An intelligent manufacturing production line virtual-real mapping control method based on vPLC and digital twin interaction

Abstract

The application discloses an intelligent manufacturing production line virtual-real mapping control method based on vPLC and digital twin interaction, relates to the technical field of intelligent manufacturing and industrial internet, and comprises the following steps: a three-dimensional geometric model of a production line equipment is established and imported into a virtual engine; a physical mechanism model is established for different equipment objects in the production line, physical mechanism model parameters are divided into a determined parameter set and an unknown parameter set, effective data segments under dynamic working conditions are extracted, and then the determined parameter set and the effective data segments are jointly input into a physical information neural network to estimate the unknown parameter set and output a prediction result; vPLC uploads collected equipment operation data to a digital twin platform, the digital twin platform feeds back the prediction result to the vPLC; the vPLC performs residual error calculation on the prediction result and the physical equipment operation data, generates a control instruction according to the residual error result, and then issues the control instruction to physical production line equipment for execution, so that virtual-real closed-loop control of the production line is realized. The application improves the production line operation efficiency.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing and industrial internet technology, specifically to a virtual-real mapping control method for intelligent manufacturing production lines based on vPLC and digital twin interaction. Background Technology

[0002] With the development of intelligent manufacturing, industrial big data, and artificial intelligence technologies, modern manufacturing production lines are placing higher demands on flexibility, predictability, and remote control capabilities. Traditional PLC control systems rely on dedicated hardware equipment. Although widely used in industrial control, they suffer from poor scalability, high upgrade costs, and insufficient predictive capabilities in complex production line environments.

[0003] Digital twin technology enables real-time monitoring and simulation analysis of production systems by constructing virtual mapping models of physical systems. However, existing digital twin systems mostly remain at the level of three-dimensional visualization, lacking deep coupling with the actual control system and making it difficult to achieve closed-loop control between the virtual and physical systems. On the other hand, with the development of virtual programmable logic controllers (vPLCs), control logic is gradually migrating from dedicated hardware to general-purpose hardware and software platforms. vPLCs have advantages such as flexible deployment, easy expansion, and low cost, but how to deeply integrate them with digital twins to build a truly virtual-physical mapping control system still lacks an effective method.

[0004] Meanwhile, traditional data-driven methods for digital twins often rely on large amounts of historical data in complex industrial scenarios, making it difficult to guarantee the stability of the model under different operating conditions. Physical Information Neural Networks (PINNs) embed physical mechanism equations into the neural network training process, enabling the model to learn data features while satisfying physical constraints, providing a new solution for modeling and predicting complex industrial systems.

[0005] Therefore, how to integrate PINN into the digital twin and vPLC interaction system to form a complete virtual-real mapping control closed loop is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0006] In order to solve the above-mentioned technical problems, this application proposes the following technical solution: This application provides a method for controlling the virtual-real mapping of a smart manufacturing production line based on vPLC and digital twin interaction, including: Establish a three-dimensional geometric model of the production line equipment and import it into the virtual engine. Construct a virtual production line scene in the digital twin platform and set the kinematic constraints and production process logic of the equipment. For different equipment objects in the production line, establish corresponding physical mechanism models, collect multi-source sensor data during the operation of production line equipment, and divide the relevant parameters of the physical mechanism models into a set of definite parameters and a set of unknown parameters according to physical measurability; The operating condition type of the multi-source sensor data is determined, and the effective data segments under dynamic operating conditions are extracted; The valid data segment and the determined parameter set are input into a pre-constructed physical information neural network. The physical information neural network estimates the unknown parameter set and outputs the prediction results of equipment operating status, energy consumption trend and failure risk. The vPLC uploads the real-time collected physical equipment operation data to the digital twin platform, and the digital twin platform feeds back the prediction results to the vPLC. The vPLC performs residual calculations between the received prediction results and the physical equipment operating data, generates control commands based on the residual results, and sends the control commands to the physical production line equipment for execution, thereby realizing virtual-physical closed-loop control of the production line.

[0007] In one possible implementation, a three-dimensional geometric model of the production line equipment is established and imported into a virtual engine. A virtual production line scene is then constructed within a digital twin platform, and the kinematic constraints and production process logic of the equipment are set, including: Create three-dimensional geometric models of each production line equipment in 3ds Max or CAD three-dimensional modeling software; The three-dimensional geometric model is imported into Unreal Engine, and a virtual production line scene is constructed in the digital twin platform using Unreal Engine. The spatial layout of the equipment in the virtual production line scene is consistent with the actual spatial layout of the equipment in the physical production line. In the virtual production line scenario, kinematic constraints and production process logic are set for each piece of equipment. The kinematic constraints include joint limits for each motion axis, upper limits for motion speed and acceleration, and spatial collision boundaries between equipment. The production process logic includes processing sequence, equipment start and stop conditions, production cycle time, and linkage triggering conditions between processes.

[0008] In one possible implementation, the step involves establishing corresponding physical mechanism models for different equipment objects in the production line, collecting multi-source sensor data during the operation of the production line equipment, and dividing the relevant parameters of the physical mechanism model into a set of deterministic parameters and a set of unknown parameters based on physical measurability, including: Establish corresponding physical mechanism models for different equipment objects in the production line, with the following expressions: in, For the generalized coordinate state variables of the equipment system, Represents the device speed state variable. Represents the device's acceleration state variable. This refers to the mass matrix or equivalent inertia matrix of the equipment system. The equivalent damping matrix is... The equivalent stiffness matrix, This is the input vector for external excitation or load. Data from multiple sensors during the operation of production line equipment is collected, and the relevant parameters of the physical mechanism model are divided into a set of definite parameters and a set of unknown parameters based on physical measurability.

[0009] In one possible implementation, the physical information neural network includes a data input layer for receiving multi-source sensor data and determining a parameter set. The output of the data input layer is connected to the input of a neural network feature extraction layer and a physical constraint layer, respectively. The outputs of the neural network feature extraction layer and the physical constraint layer are connected to the input of a PINN fusion calculation layer, and the output of the PINN fusion calculation layer is connected to a loss function optimization layer.

[0010] In one possible implementation, the loss function used during training of the physical information neural network is: in, For the total loss function, For data consistency loss, , , These are the weighting coefficients. Indicates the first The model's predicted output value for each sample. Indicates the first The actual measured value corresponding to each sample N Indicates the number of training or test samples; For dynamic constraint loss, This refers to the mass matrix or equivalent inertia matrix of the equipment system. The equivalent damping matrix is... The equivalent stiffness matrix, Indicates the first A vector of device state variables at each moment. , They represent the first First and second time derivatives of the device state variable vector at each time step Indicates the first The external excitation force or load input vector acting on the device at any given moment. For energy consumption constraint loss, This represents the equipment processing power predicted by the PINN model. Indicates cutting force. Indicates cutting speed. Indicates the no-load operating power of the equipment. For degradation constraint loss, Indicates the first The unknown parameters of the device predicted by the time-matter neural network. Indicates the first The unknown parameters of the device predicted by the time-matter neural network. T Indicates the length of the time series.

[0011] In one possible implementation, the valid data segment and the determined parameter set are jointly input into a pre-constructed physical information neural network. The physical information neural network estimates the unknown parameter set and outputs prediction results of equipment operating status, energy consumption trends, and fault risks, including: The valid data segment and the determined parameter set are input into the data input layer of the physical information neural network; The multi-source sensor data enters the neural network feature extraction layer through the data input layer, and undergoes nonlinear mapping and high-dimensional feature extraction through a multi-layer fully connected network. A high-dimensional feature representation of the device's operating status is obtained through multi-layer feature extraction, and the predicted values ​​of the unknown parameter set are output. The set of determined parameters enters the physical constraint layer through the data input layer, and physical residuals are constructed through the state vector and its time derivative. The predicted values ​​of the unknown parameter set are substituted into the physical mechanism model to calculate the dynamic residuals, and the dynamic residuals are used to perform physical consistency correction on the unknown parameter estimation results. The high-dimensional feature representation output by the feature extraction layer of the neural network and the dynamic residual output by the physical constraint layer are jointly input into the PINN fusion calculation layer. The fusion calculation is performed through the mapping function to obtain the unknown parameter estimation results that satisfy the physical mechanism constraints. Based on the estimation results of the unknown parameters, the future operating state of the equipment is calculated using a state evolution model. The calculation formula is as follows: in, The current device status. To predict the device status at any given time, Let be the state evolution function. This represents the set of unknown device parameters estimated by the PINN model. By minimizing the joint loss function, the physical information neural network simultaneously satisfies data consistency constraints and physical mechanism constraints, and outputs equipment operation status prediction results, energy consumption prediction results, and fault risk prediction results.

[0012] In one possible implementation, the multi-source sensor data enters the neural network feature extraction layer via the data input layer, and undergoes nonlinear mapping and high-dimensional feature extraction via a multi-layer fully connected network. The calculation formula is as follows: in, For the first Hidden layer features, For the first Hidden layer features, This is the network weight matrix. For bias parameters, This is the activation function.

[0013] In one possible implementation, the modeling steps of the state evolution model include: The damping matrix, stiffness matrix, and friction term parameters in the physical mechanism model are updated using the unknown parameter estimation results to obtain the equipment dynamic model under the current working condition; The updated device dynamic model is transformed into a first-order state-space form: The fourth-order Runge-Kutta method is used for time discretization, and the state update form is as follows: in, Represents the device system state vector. Represents the rate of change of state. Represents the device speed state variable. Represents the device's acceleration state variable. Represents the device state evolution function. Indicates device input variables, Indicates the first Device status at each time point Indicates the predicted state at the next time point. h Indicates the time step. These represent the rate of change of the system state under different intermediate states.

[0014] In one possible implementation, the vPLC performs residual calculations between the received prediction results and the physical equipment operating data, generates control commands based on the residual results, and sends the control commands to the physical production line equipment for execution, thereby achieving virtual-physical closed-loop control of the production line, including: The vPLC performs residual calculations by comparing the received prediction results with the physical equipment operating data; The residual result is compared with a preset safety threshold. If the absolute value of the residual is less than the preset safety threshold, the prediction result is superimposed on the control output to generate a correction control quantity. If the absolute value of the residual is greater than or equal to the preset safety threshold, the prediction result is masked and the physical feedback control quantity is used. After the vPLC converts the final control quantity into a drive signal through the real-time industrial Ethernet fieldbus, it sends it to the physical production line equipment for execution, realizing virtual-physical closed-loop control of the production line.

[0015] In one possible implementation, the vPLC interacts with the digital twin platform via OPC UA, Modbus, or Socket communication protocols.

[0016] Compared with the prior art, the beneficial effects of this application are as follows: This application utilizes CAD or 3ds Max to create a 3D model of the production line and imports it into Unreal Engine 4 to construct a virtual production line scene. This model models the production line's geometry, kinematic constraints, and process logic. Next, a digital twin prediction model is constructed based on a physical information neural network. By integrating physical mechanism equations and multi-source sensor data, the model predicts the status of production line equipment, energy consumption, and fault trends. Finally, data interaction is achieved between the vPLC and the digital twin platform, realizing closed-loop mapping control from prediction results to control commands. This application enables virtual-real mapping, predictive analysis, and closed-loop control of intelligent manufacturing production lines, improving production line operating efficiency, reducing energy consumption, and enhancing system flexibility and reliability.

[0017] This method is applicable to intelligent manufacturing production lines that involve the coordinated operation of multiple processes, including AGV scheduling, robotic arm collaboration, CNC machine tool processing, online inspection, and laser marking. Attached Figure Description

[0018] Figure 1 A flowchart illustrating a virtual-real mapping control method for an intelligent manufacturing production line based on vPLC and digital twin interaction, provided in an embodiment of this application; Figure 2 The overall method system architecture diagram provided for the embodiments of this application; Figure 3 A schematic diagram of the architecture of the physical information neural network provided in the embodiments of this application; Figure 4 This is an example diagram of the UI display of a CNC machine tool provided in an embodiment of this application. Detailed Implementation

[0019] The present solution will now be described in conjunction with the accompanying drawings and specific embodiments.

[0020] Figure 1 A flowchart illustrating a virtual-real mapping control method for an intelligent manufacturing production line based on vPLC and digital twin interaction, provided in an embodiment of this application, is shown below. Figure 1 and Figure 2 This embodiment of a method for controlling the virtual-real mapping of a smart manufacturing production line based on vPLC and digital twin interaction includes: S101: Establish a three-dimensional geometric model of the production line equipment and import it into the virtual engine. Construct a virtual production line scene in the digital twin platform and set the kinematic constraints and production process logic of the equipment.

[0021] In this embodiment, the production line includes: a CNC lathe, an industrial robotic arm, an AGV transport unit, online inspection equipment, and laser marking equipment. The CNC lathe serves as a key processing piece, used to verify the digital twin model and predictive control method. Three-dimensional geometric models of each production line device are created in 3ds Max or CAD 3D modeling software. These models are then imported into Unreal Engine UE4, and a virtual production line scene is constructed within the digital twin platform using Unreal Engine. The spatial layout of the equipment in the virtual production line scene matches the actual spatial layout of the equipment in the physical production line, ensuring that the virtual scene realistically reflects the operation of the physical production line.

[0022] In the virtual production line scenario, kinematic constraints and production process logic are set for each piece of equipment. Kinematic constraints include joint limits for each motion axis, upper limits for motion speed and acceleration, and spatial collision boundaries between equipment. Production process logic includes the processing sequence, equipment start and stop conditions, production cycle time, and linkage triggering conditions between processes.

[0023] S102, establish corresponding physical mechanism models for different equipment objects in the production line, collect multi-source sensor data during the operation of the production line equipment, and divide the relevant parameters of the physical mechanism model into a set of definite parameters and a set of unknown parameters according to physical measurability.

[0024] In this embodiment, a physical mechanism model is established for each different piece of equipment in the production line. The dynamic behavior of each piece of equipment is uniformly represented by the following second-order state equation: in, These are the generalized coordinate state variables of the equipment system, including equipment displacement, joint angles, or attitude variables. Represents the device speed state variable. Represents the device's acceleration state variable. This refers to the mass matrix or equivalent inertia matrix of the equipment system. The equivalent damping matrix is... The equivalent stiffness matrix, The input vector represents the external stimulus or load. The physical mechanism model in this application is an equivalent dynamic model constructed for intelligent manufacturing equipment, used to describe the dynamic response process of the equipment under external stimulus.

[0025] Depending on the type of equipment, the above physical quantities have different engineering meanings: for cutting tools or machine tool equipment, This indicates the spindle displacement or vibration state, for robotic arm equipment. Indicates the joint angle or end effector position, for transmission or actuator mechanisms. It represents the angle, rotation speed, or displacement variable.

[0026] For different equipment, multi-source sensor data related to their operating status is collected as model input, including: For cutting tool equipment: cutting force signal, spindle current, vibration signal, and temperature signal; for robotic arm equipment: joint angle, angular velocity, drive current, and end-effector load information; and for general equipment: energy consumption data, cycle time data, and environmental disturbance parameters. Multi-source sensor data is collected during the operation of production line equipment, and the relevant parameters of the physical mechanism model are divided into a set of deterministic parameters and a set of unknown parameters based on physical measurability. The set of deterministic parameters is directly obtained from the equipment structure or nominal parameters, while the set of unknown parameters includes friction coefficient, equivalent damping, equivalent stiffness, tool wear factor, and joint degradation factor, etc.

[0027] S103 performs operating condition type identification on multi-source sensor data and extracts valid data segments under dynamic operating conditions.

[0028] S104: The valid data segment and the determined parameter set are input into the pre-constructed physical information neural network. The physical information neural network estimates the unknown parameter set and outputs the prediction results of equipment operating status, energy consumption trend and failure risk.

[0029] In this embodiment, the physical information neural network includes a data input layer, a neural network feature extraction layer, a physical constraint layer, a PINN fusion calculation layer, and a loss function optimization layer, such as... Figure 3 As shown. The data input layer is used to receive multi-source sensor data and determine the parameter set. The multi-source sensor data includes cutting force, spindle current, vibration signal, and temperature signal in the tool scenario; joint angle, angular velocity, drive current, and end effector load in the robotic arm scenario; and power, cycle time, and environmental disturbance signals in the general equipment scenario. The determined parameter set includes equipment structural parameters, rated operating parameters, and equipment nominal parameters.

[0030] The output of the data input layer is connected to the input of the neural network feature extraction layer and the physical constraint layer, respectively. The output of the neural network feature extraction layer and the physical constraint layer are connected to the input of the PINN fusion calculation layer. The output of the PINN fusion calculation layer is connected to the loss function optimization layer.

[0031] The loss function used during training of the physical information neural network is: in, For the total loss function, For data consistency loss, , , These are the weighting coefficients. Indicates the first The model prediction output value corresponding to each sample includes equipment status prediction value, energy consumption prediction value, or fault status prediction value. Indicates the first The actual measured value corresponding to each sample N Indicates the number of training or test samples; For dynamic constraint loss, Indicates the first A vector of device state variables at each moment, including state information such as displacement, angle, or attitude. , They represent the first First and second time derivatives of the device state variable vector at each time step Indicates the first The external excitation force or load input vector acting on the device at any given moment. This refers to the mass matrix or equivalent inertia matrix of the equipment system. The equivalent damping matrix is... The equivalent stiffness matrix, Indicates the first The vector of external excitation force or load acting on the equipment at any given moment is used to describe the mass or inertial characteristics of the equipment during its motion. For energy consumption constraint loss, This represents the equipment processing power predicted by the PINN model. Indicates cutting force. Indicates cutting speed. Indicates the no-load operating power of the equipment. For degradation constraint loss, Indicates the first The unknown parameters of the device predicted by the time-matter neural network. Indicates the first The unknown parameters of the device predicted by the time-matter neural network. T Indicates the length of the time series.

[0032] Degradation and wear evolution constraint loss It is used to constrain the smooth evolution of unknown parameters over time, conforming to the physical laws of tool wear or mechanical aging. To adjust the weights. By minimizing The model learns complex data features under physical consistency constraints and is used for energy consumption prediction, fault diagnosis and scheduling optimization.

[0033] Subsequently, the physical information neural network was tested and validated using production line data. This included: obtaining historical equipment operating data from the production line database as a test dataset, which included vibration data, current data, temperature data, power data, and equipment operating status data. The test data was then input into the trained physical information neural network to obtain equipment status prediction results, energy consumption prediction results, and fault risk prediction results.

[0034] Multi-source sensor data and a defined parameter set are input into the data input layer of a physical information neural network. The multi-source sensor data then enters the feature extraction layer of the neural network, where it undergoes nonlinear mapping and high-dimensional feature extraction via a multi-layer fully connected network. The calculation formula is as follows: in, For the first Hidden layer features, For the first Hidden layer features, This is the network weight matrix. For bias parameters, This is the activation function. Neural networks use a mapping function... Estimate the set of unknown parameters, which is represented as follows: ,in, This is the equivalent damping coefficient. This is the equivalent stiffness coefficient. The coefficient of friction, This is the tool wear factor.

[0035] For robotic arm scenarios, unknown parameters also include joint degradation parameters and load variation parameters. The determined parameter set is input into the physical constraint layer via the data input layer, and physical residuals are constructed using the state vector and its time derivative. , , And residual information of the equipment dynamics equations.

[0036] Since there is a deviation between the initial estimate of the unknown parameters and the actual physical parameters, the dynamic residual is generated after substituting the output parameters of the neural network into the physical mechanism model: Ideally, it should meet the following conditions. However, in the initial stage of model training, due to the existence of unknown parameter estimation errors, the residuals are not zero. This is addressed by introducing the squared residual term into the loss function: By optimizing the network parameters in reverse, the dynamic residual is gradually reduced, thereby obtaining the estimation results of unknown parameters that satisfy the constraints of physical laws.

[0037] The high-dimensional feature representation output from the neural network feature extraction layer and the dynamic residual output from the physical constraint layer are jointly input into the PINN fusion computation layer, and then processed through a mapping function. A fusion calculation is performed, and combined with dynamic residuals, energy consumption constraints, and degradation constraints, joint optimization is carried out to obtain the estimation results of unknown parameters that satisfy the physical mechanism constraints.

[0038] Based on the estimation results of unknown parameters, the future operating state of the equipment is calculated using a state evolution model. The calculation formula is as follows: in, The current device status. To predict the device status at any given time, Let be the state evolution function. This represents the set of unknown parameters of the equipment estimated by the PINN model. By minimizing the joint loss function, the physical information neural network simultaneously satisfies the data consistency constraint and the physical mechanism constraint, and outputs the equipment operating status prediction result, energy consumption prediction result, and fault risk prediction result.

[0039] To ensure that the prediction results conform to the physical mechanism constraints, the parameters output by the neural network are embedded into the physical mechanism model, and the state evolution process is described by numerical integration.

[0040] Dynamically constrained state evolution modeling and numerical embedding, including: estimating results using unknown parameters. The damping matrix, stiffness matrix, and friction parameters in the physical mechanism model are updated respectively to obtain the dynamic model of the equipment under the current working condition; The updated device dynamic model is transformed into a first-order state-space form: The fourth-order Runge-Kutta method is used for time discretization, and the state update form is as follows: in, This represents the device system state vector, including state information such as device pose, velocity, and acceleration. Represents the rate of change of state. Represents the device speed state variable. Represents the device's acceleration state variable. Represents the device state evolution function. This represents the device's input variables, including external loads, control inputs, and environmental disturbances. Indicates the first Device status at each time point Indicates the predicted state at the next time point. h Indicates the time step. These represent the rate of change of the system state under different intermediate states, which are used to improve the accuracy of dynamic state prediction.

[0041] Through the aforementioned PINN prediction model, the digital twin platform can achieve real-time perception, dynamic prediction, and trend analysis of the operating status of physical production line equipment, providing a predictive basis for subsequent virtual-physical closed-loop control.

[0042] S105, vPLC uploads the real-time collected physical equipment operation data to the digital twin platform, and the digital twin platform feeds back the prediction results to vPLC.

[0043] like Figure 4 As shown, the digital twin platform maps physical production line equipment in real time using a virtual model, and visualizes the physical equipment's operating status, predictive analysis results, and optimized control strategies. Specifically, the vPLC on the physical equipment side collects real-time equipment operating data, including equipment operating status parameters, processing parameters, sensor monitoring data, and equipment energy consumption data, and uploads the collected data to the digital twin platform. Based on the real-time operating data, the digital twin platform obtains equipment status prediction results through a physical information neural network prediction model and feeds the prediction results back to the vPLC. In this embodiment, the vPLC and the digital twin platform interact via OPC UA, Modbus, Profint, or Socket communication protocols, enabling bidirectional data transmission of physical equipment data acquisition, digital twin status updates, and prediction result feedback.

[0044] S106, the vPLC performs residual calculations between the received prediction results and the physical equipment operating data, generates control commands based on the residual results, and sends the control commands to the physical production line equipment for execution, thereby realizing virtual-physical closed-loop control of the production line.

[0045] In this embodiment, the vPLC calculates the residual between the received prediction result and the physical equipment operating data. The residual result is then compared to a preset safety threshold. If the absolute value of the residual is less than the preset safety threshold, the prediction result is superimposed on the control output to generate a corrective control quantity. If the absolute value of the residual is greater than or equal to the preset safety threshold, the prediction result is masked, and the physical feedback control quantity is used. The vPLC converts the final control quantity into a drive signal via a real-time industrial Ethernet fieldbus and sends it to the physical production line equipment for execution, achieving virtual-physical closed-loop control of the production line. Control commands include equipment operating parameter adjustment commands, processing technology parameter optimization commands, and equipment status adjustment commands. Through this method, a digital twin virtual-physical closed-loop control based on a physical information neural network prediction model is achieved, improving the predictive capability of industrial equipment operating status and the level of intelligent control.

[0046] In this embodiment, "multiple" refers to two or more. "And / or" describes the relationship between associated objects, indicating that three relationships can exist. For example, A and / or B can represent the existence of A alone, the simultaneous existence of A and B, or the existence of B alone. A and B can be singular or plural. The character " / " generally indicates that the preceding and following associated objects have an "or" relationship.

[0047] The above description is merely a specific embodiment of this application. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the protection scope of this application.

Claims

1. A method for controlling the virtual-real mapping of an intelligent manufacturing production line based on vPLC and digital twin interaction, characterized in that, include: Establish a three-dimensional geometric model of the production line equipment and import it into the virtual engine. Construct a virtual production line scene in the digital twin platform and set the kinematic constraints and production process logic of the equipment. For different equipment objects in the production line, establish corresponding physical mechanism models, collect multi-source sensor data during the operation of production line equipment, and divide the relevant parameters of the physical mechanism models into a set of definite parameters and a set of unknown parameters according to physical measurability; The operating condition type of the multi-source sensor data is determined, and the effective data segments under dynamic operating conditions are extracted; The valid data segment and the determined parameter set are input into a pre-constructed physical information neural network. The physical information neural network estimates the unknown parameter set and outputs the prediction results of equipment operating status, energy consumption trend and failure risk. The vPLC uploads the real-time collected physical equipment operation data to the digital twin platform, and the digital twin platform feeds back the prediction results to the vPLC. The vPLC performs residual calculations between the received prediction results and the physical equipment operating data, generates control commands based on the residual results, and sends the control commands to the physical production line equipment for execution, thereby realizing virtual-physical closed-loop control of the production line.

2. The intelligent manufacturing production line virtual-real mapping control method based on vPLC and digital twin interaction according to claim 1, characterized in that, A 3D geometric model of the production line equipment is created and imported into a virtual engine. A virtual production line scene is then constructed within the digital twin platform, and the kinematic constraints and production process logic of the equipment are set, including: Create three-dimensional geometric models of each production line equipment in 3ds Max or CAD three-dimensional modeling software; The three-dimensional geometric model is imported into Unreal Engine, and a virtual production line scene is constructed in the digital twin platform using Unreal Engine. The spatial layout of the equipment in the virtual production line scene is consistent with the actual spatial layout of the equipment in the physical production line. In the virtual production line scenario, kinematic constraints and production process logic are set for each piece of equipment. The kinematic constraints include joint limits for each motion axis, upper limits for motion speed and acceleration, and spatial collision boundaries between equipment. The production process logic includes processing sequence, equipment start and stop conditions, production cycle time, and linkage triggering conditions between processes.

3. The intelligent manufacturing production line virtual-real mapping control method based on vPLC and digital twin interaction according to claim 1, characterized in that, The process involves establishing corresponding physical mechanism models for different equipment objects in the production line, collecting multi-source sensor data during the operation of the production line equipment, and dividing the relevant parameters of the physical mechanism models into a set of deterministic parameters and a set of unknown parameters based on physical measurability, including: Establish corresponding physical mechanism models for different equipment objects in the production line, with the following expressions: in, For the generalized coordinate state variables of the equipment system, Represents the device speed state variable. Represents the device's acceleration state variable. This refers to the mass matrix or equivalent inertia matrix of the equipment system. The equivalent damping matrix is... The equivalent stiffness matrix, This is the input vector for external excitation or load. Data from multiple sensors during the operation of production line equipment is collected, and the relevant parameters of the physical mechanism model are divided into a set of definite parameters and a set of unknown parameters based on physical measurability.

4. The intelligent manufacturing production line virtual-real mapping control method based on vPLC and digital twin interaction according to claim 1, characterized in that, The physical information neural network includes a data input layer for receiving multi-source sensor data and determining a parameter set. The output of the data input layer is connected to the input of the neural network feature extraction layer and the input of the physical constraint layer, respectively. The outputs of the neural network feature extraction layer and the physical constraint layer are connected to the input of the PINN fusion calculation layer. The output of the PINN fusion calculation layer is connected to the loss function optimization layer.

5. The intelligent manufacturing production line virtual-real mapping control method based on vPLC and digital twin interaction according to claim 4, characterized in that, The loss function used during training of the physical information neural network is: in, For the total loss function, For data consistency loss, , , These are the weighting coefficients. Indicates the first The model's predicted output value for each sample. Indicates the first The actual measured value corresponding to each sample N Indicates the number of training or test samples; For dynamic constraint loss, This refers to the mass matrix or equivalent inertia matrix of the equipment system. The equivalent damping matrix is... The equivalent stiffness matrix, Indicates the first A vector of device state variables at each moment. , They represent the first First and second time derivatives of the device state variable vector at each time step Indicates the first The external excitation force or load input vector acting on the device at any given moment. For energy consumption constraint loss, This represents the equipment processing power predicted by the PINN model. Indicates cutting force. Indicates cutting speed. Indicates the no-load operating power of the equipment. For degradation constraint loss, Indicates the first The unknown parameters of the device predicted by the time-matter neural network. Indicates the first The unknown parameters of the device predicted by the time-matter neural network. T Indicates the length of the time series.

6. The intelligent manufacturing production line virtual-real mapping control method based on vPLC and digital twin interaction according to claim 1, characterized in that, The valid data segment and the determined parameter set are input together into a pre-constructed physical information neural network. The physical information neural network estimates the unknown parameter set and outputs prediction results of equipment operating status, energy consumption trends, and fault risks, including: The valid data segment and the determined parameter set are input into the data input layer of the physical information neural network; The multi-source sensor data enters the neural network feature extraction layer through the data input layer, and undergoes nonlinear mapping and high-dimensional feature extraction through a multi-layer fully connected network. A high-dimensional feature representation of the device's operating status is obtained through multi-layer feature extraction, and the predicted values ​​of the unknown parameter set are output. The set of determined parameters enters the physical constraint layer through the data input layer, and physical residuals are constructed through the state vector and its time derivative. The predicted values ​​of the unknown parameter set are substituted into the physical mechanism model to calculate the dynamic residuals, and the dynamic residuals are used to perform physical consistency correction on the unknown parameter estimation results. The high-dimensional feature representation output by the feature extraction layer of the neural network and the dynamic residual output by the physical constraint layer are input together into the PINN fusion calculation layer. The fusion calculation is performed through the mapping function to obtain the unknown parameter estimation results that satisfy the physical mechanism constraints. Based on the estimation results of the unknown parameters, the future operating state of the equipment is calculated using a state evolution model. The calculation formula is as follows: in, The current device status. To predict the device status at any given time, Let be the state evolution function. This represents the set of unknown device parameters estimated by the PINN model. By minimizing the joint loss function, the physical information neural network simultaneously satisfies data consistency constraints and physical mechanism constraints, and outputs equipment operation status prediction results, energy consumption prediction results, and fault risk prediction results.

7. The intelligent manufacturing production line virtual-real mapping control method based on vPLC and digital twin interaction according to claim 6, characterized in that, The multi-source sensor data enters the neural network feature extraction layer through the data input layer, and undergoes nonlinear mapping and high-dimensional feature extraction through a multi-layer fully connected network. The calculation formula is as follows: in, For the first Hidden layer features, For the first Hidden layer features, This is the network weight matrix. For bias parameters, This is the activation function.

8. The intelligent manufacturing production line virtual-real mapping control method based on vPLC and digital twin interaction according to claim 6, characterized in that, The modeling steps of the state evolution model include: The damping matrix, stiffness matrix, and friction term parameters in the physical mechanism model are updated using the unknown parameter estimation results to obtain the equipment dynamic model under the current working condition; The updated device dynamic model is transformed into a first-order state-space form: The fourth-order Runge-Kutta method is used for time discretization, and the state update form is as follows: in, Represents the device system state vector. Represents the rate of change of state. Represents the device speed state variable. Represents the device's acceleration state variable. Represents the device state evolution function. Indicates device input variables, Indicates the first Device status at each time point Indicates the predicted state at the next time point. h Indicates the time step. These represent the rate of change of the system state under different intermediate states.

9. The intelligent manufacturing production line virtual-real mapping control method based on vPLC and digital twin interaction according to claim 1, characterized in that, The vPLC performs residual calculations between the received prediction results and the physical equipment operating data, generates control commands based on the residual results, and sends the control commands to the physical production line equipment for execution, thereby realizing virtual-physical closed-loop control of the production line, including: The vPLC performs residual calculations by comparing the received prediction results with the physical equipment operating data; The residual result is compared with a preset safety threshold. If the absolute value of the residual is less than the preset safety threshold, the prediction result is superimposed on the control output to generate a correction control quantity. If the absolute value of the residual is greater than or equal to the preset safety threshold, the prediction result is masked and the physical feedback control quantity is used. After the vPLC converts the final control quantity into a drive signal through the real-time industrial Ethernet fieldbus, it sends it to the physical production line equipment for execution, realizing virtual-physical closed-loop control of the production line.

10. The intelligent manufacturing production line virtual-real mapping control method based on vPLC and digital twin interaction according to claim 1, characterized in that, The vPLC and the digital twin platform interact with each other via OPC UA, Modbus, or Socket communication protocols.

Images