to start from the observed result (product quality status) and use a probabilistic reasoning mechanism to trace back to possible causes (abnormal process parameters), providing a scientific basis for quality control and parameter optimization in the production process. In a 5G factory monitoring platform, this algorithm can be applied as follows: when a product quality deviation or defect is detected, the system can automatically analyze and determine the most likely cause of the quality problem—an incorrect process parameter setting—thereby guiding operators or the automatic control system to make targeted parameter adjustments.

[0041] In this embodiment, by introducing modeling methods from three dimensions—structural mechanics, fluid mechanics, and electromagnetic induction—stress analysis, thermal flow simulation, and electromagnetic field simulation are achieved using the finite element method, Navier-Stokes equations, and Maxwell equations, respectively, thereby enhancing the physical realism and predictive capabilities of the digital twin model. Each model exhibits high engineering fit in boundary condition definitions and parameter settings, ensuring that the model maps to actual operating conditions and providing highly reliable simulation support for equipment design optimization, fault prediction, and operating parameter control.

[0042] Furthermore, the process of constructing the association model includes: Extract historical quality inspection results and corresponding process parameter data from the quality-process parameter association database; A Bayesian network algorithm is used to establish the conditional probability relationship between quality defect types and process parameter deviations, and the network parameters are determined by maximum likelihood estimation. A feature selection algorithm is used to identify key process parameters that have a significant impact on product quality, and a parameter importance ranking table is established. A mathematical mapping function between quality indicators and key process parameters is established based on multiple regression analysis, and the parameter sensitivity coefficients and confidence intervals are calculated to form a complete correlation model between quality and process parameters.

[0043] Specifically, the process involves extracting historical quality inspection results and corresponding process parameter data from a quality-process parameter correlation database as the foundation for modeling. A Bayesian network algorithm is used to establish the conditional probability relationship between quality defect types and process parameter deviations, and the parameters in the network are determined using maximum likelihood estimation. A feature selection algorithm is employed to identify key process parameters that significantly impact product quality from a large pool of process parameters, and a ranking table of parameter importance is established. Based on multiple regression analysis, a mathematical mapping function between quality indicators and key process parameters is established, and the sensitivity coefficients and corresponding confidence intervals for each parameter are calculated, ultimately forming a complete quality-process parameter correlation model.

[0044] In this embodiment, a correlation model between quality defects and process parameters is constructed. This model can not only quantify the impact of each process parameter on quality indicators, but also provide parameter sensitivity and confidence interval analysis, improving the scientific rigor and interpretability of the modeling. The identification of key process parameters and the tracing of the causes of quality problems provide reliable data support and intelligent decision-making basis for process optimization and quality control.

[0045] The specific steps for establishing a correlation model from quality results to process parameters using a multiphysics digital twin simulation model through back-inference algorithms include: Using product quality inspection results as known outputs and process parameters as input variables for inference, a Bayesian back-inference method is employed to establish an inverse mapping relationship between quality results and process parameters based on conditional probability theory. A Markov chain Monte Carlo sampling method is used to search the parameter space for the most probable combination of process parameters leading to a specific quality result. The inverse inference is validated by forward computation results from a multiphysics simulation model, and iterative optimization ensures accuracy. A parameter sensitivity analysis matrix is ​​constructed to quantify the impact of each process parameter on the quality result, establishing a parameter importance ranking and threshold setting mechanism to form a complete inverse correlation model from quality results to process parameters. By utilizing Bayesian back-inference and the Markov chain Monte Carlo method to achieve inverse modeling from quality results to process parameters, and combining multiphysics simulation validation and sensitivity analysis, key parameters affecting quality are identified, improving the scientific rigor and accuracy of process adjustments.

[0046] For details on the control decision execution module, please refer to [link / reference]. Figure 4 It includes a self-learning control unit and an active intervention control unit. The self-learning control unit trains the control strategy based on the correlation model and uses the quality index as the reward function, and adjusts the PLC control parameters and formula parameters. The active intervention control unit analyzes multi-dimensional time series data to predict equipment trends. Furthermore, the self-learning control unit obtains parameter sensitivity coefficients based on the correlation model, uses a deep Q-network algorithm with a comprehensive quality index composed of product qualification rate, production efficiency, and energy consumption as the reward function, determines the parameter adjustment range according to the quality impact predicted by the correlation model, updates the neural network weights through an experience playback mechanism, and outputs the PLC control parameter adjustment amount and formula parameter correction value; the active intervention control unit analyzes multi-dimensional time-series data including equipment vibration spectrum characteristics, temperature change trends, current waveform distortion rate, and sound spectrum characteristics, uses a long short-term memory network to extract time-series features and predict the probability of equipment abnormality, and issues an early warning signal and recommends intervention measures when the probability of abnormality exceeds a set threshold.

[0047] In this embodiment, a control decision execution module with self-learning and predictive capabilities is constructed to enhance the intelligence of equipment control. The self-learning control unit uses a comprehensive quality index as the reward function and dynamically optimizes PLC control parameters and recipe parameters by combining parameter sensitivity to achieve continuous adaptive optimization of the production process; the active intervention control unit, based on multi-dimensional time-series data, predicts abnormal trends in equipment in advance and intelligently triggers early warnings and interventions, reducing failure rates and energy consumption, and improving system stability.

[0048] The visualization module overlays device trends onto the 3D model in the form of a heat map for integrated display.

[0049] By constructing a spatiotemporally coupled control network, the adaptability and real-time response capability of the factory control system to complex production environments were improved. The combination of the minimum spanning tree algorithm and dynamic updates of the process weight matrix ensured network communication efficiency and the continuous effectiveness of control logic, enhancing rapid response to changing production environments. A quality-process parameter correlation database in the form of a multi-dimensional data matrix was constructed to achieve precise mapping between the production process and quality results, providing systematic traceability capabilities for quality analysis. The integration of a multi-physics digital twin simulation model enabled collaborative analysis of key factors such as stress distribution, heat conduction, and electromagnetic interference, improving the predictability of equipment operation and the realism of digital modeling. The self-learning control unit dynamically optimizes control parameters based on parameter sensitivity, achieving continuous adaptive optimization of the production process; the proactive intervention control unit triggers early warnings and intervention measures in advance, effectively reducing failure rates and energy consumption, and improving intelligence, stability, and production efficiency.

[0050] Example 2:

[0051] This invention provides a 5G-based, digital twin-based, end-to-end visual monitoring platform for beverage production plants. The technical solution, tailored to the specific process requirements of beverage production, is as follows: 5G edge computing nodes are deployed in a four-tier architecture of the beverage production plant, specifically including a factory-level control center, a workshop-level dispatch station, a production line-level control cabinet, and equipment-level sensor nodes. Addressing the specific characteristics of beverage production processes, edge computing nodes are deployed in the raw material preparation workshop, mixing and blending production line, filling and capping production line, packaging production line, and quality inspection station. These nodes construct a spatiotemporally coupled control network based on the spatiotemporal correlation of the beverage production process, fully considering the complete flow path of raw materials from water treatment to finished product packaging.

[0052] Specific parameters of the beverage production environment are collected, including workshop temperature field distribution, relative humidity gradient changes, clean air velocity vector, microbial concentration distribution, UV germicidal lamp illuminance distribution, and equipment operating noise spectrum characteristics. Based on the material flow path analysis algorithm for beverage production, spatiotemporal correlation modeling is performed on the entire process from raw water treatment, syrup preparation, carbonation, filling, capping to packaging. By analyzing the process dependencies between syrup preparation tanks and mixing tanks, the synchronization and coordination requirements between filling machines and capping machines, and the cycle time matching relationship between packaging lines and quality inspection stations, the process coupling coefficient between each piece of equipment is extracted. Based on the coupling coefficient, the process weight matrix between 5G edge computing nodes is calculated, and a weighted directed graph topology reflecting the process correlation between equipment in the beverage production line is constructed. When a product formula change (such as switching from cola to orange juice production) or a change in the status of key equipment (such as switching a filling machine from running mode to cleaning mode) is detected, a topology reconstruction algorithm is triggered. The minimum spanning tree algorithm is used to calculate the optimal path to adapt to the production of new products, ensuring the efficient operation of the production line during product switching.

[0053] To meet the quality control requirements of beverage production, key equipment operating parameters are collected. In the syrup preparation stage, data on stirrer speed, syrup viscosity, temperature control accuracy, and pressure sensor data are collected; in the carbonation process, carbon dioxide injection pressure, solubility, and temperature change trends are monitored; and in the filling stage, key parameters such as filling head lifting and lowering displacement, liquid level sensor data, filling speed, and capping torque are collected.

[0054] Environmental data collection covers key environmental factors affecting beverage quality, such as workshop cleanliness level, relative humidity control, temperature distribution uniformity, and microbial monitoring data. Product quality traceability data includes quality indicators for each bottle of beverage, such as volume accuracy, sugar content, acidity measurement results, carbon dioxide content, microbial indicators, packaging sealing test, and identification of appearance defects.

[0055] All collected data is precisely synchronized and correlated according to production batch and timestamp to establish a multi-dimensional data matrix correlation database specifically for beverage production. This database can trace the entire production process of any bottle of beverage from raw material input to finished product output, providing data support for rapid location of quality problems and optimization of process parameters.

[0056] Based on the characteristics of beverage production equipment, a multiphysics digital twin model was constructed. For the syrup mixing tank, a structural mechanics model was established to analyze the stress distribution of the stirring blades in high-viscosity syrup and the tank deformation; a fluid flow model was established to simulate the mixing process of syrup and additives, analyzing the flow field distribution and mass transfer effect; and a heat transfer model was constructed to simulate the temperature control system's control of the syrup temperature. For the filling production line, a precision mechanical motion model of the filling head was established to analyze the displacement accuracy and dynamic response characteristics of the lifting mechanism; the flow state of the beverage in the filling pipeline was simulated to optimize the filling speed and liquid level control accuracy; and a mechanical model of the capping process was established to analyze the capping pressure distribution and sealing effect.

[0057] By integrating the various sub-models using a multiphysics coupling algorithm, a unified digital twin simulation model reflecting the operating characteristics of beverage production equipment is formed. This model predicts product quality performance under different process parameters, providing a basis for formula optimization and process improvement.

[0058] Based on the quality inspection results of beverage production, a back-inference algorithm is used to establish a correlation model between product quality and process parameters. When the quality inspection finds that the sugar content of a batch of beverages exceeds the standard, this quality result is taken as a known output. The combination of process parameters that led to the quality problem is analyzed using Bayesian back-inference, including key parameters such as syrup concentration setpoint, stirring time, and temperature control accuracy.

[0059] The Markov chain Monte Carlo sampling method was used to search for the most probable combination of causal parameters in the parameter space, and the results were verified by forward computation of a multiphysics simulation model. A parameter sensitivity analysis matrix was constructed to quantify the influence of various process parameters such as syrup concentration, stirring speed, and temperature on saccharin quality indicators, and a parameter importance ranking was established to provide a scientific basis for process adjustment.

[0060] The self-learning control unit is based on a quality-process parameter correlation model and employs a deep reinforcement learning algorithm. It uses a comprehensive index comprising beverage product qualification rate, production line efficiency, and energy consumption level as the reward function. Based on the quality impact predicted by the correlation model, it automatically adjusts PLC parameters such as syrup concentration, carbonation pressure setting, and filling speed control to achieve continuous optimization of the production process.

[0061] The proactive intervention control unit analyzes multi-dimensional time-series data of the equipment, including the vibration spectrum characteristics of the filling machine, the temperature change trend of the syrup mixing tank, the current waveform characteristics of the compressor, and the sound spectrum of the conveyor belt. A long short-term memory network is used to extract time-series features and predict the probability of equipment malfunctions. When a potential blockage in the filling machine or a possible malfunction in the syrup pump is predicted, the system issues an early warning and recommends corresponding maintenance measures to avoid quality accidents and production interruptions.

[0062] The visualization module overlays the operational trends of the beverage production line equipment onto a 3D factory model using a heatmap format. Color changes reflect the real-time operational status of each area, enabling production managers to quickly identify potential problem areas and take timely action.

[0063] In this embodiment, a specially adapted digital twin monitoring platform was constructed to address the technological characteristics and quality requirements of beverage production. By monitoring key processes such as raw material preparation, mixing, filling, and packaging in real time, end-to-end quality traceability and intelligent control from raw materials to finished products are achieved. The system can automatically adjust process parameters according to the formulation requirements of different beverage products, improving product quality stability and production efficiency while reducing raw material waste and energy consumption.

[0064] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A 5G factory end-to-end visualization monitoring platform based on digital twins, characterized in that, include: The network construction module deploys 5G edge computing nodes in the factory's four-level architecture. Based on the 5G edge computing nodes, a spatiotemporal coupling control network is constructed by combining the spatiotemporal correlation of the production process and the material flow path. The spatiotemporal coupling control network dynamically adjusts its topology according to the real-time production status. The data acquisition and processing module collects equipment operating parameters, environmental data, and product quality traceability data based on a spatiotemporal coupled control network, and establishes a quality-process parameter correlation database. The digital twin model construction module constructs a multi-physics digital twin model based on the quality-process parameter correlation database, and uses the multi-physics digital twin model to establish a correlation model from quality results to process parameters through a reverse reasoning algorithm. The control decision execution module includes a self-learning control unit and an active intervention control unit. The self-learning control unit trains the control strategy based on the correlation model with the quality index as the reward function, and adjusts the PLC control parameters and recipe parameters. The active intervention control unit analyzes multi-dimensional time-series data to predict device trends; The visualization module overlays device trends onto the 3D model in the form of a heat map for integrated display.

2. The 5G factory full-process visualization monitoring platform based on digital twin as described in claim 1, characterized in that: The construction process of the spatiotemporal coupling control network includes: Collect factory environmental parameters, including temperature field distribution data, humidity gradient data, airflow velocity vector, dust concentration distribution, light intensity distribution, and noise spectrum data; Based on the material flow path analysis algorithm, the spatiotemporal correlation model of the production process is performed, and the process coupling coefficient between equipment is extracted. The process weight matrix between 5G edge computing nodes is calculated based on the process coupling coefficient, and a spatiotemporal coupling control network topology in the form of a weighted directed graph is constructed.

3. The 5G factory full-process visualization monitoring platform based on digital twin as described in claim 1, characterized in that: The topology adjustment process includes: The system monitors changes in production tasks and equipment operating status. When a work order switch and / or equipment status change is detected, a topology reconstruction algorithm is triggered. The minimum spanning tree algorithm in graph theory is used to recalculate the optimal production path. At the same time, the process weight matrix is ​​updated according to the current production load and network latency conditions to achieve dynamic reconstruction of the control network.

4. The 5G factory full-process visualization monitoring platform based on digital twin as described in claim 1, characterized in that: The process of constructing the quality-process parameter association database includes: The equipment operating parameters are collected based on a spatiotemporal coupling control network, including motor speed, torque value, power consumption, bearing temperature, vibration acceleration, and displacement deviation. Obtain factory environmental parameters; collect product quality traceability data, including product dimensional accuracy, surface roughness, hardness value, component content, and defect type identifier; synchronize and associate the above data according to timestamps to establish a multi-dimensional data matrix-based associated database.

5. The 5G factory full-process visualization monitoring platform based on digital twin as described in claim 1, characterized in that: The process of constructing the multiphysics digital twin model includes: A structural mechanical model of the equipment is constructed based on the finite element analysis method, and the stress distribution and deformation are calculated. A fluid flow model was established using computational fluid dynamics to simulate the flow and heat transfer processes of coolant. An electromagnetic induction model was constructed using electromagnetic field simulation technology to analyze the magnetic induction intensity and electromagnetic force distribution. By integrating the structural mechanics model, fluid flow model, and electromagnetic induction model through a multiphysics coupling algorithm, a multiphysics digital twin simulation model is formed.

6. The 5G factory full-process visualization monitoring platform based on digital twin as described in claim 5, characterized in that: The structural mechanics model construction process includes: acquiring the equipment's geometric parameters and material properties, and performing mesh generation; defining boundary conditions and load conditions, and calculating nodal displacement deformation and stress distribution; the fluid flow model construction process includes: establishing a fluid domain geometric model, setting inlet velocity, outlet pressure, and wall boundary conditions; using the Navier-Stokes equations to describe fluid motion, and simulating coolant flow and heat transfer processes; the electromagnetic induction model construction process includes: establishing an electromagnetic field distribution model based on Maxwell's equations, defining coil current excitation and magnetic material properties, and calculating magnetic induction intensity and electromagnetic force distribution.

7. The 5G factory full-process visualization monitoring platform based on digital twin as described in claim 1, characterized in that: The process of constructing the association model includes: Extract historical quality inspection results and corresponding process parameter data from the quality-process parameter association database; A Bayesian network algorithm is used to establish the conditional probability relationship between quality defect types and process parameter deviations, and the network parameters are determined by maximum likelihood estimation. A feature selection algorithm is used to identify key process parameters that have a significant impact on product quality, and a parameter importance ranking table is established. A mathematical mapping function between quality indicators and key process parameters is established based on multiple regression analysis, and the parameter sensitivity coefficients and confidence intervals are calculated to form a complete correlation model between quality and process parameters.

8. The 5G factory full-process visualization monitoring platform based on digital twin as described in claim 1, characterized in that: The self-learning control unit obtains parameter sensitivity coefficients based on the correlation model, uses a deep Q-network algorithm with a comprehensive quality index composed of product qualification rate, production efficiency, and energy consumption as the reward function, determines the parameter adjustment range according to the quality impact predicted by the correlation model, updates the neural network weights through an experience playback mechanism, and outputs the PLC control parameter adjustment amount and formula parameter correction value. The active intervention control unit analyzes multi-dimensional time-series data including equipment vibration spectrum characteristics, temperature change trends, current waveform distortion rate, and sound spectrum characteristics, uses a long short-term memory network to extract time-series features and predict the probability of equipment anomalies, and issues an early warning signal and recommends intervention measures when the anomaly probability exceeds a set threshold.

Images