Industrial internet digital twin model cooperative control method implemented by computer

Abstract

The invention relates to the technical field of computers, and particularly discloses a computer-implemented industrial internet digital twinning model cooperative control method, which greatly reduces the mapping deviation of a model and a physical entity through dynamic parameter correction, multi-surface feature cooperation of LSTM-Attention algorithm iteration and cooperative fusion, solves the problem of control parameter prediction lag, and improves the reliability of the model and the physical entity. The reliability of the model on equipment state reproduction and control is improved; multi-feature dynamic adaptation can avoid the problems that low-priority equipment occupies redundant resources and high-priority equipment has insufficient resources, and ensures that stable control and iteration performance is still kept under different working conditions; the whole-process traceability of the model state and the regulation and control data is realized, the multi-equipment action interference and parameter overrun risk are eliminated, and the stable and accurate global cooperative control performance is kept in the whole period of long-term operation of the production line.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a computer-implemented collaborative control method for an industrial internet digital twin model. Background Technology

[0002] Currently, with the in-depth development of industrial internet technology, digital twins, as a core technology connecting physical devices and virtual control, have become a key means to improve industrial precision and reduce operation and maintenance costs. They are widely used in fields such as intelligent manufacturing, equipment monitoring, and process optimization. Industrial manufacturers have increasingly stringent requirements for equipment operation stability, multi-device collaborative efficiency, and green production. However, the application of existing digital twin technology has the following shortcomings:

[0003] First, existing digital twin models are mostly built based on initial CAD drawings or static parameters of the equipment, lacking a dynamic correction mechanism based on real-time multidimensional data (vibration, temperature, energy consumption, etc.), resulting in insufficient model mapping accuracy and dynamic adaptability.

[0004] Secondly, in traditional solutions, the weight configuration of digital twin models is fixed and is not dynamically adjusted in combination with the actual operating characteristics of the equipment. It does not consider the "matching accuracy between model predictions and physical measured values", nor does it associate "equipment health status" and "collaborative importance", resulting in an imbalance in the allocation of computing resources and a lack of flexibility in model weight management and resource scheduling.

[0005] Meanwhile, in industrial production lines, there are close material transfer and energy coordination relationships between equipment. However, existing digital twin models do not quantify the degree of such correlation, which makes it easy for multiple devices to interfere with each other or exceed parameter thresholds when running, making it impossible to achieve global collaborative optimization and resulting in insufficient collaborative control capabilities for multiple devices.

[0006] Therefore, designing a computer-implemented collaborative control method for industrial internet digital twin models to improve model mapping accuracy, dynamic adaptability, model weight management and resource scheduling flexibility, and multi-device collaborative control capabilities is an urgent problem to be solved. Summary of the Invention

[0007] The purpose of this invention is to provide a computer-implemented collaborative control method for digital twin models in the industrial internet, aiming to solve the technical problems of insufficient model mapping accuracy and dynamic adaptability, lack of flexibility in model weight management and resource scheduling, and insufficient collaborative control capability of multiple devices in the existing technology.

[0008] To achieve the above objectives, the present invention employs a computer-implemented collaborative control method for an industrial internet digital twin model, comprising the following steps:

[0009] S1: Acquire multi-dimensional data of industrial equipment on-site, and construct physical entity models and digital twin models based on the multi-dimensional data;

[0010] S2: Manage the dynamic weights of the digital twin model through industrial control programs;

[0011] S3: Identify the anomaly types of industrial field equipment based on multi-dimensional data and calculate the optimal control parameters for industrial field equipment;

[0012] S4: Combines the blockchain module to record the status, control data and carbon emission information of the digital twin model;

[0013] S5: Send control parameters to the physical entity model to achieve collaborative control of the industrial internet digital twin model.

[0014] The multidimensional data includes vibration parameters, temperature parameters, displacement parameters, energy consumption parameters, and rotational speed parameters.

[0015] The physical entity model acquires data via the OPC UA protocol, with a data acquisition and transmission delay of ≤50ms, and simultaneously receives actuator control commands.

[0016] The digital twin model includes a core twin unit and a collaborative twin unit:

[0017] The core twin unit is used to build a basic model of the device, and dynamically corrects the parameters based on the LSTM-Attention model to ensure that the mapping error between the basic model and the physical device is ≤0.5%.

[0018] The collaborative twin unit quantifies and clarifies the degree of correlation between the core twin unit based on the material transfer and energy coordination relationships between devices.

[0019] The process involves managing the dynamic weights of the digital twin model through an industrial control program. The weights are determined based on matching accuracy, equipment health status, and collaborative importance. An initial base weight is set, and computing resources and iteration update frequency are allocated according to the real-time operating status of the equipment on site.

[0020] The process of calculating the optimal control parameters for industrial field equipment involves determining parameter priorities based on dynamic weights, simultaneously defining the constraint boundaries for parameter optimization, collecting multi-dimensional data as calculation inputs, extracting disturbance features and determining disturbance types, and combining priorities and disturbance types to determine the optimal parameter adjustment direction.

[0021] The blockchain module uses a Hyperledger Fabric 2.4 private chain, with the following on-chain frequencies: device status data every 1 minute, digital twin model parameters every 5 minutes, and carbon emission data every 60 minutes. A smart contract warning is triggered when the temperature exceeds the limit or the parameters are abnormal.

[0022] Specifically, the control parameters are sent to the physical entity model to achieve collaborative control of the industrial internet digital twin model. The digital twin model predicts the collaborative state in the next 10 to 30 minutes. When there is a risk of action interference or parameter exceeding the threshold, the control parameters are sent through the 5G-MEC edge node, and the control response time is ≤500ms.

[0023] The industrial control program uses a PLC control program.

[0024] The physical entity model is constructed based on equipment CAD drawings and finite element analysis software.

[0025] The identification of abnormal equipment types in industrial sites employs a random forest anomaly identification model.

[0026] The process of sending control parameters to the physical entity model enables collaborative control of the industrial internet digital twin model. By collecting execution data in real time from the physical entity model, the execution data is compared with the predicted parameters of the digital twin model. If the deviation rate between the two is greater than 1%, the secondary parameter correction of the digital twin model is triggered, and steps S3 to S5 are re-executed until the deviation rate is less than or equal to 1%, thereby achieving closed-loop optimization of collaborative control.

[0027] The present invention provides a computer-implemented collaborative control method for an industrial internet digital twin model, which has the following beneficial effects:

[0028] 1. It solves the limitations of low mapping accuracy and lagging control and early warning in digital twins. Through dynamic parameter correction and multi-faceted feature collaboration of LSTM-Attention algorithm iteration, it covers the full state features of equipment from initial operation to wear and aging. Collaborative fusion significantly reduces the mapping deviation between the model and the physical entity, solves the problem of lagging control parameter prediction, and improves the reliability of the model in reproducing and controlling the equipment state.

[0029] 2. Overcoming the limitations of fixed model weights, unreasonable resource allocation, and environmental interference, this system quantifies multi-device parameters by considering device matching accuracy, health status, collaborative importance, dynamic weight allocation, and attention-weighted fusion. It also responds to corresponding control commands by combining resource scheduling rules based on device priority. Multi-feature dynamic adaptation can avoid the problems of low-priority devices occupying redundant resources and high-priority devices lacking resources, ensuring stable control and iterative performance under different working conditions.

[0030] 3. It solves the problems of limitations of fixed thresholds, unreliable data, and poor coordination among multiple devices. By adapting dynamic formulas to the fatigue tolerance decay law of equipment, using blockchain for reliable storage and traceability, and quantifying the correlation between devices, it avoids frequent misadjustments or omissions caused by fixed thresholds, and achieves full-process traceability of model status and control data. At the same time, it eliminates the risk of interference from multiple device actions and parameter over-limits, and maintains stable and accurate global collaborative control performance throughout the entire cycle of long-term production line operation. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 This is a flowchart illustrating the steps of the computer-implemented collaborative control method for an industrial internet digital twin model according to the present invention. Detailed Implementation

[0033] Please see Figure 1 This invention provides a computer-implemented collaborative control method for an industrial internet digital twin model.

[0034] Includes the following steps:

[0035] S1: Acquire multi-dimensional data of industrial equipment on-site, and construct physical entity models and digital twin models based on the multi-dimensional data;

[0036] S2: Manage the dynamic weights of the digital twin model through industrial control programs;

[0037] S3: Identify the anomaly types of industrial field equipment based on multi-dimensional data and calculate the optimal control parameters for industrial field equipment;

[0038] S4: Combines the blockchain module to record the status, control data and carbon emission information of the digital twin model;

[0039] S5: Send control parameters to the physical entity model to achieve collaborative control of the industrial internet digital twin model.

[0040] Furthermore, the multidimensional data includes vibration parameters, temperature parameters, displacement parameters, energy consumption parameters, and rotational speed parameters;

[0041] The physical entity model acquires data via the OPC UA protocol, with a data acquisition and transmission delay of ≤50ms, and simultaneously receives actuator control commands.

[0042] The digital twin model includes a core twin unit and a collaborative twin unit:

[0043] The core twin unit is used to build a basic model of the device, and dynamically corrects the parameters based on the LSTM-Attention model to ensure that the mapping error between the basic model and the physical device is ≤0.5%.

[0044] The collaborative twin unit quantifies and clarifies the degree of correlation between the core twin unit based on the material transfer and energy coordination relationships between devices.

[0045] Furthermore, the dynamic weights of the digital twin model are managed through an industrial control program. The weights are determined based on matching accuracy, equipment health status, and collaborative importance. An initial basic weight is set, and computing resources and iteration update frequency are allocated according to the real-time operating status of the equipment on site.

[0046] Furthermore, the calculation of optimal control parameters for industrial field equipment involves determining parameter priorities based on dynamic weights, simultaneously clarifying the constraint boundaries for parameter optimization, collecting multi-dimensional data as calculation inputs, extracting disturbance features and determining disturbance types, and combining priorities and disturbance types to determine the optimal parameter adjustment direction.

[0047] Furthermore, the blockchain module adopts a Hyperledger Fabric 2.4 private chain, with the following on-chain frequencies: device status data every 1 minute, digital twin model parameters every 5 minutes, and carbon emission data every 60 minutes. A smart contract warning is triggered when the temperature exceeds the limit or the parameters are abnormal.

[0048] Furthermore, the control parameters are sent to the physical entity model to realize collaborative control of the industrial internet digital twin model. The digital twin model predicts the collaborative state in the next 10 to 30 minutes. When there is a risk of action interference or parameter exceeding the threshold, the control parameters are sent through the 5G-MEC edge node, and the control response time is ≤500ms.

[0049] Furthermore, the industrial control program adopts a PLC control program.

[0050] Furthermore, the physical entity model is constructed based on equipment CAD drawings and finite element analysis software.

[0051] Furthermore, the identification of industrial field equipment anomaly types adopts a random forest anomaly identification model.

[0052] Furthermore, the control parameters are sent to the physical entity model to realize collaborative control of the industrial internet digital twin model. By collecting execution data in real time from the physical entity model, the execution data is compared with the predicted parameters of the digital twin model. If the deviation rate between the two is >1%, the secondary parameter correction of the digital twin model is triggered, and steps S3 to S5 are re-executed until the deviation rate is ≤1%, thereby realizing closed-loop optimization of collaborative control.

[0053] In this embodiment,

[0054] The matching accuracy is defined as "the deviation rate between the predicted value of the digital twin model and the measured value of the physical device". For example, when the deviation rate is ≤0.5%, the matching accuracy score is 100, and the score decreases by 10 for every 0.1% increase in the deviation rate.

[0055] The importance of collaboration is assigned a value based on the proportion of equipment to the total material transfer volume of the production line;

[0056] The disturbance characteristics include vibration parameter disturbance characteristics, temperature parameter disturbance characteristics, and energy consumption parameter disturbance characteristics;

[0057] The vibration parameter disturbance characteristics are a sudden change of ≥20% in vibration amplitude within 10s or a deviation of vibration frequency from the rated value by ±5Hz.

[0058] Temperature parameter disturbance characteristics include a sudden temperature rise of ≥8℃ within 5 minutes or a temperature exceeding the rated value for more than 3 minutes.

[0059] The disturbance characteristics of energy consumption parameters are that the energy consumption fluctuation range is ≥15%;

[0060] Disturbance types include mechanical disturbances (triggered by vibration characteristics), thermal disturbances (triggered by temperature characteristics), electrical disturbances (triggered by energy consumption characteristics), and material disturbances (vibration characteristics + energy consumption characteristics). When the vibration amplitude changes abruptly by ≥20% within 10 seconds and there are no other triggering characteristics, it is determined to be a mechanical disturbance. When the temperature rises sharply by ≥8℃ within 5 minutes and the energy consumption fluctuates by ≥10%, it is determined to be a thermal disturbance.

[0061] Vibration parameters are measured using piezoelectric accelerometers (range 0~50g, frequency response 0~10kHz), installed on vibration-sensitive parts such as bearing housings and motor end covers; temperature parameters are measured using K-type thermocouples (measurement range -200℃~1300℃, accuracy ±0.5℃), deployed on motor windings, the surface of power components, and the cooling system outlet; displacement parameters are measured using laser displacement sensors (range 0~500mm, resolution 0.1μm), aligned with the reference surface of moving parts (pistons, conveyor belts, etc.); energy consumption parameters are measured using a three-phase smart meter (measurement accuracy 0.5 class, sampling frequency 1Hz), connected in series in the equipment power supply circuit; and speed parameters are measured using a photoelectric encoder (resolution 1024 lines / revolution, output type A / B phase), coaxially connected to the equipment spindle.

[0062] The OPC UA protocol enables data interaction between the sensor and the edge gateway. The edge gateway preprocesses the collected data (removing outliers beyond the measurement range and reducing noise through moving average filtering) and then sends it synchronously to the physical entity model and the digital twin model.

[0063] The physical entity model is constructed based on the equipment CAD drawings and finite element analysis software, using ANSYS Workbench 2023R2. The equipment geometric model is established by importing the equipment CAD drawings into the ANSYS Workbench 2023R2 finite element analysis software.

[0064] Mesh generation: Powered components (bearings, gears, etc.) use tetrahedral meshes with element sizes of 0.5~2mm; non-powered components (shells, etc.) use hexahedral meshes with element sizes of 5~10mm; ensure mesh quality ≥0.8 and verify using ANSYS mesh quality control tools;

[0065] Loading boundary conditions: Input the rated load of the equipment (including the rated torque of the motor, the load of the equipment, etc.) and environmental parameters (temperature, humidity). Obtain the parameter reference values ​​(including the vibration amplitude and temperature at the rated speed) of the equipment under normal operation through static and thermodynamic simulation.

[0066] Model integration: The physical entity model receives real-time data from the edge gateway through the OPC UA client, and at the same time receives control commands from actuators (including frequency converters, regulating valves, etc.). The response time for receiving control commands is ≤10ms.

[0067] Core twin unit:

[0068] The LSTM-Attention model includes:

[0069] Input layer: 5 neurons (including standardized feature values ​​of vibration, temperature, displacement, energy consumption, and rotational speed);

[0070] Hidden layers: 2 layers of LSTM units, 128 neurons per layer, activation function is tanh, dropout rate=0.2 (to prevent overfitting);

[0071] Attention layer: The Bahdanau attention mechanism is used to assign weights to the temporal features of the LSTM output;

[0072] Output layer: 5 neurons, outputting the predicted parameters of the digital twin model (corresponding one-to-one with the input parameters).

[0073] Model training:

[0074] Training data: Historical operating data of the equipment (sample size 15,000 sets, 70% normal data and 30% fault data), sampling frequency 1Hz;

[0075] Optimizer: Adam optimizer, learning rate 0.001, number of iterations 500, loss function is mean squared error (MSE).

[0076] Accuracy verification: After training, 3000 sets of data were used as a test set for verification to ensure that the single-parameter mapping error between the core twin unit and the physical device is ≤0.5%;

[0077] Collaborative twin unit:

[0078] Quantifying the degree of correlation: Calculate the correlation degree based on the "material transfer efficiency" and "energy coupling coefficient" between devices (Example: Correlation degree = 60% material transfer efficiency + 40% energy coupling coefficient).

[0079] In the formula, material transfer efficiency = receiving capacity of equipment B / conveying capacity of equipment A (material transfer efficiency from conveyor belt A to conveyor belt B).

[0080] Energy coupling coefficient = energy exchanged between devices / total energy consumption of devices (e.g., if motor A provides power to pump B, energy coupling coefficient = energy transferred from A to B / total energy consumption of A).

[0081] A correlation score of ≥0.8 indicates a strong correlation; a correlation score of 0.5~0.8 indicates a medium correlation; and a correlation score <0.5 indicates a weak correlation.

[0082] Dynamic weight management of digital twin models:

[0083] Matching accuracy is the deviation rate between the predicted value of the digital twin model and the measured value of the physical device. The matching accuracy score is calculated as follows: Matching accuracy score = 100 - (deviation rate - 0.5%) / 0.1% × 10, where the matching accuracy score is 100 when the deviation rate is ≤ 0.5%, and decreases by 10 for every 0.1% increase, with the lowest score being 0.

[0084] Equipment health status: Combining equipment operating time and failure rate, the quantitative scoring formula is: Health score = [(1 - Equipment operating time / Rated life) × 30 + (1 - Failure rate in the last 72 hours) × 70], where the rated life is provided by the equipment manufacturer, such as the rated life of a motor being 20,000 hours; Failure rate in the last 72 hours = Number of failures in the last 72 hours / Total operating time in the last 72 hours × 100%, with a score range of 0~100;

[0085] Importance of collaboration: Classified according to the proportion of equipment in the total transfer volume (material / energy) of the production line, with corresponding weighting coefficients and quantitative scores:

[0086] Percentage range Synergy Importance Level Weighting coefficient Quantitative score (0~100) ≥30% high 1.2 100 10%~30% middle 1.0 80 ＜10% Low 0.8 60

[0087] Dynamic weight total score = matching accuracy score × 0.4 + health score × 0.3 + collaboration importance score × 0.3, where the dynamic weight total score ranges from 0 to 100, and the higher the score, the higher the device priority;

[0088] Based on the dynamic weighted total score, computing resources are divided into three levels:

[0089] A dynamic weighted total score of ≥80 is considered high priority and allocated 40% of computing resources;

[0090] Dynamic weighted total scores between 50 and 80 are considered medium priority and allocated 35% of computing resources;

[0091] Dynamic weighted total score <50 is considered low priority and allocated 25% of computing resources;

[0092] Iteration update frequency:

[0093] High priority: Update frequency 10 seconds / time; Medium priority: Update frequency 30 seconds / time; Low priority: Update frequency 60 seconds / time;

[0094] The PLC control program is used to calculate weights and schedule resources. The PLC communicates with the server where the digital twin model is deployed via the Profinet protocol, and the weight update cycle is 5 seconds.

[0095] Random Forest Anomaly Detection Model:

[0096] Model structure: 100 decision trees, maximum depth 15, minimum number of sample splits 2, minimum number of sample leaf nodes 1, and feature selection method is "Gini coefficient";

[0097] Training data requirements: Sample size ≥ 10,000 groups, including 70% normal data and 30% fault data. Fault data is categorized as follows:

[0098] Fault type Triggering features Sample size (groups) Mechanical disturbance fault Vibration amplitude changes by ≥20% within 10 seconds 900 Thermal disturbance fault Temperature rise ≥8℃ within 5 minutes + energy consumption fluctuation ≥10% 900 Power disturbance fault Energy consumption fluctuation range ≥15% 900 Material disturbance fault Vibration amplitude sudden change within 10 seconds ≥20% + energy consumption fluctuation ≥10% 900

[0099] Model validation: After training, the model is validated using a test set (including 3000 sets of data). The anomaly detection accuracy is ≥95% and the false positive rate is ≤5%.

[0100] Optimal control parameter calculation:

[0101] Based on the dynamic weighted total score, the control parameters are divided into 3 levels:

[0102] Level 1: Equipment parameters corresponding to a total weight score ≥ 80;

[0103] Level 2: Equipment parameters with a total weight of 50-80;

[0104] Level 3: Equipment parameters corresponding to a total weighted score of <50;

[0105] Constraint boundary settings:

[0106] Parameter type Constraint boundaries (example) vibration amplitude ≤0.15mm (exceeding this will trigger a mechanical failure) temperature ≤120℃ (maximum withstand temperature of motor windings) rotational speed 500~1500 r / min (rated speed range of the equipment) Energy consumption ≤Rated power × 1.1 (avoid overload) Displacement ≤5mm (maximum displacement of moving parts)

[0107] Parameter adjustment direction determined: An adjustment strategy is formulated based on the exception type and parameter priority, as shown in the table below (example shown):

[0108] Exception types Parameters (Level 1) Adjust direction Mechanical disturbance vibration amplitude Reduce the equipment speed (for every 100 r / min reduction, the vibration amplitude decreases by approximately 5%). thermal disturbance temperature Increase the cooling fan speed (for every 200 rpm increase, the temperature decreases by approximately 3°C). Power disturbance Energy consumption Adjust the power supply voltage to ±5% of the rated value (to stabilize energy consumption fluctuations). Material disturbance Vibration + Energy Consumption Reduce material conveying volume (for every 10% reduction, vibration amplitude decreases by approximately 8%).

[0109] Using the fmincon function (nonlinear programming solver) in MATLAB R2023a, with "minimizing parameter deviation" as the objective function, the constraint boundary and adjustment direction are substituted to output the optimal control parameters (Example: under mechanical disturbance, the optimal speed = current speed - (current vibration amplitude - 0.1mm) / 0.05mm × 100r / min).

[0110] The blockchain module uses a Hyperledger Fabric 2.4 private chain and includes the following nodes:

[0111] Orderer node (1): Responsible for sorting transactions, using Kafka sorting service to ensure transaction order consistency;

[0112] Peer nodes (3): 1 endorsement node (responsible for verifying the legality of transactions) and 2 ledger nodes (store the blockchain ledger, using the LeveIDB database);

[0113] CA Node: Responsible for generating and managing node certificates and user certificates to ensure the trustworthiness of node identities;

[0114] Chaincode development (smart contracts): Chaincode is written in Go language to realize data on-chain, query, and anomaly warning functions. The chaincode is deployed on the endorsing node and the call response time is ≤100ms.

[0115] The rules for uploading data to the blockchain are as follows:

[0116] Data types On-chain frequency Data structure (key fields) Device status data 1 min / time Device ID, timestamp, vibration value, temperature value, displacement value, rotational speed value Digital twin model parameters 5 min / time Model ID, timestamp, predicted parameter values, mapping error, total weight score Carbon emission data 60min / time Device ID, timestamp, energy consumption value, carbon emission factor, carbon emissions

[0117] The carbon emission calculation method is as follows: carbon emission = energy consumption value (kWh) × carbon emission coefficient (kgCO2 / kWh), and the electricity carbon emission coefficient adopts the latest real-time data of State Grid (e.g., 0.5703kgCO2 / kWh).

[0118] The data on the blockchain is encrypted using AES-256, and the ledger uses "chain storage + hash verification". The hash value of each transaction contains the hash value of the previous transaction, ensuring that the data cannot be tampered with.

[0119] Smart contract alert logic:

[0120] 1. Smart contract alerts are triggered under the following conditions:

[0121] Temperature exceeds limit: Measured temperature > rated temperature +10℃ (e.g., if the rated temperature is 100℃, the trigger threshold is 110℃).

[0122] Abnormal parameters: vibration amplitude > 0.15 mm, energy consumption fluctuation ≥ 20%, speed deviation from rated value ± 10%;

[0123] Mapping error anomaly: Mapping error between the digital twin model and the physical device > 1%;

[0124] 2. Smart Contract Alert Execution Process:

[0125] Endorsing nodes verify the on-chain data in real time, and automatically execute the smart contract when the triggering conditions are met;

[0126] The contract generates early warning information (including device ID, anomaly type, anomaly value, and timestamp) and pushes it to the mobile APP of the operation and maintenance personnel (via MQTT protocol) and the large screen of the industrial control center;

[0127] At the same time, the warning log is recorded to the blockchain ledger to facilitate subsequent fault tracing.

[0128] Control parameters are sent to the physical entity model: The optimal control parameters are sent from the anomaly identification and parameter calculation module to the 5G-MEC edge node. The physical distance between the deployment location and the device is ≤1km, and the transmission delay is ≤20ms. The edge node sends the parameters to the PLC via the Profinet protocol. The PLC parses the parameters and sends them to the actuator (such as frequency converter, regulating valve, etc.). The actuator receives the parameters and executes the adjustment parameters. The control response time is ≤500ms.

[0129] The closed-loop optimization mechanism is as follows:

[0130] Deviation rate calculation: The physical entity model collects actual execution data of the actuators through sensors and compares it with the predicted parameters of the digital twin model. The deviation rate formula is:

[0131] Deviation rate = |Execution data - Prediction parameters| / Prediction parameters × 100%;

[0132] Deviation rate ≤1%: ​​The control is deemed effective; maintain the current parameters and continue real-time monitoring.

[0133] Deviation rate > 1%: Trigger secondary correction, the process is as follows:

[0134] 1. The physical entity model feeds back the deviation data to the edge nodes, and the edge nodes upload it to the anomaly identification and parameter calculation module;

[0135] 2. The anomaly identification and parameter calculation module re-executes step S3 (updating the anomaly type judgment based on the deviation data, adjusting the parameter priority and constraint boundary) and calculates the new optimal control parameters;

[0136] 3. The new parameters are redistributed through the 5G-MEC edge node, and the "distribution-execution-deviation calculation" process is repeated until the deviation rate is ≤1%.

[0137] Secondary correction step size = current parameter × (100% + X);

[0138] When the parameter is rotational speed, X=5%; when the parameter is temperature, X=3%; when the parameter is energy consumption, X=4%; when the parameter is vibration amplitude, X=8%; when the parameter is displacement, X=6%.

[0139] Example 1:

[0140] This embodiment uses a precision machining production line for automotive parts as an application scenario. The production line includes four core pieces of equipment: a material conveyor belt A (for conveying aluminum alloy blanks), a CNC machine tool B (for milling), a coolant pump C (for cooling the CNC machine tool B), and a robotic arm D (for transferring machined parts). It is necessary to solve problems such as "mismatch between conveyor belt and machine tool cycle time, machine tool overheating leading to decreased machining accuracy, and coolant pump consuming redundant computing power". The method in this application achieves full-process collaborative control.

[0141] 1.1 Multidimensional Data Acquisition Scheme

[0142] For the four devices, deploy the data acquisition equipment to ensure that the data type and transmission latency meet the requirements, as shown in the table below:

[0143] equipment Collect parameters sensor Sensor installation location sampling frequency Transmission Protocol Transmission delay Conveyor Belt A Rotation speed, displacement Photoelectric encoder, laser displacement sensor Conveyor belt spindle, belt reference surface 1Hz OPC UA ≤35ms CNC machine tool B Temperature, vibration, energy consumption K-type thermocouples, piezoelectric accelerometers, and three-phase smart meters. Machine tool spindle winding, bearing housing, power supply circuit 1Hz OPC UA ≤42ms Cooling pump C Speed, temperature, energy consumption Photoelectric encoder, K-type thermocouple, three-phase smart meter Pump shaft, coolant outlet, power supply circuit 1Hz OPC UA ≤38ms Robotic arm D Displacement, vibration Laser displacement sensor, piezoelectric accelerometer robotic arm end effector, joint 1Hz OPC UA ≤40ms

[0144] 1.2 Construction of Physical Entity Model

[0145] Import the equipment's original CAD drawings into the ANSYS Workbench 2023R2 finite element software:

[0146] Mesh generation: The spindle (power component) of CNC machine tool B uses a tetrahedral mesh (unit size 1mm, mesh quality 0.85); the conveyor belt A frame (non-power component) uses a hexahedral mesh (unit size 8mm, mesh quality 0.82).

[0147] Boundary conditions: Input CNC machine tool B with rated power of 15kW and rated spindle speed of 3000r / min; Cooling pump C with rated flow rate of 50L / min and rated speed of 2000r / min;

[0148] Data integration: Receives real-time sensor data via OPC UA client; control command reception response time ≤ 8ms.

[0149] 1.3 Construction of Digital Twin Model

[0150] Core twin unit: Each device independently constructs an LSTM-Attention model (input layer 5 neurons → 2 layers of LSTM (128 units / layer, dropout=0.2) → Bahdanau attention layer → output layer 5 neurons). The training data consists of 15,000 sets of historical data from the production line (70% normal data and 30% fault data). After 500 iterations of the Adam optimizer (learning rate 0.001), the accuracy is verified as follows: temperature mapping error of CNC machine tool B is 0.3%, speed mapping error of conveyor belt A is 0.2%, and speed mapping error of coolant pump C is 0.4% (all ≤0.5%).

[0151] Collaborative twin unit: quantified based on "association degree = 60% material transfer efficiency + 40% energy coupling coefficient":

[0152] Conveyor belt A and CNC machine tool B: Material transfer efficiency 98%, energy coupling coefficient 0, correlation degree = 60% × 98% = 58.8% (medium correlation);

[0153] Cooling pump C and CNC machine tool B: Material transfer efficiency 0, energy coupling coefficient 85%, correlation degree = 40% × 85% = 34% (weak correlation);

[0154] Robotic arm D and CNC machine tool B: Material transfer efficiency 95%, energy coupling coefficient 0, correlation degree = 60% × 95% = 57% (medium correlation);

[0155] 2.1 Based on the formula "Matching Accuracy (40%) + Health Status (30%) + Collaboration Importance (30%)", the basic parameters of the equipment are calculated as follows: operating time (conveyor belt A runs for 5000 hours, CNC machine tool runs for 8000 hours, cool pump C runs for 6000 hours, robotic arm D runs for 7000 hours), rated lifespan for 20000 hours, failure rate in the last 72 hours for 0.5%, material transfer volume of production line for conveyor belt A is 35%, for CNC machine tool B it is 35%, for cool pump C it is 8%, and for robotic arm D it is 12%. The scores of each equipment item and the total weight are as follows:

[0156] equipment Matching accuracy score (deviation rate) Health Score Synergy Importance Score Dynamic weighted total score Priority Computing resource allocation Iteration frequency Conveyor Belt A 100（0.2%） [(1-5000 / 20000)×30+(1-0.5%)×70]≈92.15 100（35%≥30%） 100×0.4+92.15×0.3+100×0.3≈97.65 high 40% 10 seconds / time CNC machine tool B 100（0.3%） [(1-8000 / 20000)×30+(1-0.5%)×70]≈87.65 100（35%≥30%） 100×0.4+87.65×0.3+100×0.3≈96.30 high 40% 10 seconds / time Cooling pump C 100（0.4%） [(1-6000 / 20000)×30+(1-0.5%)×70]≈90.65 60（8%＜10%） 100×0.4+90.65×0.3+60×0.3≈85.20 middle 15% 30 seconds / time Robotic arm D 98（0.4%） [(1-7000 / 20000)×30+(1-0.5%)×70]≈89.15 80（12%，10%-30%） 98×0.4+89.15×0.3+80×0.3≈89.95 high 5% 10 seconds / time

[0157] Management is achieved through a Siemens S7-1200 PLC (Profinet protocol communication, weight update cycle 5s / time).

[0158] 3.1 Anomaly Detection:

[0159] When the production line is running for 2 hours, the CNC machine tool B sensor collects data on the temperature rising from 85℃ to 96℃ within 5 minutes (temperature rise of 11℃ > 8℃) and energy consumption fluctuation of 12% ≥ 10%, triggering the random forest anomaly identification model (100 decision trees, accuracy of 96.5%), and outputting the anomaly type as "thermal disturbance".

[0160] 3.2 Calculation of Optimal Control Parameters

[0161] Parameter priority: CNC machine tool B (high priority, level 1 parameters: temperature, energy consumption); Cooling pump C (medium priority, level 2 parameters: speed);

[0162] Constraint boundaries: CNC machine tool B temperature ≤ 120℃, cooling pump C speed ≤ 2000r / min;

[0163] Adjustment direction: For every 200 r / min increase in the cooling pump speed, the machine tool temperature decreases by about 3℃. The target temperature is 88℃, so a temperature drop of 8℃ is required.

[0164] Numerical calculation: The current cooling pump speed is 1500 r / min. The required speed increase is 8 / 3 × 200 ≈ 533 r / min. 1500 r / min + 533 r / min = 2033 r / min, which exceeds the rated speed of 2000 r / min. Therefore, the initial optimal speed is set to 2000 r / min. The temperature that can be reduced is (2000-1500) / 200 × 3 = 7.5℃. The temperature can be reduced to 96℃ - 7.5℃ = 88.5℃.

[0165] 4.1 A Hyperledger Fabric 2.4 private blockchain (1 Orderer, 3 Peers, 1 CA node, Go language chaincode, AES-256 encryption) is used. Data upload and alerts are as follows:

[0166] Data on the blockchain:

[0167] Equipment status data (1 minute / time): A total of 4320 records were accumulated within 72 hours, such as equipment ID=B-001, timestamp=2024-10-01 10:02:00, temperature=96℃;

[0168] Digital twin model parameters (5 min / time): 864 records accumulated within 72 hours, such as model ID=DT-C-001, timestamp=2024-10-01 10:05:00, predicted rotational speed=2000 r / min, mapping error=0.4%;

[0169] Carbon emission data (60 min / time): Cooling pump C consumes 1.2 kWh per hour, total energy consumption in 72 hours = 86.4 kWh, carbon emission = 86.4 kWh × 0.5703 (State Grid coefficient) ≈ 49.27 kg CO2, a total of 72 records were recorded in 72 hours (each record corresponds to 1 hour of data, such as equipment ID = C-001, timestamp = 2024-10-01 10:00:00, energy consumption = 1.2 kWh, carbon emission ≈ 0.684 kg CO2);

[0170] Smart contract alert: When the temperature of CNC machine tool B reaches 96℃ > 95℃, the endorsing node pushes the alert to the maintenance APP and the control center screen, and the alert log is recorded on the blockchain for traceability.

[0171] 5.1 Parameter Distribution:

[0172] The optimal speed of 2000 r / min is sent to the PLC through the 5G-MEC edge node (transmission delay 18ms < 20ms). The PLC controls the cooling pump C to adjust the speed, and the control response time is 420ms < 500ms.

[0173] 5.2 Closed-loop optimization:

[0174] Initial monitoring: After 10 seconds, the actual speed of coolant pump C was 1998 r / min, with a deviation rate of |1998-2000| / 2000×100%=0.1%<1%; the temperature of CNC machine tool B decreased to 88℃ (target temperature was 88.5℃), with a deviation rate of |88-88.5| / 88.5×100%≈0.56%≤1%, indicating effective control;

[0175] Continuous stability verification: Within the next 30 minutes, the temperature remained stable at 88.5±0.5℃, with a deviation rate of ≤1%, requiring no secondary correction, and the closed-loop optimization was completed.

[0176] The above description discloses only one preferred embodiment of the present invention, and should not be construed as limiting the scope of the present invention. Those skilled in the art will understand that all or part of the processes of the above embodiments can be implemented, and equivalent changes made in accordance with the claims of the present invention are still within the scope of the invention.

Claims

1. A computer-implemented industrial internet digital twin model collaborative control method, characterized in that, It comprises the following steps: S1: Obtain the multi-dimensional data of the industrial equipment on site, and construct a physical entity model and a digital twin model according to the multi-dimensional data; S2: Manage the dynamic weight of the digital twin model through the industrial control program; S3: Identify the abnormal type of the industrial site equipment based on the multi-dimensional data, and calculate the optimal control parameters of the industrial site equipment; S4: Record the state of the digital twin model, the control data and the carbon emission information in combination with the blockchain module; S5: Issue the control parameters to the physical entity model to realize the collaborative control of the industrial internet digital twin model.

2. The computer-implemented industrial internet digital twin model collaborative control method of claim 1, wherein The multi-dimensional data includes vibration parameters, temperature parameters, displacement parameters, energy consumption parameters and rotation speed parameters; The physical entity model collects data through the OPC UA protocol, and the data collection transmission delay is ≤50ms, while receiving the actuator control instruction; The digital twin model includes a core twin unit and a collaborative twin unit: The core twin unit is used to construct a basic model for the equipment, and the mapping error between the basic model and the physical entity equipment is ≤0.5% based on the dynamic correction parameters of the LSTM-Attention model; The collaborative twin unit quantifies and clarifies the correlation between the core twin unit based on the material transfer and energy matching relationship between the equipment.

3. The computer-implemented industrial internet digital twin model collaborative control method of claim 1, wherein the dynamic weight of the digital twin model is managed through the industrial control program, and wherein The weight is confirmed according to the matching accuracy, the equipment health status and the collaborative importance, the initial basic weight is set, and the calculation resource and the iteration update frequency are allocated according to the real-time running state of the equipment on site.

4. The computer-implemented industrial internet digital twin model collaborative control method of claim 1, wherein the optimal control parameters of the industrial site equipment are calculated, and wherein Based on the dynamic weight, the priority of the parameters is determined, the constraint boundary of the parameter optimization is clarified, the multi-dimensional data is collected as the calculation input, the disturbance feature is extracted and the disturbance type is determined, the optimal parameter adjustment direction is determined in combination with the priority and the disturbance type.

5. The computer-implemented industrial internet digital twin model collaborative control method of claim 1, wherein The blockchain module uses Hyperledger Fabric2.4 private chain, the upper chain frequency is: equipment state data 1min / time, digital twin model parameters 5min / time, carbon emission data 60min / time, and the intelligent contract early warning is triggered when the temperature exceeds the limit or the parameter is abnormal.

6. The computer-implemented industrial internet digital twin model collaborative control method of claim 1, wherein the control parameters are issued to the physical entity model to realize the collaborative control of the industrial internet digital twin model, and wherein The digital twin model predicts the collaborative state in the next 10-30 minutes, and when there is a risk of action interference or parameter threshold value, the control parameters are issued through the 5G-MEC edge node, and the control response time is ≤500ms.

7. The computer-implemented industrial internet digital twin model collaborative control method of claim 1, wherein, the industrial control program is a PLC control program.

8. The computer-implemented industrial internet digital twin model collaborative control method of claim 1, wherein, the physical entity model is constructed based on equipment CAD drawings and finite element analysis software.

9. The computer-implemented industrial internet digital twin model collaborative control method of claim 1, wherein, the identification of the industrial field device abnormal type uses a random forest anomaly identification model.

10. The computer-implemented industrial internet digital twin model collaborative control method of claim 1, wherein, by comparing the execution data collected by the physical entity model in real time with the predicted parameters of the digital twin model, if the deviation rate is > 1%, the secondary parameter correction of the digital twin model is triggered, and steps S3-S5 are re-executed until the deviation rate is ≤ 1%, achieving closed-loop optimization of collaborative control.

Images