Digital twin monitoring system for a precision servo press line

Abstract

The application relates to the technical field of digital twinning, in particular to a digital twinning monitoring system for a precision servo press production line, which comprises a physical production line module, an edge computing module, a cloud digital twinning module and a visual monitoring module; wherein the physical production line module is used for real-time monitoring and collecting production line data; the edge computing module is used for pre-processing and analyzing the collected original data, judging product defects, equipment faults and abnormal states of employees in real time to trigger edge early warning and output multidimensional data sets; the cloud digital twinning module is used for constructing a digital twinning body, establishing a real-time coupling simulation model integrating electromagnetism-heat-force-sound to simulate the internal physical process of equipment in real time, predicting equipment performance changes and potential faults, locating fault causes and predicting residual life; and the visual monitoring module is used for multidimensional visual presentation and decision support. Thus, the problems of insufficient monitoring of production line employees, single physical field monitoring and lack of fault predictability are solved.

Description

Technical Field

[0001] This application relates to the field of digital twin technology, and in particular to a digital twin monitoring system for a precision servo press production line. Background Technology

[0002] In the field of modern precision manufacturing, precision servo presses are key production equipment, and their operating status directly affects product quality and production efficiency. With the deepening of Industry 4.0, traditional equipment monitoring methods are no longer sufficient to meet the production demands for high precision and high reliability.

[0003] Traditional monitoring relies heavily on single physical field data, and data acquisition is often delayed, making multi-parameter collaborative analysis impossible. Furthermore, the complex operating environment of presses, characterized by strong electromagnetic interference and multi-physical field coupling, leads to low accuracy in fault prediction and untimely fault diagnosis. Sudden equipment failures frequently cause production line downtime, resulting in significant economic losses. Equipment failures are often addressed through reactive repairs, lacking foresight and exhibiting slow response times, resulting in prolonged downtime and severely impacting production line efficiency.

[0004] Furthermore, traditional digital twin monitoring of production lines mostly focuses on monitoring equipment, believing that solving equipment malfunctions can improve production line efficiency, thus ignoring the impact of production line employees on production efficiency. If production line employees leave their workstations for extended periods or engage in abnormal operations, it can lead to problems such as idle stockpiling on the production line and increased product defect rates, thereby affecting production line efficiency. Relying solely on manual monitoring of abnormal employee behavior can cause response delays and potentially lead to safety accidents.

[0005] In addition, with the acceleration of digital transformation in the manufacturing industry, enterprises have an increasingly urgent need for equipment lifecycle management and virtual simulation optimization. There is a pressing need to build an intelligent monitoring system that integrates multi-physics perception, real-time data analysis, and virtual-real interactive mapping to improve the intelligence level of production lines. Summary of the Invention

[0006] This application provides a digital twin monitoring system for a precision servo press production line to address issues such as insufficient monitoring by production line staff, limited physical field monitoring, and lack of fault prediction.

[0007] The first aspect of this application provides a digital twin monitoring system for a precision servo press production line, comprising: a physical production line module, an edge computing module, a cloud-based digital twin module, and a visualization monitoring module; wherein...

[0008] The physical production line module is used to monitor and collect real-time data on the operating status of production line equipment, product data, and video data on the actions and workstation status of production line employees.

[0009] The edge computing module uses an edge computing server to preprocess and analyze the collected raw data, and judges product defects, equipment failures and abnormal employee status in real time to trigger edge warnings. Combining equipment operating status, product and employee data, it extracts key information and outputs multi-dimensional datasets.

[0010] The cloud-based digital twin module is used to construct a digital twin, reflect the production line's operating status in real time, establish a real-time coupled simulation model integrating electromagnetic, thermal, mechanical, and acoustic technologies to simulate the internal physical processes of the equipment in real time, predict equipment performance changes and potential faults, and locate the root cause of faults and predict remaining lifespan based on the output and coupled simulation of the edge computing module.

[0011] The visualization monitoring module is used to present the simulation results of the cloud-based digital twin module, the real-time operating status of the equipment, fault warning information and remaining life prediction data in a multi-dimensional visualization and provide decision support.

[0012] Optionally, the physical production line module includes: a mechanical sensing sensor, a thermal sensing sensor, an electrical sensing sensor, a photoacoustic sensing sensor, and a position sensing sensor. The mechanical sensing sensor is used to monitor mechanical stress in key components and collect pressure values ​​at key workstations. The thermal sensing sensor is used to monitor temperature distribution and capture microscopic temperature fluctuations in core components. The electrical sensing sensor is used to collect current harmonics and analyze them using THD (Total Harmonic Distortion) to reflect the motor's operating status. The photoacoustic sensing sensor is used to collect product images, personnel actions and workstation status, and production line ambient sound. The position sensing sensor is used to record the slider's full-stroke position data.

[0013] Optionally, determining product defects, equipment malfunctions, and abnormal employee states to trigger edge warnings includes: constructing a defect detection model, a voiceprint diagnosis model, and a deep learning model at the edge; the defect detection model extracts features from product images to identify product defects; the voiceprint diagnosis model uses deep learning algorithms to analyze equipment operating sounds in real time to determine if the equipment is malfunctioning; the deep learning model uses human skeleton key point detection technology and target detection and tracking technology to identify employee violations, determine absence from duty, and monitor abnormal workstation states; and the edge warning immediately triggers an emergency stop or speed reduction of the production line equipment.

[0014] Optionally, the cloud-based digital twin module includes: a twin construction unit, a simulation unit, and a deep analysis and prediction unit. The twin construction unit constructs a production line twin based on the equipment's 3D model and real-time operating data, including the equipment's position, attitude, operating parameters, and personnel activities. The simulation unit establishes a real-time coupled simulation model integrating electromagnetic, thermal, mechanical, and acoustic domains. Based on multiphysics theory, it uses a reduced-order model to compress finite element equations, achieving rapid and high-precision multiphysics simulation. The deep analysis and prediction unit uses a dynamic risk scoring model to output a real-time risk score of 0-100 to predict potential faults. Based on multi-data fusion analysis and multiphysics coupling inversion, it locates the root cause of faults. Based on Paris crack propagation law and the Wiener process degradation model, it calculates the remaining life of key components.

[0015] Optionally, the real-time coupled simulation model integrating electromagnetic-thermal-mechanical-acoustic fields includes: an electromagnetic field simulation sub-model, a thermodynamic simulation sub-model, a mechanical stress simulation sub-model, an acoustic simulation sub-model, and a coupled solver, wherein the coupled solver uses the alternating direction multiplier method to achieve multi-field coupled iterative calculation.

[0016] Optionally, the visualization monitoring module includes: a digital twin visualization interface and a real-time monitoring screen. The digital twin visualization interface presents a digital twin model of the production line using AR (Augmented Reality) technology, allowing users to immerse themselves in and interactively view real-time equipment parameters, personnel operations, and material flow. The real-time monitoring screen displays real-time data and video footage. The interface presents key parameters and production indicators in chart form and supports historical data query and playback, facilitating data analysis and problem tracing.

[0017] The second aspect of this application provides a method for a digital twin monitoring system for a precision servo press production line, comprising: acquiring equipment operating status data, employee actions, and workstation status video data of the precision servo press production line; preprocessing and edge analysis of the collected raw data, real-time judgment of equipment faults and abnormal employee states to trigger edge warnings to trigger emergency stops or speed reductions of the production line equipment, extracting key features and outputting a multi-dimensional dataset; constructing a production line twin based on the equipment's three-dimensional model and real-time operating data, establishing a real-time coupled simulation model integrating electromagnetic, thermal, mechanical, and acoustic technologies to simulate the internal physical processes of the equipment in real time, predicting equipment performance changes and potential faults, and locating the root cause of the fault and predicting the remaining lifespan based on the output of the edge computing module and the coupled simulation; and presenting the simulation results, real-time equipment operating status, fault warning information, and remaining lifespan prediction data in a multi-dimensional visualization and providing decision support, wherein the decision support includes providing equipment maintenance suggestions, taking early production line maintenance strategies, and providing personnel management plans.

[0018] A third aspect of this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the program to implement the digital twin monitoring method for a precision servo press production line described in the above embodiments.

[0019] A fourth aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement a digital twin monitoring method for a precision servo press production line as described in the above embodiments.

[0020] A fifth aspect of this application provides a computer program product storing a computer program that, when executed by a processor, implements a digital twin monitoring method for a precision servo press production line as described in the above embodiments.

[0021] Therefore, this application has the following beneficial effects:

[0022] This invention utilizes edge computing to determine product defects, equipment malfunctions, and abnormal employee states in real time to trigger edge warnings. A defect detection model enables rapid, intelligent, and accurate detection of product defects. A voiceprint diagnostic model analyzes equipment operating sounds in real time to detect minor malfunctions in internal components. A deep learning model identifies employee violations, unauthorized absences, and various abnormal states in real time. A rapid response mechanism is designed so that when these issues are identified, the edge side can immediately issue an alarm and take measures such as stopping or slowing down the machine to prevent further production of defective products, aggravation of equipment malfunctions, and further threats to employee safety. This improves product quality, reduces scrap rates, and increases the overall operational efficiency of the production line. Furthermore, a real-time coupled simulation model integrating electromagnetic, thermal, mechanical, and acoustic fields is established to simulate the internal physical processes of the equipment, overcoming the limitations of single-physics field simulations, which often neglect the interaction between fields. The coupling effect between multiple physical fields can cause results to deviate from reality. However, real-time coupled simulation of multiple fields can capture the chain reaction of physical processes, improve simulation accuracy, and more closely resemble actual physical processes. Equipment failures are often the result of abnormal coupling of multiple physical fields, rather than caused by a single factor. Real-time coupled simulation can better achieve early warning and root cause location of failures, avoid post-failure maintenance of equipment failures, and improve production line efficiency. Based on the Paris crack propagation law and the Wiener process degradation model, the remaining life of key components is calculated, which solves the limitations of single models in terms of "interpretability, uncertainty quantification, and full life cycle adaptability". It provides more accurate, reliable, and practical results for the prediction of the remaining life of key components, supporting the risk management of equipment throughout its entire life cycle from design to operation and maintenance. Through the identification of employee actions and the detection of employee positions on the edge side, the final decision support for personnel management solutions is provided, realizing intelligent, refined, and efficient personnel management.

[0023] This solved technical problems such as insufficient monitoring by production line staff, limited physical field monitoring, and lack of fault prediction.

[0024] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0025] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein...

[0026] Figure 1 This is a schematic diagram of the structure of a digital twin monitoring system for a precision servo press production line according to an embodiment of this application.

[0027] Figure 2 This is a schematic diagram of the cloud-based digital twin module of a digital twin monitoring system for a precision servo press production line according to an embodiment of this application.

[0028] Figure 3 This is a flowchart of a digital twin monitoring method for a precision servo press production line according to an embodiment of this application.

[0029] Figure 4 This is a schematic diagram of the structure of an electronic device provided according to an embodiment of this application. Detailed Implementation

[0030] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0031] The following description, with reference to the accompanying drawings, describes a digital twin monitoring system for a precision servo press production line according to an embodiment of this application. Addressing the issues of insufficient monitoring of production line employees, limited physical field monitoring, and lack of fault prediction mentioned in the background art, this application provides a digital twin monitoring system for a precision servo press production line. It uses edge computing to determine product defects, equipment malfunctions, and abnormal employee states in real time to trigger edge warnings. A defect detection model enables rapid, intelligent, and accurate detection of product defects. A voiceprint diagnostic model allows for real-time analysis of equipment operating sounds to detect minor faults in internal parts. A deep learning model can identify employee violations, unauthorized absences, and various abnormal states in real time. Furthermore, a rapid response mechanism is designed so that when the above problems are identified, the edge side can immediately issue an alarm and take measures such as stopping the machine or reducing its speed to prevent further production of defective products, further aggravation of equipment malfunctions, and further threats to employee safety, thereby improving product quality, reducing scrap rates, and increasing the overall operating efficiency of the production line. A real-time coupled simulation model integrating electromagnetic, thermal, mechanical, and acoustic technologies is established to simulate the internal physical processes of the equipment in real time. This approach overcomes the limitations of single-physics field simulation, which often ignores the coupling effect between fields, leading to results that deviate from reality. Real-time multi-field coupled simulation captures the chain reactions of physical processes, improving simulation accuracy and more closely resembling actual physical processes. Equipment failures are often the result of abnormal coupling of multiple physics fields, rather than a single factor. Real-time coupled simulation can better achieve early warning and root cause localization of failures, avoiding post-failure maintenance and improving production line efficiency. Based on Paris' crack propagation law and the Wiener process degradation model, the remaining lifespan of key components is calculated, overcoming the limitations of single models in terms of interpretability, uncertainty quantification, and full-cycle adaptability. This provides more accurate, reliable, and practical results for predicting the remaining lifespan of key components, supporting full lifecycle risk management of equipment from design to operation and maintenance. Through edge-side recognition of employee actions and detection of employee workstations, the final personnel management solution provides decision support, achieving intelligent, refined, and efficient personnel management. This solves problems such as insufficient monitoring of production line employees, single-physics field monitoring, and lack of fault prediction.

[0032] Specifically, Figure 1 This is a schematic diagram of the structure of a digital twin monitoring system for a precision servo press production line provided in an embodiment of this application.

[0033] This application provides a digital twin monitoring system for a precision servo press production line. The monitoring system 10 includes:

[0034] The system includes 100 physical production line modules, 200 edge computing modules, 300 cloud-based digital twin modules, and 400 visualization monitoring modules.

[0035] The physical production line module 100 is used to monitor and collect real-time data on the operating status of production line equipment, product data, and video data on the actions and workstations of production line employees. The edge computing module 200 uses an edge computing server to preprocess and analyze the collected raw data, and to judge product defects, potential equipment failures, and abnormal employee states in real time to trigger edge warnings. It combines equipment operating status, product and employee data to extract key information and output multi-dimensional datasets. The cloud-based digital twin module 300 is used to construct a digital twin to reflect the operating status of the production line in real time. It establishes a real-time coupled simulation model that integrates electromagnetic, thermal, mechanical and acoustic technologies to simulate the internal physical processes of the equipment in real time, predict equipment performance changes and potential failures, and locate the root cause of failures and predict the remaining lifespan based on edge warnings and coupled simulations. The visualization monitoring module 400 is used to present the simulation results of the cloud-based digital twin module, the real-time operating status of the equipment, the fault warning information and the remaining lifespan prediction data in a multi-dimensional visualization and provide decision support.

[0036] It is understood that this application embodiment uses edge computing to judge product defects, equipment failures, and abnormal employee states in real time to trigger edge early warnings. A defect detection model can achieve rapid, intelligent, and accurate detection of product defects; a voiceprint diagnostic model can analyze equipment operating sounds in real time to detect minor faults in internal parts; and a deep learning model can identify employee violations, unauthorized absences, and various abnormal states in real time. A rapid response mechanism is designed so that when the above problems are identified, the edge side can immediately issue an alarm and take measures such as stopping or slowing down the machine to prevent further production of product defects, further aggravation of equipment failures, and further threats to employee safety, thereby improving product quality, reducing scrap rates, and increasing the overall operating efficiency of the production line. A real-time coupled simulation model integrating electromagnetic, thermal, mechanical, and acoustic fields is established to simulate the internal physical processes of the equipment in real time, overcoming the limitations of single-physics field simulation. Single-physics field simulation often ignores the coupling effect between fields, leading to results that deviate from reality. Multi-field real-time coupled simulation can capture the chain reaction of physical processes, improving simulation accuracy and more closely resembling actual physical processes. Equipment failures are often the result of abnormal coupling of multiple physical fields. The failures are not caused by a single factor. Real-time coupled simulation can better achieve early warning and root cause location of faults, avoiding post-failure maintenance and improving production line efficiency. The cloud-based digital twin module, based on Paris crack propagation law and Wiener process degradation model, calculates the remaining life of key components, overcoming the limitations of single models in terms of "interpretability, uncertainty quantification, and full-cycle adaptability". It provides more accurate, reliable, and practical results for the remaining life prediction of key components, supporting risk management throughout the entire life cycle of equipment from design to operation and maintenance. The multi-dimensional visualization and decision support of the visualization monitoring module integrates real-time operating data, simulation results, life prediction, and historical fault records, making it easier to discover hidden patterns. The visualization of fault warnings and remaining life predictions allows managers to quickly capture high-priority risks on the monitoring screen, avoiding being overwhelmed by massive amounts of information. Warning tracing and impact simulation can help managers weigh "immediate shutdown for maintenance" versus "operation with warning until the end of the shift", reducing decision-making errors. Decision support can provide better equipment maintenance suggestions, take production line maintenance strategies in advance, and provide personnel management solutions, making managers' decisions more scientific.

[0037] In this embodiment, the physical production line module includes: a mechanical sensing sensor, a thermal sensing sensor, an electrical sensing sensor, a photoacoustic sensing sensor, and a position sensing sensor.

[0038] Among them, the mechanical sensing sensor is used to monitor the mechanical stress of key parts and collect the pressure value of key workstations; the thermal sensing sensor is used to monitor the temperature distribution and capture the micro temperature fluctuations of core components; the electrical sensing sensor is used to collect current harmonics and reflect the motor operating status through THD analysis; the photoacoustic sensing sensor is used to collect product images, personnel actions and workstation status and production line ambient sound; and the position sensing sensor is used to record the slider's full stroke position data.

[0039] Specifically, FBG fiber optic strain gauges are attached to the press slide and mold base, and the deformation signal is converted into a digital quantity by a wavelength demodulator for monitoring mechanical stress; an infrared thermopile array is used to non-contact scan the lead screw nut, and a PT100 thin-film temperature sensor is embedded in the servo motor winding for monitoring temperature distribution; a Rogowski coil is installed at the output of the driver to collect current harmonics THD and common-mode noise; cameras and microphones are installed at the conveyor belt inlet or outlet, next to the operating station, and above the workpiece placement platform to collect product images, personnel actions and station status, and production line ambient sound.

[0040] Understandably, placing multiple sensors at key locations on a device allows for real-time collection of operational status data, providing a data source foundation for subsequent multi-dimensional data mining to uncover underlying connections and patterns. When system problems occur, multi-dimensional data can help quickly pinpoint the root cause. In terms of prediction, multi-dimensional data can provide richer input features for the prediction model, improving prediction accuracy.

[0041] In this embodiment, the edge computing module uses an edge computing server to preprocess and analyze the collected raw data, and in real time judges product defects, potential equipment failures and abnormal employee status to trigger edge warnings. Combining equipment operating status, product and employee data, it extracts key information and outputs a multi-dimensional dataset.

[0042] Specifically, edge computing servers receive raw data collected by sensor networks in real time via industrial buses or wireless transmission. This includes equipment data, product data, and employee data, such as servo motor current, temperature, pressure head pressure, displacement, and transmission system vibration amplitude; workpiece dimensions, pressing accuracy, and surface defects; and video streams from workstation cameras, operation sequence, and on-duty or off-duty status. Preprocessing includes data cleaning, standardization, and dimensionality reduction. Data cleaning removes sensor noise, such as environmental interference signals from vibration sensors, using Kalman filtering to eliminate outliers and retain valid text. Standardization unifies data formats, such as standardizing temperature units to °C and synchronizing timestamps to UTC (Universal Time Coordinated), and extracts frames from unstructured data such as video streams to convert them into structured features. Dimensionality reduction compresses the dimensions of high-dimensional data using PCA (Principal Component Analysis) algorithms, retaining over 95% of key information and reducing the computational cost of subsequent analysis.

[0043] Based on the preprocessed data, the edge computing server deploys a lightweight model and a rule engine to achieve three types of core analysis, all of which are performed locally.

[0044] Specifically, the core analysis involves identifying product defects, potential equipment malfunctions, and abnormal employee states to trigger edge warnings. This includes building a defect detection model, a voiceprint diagnosis model, and a deep learning model at the edge. The defect detection model extracts features from product images to identify defects; the voiceprint diagnosis model uses deep learning algorithms to analyze equipment operating sounds in real time to determine if there are potential equipment malfunctions; and the deep learning model uses human skeleton key point detection technology and target detection and tracking technology to identify employee violations, determine absences from work, and monitor abnormal workstation states. Once an edge warning is triggered, it immediately triggers an emergency stop or speed reduction of the production line equipment.

[0045] It should be explained that the above model is pre-trained in the cloud and then deployed to the edge, supporting local incremental updates. The defect detection model uses a lightweight convolutional neural network, pre-trained in the cloud and then deployed to the edge server. The model input is a local image of the product captured by a camera, and the output is the defect type and confidence level (0-100%). After the image is converted to grayscale, edge and texture features are extracted through convolutional layers, and an attention mechanism is used to focus on high-risk areas. The Hough transform is used to detect key geometric features of the product, and the deviation between the actual size and the standard CAD model is calculated. If the deviation is > ±0.03mm, it is judged as a "size defect". The deviation threshold is set according to the product precision. Texture features include the linear edges of scratches and the circular outline of pores. When the confidence level of a defect in a single frame image is ≥90%, or the same defect is detected in 3 consecutive frames (i.e., confidence level ≥70%), a "product defect warning" is triggered. If the defect is "fatal," such as a crack on the press surface, the edge server directly sends a signal to the PLC (Programmable Controllers) to trigger an emergency stop of the equipment in the current process. If the defect is "minor," such as a scratch on a non-critical surface, the equipment is slowed down and a manual review is prompted.

[0046] The voiceprint diagnosis model employs a hybrid "CNN+LSTM" model. CNN (Convolutional Neural Network) extracts the spectral features of the sound signal, while LSTM (Long Short-Term Memory) captures temporal variation patterns. The model input is a 10-second audio clip captured by a microphone, and the output is the fault type and probability. Fault types include "abnormal noise from motor bearings" and "low oil in gearbox." The analysis logic involves converting the audio signal into a Mel-ray spectrogram, highlighting the characteristic frequencies of equipment faults. By comparing the real-time spectrum with a normal voiceprint database, if the amplitude of a characteristic frequency exceeds the historical average by 40%, or if a new abnormal frequency appears, the model identifies it as a "potential fault." When the fault probability is ≥80%, a "Level 1 Equipment Warning" is triggered, and the edge server immediately cuts off the equipment's power supply and pushes fault location information to the maintenance terminal. When the fault probability is between 50% and 80%, a "Level 2 Equipment Warning" is triggered, the equipment speed is reduced to 30% of its rated speed, and a "Prepare for Shutdown and Maintenance" message is displayed.

[0047] The deep learning model uses object detection technology to locate employee positions and workstation areas in real time, and uses skeletal keypoint technology to extract the coordinates of 17 key nodes of the human body to analyze movement trajectories. The behavior analysis logic is as follows: the compliance of actions is judged by the coordinates of skeletal keypoints. For example, if an employee does not wear protective gloves as required, i.e., there is no protective equipment feature at the wrist keypoint, or if the employee puts their hand into the pressing area while the equipment is running, i.e., the hand coordinates enter the preset danger zone, it is judged as "operation violation"; if no human target is detected in the workstation area for 15 consecutive seconds and the "leave post reporting" button signal is not triggered, it is judged as "unauthorized departure from the workstation"; if an employee falls in the workstation area, i.e., the verticality deviation of the skeletal keypoints is >45°, or remains still for a long time, i.e., the joint movement distance is <5cm within 10 seconds, it is judged as "abnormal state". When "unauthorized entry into a danger zone" is detected, the edge server immediately triggers an emergency stop of the equipment and activates an audible and visual alarm. When "unauthorized departure from the workstation" or "improper operation" is detected, the equipment is slowed down and a correction prompt is displayed on the workstation screen. When "abnormal employee status" is detected, in addition to the emergency stop of the equipment, an alarm message is simultaneously pushed to the safety monitoring center.

[0048] Furthermore, while providing real-time analysis and early warning, the edge computing server combines data from devices, products, and employees to extract key information and integrate it into a multi-dimensional dataset, which is then uploaded to the cloud via an encrypted transmission protocol. Moreover, the aforementioned edge model is not statically deployed but continuously optimized through a closed loop of "cloud training - edge inference - data feedback." The edge server periodically uploads early warning cases to the cloud; the cloud retrains the model based on new data, generates a lightweight version, and pushes it to the edge for updates; the iteration cycle can be set to once a week to ensure the model adapts to changes in production line conditions.

[0049] Understandably, edge-side localized early warning avoids the long delays of "data upload to the cloud - cloud analysis - feedback instructions," especially for equipment security risks and serious product defects, enabling responses within milliseconds and reducing losses. The edge side only uploads "critical information" rather than raw data, reducing data transmission volume by more than 70% and lowering bandwidth costs. Simultaneously, multi-dimensional datasets provide precise input for deep simulation and global optimization of cloud-based digital twins, achieving synergy between "real-time edge response + global cloud decision-making." Edge analysis and early warning can complete data processing, analysis, early warning, and response without relying on the cloud, ensuring stable operation of the production line during network fluctuations and improving system reliability.

[0050] In this embodiment, the cloud-based digital twin module includes: a twin construction unit, a simulation unit, and a deep analysis and prediction unit. Combined with... Figure 2 This is a schematic diagram of the cloud-based digital twin module of a digital twin monitoring system for a precision servo press production line, provided in an embodiment of this application.

[0051] The twin construction unit constructs a production line twin based on the equipment's 3D model and real-time operating data, including the equipment's position, attitude, operating parameters, and personnel activities. The simulation unit establishes a real-time coupled simulation model integrating electromagnetic, thermal, mechanical, and acoustic fields. Based on multiphysics theory, it uses a reduced-order model to compress finite element equations, achieving rapid and high-precision multiphysics simulation. The in-depth analysis and prediction unit uses a dynamic risk scoring model to output a real-time risk score of 0-100 to predict potential failures. Based on multi-data fusion analysis and multiphysics coupling inversion, it locates the root cause of failures. Based on Paris crack propagation law and Wiener process degradation model, it calculates the remaining life of key components.

[0052] Specifically, the steps for building a production line twin are as follows: collect detailed design drawings, equipment parameters, process flow, personnel status, and other information of the precision servo press production line; use 3D modeling software to build an accurate 3D model of the production line on the Uno Thing-JS platform based on the collected information, including various equipment, working environment, personnel status, etc.; optimize and calibrate the model to ensure that it is consistent with the geometry, size, and position of the actual production line.

[0053] The multi-dimensional dataset uploaded by the edge computing module is injected into the twin through the data interface: based on real-time data from displacement sensors, the coordinates of the corresponding devices in the twin are updated with an accuracy controlled within ±0.1mm to ensure complete synchronization with the movement trajectory of the physical devices; at the same time, parameters such as motor current, pressing force, and temperature are superimposed on the twin in the form of "dynamic labels + color mapping", such as the model displaying red when the motor temperature exceeds the threshold; combined with the skeletal key point data of the camera, a virtual motion image of the employee is generated in the twin, which restores the operation posture in real time, such as workpiece placement and button operation, and marks the on-duty status of the workstation, such as the virtual workstation displaying gray when the employee leaves the workstation.

[0054] The real-time coupled simulation model integrating electromagnetic, thermal, mechanical, and acoustic fields includes: an electromagnetic field simulation sub-model, a thermodynamic simulation sub-model, a mechanical stress simulation sub-model, an acoustic simulation sub-model, and a coupled solver. The coupled solver uses the alternating direction multiplier method to achieve multi-field coupled iterative calculation.

[0055] Specifically, the electromagnetic field simulation sub-model takes the motor current as input and is built based on Maxwell's equations. By solving the equations, it calculates the distribution of magnetic field strength and magnetic flux density in the servo motor windings, thereby obtaining the magnitude and direction of the electromagnetic force. For example, when analyzing the operating state of the servo motor, this sub-model can accurately determine the magnetic field distribution at various locations inside the motor under different currents, and the calculation results of the electromagnetic force provide key load inputs for subsequent mechanical stress simulations.

[0056] Furthermore, the governing equations for the electromagnetic field simulation are:

[0057]

[0058] in, Let be the cleavage operator, representing the sum of first-order partial differential vectors in three directions of space. For the magnetic permeability of silicon steel sheets, The conductivity of the motor windings. and To solve for the variables, This is a magnetic misalignment. To excite the current density, For time.

[0059] Used to calculate the magnetic field strength B, where Finally, Joule heat is input into the thermal simulation sub-model, and electromagnetic force density is input into the mechanical stress simulation sub-model.

[0060] The formula for Joule heating is:

[0061]

[0062] in, It is Joule fever. The conductivity of the motor windings. denoted as electric field strength.

[0063] The formula for deriving electromagnetic force density is:

[0064]

[0065] in, Electromagnetic force density, For conduction current density, is the magnetic flux density.

[0066] The thermodynamic simulation sub-model is established based on the heat conduction equation, combining motor loss and cooling system data. Electromagnetic losses are calculated by the electromagnetic field simulation sub-model, while mechanical losses are determined based on the transmission system operating parameters. The model simulates temperature field distribution, such as bearing temperature rise rate and winding temperature changes. For example, when the motor load changes, it can simulate the dynamic temperature changes of various components in real time, and the temperature calculation results can be used as the basis for material property changes in mechanical stress simulation.

[0067] Furthermore, the governing equations for the thermodynamic simulation are:

[0068]

[0069] in, For material density, For specific heat capacity, For temperature, For time, Thermal conductivity, Electromagnetic loss (from electromagnetic field simulation). The heat loss is due to friction (from mechanical stress simulation).

[0070] The output coupling terms are thermal strain and temperature-dependent material properties. Thermal strain is input into the mechanical stress simulation sub-model, and temperature-dependent material properties are fed back into the electromagnetic force field simulation sub-model.

[0071] The mechanical stress simulation sub-model is built based on the pressure of the indenter, the load of the transmission system, and the electromagnetic force provided by the electromagnetic field simulation sub-model, according to the equations of elasticity. It is used to calculate the stress distribution of components such as the lead screw and slider, including stress concentration at the root of the thread. For example, during the press-fitting process, the stress values ​​at the contact points between the indenter and the workpiece, as well as at key locations on the lead screw, can be obtained in real time. Simultaneously, the component vibration data calculated by this sub-model provides a vibration excitation source for the acoustic simulation sub-model.

[0072] The mechanical stress simulation control equation is as follows:

[0073]

[0074] in, For density, For acceleration vectors, For stress tensor, This is the electromagnetic force vector (from electromagnetic force field simulation). This is the volume force vector related to thermal stress.

[0075] Simultaneously, vibration displacement is output to the acoustic simulation sub-model, and stress intensity factor is output to the depth analysis unit.

[0076] The acoustic simulation sub-model takes the mechanical vibrations (such as gear meshing vibration and lead screw vibration) obtained from the mechanical stress simulation sub-model as input and constructs it through the sound radiation model. This model associates vibration characteristics with acoustic signature characteristics to simulate the sound of equipment operation. It can simulate parameters such as sound frequency and sound pressure level generated by the equipment under different vibration states, and then compare them with the acoustic signature data collected from the edge side to help determine the wear status of components.

[0077] The acoustic simulation control equations are as follows:

[0078]

[0079] in, For sound pressure, For the speed of sound, The vibration velocity (from mechanical stress simulation). This is the bearing vibration surface.

[0080] Simultaneously, simulated voiceprint features are output to the voiceprint diagnostic model of the edge computing module for comparison with the measured voiceprint.

[0081] The coupled solver employs the alternating direction multiplier method to achieve multi-field coupled iterative calculations. Due to the close parameter relationships between the sub-models—such as the heat generated by the electromagnetic field affecting the temperature field, and temperature changes altering material properties and thus influencing mechanical stress distribution—the coupled solver uses an alternating iterative approach, taking the calculation results of one sub-model as input parameters for other related sub-models, continuously iterating until the calculation results of all sub-models converge, thereby achieving real-time coupled simulation of multiphysics.

[0082] It should be explained that the coupling distribution of the magnetic field and temperature inside the motor can predict the risk of winding overheating; the stress change trend of the lead screw under different pressing forces can identify the location of fatigue damage; and the correlation curve between gearbox vibration and sound patterns can match fault sound pattern characteristics. These results are superimposed on the corresponding components of the twin in the form of color cloud maps, dynamic curves, etc., to achieve "virtual perspective".

[0083] Understandably, the alternating iterative method decomposes high-dimensional coupled problems into low-dimensional sub-problems and supports parallel computing, enabling real-time simulations integrating electromagnetic, thermal, mechanical, and acoustic systems to be completed within 50ms, meeting the real-time requirements of digital twins. However, the full-order finite element model has a large computational load and cannot meet real-time requirements. Therefore, a reduced-order model technique is adopted, extracting low-order information such as key vibration modes and temperature distribution characteristics through modal analysis, compressing the dimensionality of the finite element equations while retaining over 95% of the computational accuracy. This reduces the simulation time from 5 minutes to 50ms, meeting the real-time coupling requirements.

[0084] The deep analysis and prediction unit uses a dynamic risk scoring model to output a real-time risk score from 0 to 100 to predict potential failures. Based on multi-data fusion analysis and multi-physics coupling inversion, it locates the root cause of failures and calculates the remaining life of key components based on Paris' crack propagation law and the Wiener process degradation model. Specifically, it constructs a dynamic risk scoring model based on multi-dimensional data of "equipment-product-personnel," with a score from 0 to 100; the lower the score, the higher the risk. The weights for equipment, product, and personnel are set at 40%, 30%, and 30%, respectively. A point deduction system is implemented based on the stress or temperature margin of the input simulation unit, the frequency of edge warnings, the defect rate of the input defect monitoring model, the frequency of violations, and the duration of absence from duty. For example, 20 points are deducted when the lead screw stress reaches 80% of the rated value, 15 points are deducted for 3 level-two warnings within 1 hour, 25 points are deducted for a defect rate exceeding 5%, 10 points are deducted for a deviation exceeding 0.02mm, 20 points are deducted for 2 violations within 1 hour, and 15 points are deducted for absence from duty exceeding 10 minutes. The model updates the score every 10 seconds, and a cloud warning is triggered when the score is less than 60 points.

[0085] When a risk score anomaly or an edge warning is triggered, the root cause is located through "multi-data fusion + multi-physics inversion": fusing equipment parameters, product defect data, and multi-physics results from the simulation unit at the edge, and performing inversion based on a multi-physics coupling model. For example, if "pressing accuracy deviation + abnormal screw vibration" is detected, the simulation inversion shows "uneven stress distribution caused by screw thread wear". Combining this with the screw acoustic signature characteristics uploaded from the edge (the 200Hz characteristic frequency corresponding to wear), the root cause is ultimately located as "excessive wear of the screw".

[0086] For critical components (such as bearings and lead screws), the remaining life is calculated by combining damage mechanism models and data-driven models: the Paris crack propagation law calculates the crack propagation rate based on the stress intensity factor of the simulation unit and the vibration fatigue data of the edge side, and predicts the number of cycles from the "initial length" to the "critical fracture length" of the crack; the Wiener process degradation model uses historical degradation data as input and predicts the performance degradation trajectory through stochastic process modeling; the final life is taken as the weighted average of the two models (each with a weight of 50%), and is marked on the corresponding component of the twin (e.g., "Remaining life of lead screw: 120 hours").

[0087] Understandably, the deep analysis and prediction unit has been upgraded from "passive monitoring" to "proactive prediction". Dynamic risk scoring enables the quantification of global safety status, root cause location of faults reduces troubleshooting time by 70%, and remaining life prediction supports "preventive maintenance".

[0088] In this embodiment, the visualization monitoring module includes a digital twin visualization interface and a real-time monitoring screen.

[0089] Among them, the digital twin visualization interface uses AR technology to present a digital twin model of the production line, allowing users to immerse themselves in and interactively view real-time equipment parameters, personnel operations, and material flow. The real-time monitoring screen displays real-time data and video footage, and the interface shows key parameters and production indicators in the form of charts and other formats. It also supports historical data query and playback, facilitating data analysis and problem tracing.

[0090] Specifically, the digital twin visualization interface relies on AR technology to precisely overlay the digital twin model of the production line with the actual physical production line, creating an immersive interactive experience for users. Staff wearing AR glasses or using an AR display screen can intuitively see a virtual twin model that corresponds 1:1 to the physical scene. Elements in the model, such as equipment, personnel, and materials, are synchronized with the actual production status in real time.

[0091] In terms of interactive functions, users can interact with the virtual model through gestures, voice, or touch. For example, clicking on a virtual servo motor will immediately display the motor's real-time operating parameters, such as current, speed, and temperature, and can also display simulation results such as the electromagnetic field distribution and temperature field cloud map inside the motor. To view the material flow, users can issue the voice command "display material transport path," and the virtual model will dynamically mark the complete trajectory of the material from feeding to processing to discharging with highlighted lines. For personnel operations, users can select a virtual personnel model to view information such as their operation standard score and on-duty time. If there is any violation, the virtual model will mark the specific action part with a flashing red light.

[0092] In addition, when the system detects equipment failure or product defect, the AR visualization interface will actively mark the corresponding virtual model part. For example, it will display a prompt "Abnormal bearing temperature, estimated remaining life of 24 hours" at the location of the faulty bearing, and provide virtual arrows for maintenance guidance to help staff quickly locate the problem and take measures.

[0093] The real-time monitoring interface is primarily presented through a web-based interface, integrating real-time production line data and video feeds to provide managers with a comprehensive production status monitoring tool. For data display, the interface uses various chart formats to intuitively present key parameters and production indicators. For example, a line graph shows the temperature change trend of the servo motor over the past 24 hours, a bar chart compares the product defect rate at different workstations, and the dashboard displays the current pressing force value in real time with safety threshold ranges marked. When a parameter exceeds the threshold, the chart automatically changes color to trigger an alarm. Simultaneously, the interface also displays core production indicators in the form of data cards, such as total output, pass rate, and equipment utilization rate, allowing managers to quickly grasp the overall production situation.

[0094] In terms of video display, it can simultaneously receive high-definition camera feeds from multiple workstations, allowing users to freely switch between different perspectives. It can also use the picture-in-picture function to view the real-time status of multiple key workstations simultaneously. The video feed can also be linked with data; when a product defect warning occurs at a workstation, the corresponding video window will automatically zoom in and mark the defect location.

[0095] This interface supports historical data query and playback functions. Managers can select a time range to retrieve production data and video footage from any past period. For example, entering "August 17, 2025, 10:00-11:00" will display equipment operating parameter curves, product quality data, and corresponding video recordings for that period. This facilitates the traceability and analysis of past production issues, allowing for the summarization of lessons learned to optimize production processes.

[0096] Understandably, the digital twin visualization interface and real-time monitoring screen will also feed back user interaction commands and detected anomalies to other units, forming a data loop; and it can make data and results more intuitive and efficient for users to obtain and utilize; the collected user operation data and problem feedback can also be used to optimize the interface display and functions, continuously improving the system's practicality and ease of use.

[0097] Taking the daily operation scenario of the "automotive gearbox gear pressing station" in a precision servo press production line as an example, this application provides a detailed explanation of a digital twin monitoring system for a precision servo press production line. The scenario background of this production line is as follows: This station needs to press the gear blank and the drive shaft together using a servo press, with a pressing accuracy requirement of ±0.01mm. The core components of the equipment include a servo motor (power 15kW), a ball screw (diameter 50mm), and a press head (hardness HRC58). It processes an average of 500 sets of workpieces per day and needs to monitor the pressing force, motor status, workpiece quality, and operator specifications in real time.

[0098] Firstly, the physical production line module achieves multi-dimensional data acquisition: strain gauge sensors at the pressure head collect the pressing force in real time; stress sensors at the lead screw nut monitor the stress of the threaded pair; thermocouple sensors on the motor housing monitor the temperature distribution; infrared temperature probes are attached to the surface of the pressure head to capture microscopic temperature fluctuations caused by friction during the pressing process; current transformers in the motor control cabinet collect three-phase current and reflect the motor's operating status through THD analysis; a 2K industrial camera (30fps) above the workstation captures images of the workpiece pressing surface and operator actions, and a microphone array collects the sound of equipment operation; and a magnetic scale at the slider records the position data of the entire stroke.

[0099] Next, the edge computing module performs real-time analysis and early warning triggering: The edge computing server is deployed locally in the workshop with a latency of ≤20ms. It preprocesses the collected data and uses the YOLOv8 defect detection model to analyze the gear press-fit surface image, extracting features such as "indentation depth" and "skewness". When the indentation depth is detected to be >0.1mm, it is judged as a product defect. The voiceprint diagnosis model is based on the CNN-LSTM architecture to analyze the motor running sound in real time. If the 200-300Hz characteristic frequency corresponding to "ball screw wear" is found to deviate from the simulated voiceprint by more than 35%, an equipment fault warning is triggered. The deep learning model uses camera skeleton key point detection to identify abnormal states of operators such as "not wearing protective gloves" and "being away from the post for more than 10 seconds". When any of the above warnings is triggered, the edge side immediately performs the following actions: if it is a product defect, it triggers the pressure head to slow down (from 50mm / s to 10mm / s); if it is an equipment failure or employee abnormality, it triggers an emergency stop of the production line (response time < 1s), and at the same time packages key information such as "pressing force-position curve" and "acoustic spectrum" into a multi-dimensional dataset (approximately 10MB / time) and uploads it to the cloud.

[0100] Then, the digital twin module constructs a twin and performs in-depth analysis: the twin construction unit, based on the 3D model of the workstation, injects real-time data uploaded from the edge, drives the virtual slider movement through magnetic grating ruler data, displays the motor temperature using color mapping (red indicates >80℃), and overlays the virtual action image of the operator; the simulation unit calls the coupled model integrating electromagnetic-thermal-mechanical-acoustic: the electromagnetic field sub-model takes current data as input, calculates the magnetic field distribution of the motor winding through Maxwell's equations, and outputs electromagnetic losses; the thermodynamic sub-model uses electromagnetic losses and lead screw friction losses as heat sources, and obtains the true temperature of the motor winding through heat conduction equation simulation; the mechanical stress sub-model combines pressing force and temperature data to calculate the stress at the root of the lead screw thread; the coupled solver iteratively calculates using the alternating direction multiplier method, and completes the multi-field coupling results output within 50ms. The dynamic risk scoring model integrates equipment stress margin (10%), product defect rate (2%), and employee violation frequency (1 time / hour), outputting a real-time risk score of 58 points (<60 points). Based on multi-data fusion inversion: combining data on "lead screw stress concentration," "abnormal sound signature," and "temperature gradient," the root cause of the failure is located as "lead screw thread wear." The Paris crack propagation law (stress intensity factor) is then applied. Using the Wiener process model (inputting temperature degradation data for the past 30 days), the remaining life of the lead screw is calculated to be approximately 80 hours.

[0101] Finally, the multi-dimensional visualization monitoring module provides multi-dimensional presentation and decision support: workshop managers wearing AR glasses see a virtual twin model superimposed on the physical workstation. Clicking on the virtual lead screw brings up a real-time parameter panel (stress 180MPa, remaining life 80 hours). By using gestures to zoom, they can view the internal temperature field cloud map of the motor (the red area corresponds to the overheated location of the winding). The web interface displays the trend of pressing force changes over the past hour in a line graph (marked with a 120kN threshold line), and a bar chart compares the defect rates of each workstation (this workstation has a defect rate of 2.3%, higher than the average of 1.5%). The system also embeds the workstation camera footage (automatically zooming in on the lead screw area when a fault warning is issued). Based on the analysis results, the system pushes suggestions: "Immediately stop the machine and replace the lead screw (recommended spare part model: SFU5010)", "Adjust the pressing force to 110kN to reduce stress", and "Provide operators with training on protective procedures".

[0102] Understandably, in this scenario, the system transforms the traditional "post-incident maintenance" into "pre-incident prediction," predicting lead screw failures 80 hours in advance through digital twin simulation, thus avoiding a production loss of 50,000 yuan per day due to sudden downtime, while reducing the product defect rate from 3% to 0.8%.

[0103] According to the embodiments of this application, a digital twin monitoring system for a precision servo press production line is proposed. This system uses edge computing to judge product defects, equipment malfunctions, and abnormal employee states in real time to trigger edge early warnings. A defect detection model enables rapid, intelligent, and accurate detection of product defects. A voiceprint diagnostic model analyzes the operating sounds of the equipment in real time to detect minor faults in internal parts. A deep learning model identifies employee violations, unauthorized absences, and various abnormal states in real time. A rapid response mechanism is designed so that when the above problems are identified, the edge side can immediately issue an alarm and take measures such as stopping or slowing down the machine to prevent further production of product defects, further aggravation of equipment malfunctions, and further threats to employee safety, thereby improving product quality, reducing scrap rates, and increasing the overall operating efficiency of the production line. A real-time coupled simulation model integrating electromagnetic, thermal, mechanical, and acoustic fields is established to simulate the internal physical processes of the equipment in real time, overcoming the limitations of single-physics field simulation. Single-physics simulations often overlook the coupling effects between fields, leading to results that deviate from reality. Multi-field real-time coupled simulations, however, can capture the chain reactions of physical processes, improving simulation accuracy and more closely resembling actual physical processes. Equipment failures are often the result of abnormal coupling of multiple physical fields, rather than a single factor. Real-time coupled simulations can better achieve early warning and root cause localization of failures, avoiding post-failure repairs and improving production line efficiency. Based on Paris's crack propagation law and the Wiener process degradation model, the remaining lifespan of critical components is calculated, overcoming the limitations of single models in terms of interpretability, uncertainty quantification, and full-cycle adaptability. This provides more accurate, reliable, and practical results for predicting the remaining lifespan of critical components, supporting full lifecycle risk management of equipment from design to operation and maintenance. Through edge-side recognition of employee actions and detection of employee workstations, the final personnel management solution provides decision support, achieving intelligent, refined, and efficient personnel management. This solves technical problems such as insufficient monitoring of production line employees, single-field physical monitoring, and lack of fault predictability.

[0104] Figure 3 A flowchart illustrating a digital twin monitoring method for a precision servo press production line provided in this application embodiment.

[0105] like Figure 3 As shown, the method of the digital twin monitoring system for this precision servo press production line includes the following steps:

[0106] In step S101, the equipment operation status data, employee actions, and workstation status video data of the precision servo press production line are acquired.

[0107] The equipment operation data includes servo motor current, temperature, pressure head pressure, displacement, ball screw vibration amplitude, slider position, etc. The video data of employee actions and workstation status includes workstation camera video stream, operation sequence, and whether the employee is on duty or off duty.

[0108] Understandably, real-time collection of equipment operating status data provides a data source foundation for subsequent multi-dimensional data mining to uncover intrinsic connections and deep patterns. When system problems occur, multi-dimensional data can help quickly locate the root cause. In terms of prediction, multi-dimensional data can provide richer input features for prediction models, improving prediction accuracy.

[0109] In step S102, the collected raw data is preprocessed and edge analysis is performed to determine potential equipment failures and abnormal employee states in real time to trigger edge warnings, thereby triggering emergency stops or speed reductions of production line equipment. Key features are extracted and multi-dimensional datasets are output.

[0110] The preprocessing includes data cleaning, standardization, and dimensionality reduction. Data cleaning removes sensor noise, such as environmental interference signals from vibration sensors, using Kalman filtering to eliminate outliers and retain valid text. Standardization unifies data formats, such as standardizing temperature units to °C and synchronizing timestamps to the UTC (Universal Time Coordinated) time zone, and extracts frames from unstructured data such as video streams to convert them into structured features. Dimensionality reduction compresses the dimensions of high-dimensional data using the PCA (Principal Component Analysis) algorithm, retaining more than 95% of key information and reducing the computational cost of subsequent analysis.

[0111] The system identifies product defects, potential equipment malfunctions, and abnormal employee states to trigger edge warnings. Specifically, it involves building a defect detection model, a voiceprint diagnosis model, and a deep learning model at the edge. The defect detection model extracts features from product images to identify product defects; the voiceprint diagnosis model uses deep learning algorithms to analyze equipment operating sounds in real time to determine if there are potential equipment malfunctions; and the deep learning model uses human skeleton key point detection technology and target detection and tracking technology to identify employee violations, determine absences from work, and monitor abnormal workstation states. Once an edge warning is triggered, it immediately triggers an emergency stop or speed reduction of the production line equipment.

[0112] Specifically, the aforementioned model is pre-trained in the cloud and then deployed to the edge, supporting local incremental updates. The defect detection model employs a lightweight convolutional neural network, pre-trained in the cloud and then deployed to the edge server. The model input is a local image of the product captured by a camera, and the output is the defect type and confidence level (0-100%). After grayscale processing of the image, edge and texture features are extracted through convolutional layers, and an attention mechanism is used to focus on high-risk areas. The Hough transform is used to detect key geometric features of the product, and the deviation between the actual size and the standard CAD model is calculated. If the deviation is > ±0.03mm, it is judged as a "size defect". The deviation threshold is set according to the product precision. Texture features include the linear edges of scratches and the circular outline of pores. When the confidence level of a defect in a single frame image is ≥90%, or the same defect is detected in 3 consecutive frames (i.e., confidence level ≥70%), a "product defect warning" is triggered. If the defect is "fatal," such as a crack on the press surface, the edge server directly sends a signal to the PLC (Programmable Controllers) to trigger an emergency stop of the equipment in the current process. If the defect is "minor," such as a scratch on a non-critical surface, the equipment is slowed down and a manual review is prompted.

[0113] The voiceprint diagnosis model employs a hybrid "CNN+LSTM" model. CNN (Convolutional Neural Network) extracts the spectral features of the sound signal, while LSTM (Long Short-Term Memory) captures temporal variation patterns. The model input is a 10-second audio clip captured by a microphone, and the output is the fault type and probability. Fault types include "abnormal noise from motor bearings" and "low oil in gearbox." The analysis logic involves converting the audio signal into a Mel-ray spectrogram, highlighting the characteristic frequencies of equipment faults. By comparing the real-time spectrum with a normal voiceprint database, if the amplitude of a characteristic frequency exceeds the historical average by 40%, or if a new abnormal frequency appears, the model identifies it as a "potential fault." When the fault probability is ≥80%, a "Level 1 Equipment Warning" is triggered, and the edge server immediately cuts off the equipment's power supply and pushes fault location information to the maintenance terminal. When the fault probability is between 50% and 80%, a "Level 2 Equipment Warning" is triggered, the equipment speed is reduced to 30% of its rated speed, and a "Prepare for Shutdown and Maintenance" message is displayed.

[0114] The deep learning model uses object detection technology to locate employee positions and workstation areas in real time, and uses skeletal keypoint technology to extract the coordinates of 17 key nodes of the human body to analyze movement trajectories. The behavior analysis logic is as follows: the compliance of actions is judged by the coordinates of skeletal keypoints. For example, if an employee does not wear protective gloves as required, i.e., there is no protective equipment feature at the wrist keypoint, or if the employee puts their hand into the pressing area while the equipment is running, i.e., the hand coordinates enter the preset danger zone, it is judged as "operation violation"; if no human target is detected in the workstation area for 15 consecutive seconds and the "leave post reporting" button signal is not triggered, it is judged as "unauthorized departure from the workstation"; if an employee falls in the workstation area, i.e., the verticality deviation of the skeletal keypoints is >45°, or remains still for a long time, i.e., the joint movement distance is <5cm within 10 seconds, it is judged as "abnormal state". When "unauthorized entry into a danger zone" is detected, the edge server immediately triggers an emergency stop of the equipment and activates an audible and visual alarm. When "unauthorized departure from the workstation" or "improper operation" is detected, the equipment is slowed down and a correction prompt is displayed on the workstation screen. When "abnormal employee status" is detected, in addition to the emergency stop of the equipment, an alarm message is simultaneously pushed to the safety monitoring center.

[0115] Furthermore, while providing real-time analysis and early warning, the edge computing server combines data from devices, products, and employees to extract key information and integrate it into a multi-dimensional dataset, which is then uploaded to the cloud via an encrypted transmission protocol. Moreover, the aforementioned edge model is not statically deployed but continuously optimized through a closed loop of "cloud training - edge inference - data feedback." The edge server periodically uploads early warning cases to the cloud; the cloud retrains the model based on new data, generates a lightweight version, and pushes it to the edge for updates; the iteration cycle can be set to once a week to ensure the model adapts to changes in production line conditions.

[0116] Understandably, edge-side localized early warning avoids the long delays of "data upload to the cloud - cloud analysis - feedback instructions," especially for equipment security risks and serious product defects, enabling responses within milliseconds and reducing losses. The edge side only uploads "critical information" rather than raw data, reducing data transmission volume by more than 70% and lowering bandwidth costs. Simultaneously, multi-dimensional datasets provide precise input for deep simulation and global optimization of cloud-based digital twins, achieving synergy between "real-time edge response + global cloud decision-making." Edge analysis and early warning can complete data processing, analysis, early warning, and response without relying on the cloud, ensuring stable operation of the production line during network fluctuations and improving system reliability.

[0117] In step S103, a production line twin is constructed based on the equipment's 3D model and real-time operating data. A real-time coupled simulation model integrating electromagnetic, thermal, mechanical, and acoustic technologies is established to simulate the internal physical processes of the equipment in real time, predict equipment performance changes and potential faults, and the cloud data platform locates the root cause of the fault and predicts the remaining lifespan based on edge early warning and coupled simulation.

[0118] The specific steps for constructing a production line twin are as follows: collect detailed design drawings, equipment parameters, process flow, personnel status, and other information of the precision servo press production line; use 3D modeling software to construct an accurate 3D model of the production line on the Uno Thing-JS platform based on the collected information, including various equipment, working environment, personnel status, etc.; optimize and calibrate the model to ensure that it is consistent with the geometry, size, and position of the actual production line.

[0119] The real-time coupled simulation model integrating electromagnetic, thermal, mechanical, and acoustic fields includes: an electromagnetic field simulation sub-model, a thermodynamic simulation sub-model, a mechanical stress simulation sub-model, an acoustic simulation sub-model, and a coupled solver. The coupled solver uses the alternating direction multiplier method to achieve multi-field coupled iterative calculation.

[0120] Specifically, the electromagnetic field simulation sub-model takes the motor current as input and is built based on Maxwell's equations. By solving the equations, it calculates the distribution of magnetic field strength and magnetic flux density in the servo motor windings, thereby obtaining the magnitude and direction of the electromagnetic force. For example, when analyzing the operating state of the servo motor, this sub-model can accurately determine the magnetic field distribution at various locations inside the motor under different currents, and the calculation results of the electromagnetic force provide key load inputs for subsequent mechanical stress simulations.

[0121] Furthermore, the governing equations for the electromagnetic field simulation are:

[0122]

[0123] in, Let be the cleavage operator, representing the sum of first-order partial differential vectors in three directions of space. For the magnetic permeability of silicon steel sheets, The conductivity of the motor windings. and To solve for the variables, This is a magnetic misalignment. To excite the current density, For time.

[0124] Used to calculate the magnetic field strength B, where Finally, Joule heat is input into the thermal simulation sub-model, and electromagnetic force density is input into the mechanical stress simulation sub-model.

[0125] The formula for Joule heating is:

[0126]

[0127] in, It is Joule fever. The conductivity of the motor windings. denoted as electric field strength.

[0128] The formula for deriving electromagnetic force density is:

[0129]

[0130] in, Electromagnetic force density, For conduction current density, is the magnetic flux density.

[0131] The thermodynamic simulation sub-model is established based on the heat conduction equation, combining motor loss and cooling system data. Electromagnetic losses are calculated by the electromagnetic field simulation sub-model, while mechanical losses are determined based on the transmission system operating parameters. The model simulates temperature field distribution, such as bearing temperature rise rate and winding temperature changes. For example, when the motor load changes, it can simulate the dynamic temperature changes of various components in real time, and the temperature calculation results can be used as the basis for material property changes in mechanical stress simulation.

[0132] Furthermore, the governing equations for the thermodynamic simulation are:

[0133]

[0134] in, For material density, For specific heat capacity, For temperature, For time, Thermal conductivity, Electromagnetic loss (from electromagnetic field simulation). The heat loss is due to friction (from mechanical stress simulation).

[0135] The output coupling terms are thermal strain and temperature-dependent material properties. Thermal strain is input into the mechanical stress simulation sub-model, and temperature-dependent material properties are fed back into the electromagnetic force field simulation sub-model.

[0136] The mechanical stress simulation sub-model is built based on the pressure of the indenter, the load of the transmission system, and the electromagnetic force provided by the electromagnetic field simulation sub-model, according to the equations of elasticity. It is used to calculate the stress distribution of components such as the lead screw and slider, including stress concentration at the root of the thread. For example, during the press-fitting process, the stress values ​​at the contact points between the indenter and the workpiece, as well as at key locations on the lead screw, can be obtained in real time. Simultaneously, the component vibration data calculated by this sub-model provides a vibration excitation source for the acoustic simulation sub-model.

[0137] The mechanical stress simulation control equation is as follows:

[0138]

[0139] in, For density, For acceleration vectors, For stress tensor, This is the electromagnetic force vector (from electromagnetic force field simulation). This is the volume force vector related to thermal stress.

[0140] Simultaneously, vibration displacement is output to the acoustic simulation sub-model, and stress intensity factor is output to the depth analysis unit.

[0141] The acoustic simulation sub-model takes the mechanical vibrations (such as gear meshing vibration and lead screw vibration) obtained from the mechanical stress simulation sub-model as input and constructs it through the sound radiation model. This model associates vibration characteristics with acoustic signature characteristics to simulate the sound of equipment operation. It can simulate parameters such as sound frequency and sound pressure level generated by the equipment under different vibration states, and then compare them with the acoustic signature data collected from the edge side to help determine the wear status of components.

[0142] The acoustic simulation control equations are as follows:

[0143]

[0144] in, For sound pressure, For the speed of sound, The vibration velocity (from mechanical stress simulation). This is the bearing vibration surface.

[0145] Simultaneously, simulated voiceprint features are output to the voiceprint diagnostic model of the edge computing module for comparison with the measured voiceprint.

[0146] The coupled solver employs the alternating direction multiplier method to achieve multi-field coupled iterative calculations. Due to the close parameter relationships between the sub-models—such as the heat generated by the electromagnetic field affecting the temperature field, and temperature changes altering material properties and thus influencing mechanical stress distribution—the coupled solver uses an alternating iterative approach, taking the calculation results of one sub-model as input parameters for other related sub-models, continuously iterating until the calculation results of all sub-models converge, thereby achieving real-time coupled simulation of multiphysics.

[0147] Furthermore, for key components, the remaining life is calculated by combining the damage mechanism model and the data-driven model: the Paris crack propagation law calculates the crack propagation rate based on the stress intensity factor of the simulation unit and the vibration fatigue data of the edge side, and predicts the number of cycles from the "initial length" to the "critical fracture length" of the crack; the Wiener process degradation model uses historical degradation data as input and predicts the performance degradation trajectory through stochastic process modeling; the final life is taken as the weighted average of the two models (each with a weight of 50%), and is marked on the corresponding component of the twin.

[0148] Understandably, the alternating iterative method decomposes high-dimensional coupled problems into low-dimensional sub-problems and supports parallel computing, enabling real-time simulations integrating electromagnetic, thermal, mechanical, and acoustic systems to be completed within 50ms, meeting the real-time requirements of digital twins. However, the full-order finite element model has a large computational load and cannot meet real-time requirements. Therefore, a reduced-order model technique is adopted, extracting low-order information such as key vibration modes and temperature distribution characteristics through modal analysis, compressing the dimensionality of the finite element equations while retaining over 95% of the computational accuracy. This reduces the simulation time from 5 minutes to 50ms, meeting the real-time coupling requirements.

[0149] In step S104, the simulation results, real-time equipment operating status, fault warning information and remaining life prediction data are presented in a multi-dimensional visualization and decision support is provided. The decision support includes providing equipment maintenance suggestions, taking production line maintenance strategies in advance, and providing personnel management solutions.

[0150] The aforementioned visualization and decision support are achieved based on a digital twin visualization interface and real-time monitoring screens.

[0151] Specifically, the digital twin visualization interface uses AR technology to present a digital twin model of the production line, allowing users to immerse themselves in and interactively view real-time equipment parameters, personnel operations, and material flow. The real-time monitoring screen displays real-time data and video footage, and the interface shows key parameters and production indicators in the form of charts and other formats. It also supports historical data query and playback, facilitating data analysis and problem tracing.

[0152] Understandably, the digital twin visualization interface and real-time monitoring screen will also feed back user interaction commands and detected anomalies to other units, forming a data loop; and it can make data and results more intuitive and efficient for users to obtain and utilize; the collected user operation data and problem feedback can also be used to optimize the interface display and functions, continuously improving the system's practicality and ease of use.

[0153] Taking the "main bearing cap press-fitting station" of an automotive engine cylinder block as an example, this application details a digital twin monitoring method for a precision servo press production line. This station requires pressing aluminum alloy bearing caps into the cylinder block with a pressure of 12-15 kN and an accuracy of ±0.02 mm. The core equipment is a servo press (equipped with an 18 kW servo motor and a 40 mm diameter ball screw). An average of 800 cylinder blocks are press-fitted daily, requiring strict monitoring of press-fitting quality, equipment health, and operator safety.

[0154] First, multi-source data acquisition is performed: A current transformer collects the servo motor current, which fluctuates between 8-12A during normal pressing, and is recorded every 5ms; a strain gauge sensor collects the pressure head pressure, with the peak pressure stabilizing at 13kN±0.5kN, and a sampling frequency of 1kHz; an accelerometer collects the ball screw vibration, with a normal amplitude ≤0.2g (g is gravitational acceleration), and high-frequency sampling (500Hz) captures impact signals; a magnetic scale records the slider position, with a pressing stroke of 50-150mm and an accuracy of ±0.01mm, providing real-time feedback on pressing progress. A workstation camera collects operator actions at 30 frames per second, focusing on "hand protection" and "operation trajectory"; millimeter-wave radar monitors personnel presence to determine if unauthorized personnel are approaching.

[0155] Edge preprocessing and analysis: The ball screw vibration signal is processed to filter out environmental noise from the workshop air compressor (80Hz), retaining the 200-300Hz characteristic frequency generated by the pressing impact; the temperature unit is standardized to ℃ (converted from the original data of the motor temperature sensor); one frame is extracted from every three frames of the video stream and converted into structured data of "personnel position - equipment status"; dimensionality reduction is performed on three types of high-frequency data (motor current, pressure, and vibration, totaling 2000+ dimensions), retaining 95% of the information (compressed to 30 dimensions) to reduce edge computing power consumption. For local images of the bearing cover pressing surface captured by the camera, the convolutional layer identifies the "linear edge" of the scratch (length > 2mm, width > 0.1mm), and the flatness of the pressing surface is detected by combining Hough transform (deviation > ±0.03mm is used to determine dimensional defects). If a scratch is detected in two consecutive frames (confidence level 85%), a "product defect warning" is triggered, the equipment speed is reduced to 50% of the rated speed, and the workstation screen displays "Scratches on the bearing cover of the 3rd cylinder block, manual review required."

[0156] The 10-second audio (20-20kHz) collected by the microphone is converted into a Mel spectrogram. The probability of a fault is detected to be 82%, triggering a "Level 1 Equipment Warning". The edge server immediately cuts off the motor power supply and pushes "Motor bearing abnormal, estimated remaining lifespan 2 hours" to the maintenance terminal. At the same time, the PLC controls the equipment to stop urgently (response delay 80ms).

[0157] The camera extracts the coordinates of 17 joints of the human body. During the pressing process (50-150mm from the slider position), if the hand coordinates fall into the preset danger zone (100mm in front of the pressing area), it is determined that "the person has entered the danger zone in violation of regulations". The equipment is immediately stopped and the audible and visual alarm is activated (80dB alarm in the workshop). At the same time, the system pushes the message "Employee A has violated regulations and the equipment has been stopped" to the safety monitoring center.

[0158] The edge server integrates key information such as "bearing cover scratch coordinates (X: 12mm, Y: 5mm), motor bearing failure frequency (200Hz), and employee violation time (10:23:45)," and uploads it to the cloud via 5G after encryption (AES-256) (data packet size 12MB, upload time 20ms).

[0159] Cloud-based twin construction and simulation analysis: Based on the equipment's 3D model (CAD conversion), real-time data uploaded from the edge is injected. The real-time data includes motor temperature (85℃, the motor casing in the virtual model is highlighted in red), slider position (120mm, the virtual slider moves synchronously), and employee violation location (danger zone red warning box, the virtual personnel model's actions are restored). In the electromagnetic field simulation sub-model, the motor current (12A) is input, the winding magnetic field distribution is calculated, and the electromagnetic loss (150W, as a heat source) is output. In the thermodynamic simulation sub-model, the motor temperature field is simulated by combining electromagnetic loss and bearing friction heat (50W), and the actual winding temperature (92℃, 7℃ higher than the surface sensor, explaining the overheating risk) is obtained. In the mechanical stress simulation sub-model, the pressing force (13kN) and temperature (92℃) are input, and the stress distribution of the motor bearing is calculated (stress concentration in the bearing rollers reaches 250MPa, close to the material fatigue limit of 280MPa). In the acoustic sub-model, the bearing vibration (acceleration 0.3g) is converted into acoustic signature features (200Hz abnormal frequency), and compared with the edge-collected acoustic signature (deviation 35%, verifying the fault). The coupled solver uses the alternating direction multiplier method to iterate for 50ms, converges the multi-field results, and locates the root cause as "insufficient lubrication of the motor bearing leads to wear, and stress concentration accelerates failure".

[0160] Based on Paris's crack propagation law, input the stress intensity factor ( Vibration fatigue data were used to calculate the time required for a crack to propagate from 0.1 mm to the critical length (0.5 mm) in 80 hours. Bearing temperature degradation data (average 85℃ to 92℃) for the past 30 days were input into the Wiener process model to predict the performance degradation trajectory. The results of the two models were weighted (50% each) to output the remaining life of the motor bearing of 78 hours (matching the actual working conditions).

[0161] Maintenance personnel wearing AR glasses see a virtual twin superimposed on the physical workstation. Clicking on the virtual motor brings up a parameter panel (temperature 92℃, remaining lifespan 78 hours, stress distribution cloud map). Gestures zoom in and out to view the motor's internal structure, highlighting the worn bearing area in red and displaying maintenance instructions: "Replace bearing model XX, recommended shutdown after 16:00 today." On the workshop's central control screen, a line graph shows the motor temperature change over 72 hours (92℃ exceeds the threshold of 85℃, red warning), and a bar chart compares the defect rates of each workstation (this workstation has a 2.3% defect rate, higher than the average of 1.5%). When a fault occurs, the camera automatically switches to a close-up of the motor bearing, marking the wear location. Suggestions for equipment maintenance, production line maintenance, and personnel management are provided, including: "Immediately stockpile spare motor bearings, as their lifespan will expire in 80 hours," "Shut down for maintenance from 16:00 to 16:30 today, replace the bearing, optimize the lubrication system," and "Provide safety training to operator A (one violation), standardizing hand movement techniques."

[0162] This scenario fully demonstrates the closed loop of the system from "data collection - edge response - cloud analysis - decision implementation", allowing the value of digital twin monitoring to be transformed from technical logic into quantifiable production benefits.

[0163] According to the embodiments of this application, a digital twin monitoring method for a precision servo press production line is proposed. This method uses edge computing to judge product defects, equipment malfunctions, and abnormal employee states in real time to trigger edge early warnings. A defect detection model enables rapid, intelligent, and accurate detection of product defects. A voiceprint diagnostic model analyzes the operating sound of the equipment in real time to detect minor faults in internal parts. A deep learning model identifies employee violations, unauthorized absences, and various abnormal states in real time. A rapid response mechanism is designed so that when the above problems are identified, the edge side can immediately issue an alarm and take measures such as stopping or slowing down the machine to prevent further production of product defects, further aggravation of equipment malfunctions, and further threats to employee safety, thereby improving product quality, reducing scrap rates, and increasing the overall operating efficiency of the production line. A real-time coupled simulation model integrating electromagnetic, thermal, mechanical, and acoustic fields is established to simulate the internal physical processes of the equipment in real time, overcoming the limitations of single-physics field simulation. Single-physics simulations often overlook the coupling effects between fields, leading to results that deviate from reality. Multi-field real-time coupled simulations, however, can capture the chain reactions of physical processes, improving simulation accuracy and more closely resembling actual physical processes. Equipment failures are often the result of abnormal coupling of multiple physical fields, rather than a single factor. Real-time coupled simulations can better achieve early warning and root cause localization of failures, avoiding post-failure repairs and improving production line efficiency. Based on Paris's crack propagation law and the Wiener process degradation model, the remaining lifespan of critical components is calculated, overcoming the limitations of single models in terms of interpretability, uncertainty quantification, and full-cycle adaptability. This provides more accurate, reliable, and practical results for predicting the remaining lifespan of critical components, supporting full lifecycle risk management of equipment from design to operation and maintenance. Through edge-side recognition of employee actions and detection of employee workstations, the final personnel management solution provides decision support, achieving intelligent, refined, and efficient personnel management. This solves technical problems such as insufficient monitoring of production line employees, single-field physical monitoring, and lack of fault predictability.

[0164] It should be noted that the foregoing explanation of the system embodiment of a digital twin monitoring system for a precision servo press production line also applies to the method of the digital twin monitoring system for a precision servo press production line in this embodiment, and will not be repeated here.

[0165] Figure 4 A schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include:

[0166] The memory 401, the processor 402, and the computer program stored on the memory 401 and capable of running on the processor 402.

[0167] When the processor 402 executes the program, it implements the digital twin monitoring method for a precision servo press production line provided in the above embodiments.

[0168] Furthermore, electronic devices also include:

[0169] Communication interface 403 is used for communication between memory 401 and processor 402.

[0170] The memory 401 is used to store computer programs that can run on the processor 402.

[0171] Memory 401 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0172] If the memory 401, processor 402, and communication interface 403 are implemented independently, then the communication interface 403, memory 401, and processor 402 can be interconnected via a bus to complete communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, Figure 4 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0173] Optionally, in a specific implementation, if the memory 401, processor 402, and communication interface 403 are integrated on a single chip, then the memory 401, processor 402, and communication interface 403 can communicate with each other through an internal interface.

[0174] Processor 402 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0175] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described digital twin monitoring method for a precision servo press production line.

[0176] This application also provides a computer program product that stores a computer program that, when executed by a processor, implements the above-described digital twin monitoring method for a precision servo press production line.

[0177] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0178] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0179] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0180] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0181] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

Claims

1. A digital twin monitoring system for a precision servo press line, the system comprising: Comprise: Physical production line module, edge computing module, cloud digital twin module, visual monitoring module; wherein, The physical production line module is used for real-time monitoring and collecting production line equipment operation state data, product data, action and station situation video data of production line employees; The edge computing module adopts an edge computing server to pre-process and analyze the collected raw data, judges product defects, equipment failures and employee abnormal states in real time to trigger edge early warning, extracts key information to output multi-dimensional data sets in combination with equipment operation state, product and employee data, wherein the judging product defects, equipment failures and employee abnormal states to trigger edge early warning comprises: constructing a defect detection model, a voiceprint diagnosis model and a deep learning model on the edge side, the defect detection model extracts features from product images to identify product defects; the voiceprint diagnosis model uses a deep learning algorithm to analyze equipment operation sound in real time to determine whether the equipment has a fault; the deep learning model uses human body skeleton key point detection technology and target detection and tracking technology to identify employee violations, determine off-duty and monitor abnormal station state; the edge early warning triggers production line equipment emergency stop or speed reduction immediately; The cloud digital twin module is used for constructing a digital twin, reflecting production line operation state in real time, establishing a real-time coupling simulation model that fuses electromagnetic-thermal-mechanical-acoustic to simulate equipment internal physical process in real time, predicting equipment performance change and potential failure, positioning fault root cause and predicting remaining life based on the output of the edge computing module and coupling simulation, wherein the cloud digital twin module comprises: a twin construction unit, a simulation unit and a deep analysis and prediction unit, wherein the twin construction unit constructs a production line twin containing equipment position, posture, operation parameters and personnel activity based on equipment three-dimensional model and real-time operation data; the simulation unit establishes a real-time coupling simulation model that fuses electromagnetic-thermal-mechanical-acoustic, adopts reduced-order model to compress finite element equations based on multi-physical field theory, realizes fast and high-precision multi-physical field simulation; the deep analysis and prediction unit adopts a dynamic risk scoring model to output real-time risk score of 0-100, predicts potential failure, positions fault root cause based on multi-data fusion analysis and multi-physical field coupling inversion, calculates remaining life of key components based on Paris crack propagation law and Wiener process degradation model; The visual monitoring module is used for multi-dimensional visual presentation of simulation results, equipment real-time operation state, fault early warning information and remaining life prediction data of the cloud digital twin module and provides decision support.

2. The digital twin monitoring system for a precision servo press line of claim 1, wherein, The physical production line module comprises a mechanical perception sensor, a thermal perception sensor, an electrical perception sensor, an opto-acoustic perception sensor and a position perception sensor, wherein the mechanical perception sensor is used to monitor the mechanical stress of key positions and collect the pressure value of the key workstations; the thermal perception sensor is used to monitor the temperature distribution and capture the microscopic temperature fluctuations of the core components; the electrical perception sensor is used to collect the current harmonics and reflect the motor operating state through THD analysis; the opto-acoustic perception sensor is used to collect product images, personnel actions, workstation states and production line environment sounds; and the position perception sensor is used to record the full-stroke position data of the slider.

3. The digital twin monitoring system for a precision servo press line of claim 1, wherein, The real-time coupling simulation model of the fusion of electromagnetic-thermal-mechanical-acoustic comprises an electromagnetic field simulation sub-model, a thermodynamic simulation sub-model, a mechanical stress simulation sub-model, an acoustic simulation sub-model and a coupling solver, wherein the coupling solver realizes multi-field coupling iterative calculation by using an alternating direction multiplier method.

4. The digital twin monitoring system for a precision servo press line of claim 1, wherein, The visual monitoring module comprises a digital twin visualization interface and a real-time monitoring picture, wherein the digital twin visualization interface presents the production line digital twin model through AR technology, and a user can immerse in and interactively view the real-time parameters of the equipment, personnel operations and material flow conditions; and the real-time monitoring picture is used to display real-time data and video pictures, the interface displays key parameters and production indexes in the form of charts, supports historical data query and playback, and is convenient for data analysis and problem tracing.

5. A method for applying a digital twin monitoring system to a line of precision servo presses according to any one of claims 1-4, characterized in that, The method comprises: acquiring equipment operating state data, personnel action and workstation condition video data of the precision servo press production line; preprocessing and edge analysis are performed on the collected raw data, equipment failure and personnel abnormal state are judged in real time, edge early warning is triggered to trigger production line equipment emergency stop or speed reduction, key features are extracted and a multi-dimensional data set is output; a production line twin is constructed based on the three-dimensional model of the equipment and real-time operating data, a real-time coupling simulation model of the fusion of electromagnetic-thermal-mechanical-acoustic is established to simulate the internal physical process of the equipment in real time, predict the performance change and potential failure of the equipment, and locate the root cause of the failure and predict the remaining life based on the output of the edge computing module and the coupling simulation; the simulation results, real-time operating state of the equipment, failure early warning information and remaining life prediction data are presented in multi-dimensions and decision support is provided, wherein the decision support comprises giving equipment maintenance suggestions, taking production line maintenance strategies in advance and giving personnel management schemes.

6. An electronic device, comprising: The computer program or instructions are executed to realize the digital twin monitoring method of the precision servo press production line of claim 5.

7. A computer readable storage medium having stored thereon a computer program or instructions, characterized in that, The computer program or instructions are executed to realize the digital twin monitoring method of the precision servo press production line of claim 5.

8. A computer program product comprising computer programs or instructions, characterized in that, The computer program or instructions are executed to realize the digital twin monitoring method of the precision servo press production line of claim 5.

Images