Mechanical and electrical equipment measurement and control data driven digital twin linkage management method

Abstract

This invention belongs to the field of equipment fault monitoring technology and discloses a digital twin linkage management method driven by electromechanical equipment measurement and control data. It uses a low-latency communication link to synchronize pre-processed core measurement and control data from the edge to a four-layer linkage digital twin architecture in real time, achieving millisecond-level synchronous operation between the physical entity and the virtual model. The synchronization accuracy verification module can automatically trigger model parameter correction, continuously and stably ensuring mapping accuracy. The reverse link directly sends fault warning commands, predictive maintenance strategies, and operating condition optimization parameters to the physical entity measurement and control system through encrypted communication, driving the equipment to execute corresponding control actions. The forward link collects data after execution in real time to verify the effect. A multi-source sensor network covering the core components of the equipment, the measurement and control link, and the operating environment is deployed to collect all heterogeneous measurement and control data. Hardware timestamps combined with linear interpolation and resampling achieve a unified spatiotemporal reference for the data.

Description

Technical Field

[0001] This invention belongs to the field of equipment fault monitoring technology, specifically a digital twin linkage management method driven by electromechanical equipment measurement and control data. Background Technology

[0002] High-end electromechanical equipment generates a large amount of multi-source heterogeneous measurement and control data during operation, including operating status parameters such as vibration, temperature, pressure, current, speed, and displacement, as well as measurement and control link data such as control commands, action feedback, and operating parameters. Digital twin technology, as a core technology for real-time mapping between physical entities and virtual spaces, has been widely applied in the fields of electromechanical equipment condition monitoring, fault diagnosis, and operation management. However, existing related technologies still have the following technical problems: Existing technologies mostly adopt a fixed pattern of mechanism modeling and one-way data mapping, which can only achieve one-way synchronization of field measurement and control data to the virtual model. It cannot build a fully closed-loop two-way linkage between the physical entity and the digital twin. The simulation optimization results and fault warning strategies of the virtual model cannot directly drive the measurement and control system of the physical entity in real time, resulting in the digital twin only remaining at the state visualization level.

[0003] The existing technology for measurement and control data processing is severely disconnected from the twin modeling process. It generally adopts a mode of uploading all data and processing it centrally in the cloud. It has not optimized the pain points of multi-source heterogeneity, large differences in sampling frequency, and spatiotemporal asynchrony of electromechanical equipment measurement and control data, resulting in high data transmission bandwidth consumption, large processing latency, and insufficient synchronization accuracy between twin models and physical entities.

[0004] Existing technologies lack a closed-loop iterative optimization mechanism for the entire process, and cannot fully integrate and feedback equipment operation data, maintenance execution data, and strategy execution effect data. As a result, twin models and management strategies cannot achieve adaptive iterative optimization. Summary of the Invention

[0005] The purpose of this invention is to provide a digital twin linkage management method driven by electromechanical equipment measurement and control data, so as to solve one or more problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a digital twin linkage management method driven by electromechanical equipment measurement and control data, comprising the following specific steps: Furthermore, in the measurement, control, and screening stage, a multi-source sensor network covering the core components of the equipment, the measurement and control link, and the operating environment is deployed to collect vibration, temperature, pressure, current, speed, displacement, control commands, action feedback data, and on-site video data for AR color recognition. The video stream is preferentially transmitted via wired link. A spatiotemporal synchronization mechanism based on hardware timestamps is built at the edge. For sensors with different sampling frequencies, including high-frequency vibration sensors and low-frequency temperature and pressure sensors, a combination of linear interpolation and resampling is used to achieve a unified spatiotemporal reference for all measurement and control data, distinguishing between high-speed measurement and control data and low-speed consumable wear data, and adapting them to the transmission links respectively. A lightweight time-series feature extraction network is deployed at the edge to pre-screen the synchronized measurement and control data. Based on data-driven dynamic thresholds, redundant data is filtered out, and core feature data that is strongly correlated with the equipment's operating status is retained. At the same time, the sensor sampling frequency and screening threshold are dynamically adjusted according to the real-time operating conditions of the equipment. For wear signals of consumables such as pulleys, a 10-second upload cycle is set, and NB-IoT narrowband low-speed transmission is adopted.

[0007] The lightweight temporal feature extraction network is specifically adapted to the limited computing power of edge hardware. It has a small model size, fast inference speed, and low memory usage. It can run stably on embedded edge computing units and can complete real-time feature extraction and filtering without relying on cloud computing power. The entire data processing is completed locally, further reducing data transmission latency.

[0008] Furthermore, the twin modeling stage constructs a four-layer linked digital twin architecture based on the core measurement and control data screened in the measurement and control data collection and screening stage. The digital twin architecture includes a physical entity layer twin, an operational status layer twin, a performance evaluation layer twin, and a decision management layer twin. The physical entity layer twin maps the geometric structure, physical attributes, hardware interfaces, and measurement and control links of the electromechanical equipment, replicating the physical characteristics of the equipment; the operation status layer twin maps the operating conditions, action status, and environmental parameters of the equipment in real time based on core measurement and control data, achieving real-time synchronization with the physical entity; the performance evaluation layer twin is used for quantitative evaluation and fault simulation of equipment health, operating efficiency, and reliability; the decision management layer twin is used for the generation and simulation verification of operation and maintenance strategies, operating condition optimization, and scheduling instructions. The modeling process adopts a data-driven incremental learning mode, using pre-screened measurement and control data as the core input, combined with prior knowledge of equipment mechanisms, to dynamically correct the model parameters of each twin layer. When the equipment operating conditions change or components wear out, the model is automatically updated incrementally based on real-time measurement and control data.

[0009] Furthermore, the bidirectional linkage stage is based on the four-layer digital twin architecture constructed in the twin modeling stage, and constructs a bidirectional mapping linkage mechanism of forward real-time mapping and reverse control. The forward mapping link synchronizes the core measurement and control data processed at the edge of the measurement and control data acquisition and screening stage to the digital twin in real time via a hybrid low-latency communication link of 5G and industrial Ethernet. Wired transmission is prioritized to reduce industrial wireless interference, while consumable wear data is transmitted using a narrowband low-speed NB link. The total end-to-end latency is ≤50ms, and the single packet transmission latency of core measurement and control data is ≤10ms. This drives the twin at each layer to achieve millisecond-level synchronous operation with the physical entity. At the same time, a synchronization accuracy verification module is set up to monitor transmission packet loss and link interruption. When the mapping synchronization error or transmission anomaly exceeds the preset threshold, the dynamic correction process of model parameters in the twin modeling stage is automatically triggered and a transmission fault warning is issued. If any of the following conditions is met: timing matching degree <99%, numerical deviation rate >1%, or state consistency <99%, the preset threshold is exceeded. The criteria for millisecond-level synchronization are that the time difference between the physical entity's action and the twin's mapped action does not exceed 10 milliseconds, and the numerical deviation is within the allowable range of the device. The system ensures stability through three measures: unified clock real-time calibration, efficient data compression, and edge local caching acceleration.

[0010] The data transmitted via the forward link adopts a standardized lightweight format. After data compression, noise reduction, and normalization preprocessing, it is transmitted. The data format is compatible with the reception requirements of each layer of the four-layer twin. There is no data loss or format disorder during the transmission process, ensuring that the core data received by the twin can be directly used for mapping calculation.

[0011] The reverse mapping link sends the simulation optimization results, fault warning instructions, operation and maintenance scheduling strategies, and operating condition optimization parameters of the digital twin to the measurement and control system of the physical entity through an encrypted communication link, driving the measurement and control system to execute corresponding control actions. At the same time, the forward mapping link collects the measurement and control data after the action is executed in real time to verify the execution effect of the reverse control.

[0012] The effectiveness verification uses the degree of recovery of equipment operating parameters, the extent of improvement in health, the level of energy consumption reduction, and the status of fault elimination as the core judgment indicators. All indicators are based on objective measured data.

[0013] Furthermore, the status assessment stage is based on the bidirectional mapping data synchronized in the bidirectional linkage stage. In the performance assessment layer twin, a data-driven equipment health quantification assessment model is constructed. A multi-scale feature fusion algorithm is adopted to integrate the time domain, frequency domain, and time-frequency domain features of multi-dimensional measurement and control data. The health quantification score of the core components of the equipment, consumable materials such as pulleys, and the overall machine operating status is evaluated, and four levels are simultaneously divided: healthy, alert, abnormal, and faulty. The health measurement and assessment model uses multi-scale fusion features as the sole input, combines historical health status data of the equipment with normal operating benchmarks to make a comprehensive score. The scoring process fully considers the impact of component wear, environmental interference, and load changes, and the scoring results objectively reflect the true health level of the equipment.

[0014] When the health score falls below the abnormal threshold, the fault tracing and reasoning module is automatically triggered. Based on the historical measurement and control data and fault simulation database of the twin, the causal reasoning algorithm is used to locate the location, type, and cause of the transmission anomaly of the fault. The wear status of the pulley is monitored in particular, and the root cause of the fault is traced. The evolution trend of the fault is simulated in the twin at the same time, generating early warning information and emergency response strategies for equipment faults and transmission anomalies. These are then sent to the measurement and control system of the physical entity through the reverse mapping link of the two-way linkage stage, so as to realize early warning and handling of faults.

[0015] The fault simulation database is constructed from typical equipment fault cases, historical fault handling records, and simulated fault data. The database covers common fault modes of core components such as motors, bearings, sensors, and actuators. Each type of fault corresponds to complete characteristics, evolution process, and handling solutions.

[0016] Furthermore, the operation and maintenance optimization phase, based on the health assessment results and fault tracing data from the status assessment phase, constructs a data-driven remaining service life prediction model in the decision management twin. Combining the equipment's historical operating data, service cycle, and operation and maintenance records, it predicts the remaining service life of core components, consumables such as pulleys, and the entire machine, generating a dedicated predictive operation and maintenance strategy that includes optimal operation and maintenance time, operation and maintenance content, consumable replacement, spare parts requirements, and operation and maintenance process. The model input features also include real-time health scores, component wear trends, cumulative operating load values, and environmental impact coefficients. All features are highly compatible with the long-term performance degradation patterns of electromechanical equipment and can comprehensively reflect the operating status of the equipment throughout its entire life cycle.

[0017] Simultaneously, a working condition optimization simulation module is built in the twin. Based on the real-time measurement and control data of the equipment, the operating condition parameters of the equipment are optimized by simulating the optimization of the operating efficiency, energy consumption, and life loss as a multi-objective optimization function, and the optimal working condition adjustment scheme is generated. Predictive operation and maintenance strategies and operating condition optimization schemes are first simulated and verified in a twin to confirm that there are no operational risks. Then, they are sent to the physical entity's measurement and control system and operation and maintenance management platform through the reverse mapping link of the two-way linkage stage to realize the linkage execution of operation and maintenance actions and operating condition adjustments. At the same time, the operation data after execution is collected through the forward mapping link to verify the execution effect of the strategies and schemes.

[0018] Furthermore, the cluster collaboration stage targets multi-electromechanical equipment clusters at the production line and workshop levels, constructing an independent digital twin for each device within the cluster to form a distributed twin cluster. Simultaneously, a cluster-level collaborative management architecture is built, extending the control scope from a single device to the production line / workshop level system. First, real-time data communication and status synchronization between the twins of each device are achieved. Based on the production process of the production line and the linkage relationship between upstream and downstream equipment, a cluster collaborative constraint model is constructed. Then, based on the full measurement and control data of all devices in the cluster, a global operation status assessment and collaborative simulation are carried out in the cluster-level twin. With the overall cluster production efficiency, energy consumption, and failure risk as the global optimization objectives, a cluster collaborative scheduling strategy is generated to dynamically adjust the operating parameters, measurement and control strategies, and production cycle of each device. The scheduling strategy is generated based on the real-time operating status of the cluster, the progress of production orders, the load balancing of equipment, and the fault warning information. When adjusting, priority is given to ensuring the stable operation of key equipment and balancing the production cycle of upstream and downstream equipment, so as to achieve the optimal operation of the cluster as a whole without affecting production.

[0019] Finally, the collaborative scheduling strategy is distributed to the measurement and control systems of each device through the reverse mapping link of the two-way linkage stage, realizing the collaborative operation and distributed measurement and control management of multi-device clusters.

[0020] Furthermore, the iterative optimization phase is based on the multi-device cluster collaborative management and control effect achieved in the cluster collaboration phase, constructs a full-process measurement and control data iterative optimization architecture, collects physical entity operation measurement and control data, reverse control execution data, operation and maintenance action execution data, twin simulation data, and management strategy execution effect data, forming a full-dimensional closed-loop dataset. Based on the closed-loop dataset, a deep reinforcement learning optimization model is deployed in the cloud. The global optimization goal is to achieve the highest twin mapping accuracy, the best device health, the highest cluster operation efficiency, and the lowest operation and maintenance cost. The model parameters, spatiotemporal synchronization thresholds, feature selection rules, health assessment models, RUL prediction models, operation and maintenance and optimization strategies, and cluster collaborative scheduling rules of each layer of twins are iteratively optimized. The model performs iterative optimization at a fixed cycle. After each iteration, the optimization effect is quantitatively evaluated. When the four indicators of mapping accuracy, device health, cluster efficiency, and operation and maintenance cost all reach the preset optimal value and remain stable without fluctuation, the iteration is considered complete, and optimization is paused until the device status changes. At the same time, the optimized parameters and rules will be synchronously distributed to the edge terminal and the measurement and control system of each device via OTA upgrade.

[0021] The beneficial effects of this invention are as follows: 1. This invention uses a low-latency communication link to synchronize the pre-processed core measurement and control data at the edge to a four-layer linked digital twin architecture in real time, achieving millisecond-level synchronous operation between the physical entity and the virtual model. The synchronization accuracy verification module can automatically trigger model parameter correction, continuously and stably ensuring mapping accuracy. The reverse link directly sends fault warning commands, predictive operation and maintenance strategies, and operating condition optimization parameters to the physical entity measurement and control system through encrypted communication, driving the equipment to execute corresponding control actions. The forward link collects data after execution in real time to verify the effect. This improves the efficiency of fault warning response and emergency handling, and achieves deep collaboration between the physical entity and the virtual model.

[0022] 2. This invention deploys a multi-source sensor network covering the core components of the equipment, the measurement and control link, and the operating environment to collect all heterogeneous measurement and control data. It achieves a unified spatiotemporal reference for the data through hardware timestamps combined with linear interpolation and resampling. A lightweight temporal feature extraction network and dynamic thresholds are used to filter core features and redundant data. The sensor sampling frequency can be dynamically adjusted according to the real-time operating conditions of the equipment, reducing data transmission bandwidth usage and cloud computing load. In the twin modeling stage, a four-layer linked twin architecture is constructed. An incremental learning mode combined with prior knowledge of the equipment mechanism is used to dynamically correct model parameters. The model can be automatically updated in real time as the equipment's operating conditions change and components wear out, improving the efficiency and accuracy of twin model construction.

[0023] 3. This invention comprehensively collects data from all dimensions of physical entity operation, control execution, maintenance actions, and twin simulation, forming a closed-loop dataset. Through a cloud-based deep reinforcement learning model, it iteratively optimizes model parameters, feature selection rules, scheduling strategies, and maintenance solutions with the goals of optimal mapping accuracy, best equipment health, highest cluster operating efficiency, and lowest maintenance cost. The optimization results are synchronized to the edge and measurement and control system via OTA upgrades. Simultaneously, it constructs a distributed twin cluster and cluster-level collaborative architecture for multi-device clusters, enabling data interoperability and global collaborative scheduling between devices. It dynamically adapts to production line processes to optimize equipment operating parameters, improving cluster operating efficiency and reducing maintenance costs. Furthermore, it employs a wired-first hybrid transmission link to enhance anti-interference capabilities and transmission stability, avoiding interruptions and packet loss that could affect twin operation. Attached Figure Description

[0024] Figure 1 This is the overall flowchart of the electromechanical equipment measurement and control data-driven digital twin linkage management of the present invention; Figure 2 This is a flowchart of the measurement, control, sampling, screening, and twin modeling sub-process of the present invention; Figure 3 This is a flowchart of the bidirectional linkage and closed-loop iterative optimization sub-process of the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] like Figures 1 to 3 As shown, this embodiment of the invention provides a digital twin linkage management method driven by electromechanical equipment measurement and control data, including the following specific steps: In this embodiment of the invention, in response to the core measurement and control requirements of the target electromechanical equipment, the measurement and control screening stage deploys a multi-source sensor network covering the core components of the equipment, the measurement and control link, and the operating environment to collect vibration, temperature, pressure, current, speed, displacement, control commands, action feedback data, and on-site video data for AR color recognition. The video stream is preferentially transmitted via wired link. The sensor network is deployed at key locations based on the equipment's critical fault-prone areas, core power components, and key data acquisition nodes. High-sensitivity sensors are deployed at core components, fast-response sensors are deployed at key nodes of the measurement and control link, and highly stable sensors are deployed at key locations in the operating environment. All sensor nodes are uniformly connected to the edge acquisition unit to form a fully covered measurement and control data acquisition network.

[0027] A spatiotemporal synchronization mechanism based on hardware timestamps is built at the edge. For sensors with different sampling frequencies, including kilohertz-level high-frequency vibration sensors and hertz-level low-frequency temperature and pressure sensors, a combination of linear interpolation and resampling is used to achieve a unified spatiotemporal reference for all measurement and control data, distinguishing between high-speed measurement and control data and low-speed consumable wear data, and adapting them to the transmission links respectively. The hardware timestamp is generated by the high-precision constant-temperature clock module built into the sensor. The time base is synchronized with the unified clock source in the industrial field, and the time stamp accuracy reaches the microsecond level. It can accurately bind the acquisition time, acquisition component and acquisition location of each set of measurement and control data. Linear interpolation smoothly fills the time gap caused by sensors with different sampling frequencies, and focuses on handling the sampling interval differences between high-frequency vibration sensors and low-frequency temperature, humidity and pressure sensors to ensure that all time series data are continuous without breaks. The resampling module converts all types of heterogeneous measurement and control data into a fixed sampling frequency that matches the core control cycle of electromechanical equipment, and completes the dual normalization of time axis and spatial coordinates.

[0028] A lightweight time-series feature extraction network is deployed at the edge to pre-screen the synchronized measurement and control data. Based on data-driven dynamic thresholds, redundant data is filtered out, and core feature data that is strongly correlated with the equipment's operating status is retained. At the same time, the sensor sampling frequency and screening threshold are dynamically adjusted according to the real-time operating conditions of the equipment. For wear signals of consumables such as pulleys, a 10-second upload cycle is set, and NB-IoT narrowband low-speed transmission is adopted to reduce the data transmission bandwidth occupation and cloud computing pressure while ensuring the validity of measurement and control data.

[0029] The temporal feature extraction network adopts a three-level cascaded structure consisting of a one-dimensional convolutional feature extraction module, a bidirectional GRU temporal coding module, and an adaptive feature filtering fully connected module. Gradient stability is achieved through residual connections between modules, balancing feature extraction efficiency and edge inference speed. The input consists of multi-source heterogeneous temporal measurement and control data unified by a spatiotemporal reference. The data dimension is defined as N×T×C, where N is the number of samples in a single sampling, T is the equipment operating time step, and C is the number of sensor acquisition channels. The input data covers all types of time series such as vibration, temperature, pressure, current, rotational speed, and displacement. The output is a low-dimensional core feature vector that is strongly correlated with the equipment operating status, automatically removing noise and redundant features.

[0030] The network training uses a mini-batch stochastic gradient descent optimizer with an initial learning rate of 1×10^-4, a batch size of 32, and a maximum of 500 training iterations. The dual optimization objectives are feature selection accuracy and feature retention rate. To address the non-stationary nature of the time series data for electromechanical equipment measurement and control, a time series smoothing regularization constraint is added to avoid overfitting.

[0031] The dynamic threshold is automatically generated by the edge computing unit through iterative calculation based on multi-dimensional information such as the long-term historical operating baseline of the equipment, real-time load conditions, changes in ambient temperature and humidity, and the degree of equipment aging. The threshold will be adaptively adjusted in real time according to the equipment start-up and shutdown switching, load rise and fall fluctuations, ambient temperature drift, and operation mode conversion. It filters out steady-state data that has remained unchanged for a long time, noise interference data with no practical diagnostic value, and abnormal outlier data that exceeds the effective measurement range, and only retains high-value effective data that can truly reflect sudden changes in equipment status, the emergence of potential faults, and the trend of performance degradation.

[0032] In this embodiment of the invention, the twin modeling stage is based on the core measurement and control data screened in the measurement and control acquisition and screening stage to construct a four-layer linked digital twin architecture. The digital twin architecture includes a physical entity layer twin, an operation status layer twin, a performance evaluation layer twin, and a decision management layer twin. The physical entity layer twin maps the geometric structure, physical attributes, hardware interfaces, and measurement and control links of the electromechanical equipment, replicating the physical characteristics of the equipment; the operation status layer twin maps the operating conditions, action status, and environmental parameters of the equipment in real time based on core measurement and control data, achieving real-time synchronization with the physical entity; the performance evaluation layer twin is used for quantitative evaluation and fault simulation of equipment health, operating efficiency, and reliability; the decision management layer twin is used for the generation and simulation verification of operation and maintenance strategies, operating condition optimization, and scheduling instructions. The four-layer interconnected digital twin adopts standardized hierarchical data interfaces and a two-way closed-loop data flow relationship from top to bottom and bottom to top. The physical entity layer twin synchronizes basic static data such as equipment geometry, physical attributes, hardware interfaces, and measurement and control links to the operation status layer twin in real time. Based on core measurement and control data, the operation status layer twin continuously pushes dynamic data such as real-time equipment operating conditions, action status, and environmental parameters to the performance evaluation layer twin. The performance evaluation layer twin stably transmits quantitative evaluation results such as equipment health, operating efficiency, fault risks, and reliability to the decision management layer twin. The decision management layer twin transmits simulation-verified operation and maintenance strategies, operating condition optimization schemes, and collaborative scheduling instructions back to the physical entity layer twin.

[0033] The modeling process adopts a data-driven incremental learning model, using pre-screened measurement and control data as the core input, combined with a small amount of prior knowledge of equipment mechanisms, to dynamically correct the model parameters of each twin layer. When the equipment operating conditions change or components wear out, the model is automatically updated incrementally based on real-time measurement and control data, without the need to reconstruct the overall model.

[0034] The triggering conditions for incremental learning updates include five typical scenarios: active switching of equipment production conditions, significant fluctuations in load amplitude, wear and tear of core components reaching a preset threshold, continuous deviation of operating parameters from the historical benchmark range, and slight deviations in the synchronization accuracy of the twin. During the update execution, only local parameter fine-tuning and correction are performed on the component sub-model, operating state sub-model, and performance evaluation sub-model that have undergone state changes, without altering the basic geometric structure model and fixed physical attribute model of the equipment.

[0035] The incremental update data comes from the core measurement and control data and historical operating benchmark data uploaded in real time from the edge terminal. The correction scope only covers the dynamic parameters and state mapping parameters of each layer of twins. After the update, the model can quickly adapt to the current operating conditions of the equipment.

[0036] In this embodiment of the invention, the bidirectional linkage stage is based on the four-layer digital twin architecture constructed in the twin modeling stage, and constructs a bidirectional mapping linkage mechanism of forward real-time mapping and reverse control. The forward mapping link synchronizes the core measurement and control data processed at the edge of the measurement and control screening stage to the digital twin in real time through a 5G and industrial Ethernet wired hybrid low-latency communication link. Wired transmission is prioritized to reduce industrial wireless interference, while consumable wear data is transmitted using NB narrowband low-speed links. This drives the twins at each layer to achieve millisecond-level synchronous operation with the physical entity. At the same time, a synchronization accuracy verification module is set up to monitor transmission packet loss and link interruption. When the mapping synchronization error or transmission anomaly exceeds the preset threshold, the dynamic correction process of model parameters in the twin modeling stage is automatically triggered and a transmission fault warning is issued. The synchronization accuracy verification module continuously compares the measured control data of the physical entity with the virtual twin mapping data at a fixed frequency, and calculates three core verification indicators in real time: timing matching degree, numerical deviation rate, and state consistency. When any indicator exceeds the preset qualified threshold, the model automatic correction process is immediately triggered. The correction process is executed in an orderly manner according to hierarchical priority. First, the real-time mapping parameters of the running state layer are adjusted to ensure the synchronization accuracy of the basic state. Then, the calculation logic and parameter configuration of the performance evaluation layer are optimized. Finally, the strategy generation configuration parameters of the decision management layer are updated. After the correction is completed, the synchronization accuracy verification is automatically restarted. The correction process ends after three consecutive verifications that meet the standard.

[0037] Formula for calculating numerical deviation rate: In the formula: The numerical deviation rate is a core verification indicator of the synchronization accuracy between the digital twin and the physical entity. When this value is greater than 1%, the system automatically triggers the dynamic correction process of the twin model parameters. This represents the instantaneous value of the measured and control data collected by the sensors at the physical entity end; This represents the instantaneous value of the corresponding data output from the mapping of the digital twin virtual model; It represents the standard rated value of the physical entity measurement and control data, which is the reference data under normal operating conditions of the equipment.

[0038] The low-latency communication link adopts a dedicated frequency band for industrial sites and an independent physical transmission channel. It has a built-in data priority scheduling and bandwidth guarantee mechanism, which sets the core measurement and control data as the highest transmission level to avoid non-critical data occupying the channel and causing delays. The link is equipped with dual redundant channels, and the backup channel is automatically switched in milliseconds when the main channel is interrupted or packets are lost.

[0039] The reverse mapping link sends the simulation optimization results, fault warning instructions, operation and maintenance scheduling strategies, and operating condition optimization parameters of the digital twin to the measurement and control system of the physical entity through an encrypted communication link, driving the measurement and control system to execute corresponding control actions. At the same time, the forward mapping link collects the measurement and control data after the action is executed in real time to verify the execution effect of the reverse control.

[0040] The reverse communication link adopts a dual security mechanism of industrial-grade symmetric encryption and identity authentication. The transmission channel is bound to a unique hardware identifier of the physical device, and only accepts legitimate control commands issued by the corresponding twin, rejecting all unauthenticated and unauthorized commands. All issued fault warnings, operation and maintenance strategies, and operating condition optimization parameters carry exclusive verification codes and timestamps. The transmission process supports breakpoint resumption, data integrity verification, and command retransmission mechanisms, which can resist security risks such as data loss, command tampering, signal interference, and illegal intrusion.

[0041] The effect verification process involves collecting all measurement and control data before and after command execution via a forward mapping link. By comparing and analyzing changes in equipment status indicators, operating efficiency, energy consumption levels, and health scores before and after the implementation of control actions, operation and maintenance strategies, and operating condition adjustments, the actual execution effect and optimization value of the control strategy can be intuitively determined. If the effect meets the standard, the complete verification data is automatically recorded in the closed-loop dataset. If the effect does not meet the standard, abnormal information is immediately marked and a secondary optimization process for the strategy is triggered. All verification data, execution data, and abnormal data are transmitted back to the cloud database in real time.

[0042] In this embodiment of the invention, the status assessment stage is based on the bidirectional mapping data synchronized in the bidirectional linkage stage. In the performance assessment layer twin, a data-driven equipment health quantification assessment model is constructed. A multi-scale feature fusion algorithm is adopted to integrate the time domain, frequency domain, and time-frequency domain features of multi-dimensional measurement and control data. The health quantification score of the core components, consumable materials such as pulleys, and the overall operating status of the equipment is evaluated. The system is simultaneously divided into four levels: healthy, alert, abnormal, and fault, so as to achieve a refined quantitative assessment of the equipment operating status. Equipment health measurement scoring formula: In the formula: This represents a quantitative score for equipment health. Based on this score, the system classifies the equipment status into four levels: healthy, alert, abnormal, and faulty. This represents the weighting coefficients of time-domain features, which are dynamically allocated through an adaptive attention mechanism to weigh the contribution of time-domain features to health. The comprehensive score representing the time-domain characteristics of the measurement and control data is calculated from time-domain statistical characteristics such as mean, variance, kurtosis, and waveform factor. This represents the frequency domain feature weighting coefficients, which are dynamically allocated through an adaptive attention mechanism to weigh the contribution of frequency domain features to health. The comprehensive score representing the frequency domain characteristics of the measurement and control data is calculated using the Fast Fourier Transform (FFT). This represents the weighting coefficients of time-frequency domain features, which are dynamically allocated through an adaptive attention mechanism to weigh the contribution of time-frequency domain features to health. The comprehensive score representing the time-frequency domain characteristics of the telemetry and control data is calculated using the wavelet packet decomposition algorithm. + + =1, the sum of the three weighting coefficients is always 1, ensuring the normalization property of the health score. The multi-scale feature fusion algorithm adopts a three-level linkage structure consisting of a multi-scale parallel feature extraction module, a cross-domain feature concatenation module, and an adaptive attention weighting module. Each module works together to complete the deep fusion of multi-dimensional features. The input data is divided into three types of core features: time-domain statistical features of measurement and control data, frequency-domain FFT transform features, and time-frequency-domain wavelet packet decomposition features. The time-domain statistical features include mean, variance, kurtosis, and waveform factor. These three types of features correspond to the static, dynamic, and transient representation dimensions of the equipment's operating status, respectively. The output is a high-dimensional compact fusion feature vector, which is directly input into the health measurement and evaluation model. The algorithm sets up three parallel scales (1, 3, and 5) to complete feature extraction. Attention weights are dynamically allocated through the Softmax function. The weight allocation rules are strongly bound to the fault-sensitive features of the core components of electromechanical equipment, with a focus on strengthening the weight ratio of key fault features such as vibration and temperature.

[0043] The four health levels are strictly divided based on the quantitative scoring range and the equipment's operating status. The health level corresponds to the optimal operating state where all operating parameters of the equipment are within the standard range, performance is not degraded, and there are no potential faults. The attention level corresponds to a warning state where parameters fluctuate slightly, performance is slightly degraded, but normal operation is not affected. The abnormal level corresponds to a dangerous state where key parameters deviate from the normal range, there are obvious potential faults, and the stable operation of the equipment may be affected. The fault level corresponds to a shutdown state where core parameters are seriously out of standard, equipment functions are limited, and a substantial fault has occurred. The thresholds for each level can be flexibly configured and adjusted according to the model of the electromechanical equipment, the characteristics of core components, operating conditions, and service life.

[0044] When the health score falls below the abnormal threshold, the fault tracing and reasoning module is automatically triggered. Based on the historical measurement and control data and fault simulation database of the twin, the causal reasoning algorithm is used to locate the location, type, and cause of the transmission anomaly of the fault. The wear status of the pulley is monitored in particular, and the root cause of the fault is traced. The evolution trend of the fault is simulated in the twin at the same time, generating early warning information and emergency response strategies for equipment faults and transmission anomalies. These are then sent to the measurement and control system of the physical entity through the reverse mapping link of the two-way linkage stage, so as to realize early warning and handling of faults.

[0045] The causal reasoning algorithm adopts a three-level combined structure of a fault cause-effect graph construction module, a conditional probability reasoning module, and a root cause ranking verification module, which closely matches the physical mechanism of fault transmission in electromechanical equipment. The input data includes real-time equipment monitoring and control data, historical fault case database, and twin fault simulation data, covering all dimensions of parameters of core equipment components, monitoring and control links, and operating environment. The output is a triplet result of fault location, fault type, and root cause, and a fault confidence score is output simultaneously. The nodes in the cause-effect graph correspond to core components of the equipment such as motors, bearings, sensors, and actuators. The directed edges accurately represent the fault propagation relationships between components. The inference confidence threshold is set to 0.85. If the value is lower than the threshold, secondary inference is automatically triggered. The root cause is locked by combining the typical fault mode library of electromechanical equipment, ensuring that the fault tracing results are consistent with the actual equipment operation and fault occurrence scenarios.

[0046] Emergency response strategies are automatically generated based on the fault type, severity level, scope of impact, and production process constraints. Minor faults trigger parameter fine-tuning, reduced load operation, and continuous monitoring. Moderate faults trigger protective shutdown, partial isolation, and audible and visual alarms. Severe faults trigger emergency power outage, power cut-off, and activation of backup equipment to prevent the fault from escalating and spreading.

[0047] In this embodiment of the invention, the operation and maintenance optimization phase is based on the health assessment results and fault tracing data from the status assessment phase. A data-driven remaining service life prediction model is constructed in the decision management twin. Combining the equipment's historical operating data, service cycle, and operation and maintenance records, the remaining service life of core components, consumables such as pulleys, and the entire machine is predicted, generating a dedicated predictive operation and maintenance strategy that includes optimal operation and maintenance time, operation and maintenance content, consumable replacement, spare parts requirements, and operation and maintenance process. The remaining service life prediction model adopts a three-level cascaded structure consisting of an LSTM time-series feature extraction layer, a multi-head attention weighting layer, and a fully connected prediction output layer. Batch normalization is used between layers to improve training stability and convergence speed. The input data consists of historical equipment operation data, service cycle data, maintenance record data, and real-time health fusion features. The input data dimension is uniformly 1×T×8, where T is the time series length and 8 is the core feature dimension. The input features cover the key parameters of the equipment's entire life cycle operation. The output is the remaining service life value of core components and the whole machine, with the unit uniformly in hours.

[0048] The model training uses the mean squared error loss function, with an initial learning rate of 5×10^-4 and a maximum of 800 training iterations. The model convergence condition is a prediction error rate of less than 5%. To address the slow degradation of electromechanical equipment performance, a lifespan degradation regularization term is added to improve the stability and accuracy of long-cycle lifespan prediction.

[0049] Once generated, the predictive maintenance strategy is simultaneously pushed to the maintenance management platform and mobile terminals, notifying personnel in advance to prepare tools, requisition spare parts, set up on-site, and apply for permissions. The strategy clearly marks standardized operating procedures, safety precautions, quality acceptance standards, and restart verification processes to reduce errors and unplanned downtime.

[0050] Simultaneously, a working condition optimization simulation module is built in the twin. Based on the real-time measurement and control data of the equipment, the operating condition parameters of the equipment are optimized by simulating the optimization of the operating efficiency, energy consumption, and life loss as a multi-objective optimization function, and the optimal working condition adjustment scheme is generated. The operating condition optimization simulation module sets multiple constraints such as upper limits for equipment operating parameters, safety protection thresholds, and production process requirements. The optimization process is carried out within the constraints to ensure that the generated operating condition scheme not only meets the optimization goals of efficiency, energy consumption, and lifespan, but also meets the hard requirements of equipment safety operation and production process, and can be directly issued for execution.

[0051] Predictive operation and maintenance strategies and operating condition optimization schemes are first simulated and verified in a twin to confirm that there are no operational risks. Then, they are sent to the physical entity's measurement and control system and operation and maintenance management platform through the reverse mapping link of the two-way linkage stage to realize the linkage execution of operation and maintenance actions and operating condition adjustments. At the same time, the operation data after execution is collected through the forward mapping link to verify the execution effect of the strategies and schemes.

[0052] The full-process simulation verification completely replicates the actual operating environment, load conditions, process constraints, safety specifications, and upstream and downstream linkages of the physical equipment. It sequentially simulates the entire process of operation and maintenance action execution, operating parameter adjustment, system linkage response, and equipment status changes, while simultaneously monitoring all key measurement and control indicators such as vibration, temperature, pressure, current, speed, energy consumption, and displacement in real time. During the simulation, five judgment criteria must be met: no parameter exceeding the standard, no equipment conflict, no safety hazards, no production capacity loss, and no linkage anomalies. Only after all criteria are met can the verification be judged as passed, avoiding the risks of safety accidents, equipment damage, and production interruptions that may be caused by the physical equipment directly executing the strategy.

[0053] In this embodiment of the invention, the cluster collaboration stage targets multi-electromechanical equipment clusters at the production line and workshop levels. It constructs an independent digital twin for each device in the cluster, forming a distributed twin cluster, and simultaneously builds a cluster-level collaborative management architecture, extending the control scope from a single device to the production line / workshop level system. The distributed twin cluster adopts a master-slave collaborative networking, with a central control management node responsible for global scheduling, and other device twins acting as slave nodes to report status, receive instructions, and provide feedback results in real time. The cluster supports flexible device access and offline exit, and newly added devices can be quickly modeled, authenticated, and added to the network. Faulty devices can be isolated with one click without affecting the overall operation.

[0054] First, real-time data communication and status synchronization between the twins of each device are achieved. Based on the production process of the production line and the linkage relationship between upstream and downstream equipment, a cluster collaborative constraint model is constructed. Then, based on the full measurement and control data of all devices in the cluster, a global operation status assessment and collaborative simulation are carried out in the cluster-level twin. With the overall cluster production efficiency, energy consumption, and failure risk as the global optimization objectives, a cluster collaborative scheduling strategy is generated to dynamically adjust the operating parameters, measurement and control strategies, and production cycle of each device. The cluster collaborative constraint model is built based on the production line cycle time, the linkage relationship between upstream and downstream equipment, capacity matching requirements, safe operation specifications, and process execution standards. The core constraints include equipment start-up and shutdown sequence, upper and lower limits of operating parameters, data interaction frequency, fault linkage response rules, and production cycle synchronization requirements. Each independent twin of the equipment achieves real-time data communication with a unified data protocol through the industrial intranet. The communication data adopts a standardized format and includes equipment status data, measurement and control acquisition data, command execution data, and fault alarm data.

[0055] Finally, the collaborative scheduling strategy is distributed to the measurement and control systems of each device through the reverse mapping link of the two-way linkage stage, realizing the collaborative operation and distributed measurement and control management of multi-device clusters.

[0056] In this embodiment of the invention, the iterative optimization stage is based on the multi-device cluster collaborative management and control effect achieved in the cluster collaboration stage, constructs a full-process measurement and control data iterative optimization architecture, collects physical entity operation measurement and control data, reverse control execution data, operation and maintenance action execution data, twin simulation data, and management strategy execution effect data, forming a full-dimensional closed-loop dataset. The closed-loop dataset is stored in a multi-level classification system based on data type, device number, time node, execution scenario, and fault type. It adopts an industrial-grade distributed database to achieve permanent storage and high-speed retrieval. When accessing the dataset, data can be filtered by device model, fault type, and maintenance event.

[0057] Based on the closed-loop dataset, a deep reinforcement learning optimization model is deployed in the cloud. The global optimization goal is to achieve the highest twin mapping accuracy, the best device health, the highest cluster operation efficiency, and the lowest operation and maintenance cost. The model parameters, spatiotemporal synchronization thresholds, feature selection rules, health assessment models, RUL prediction models, operation and maintenance and optimization strategies, and cluster collaborative scheduling rules of each layer of twins are iteratively optimized. The deep reinforcement learning optimization model adopts the DDPG deep deterministic policy gradient architecture, which includes three core modules: actor network, critic network, and experience replay buffer. The network uses an asynchronous update mechanism to improve the global optimization efficiency. The agent corresponds to the global optimization strategy of the digital twin linkage management system, and the environment is a scenario of coordinated operation between a distributed twin cluster of electromechanical equipment and physical entities. The state space includes four core state parameters: twin mapping error, equipment health score, cluster operation efficiency, and operation and maintenance cost. The action space includes all optimizable variables such as twin model parameters, spatiotemporal synchronization threshold, feature selection rules, and cluster collaborative scheduling rules.

[0058] The model training adopts an experience replay mechanism, with the buffer size set to 10,000 and the target network update step size set to 100 rounds. Continuous iterative optimization enables the system to achieve end-to-end self-learning and self-optimization.

[0059] At the same time, the optimized parameters and rules will be synchronously distributed to the edge terminal and the measurement and control system of each device via OTA upgrade.

[0060] Deep reinforcement learning globally optimizes the reward function: This represents the global optimization reward value of the deep reinforcement learning model. A higher reward value indicates a better overall system performance. It indicates the accuracy of digital twin mapping, characterizing the millisecond-level precision of synchronization between the physical entity and the virtual model; This represents the equipment health score, which is the final quantitative result output by the status assessment module. It represents the operating efficiency of a multi-device cluster and is a core operational indicator in the cluster collaborative management and control phase; This represents the total lifecycle cost of equipment maintenance, including comprehensive costs such as maintenance manpower, spare parts, and downtime losses. The coefficients in the formula are fixed optimization weights, which correspond to the optimization priorities of mapping accuracy, device health, cluster operating efficiency, and operation and maintenance costs, respectively, and are used to balance the weight allocation of multi-objective optimization.

[0061] The OTA upgrade adopts a phased verification and steady push implementation mode. First, the optimized model parameters and rule configurations are pushed to the edge to complete local function and compatibility verification. After the verification is passed, they are pushed to the single device monitoring and control system for small-scale testing. The upgrade process supports breakpoint resume, real-time progress monitoring, and automatic rollback mechanism for anomalies. The optimized parameters are synchronously updated to the edge cache and the device's local permanent storage. The device can complete the parameter loading without stopping the device, and the changes take effect immediately after restarting.

[0062] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0063] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A digital twin linkage management method driven by electromechanical equipment measurement and control data, characterized in that, The specific steps include the following: During the measurement, control, data acquisition and screening phase, a multi-source sensor network covering electromechanical equipment is deployed to collect full multi-source heterogeneous measurement and control data. A spatiotemporal synchronization mechanism is built at the edge to achieve spatiotemporal benchmark unification of measurement and control data. A lightweight time-series feature extraction network is deployed to perform feature pre-screening on the spatiotemporally unified measurement and control data. Redundant data is filtered based on dynamic thresholds and the acquisition and screening parameters are dynamically adjusted. In the twin modeling stage, a multi-level interconnected digital twin architecture is constructed; A data-driven incremental learning model is adopted, which combines prior knowledge of equipment mechanism to dynamically correct the parameters of the twin model and completes the incremental update of the model based on real-time measurement and control data. In the two-way linkage phase, a fully closed-loop two-way linkage mechanism of forward real-time mapping and reverse control is constructed; the forward link synchronizes the processed core data to the twin, sets up a synchronization accuracy verification module to trigger model correction, and the reverse link sends the twin's decision to the measurement and control system and verifies the execution effect; During the status assessment phase, a quantitative assessment model for equipment health is constructed, multi-dimensional data features are integrated to complete the scoring and grading, fault tracing reasoning is triggered and early warning and emergency strategies are generated, and fault early warning and handling are realized through reverse link; During the operation and maintenance optimization phase, a remaining service life prediction model is built to generate predictive operation and maintenance strategies. A working condition optimization simulation module is built to generate the optimal working condition adjustment plan. After the plan is verified to be risk-free through full-process simulation using a twin, it is deployed and executed, and the execution effect is verified. In the cluster collaboration phase, a distributed twin cluster and cluster-level collaborative management architecture are built for multi-electromechanical equipment clusters to achieve data synchronization and global collaborative scheduling between devices, and to issue scheduling policies to realize cluster collaborative operation and distributed measurement and control management. During the iterative optimization phase, data from the entire closed-loop operation is collected to form a closed-loop dataset. A deep reinforcement learning model is deployed to iteratively optimize system parameters and rules, and the data is simultaneously updated to the edge and the measurement and control system.

2. The digital twin linkage management method driven by electromechanical equipment measurement and control data according to claim 1, characterized in that, The multi-source sensor network in the measurement, control, and screening stage covers the core components of the equipment, the measurement and control link, and the operating environment. The collected measurement and control data includes vibration, temperature, pressure, current, speed, displacement, control commands, action feedback data, and on-site video data, which are used for AR color recognition. The video stream is preferentially transmitted via wired link. The spatiotemporal synchronization mechanism is implemented using hardware timestamps. For sensors with different sampling frequencies, a combination of linear interpolation and resampling is used to unify the spatiotemporal reference of the measurement and control data. The feature pre-screening retains core feature data that are strongly correlated with the equipment's operating status and dynamically adjusts the sensor sampling frequency and feature screening threshold according to the real-time operating conditions of the equipment.

3. The method of claim 2, wherein the method further comprises: The multi-level interconnected digital twin architecture in the twin modeling stage is a four-layer architecture, including a physical entity layer, an operational status layer, a performance evaluation layer, and a decision management layer twin. The physical entity layer twin maps the device's geometry, physical attributes, hardware interfaces, and measurement and control links; the operational status layer twin maps the device's operating conditions, action states, and environmental parameters in real time; the performance evaluation layer twin is used for quantitative evaluation of the device's operating status and fault simulation; and the decision management layer twin is used for generating and simulating control strategies. The incremental model update is based on real-time measurement and control data and is automatically triggered when the device's operating conditions change or components wear out.

4. The method of claim 3, wherein the method further comprises: In the forward link of the bidirectional linkage stage, the processed core measurement and control data is synchronized to the digital twin in real time through a low-latency communication link, driving the twins at each layer to achieve millisecond-level synchronous operation with the physical entity; the synchronization accuracy verification module simultaneously monitors transmission packet loss and link interruption, and when the mapping synchronization error or transmission anomaly exceeds the preset threshold, it automatically triggers the dynamic correction process of model parameters in the twin modeling stage and issues a transmission fault warning.

5. The method of claim 4, wherein the method further comprises: In the reverse link of the bidirectional linkage stage, the simulation optimization results, fault warning instructions, operation and maintenance scheduling strategies, and operating condition optimization parameters generated by the twin are sent to the measurement and control system of the physical entity through an encrypted communication link, driving the measurement and control system to execute corresponding control actions. Simultaneously, the measurement and control data after the action is executed are collected through the forward link to verify the execution effect of the reverse control.

6. The method of claim 5, wherein the method further comprises: The equipment health quantification assessment model in the state assessment phase is built in the performance assessment layer twin. It adopts a multi-scale feature fusion algorithm to integrate the time domain, frequency domain, and time-frequency domain features of the measurement and control data to complete the health quantification score and classify it into four levels: healthy, alert, abnormal, and fault. When the health score is lower than the abnormal threshold, the fault tracing and reasoning module is triggered. Based on historical measurement and control data and the fault simulation database, the fault location, type, and transmission anomaly cause are located and the root cause is traced. The fault evolution trend is simulated simultaneously to generate equipment fault and transmission anomaly early warning information and emergency response strategies.

7. The method of claim 6, wherein the method further comprises: The remaining service life prediction model in the operation and maintenance optimization phase is built in the decision-making management twin. Combining historical equipment operation data, service cycle, and operation and maintenance records, it predicts the remaining service life of core components and the whole machine, and generates a predictive operation and maintenance strategy that includes the optimal operation and maintenance time, operation and maintenance content, spare parts requirements, and operation and maintenance process. The operating condition optimization simulation module uses the highest operating efficiency, lowest energy consumption, and minimum life loss as multi-objective optimization functions to simulate and generate the optimal operating condition adjustment scheme. The operation and maintenance strategy and adjustment scheme are first verified by the twin's full-process simulation to ensure there is no risk before being issued for execution.

8. The method of claim 7, wherein the method further comprises: In the cluster collaboration phase, the distributed twin cluster consists of independent digital twins of individual devices within the cluster. The cluster-level collaborative management architecture covers multiple device systems at the production line and workshop levels. First, real-time data communication and status synchronization between the device twins are achieved. Then, a cluster collaboration constraint model is constructed based on the production process and upstream and downstream linkages. Next, the global optimization goal is to maximize the overall production efficiency of the cluster, minimize energy consumption, and minimize failure risk. Based on the global optimization goal of the cluster, a cluster collaborative scheduling strategy is generated to dynamically adjust the operating parameters, measurement and control strategies, and production cycle of each device.

9. The electromechanical device measurement and control data-driven digital twin linkage management method of claim 8, wherein, The closed-loop dataset in the iterative optimization phase includes physical entity operation measurement and control data, reverse control execution data, operation and maintenance action execution data, twin simulation data, and management and control strategy execution effect data. The deep reinforcement learning model takes the highest twin mapping accuracy, optimal equipment health, highest cluster operation efficiency, and lowest operation and maintenance cost as the global optimization objectives. Iteratively optimizes the twin model parameters, spatiotemporal synchronization threshold, feature selection rules, state assessment and life prediction models, management and control strategies, and collaborative scheduling rules. The optimized parameters and rules are synchronously distributed to the edge terminal and physical entity measurement and control system through OTA upgrade.

10. A digital twin linkage management method driven by electromechanical equipment measurement and control data according to claim 9, characterized in that, All control commands, operation and maintenance scheduling strategies, and operating condition optimization parameters sent to the measurement and control system via the reverse link are collected in real time through the forward link, and all measurement and control data after execution are collected to complete the closed-loop verification of the reverse control and strategy execution effect. The verification results are simultaneously incorporated into the closed-loop dataset of the iterative optimization phase for iterative optimization.

Images