Industrial internet-based device digital twin dynamic modeling method and system

Abstract

The application discloses a device digital twin dynamic modeling method and system based on an industrial internet, relates to the field of industrial internet and digital twin technology, and has the technical points that: by mining the implicit causal relationship in the device runtime sequence data which is not covered in the design stage, a complete parameter correlation network containing the causal direction, time delay and influence weight is constructed, the limitation that the prior art only depends on the explicit physical correlation is solved; and a time weight correction mechanism is introduced on the basis, the contribution degree in the tracing process is adjusted according to the aging degree of the causal relationship, and the causal correlation interference which has been weakened or failed is effectively excluded; secondly, when the device has running deviation, multidimensional tracing can be carried out along all possible causal paths, the core root cause leading to the anomaly is screened, and the cycle of fault positioning and processing is shortened.

Description

Technical Field

[0001] This invention relates to the field of industrial internet and digital twin technology, specifically to a dynamic modeling method and system for digital twins of equipment based on industrial internet. Background Technology

[0002] Digital twin technology, as one of the core supporting technologies of the Industrial Internet, enables real-time monitoring, simulation analysis, predictive maintenance, and optimized control of physical equipment by constructing virtual images of physical equipment. It has been widely used in many industrial fields such as power, chemical industry, and manufacturing.

[0003] The standard implementation process for existing digital twin dynamic modeling technology is as follows: By collecting basic operational data of the equipment through various sensors, and combining the equipment design drawings and known physical principles, an initial digital twin model containing geometric and physical dimensions is constructed. During the operation of the equipment, the real-time collected equipment operation data is periodically compared with the simulation output results of the model. When the deviation between the two exceeds a preset threshold, all parameters of the model are fully updated to correct the model error and enable it to reflect the actual operating status of the equipment.

[0004] However, during the implementation of the above technical solution, at least the following technical problems were discovered: First, there is an over-reliance on explicit physical relationships determined during the equipment design phase for root cause analysis, which only considers the parameter influence relationships pre-defined by engineers. However, in actual operation of industrial equipment, due to the complex coupling between components, the long-term cumulative effects of environmental factors, and the characteristic changes caused by equipment aging, a large number of undiscovered implicit parameter relationships are generated. These implicit relationships are often the key reasons for abnormal equipment operation, but existing technologies cannot effectively identify and utilize these relationships, resulting in low root cause localization accuracy, high misjudgment rate, and even failure to locate the true root cause of the fault. Secondly, when building models, the causal relationships between equipment parameters are generally assumed to be constant, thus ignoring the objective law that the causal relationships between parameters will gradually weaken, shift, or even completely fail as the equipment undergoes processes such as component wear, material aging, and maintenance and replacement. However, existing solutions only adopt the method of updating all parameters, that is, uniformly adjusting all parameters of the model. This not only leads to a large amount of computation and poor real-time performance in model updates, but also introduces unnecessary errors due to the invalid updates of a large number of unrelated parameters. As a result, the model accuracy decreases rapidly with the extension of running time, making it impossible to maintain stable and reliable performance throughout the entire life cycle of the equipment. To this end, this invention proposes a digital twin dynamic modeling scheme based on implicit causal relationship mining and time decay memory. Summary of the Invention

[0005] To achieve the above objectives, the present invention provides the following technical solution: A dynamic modeling method for equipment digital twins based on the Industrial Internet, the steps of which are as follows: After acquiring multi-source heterogeneous operational data of target equipment in industrial sites and obtaining a unified format dataset through standardized preprocessing, an initial digital twin model containing a four-layer structure of geometry, physics, behavior, and business rules is constructed. Real-time acquisition of full runtime sequence data of the target device, mining of implicit causal relationships between different parameters, construction of implicit causal relationship network of device parameters, and comparison of real-time running data of the device with the simulation output of the initial digital twin model at time intervals to calculate the real-time deviation value of the target parameters; Establish a causal chain time decay memory model for each causal link in the network, and record the first discovery time, historical trigger time series, single trigger impact weight and intensity fluctuation characteristics; Multi-path root cause tracing based on the causal chain time decay memory model performs time weight correction on real-time deviation values, locates core root cause parameters and all related causal chains, and quantifies the deviation contribution ratio of each causal chain. Based on the root cause tracing results, a hierarchical incremental model update instruction based on the aging degree of the causal chain is generated. The initial model is then adjusted layer by layer and its local structure is optimized according to the causal transmission direction and update priority to generate a real-time digital twin model. The effectiveness of the real-time digital twin model is then verified, and a verification report is obtained. The validated model is synchronously installed on the edge computing nodes and cloud servers of the industrial internet platform, and the installed model is monitored throughout its entire lifecycle. Data on model operation, actual equipment operation, and environmental changes are continuously collected at preset time intervals and updated accordingly.

[0006] Furthermore, when acquiring multi-source heterogeneous operational data, an encrypted transmission channel is established between the industrial internet platform edge gateway and various acquisition terminals of the target device to simultaneously collect the following data: A 3D laser scanner acquires the equipment's external dimensions, component positions, assembly gaps, and surface defect geometric parameters; built-in temperature, pressure, vibration, speed, and current sensors collect physical property data; the PLC controller outputs real-time operating status, including start / stop status, runtime, operation commands, valve opening degree, and motor power; industrial cameras capture equipment operation images and component movement processes; and environmental monitoring devices collect environmental perception data such as regional temperature and humidity, dust concentration, noise intensity, and vibration intensity.

[0007] Furthermore, the standardization preprocessing steps are as follows: Statistical outlier detection methods are used to remove erroneous data that exceeds a reasonable threshold, as well as incomplete data with a missing rate exceeding a preset missing threshold. The raw data of different formats are uniformly converted into a columnar storage format compatible with the industrial internet platform, and all data are mapped to the [0, 1] interval through the linear normalization method; A feature selection method is used to extract key features related to equipment operating status and model construction, and redundant information is removed. Interpolation is used to align key features in the time and space, ensuring that the temporal error of data from different sources does not exceed a preset time threshold and the spatial error does not exceed a preset spatial threshold, thus forming a dataset in a unified format.

[0008] Furthermore, when constructing the initial digital twin model, a three-dimensional geometric model is built based on the geometric parameter data to restore the equipment's external structure, component composition, assembly relationship, and surface texture; and a physical model is built by combining physical property data and equipment design parameters to define the material properties, mechanical properties, and thermodynamic properties of each component. Based on historical operational data and process data, a behavioral model is trained using machine learning algorithms to simulate the operating patterns and response characteristics of equipment under different working conditions. In accordance with the enterprise's production management standards and equipment operating procedures, a business rule model is constructed through a rule engine to set equipment safe operation thresholds, fault alarm conditions, and emergency shutdown logic. By integrating and associating the four-layer model, an explicit data mapping relationship based on the unique identifier of the device is established, forming an initial digital twin model.

[0009] Furthermore, a constraint-based causal discovery algorithm is employed to mine implicit causal relationships in runtime sequence data, construct an implicit causal relationship network for device parameters, and establish a causal chain time decay memory model for each causal relationship edge, as detailed below: Assign a unique identifier to each edge and record the first discovery timestamp; count the number of times each edge is triggered in real time, save the timestamp of each trigger and the corresponding impact weight value, generate an impact weight time series, and calculate the intensity decay coefficient; Set an aging threshold that is dynamically adjusted based on the equipment's operating status and maintenance records. When the intensity attenuation coefficient exceeds the dynamic aging threshold, the edge is marked as a state to be verified. Perform conditional independence checks on the edges to be verified, remove associated edges whose strength has decayed to an invalid level, and update the implicit causal relationship network of the device parameters.

[0010] Furthermore, when performing multi-path root cause tracing with time weight correction, the node position of the target parameter that caused the deviation in the implicit causal relationship network is first determined; a breadth-first traversal of a preset depth is performed along the reverse causal relationship path to generate all causal chain paths that may cause deviation of the target parameter. Obtain the strength attenuation coefficient of all causal edges in each path, and calculate the time correction weight of the path and the correction contribution of each path to the deviation of the target parameters. Paths whose correction contribution exceeds a preset threshold are marked as valid causal chains. The path with the largest correction contribution is selected from all valid causal chains, its starting node is marked as the core root cause parameter, and the complete causal chain of the path is extracted.

[0011] Furthermore, after generating hierarchical incremental model update instructions based on the aging degree of the causal chain, hierarchical parameter updates or local structure updates are performed according to the type of the core root cause parameters and the hierarchical structure of the associated causal chain. The aging degree of each edge in the causal chain is calculated, and the update task is divided into three priority levels: high, medium and low. The aging degree is the ratio of the intensity decay coefficient to the dynamic aging threshold. Following the order of priority from high to low, starting from the level where the core root cause parameter is located, the adjustment range of each level of related parameters is set sequentially along the causal forward transmission direction, and the parameter values ​​on the causal chain path are updated, while non-related parameters remain unchanged. When performing local structural updates, the corresponding local geometry and component composition in the initial model are modified according to the actual structural changes of the equipment or the replacement of components. The adjusted parameters and optimized structure are logically consistent to ensure that the causal relationships within the model are consistent with the updated implicit causal relationship network; and the results of the successful verification are synchronized to the initial model to generate a real-time digital twin model.

[0012] A dynamic modeling system for digital twins of equipment based on the Industrial Internet, comprising: The data acquisition and preprocessing module acquires multi-source heterogeneous operating data of target equipment in the industrial field and generates a unified format dataset through standardized preprocessing. The initial model building module constructs an initial digital twin model with a four-layer structure, including geometry, physics, behavior, and business rules, based on a unified format dataset. The implicit causal association mining module collects all runtime sequence data of the target device in real time, and uses a constraint-based causal discovery algorithm to mine the implicit causal relationships between parameters and construct an implicit causal association network of device parameters. The causal chain time decay memory module establishes a causal chain time decay memory model for each causal connection edge in the network, records its first discovery time, historical trigger time sequence, single trigger impact weight and intensity fluctuation characteristics, and dynamically updates the causal connection network. The time-weighted root cause tracing module compares the real-time operating data of the equipment with the initial model simulation output to calculate the deviation value, and performs multi-path root cause tracing based on the causal chain time decay memory model to perform time-weight correction, and locate the core root cause parameters and all related causal chains. The hierarchical incremental update module generates hierarchical incremental model update instructions based on the aging degree of the causal chain according to the root cause tracing results. It adjusts the parameters and optimizes the local structure of the initial model layer by layer according to the causal transmission direction and update priority, and generates a real-time digital twin model. The model verification and installation module verifies the effectiveness of real-time digital twin models and generates verification reports. It then synchronously installs compliant models to the edge computing nodes and cloud servers of the industrial internet platform. The full lifecycle monitoring and optimization module monitors the installed model throughout its entire lifecycle, continuously collecting various types of data at preset time intervals and providing feedback updates.

[0013] Furthermore, the causal chain time decay memory module includes: The causal edge identification unit assigns a unique identifier to each causal edge and records the first discovery timestamp; Trigger statistics unit to count the number of triggers and their corresponding impact weights in real time, and generate time series of impact weights; The attenuation coefficient calculation unit calculates the intensity attenuation coefficient based on the influence weight time series and dynamically adjusts the aging threshold based on the equipment operating status and maintenance records. When the intensity attenuation coefficient exceeds the aging threshold, the edge is marked as a state to be verified. The causal network update unit periodically performs conditional independence checks on the edges to be verified, removes invalid association edges, and updates the implicit causal network of device parameters.

[0014] Furthermore, the hierarchical incremental update module includes: The update instruction generation unit generates corresponding update instructions based on the type of the core root cause parameters and the hierarchical structure of the associated causal chain. Priority partitioning unit: calculates the aging degree of each edge in the associated causal chain and partitions the update priority; The layer-by-layer parameter adjustment unit, upon receiving a layer-by-layer parameter update instruction, determines the adjustment range of the associated parameters at each level in sequence according to priority, starting from the level where the core root cause parameter is located, and updates the parameters on the causal chain path. The local structure optimization unit, upon receiving a local structure update instruction, modifies the local geometry and component composition of the initial model according to the actual changes in the equipment, and performs logical consistency verification between the adjusted parameters and the optimized structure. The model synchronization unit synchronizes the verified results to the initial model, generating a real-time digital twin model.

[0015] This invention provides a method and system for dynamic modeling of digital twins of equipment based on the Industrial Internet, which has the following beneficial effects: 1. By mining implicit causal relationships in equipment runtime sequence data that were not covered in the design phase, a complete parameter association network including causal direction, time delay, and influence weight is constructed to overcome the limitation of existing technologies that rely only on explicit physical relationships. Furthermore, based on this, a time weight correction mechanism is introduced to adjust the contribution of causal relationships in the tracing process according to their aging degree, effectively eliminating interference from weakened or invalid causal relationships. Secondly, when equipment deviates from its operating parameters, it can trace the source in multiple dimensions along all possible causal paths, screen out the core root cause of the anomaly, and shorten the cycle of fault location and handling.

[0016] 2. A time decay memory model is established for each causal relationship to record the generation time, trigger history, and intensity change trend of the causal relationship, thereby dynamically identifying and eliminating invalid causal relationships; a hierarchical incremental update strategy is adopted, which only makes targeted adjustments to the correlation parameters on the effective causal chain without the need for a full parameter update, thereby improving the real-time performance of the model update; in addition, since only parameters directly related to the deviation are adjusted, the model can maintain stable accuracy throughout the entire life cycle of the equipment, reducing the workload and cost of frequent remodeling.

[0017] 3. By quantifying the aging degree of the causal chain and prioritizing updates, new causal relationships that have a significant impact on equipment operation are updated first, thus making the most rational use of limited computing resources. It is precisely because of this differentiated update mechanism that it better adapts to the dynamic changes in equipment operating status and further improves operating efficiency. Secondly, by constructing an effectiveness verification system that includes causal chain consistency and time decay characteristics, the model quality is comprehensively evaluated from multiple dimensions such as geometric accuracy, physical characteristics, behavioral patterns, and business rules, so that the installed model can accurately reflect the actual operating status of the equipment. Attached Figure Description

[0018] Figure 1 This is an overall flowchart of Embodiment 1 of the present invention; Figure 2 This is a cloud-based collaborative architecture diagram in an embodiment of the present invention; Figure 3 This is a simplified diagram of the causal relationship network in an embodiment of the present invention. Detailed Implementation

[0019] 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 of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0020] Example 1: like Figure 1 As shown, this embodiment addresses the need for digital twin dynamic modeling of rotating equipment commonly used in the chemical industry, such as centrifugal pumps, compressors, and heat exchangers. Taking a raw material liquid conveying centrifugal pump as an example, it is assumed that the pump has been running continuously for 3 years and has problems such as frequent abnormal vibration alarms, difficulty in locating the root cause of the fault, and rapid decay of the accuracy of the digital twin model.

[0021] like Figure 2 The diagram shows a three-tier architecture based on edge and cloud collaboration. First, data is collected and preprocessed. In the data collection phase, the edge gateway is used as the core device. It connects to various collection terminals of the centrifugal pump through industrial protocols. All data transmission is encrypted to prevent data from being leaked or tampered with during transmission. Then, preprocessing can be performed. Data preprocessing is completed locally at the edge gateway, which reduces the amount of data transmitted to the cloud and saves bandwidth. Then, the initial digital twin model is built and trained using 3D modeling software, finite element analysis software and machine learning framework installed on the cloud server. After the model is built, the constrained PC causal discovery algorithm on the edge computing node is used to mine the implicit causal relationships in the device runtime time series data in real time. We initially tried other causal discovery algorithms such as the LiNGAM algorithm, but found that the PC algorithm performed better when processing industrial time series data, especially in noisy conditions. After the data processing is complete, the temporal characteristic data of the causal relationship edges can be stored in the Redis database. Because Redis has a fast read and write speed, it can perform data read and write operations at the millisecond level. When a deviation in device operation is detected, the root cause tracing process is triggered to ensure the real-time response to the anomaly. Then, based on the results of the root cause tracing, a model update instruction is generated to adjust the locally running digital twin model in real time. To ensure data availability, the updated models undergo comprehensive validity verification, and the verified models are synchronously installed on edge nodes and the cloud. Meanwhile, the Grafana visualization platform is used to display the operating status of devices and models, and operation reports are generated regularly. The collected data is fed back to various front-end processes to facilitate optimization.

[0022] Based on the above, a full-process anomaly handling mechanism is added. First, when a single frame of data is missing and the missing percentage does not exceed 30%, the Kriging interpolation algorithm based on causal completion is used to complete the data by combining the causal correlation features of adjacent time series data. When the missing percentage exceeds 30%, the effective data of the two adjacent frames at the same location is retrieved for mean completion, and the data anomaly log is recorded. Second, when a sensor is detected to have no signal or the signal exceeds the range for more than 3 seconds, the backup detection channel pre-deployed at the same location is automatically switched, and a device fault warning is generated to the central control system without interrupting the current detection process. Third, when the network experiences a brief interruption, the edge gateway automatically caches local data and resumes transmission to the cloud after the network is restored. Fourth, when the model output results for 3 consecutive frames deviate from the measured values ​​by more than 15%, the model inference is paused, and the most recent 100 sets of effective samples in the same scenario are retrieved to complete the temporary fine-tuning and calibration of the model. After calibration, inference is resumed, and the model deviation log is recorded.

[0023] When in use, an encrypted transmission channel is established with various acquisition terminals of the centrifugal pump through the edge gateway, and the following data are collected synchronously: (1) Geometric parameter data: the centrifugal pump is scanned in the whole field using a three-dimensional laser scanner to obtain the external dimensions, component positions, assembly gaps and surface corrosion defects of the pump body, impeller, shaft, seals and other components. After the scan is completed, the point cloud data is imported into Geomagic Wrap software for noise reduction and splicing to remove noise points and redundant data generated during the scanning process. (2) Physical property data: The pump body temperature, inlet pressure, outlet pressure, bearing vibration acceleration, speed and motor current data are collected by the PT100 temperature sensor, diffused silicon pressure sensor, piezoelectric vibration sensor, magnetoelectric speed sensor and current transformer built into the centrifugal pump at a preset sampling frequency of 100Hz. The sampling frequency can be adjusted in the range of 50Hz to 200Hz according to the equipment type and monitoring requirements. (3) Real-time operating status data: The centrifugal pump's start-stop status, cumulative running time, start-stop operation commands, outlet valve opening degree and motor power data are collected through the PLC controller; (4) Process data: The centrifugal pump operation screen and impeller rotation process data are collected by an industrial camera at a preset frame rate of 25 frames per second. The video stream is compressed using H.265 encoding before being transmitted to the edge gateway, which can reduce the transmission bandwidth of the video data. (5) Environmental perception data: The ambient temperature, ambient humidity, dust concentration, noise intensity and ground vibration intensity data in the pump room are collected by temperature and humidity sensors, dust concentration sensors, noise sensors and vibration sensors installed in the pump room.

[0024] Because the collected raw data has issues such as inconsistent format, outliers, and spatiotemporal asynchrony, preprocessing is required before use. Outlier detection is performed using the 3σ criterion: first, the mean and standard deviation of each parameter over the past 24 hours are calculated; then, erroneous data exceeding the mean plus or minus three times the standard deviation, as well as incomplete data with a missing rate exceeding 30%, are removed. For data with a missing rate below 30%, linear interpolation is used for completion. This approach aims to preserve data integrity as much as possible. Next, format conversion can be performed to convert the various types of collected data into the Parquet columnar storage format compatible with the industrial internet platform. The Parquet format has a high compression ratio and query efficiency, making it very suitable for storing and processing large-scale industrial time-series data. The collected data typically includes sensor data in binary format, PLC data in CSV format, camera data in JSON format, and environmental monitoring data in XML format. After format conversion, normalization processing can be performed. Here, we mainly use the min-max normalization method to map all data to the range of 0 to 1, eliminating the difference in units between different data types. The calculation method is to subtract the historical minimum value of the parameter from the original data value, and then divide by the difference between the historical maximum value and the minimum value. At this point, the mutual information method can be used to extract key feature parameters related to the centrifugal pump's operating status and model construction. Mutual information can measure the nonlinear correlation between two variables. Calculate the mutual information value between each feature and key state parameters such as equipment vibration and temperature, and retain features with mutual information values ​​greater than 0.1. Finally, we perform spatiotemporal alignment on this data. Here, we use linear interpolation to align the key feature parameters spatiotemporally. In the time dimension, we use the PLC's system time as the benchmark and interpolate the data from other acquisition terminals onto a unified time axis to ensure that the time error of data from different sources does not exceed 10 milliseconds. In the spatial dimension, based on the equipment coordinate system obtained from 3D laser scanning, we convert the measurement positions of each sensor into a unified equipment coordinate system with a spatial error of no more than 1 millimeter. After preprocessing, a dataset in a unified format is generated for subsequent model building.

[0025] Based on the preprocessed unified format dataset, an initial digital twin model is constructed, comprising a four-layer structure of geometry, physics, behavior, and business rules. The specific steps are as follows: The first step is to construct a geometric model. The preprocessed 3D point cloud data is imported into SolidWorks software to construct a 3D geometric model of the centrifugal pump, restoring the external structure, assembly relationship and surface texture of components such as pump body, impeller, shaft, and seals. For internal structures that cannot be scanned, modeling is performed with reference to the equipment design drawings. The second step is to construct a physical model. Combining physical property data with parameters in the centrifugal pump design manual, the material properties, mechanical properties, and thermodynamic properties of each component are defined. The pump body is made of cast iron, for example, with a density of 7200 kg per cubic meter, an elastic modulus of 1.2 x 10^11 Pascals, a Poisson's ratio of 0.25, and an allowable stress of 200 MPa for the impeller, etc. This is just an example. The third step is to build a behavioral model. Based on the historical operating data and process data of the centrifugal pump over the past year, a long short-term memory network is used to train the behavioral model. Flow rate, speed, and ambient temperature are used as inputs to the model, and pump body temperature and vibration acceleration are used as outputs. The ratio of training set to test set is set to 8:2. After training, the model can simulate the operating rules and response characteristics of the centrifugal pump under different loads, different speeds, and different environmental conditions. The fourth step is to build a business rule model. Based on the company's production management specifications and centrifugal pump operating procedures, the Drools rule engine is used to build a business rule model. For example, the safe operating thresholds of the equipment are set as pump body temperature not exceeding 85℃ and vibration speed not exceeding 4.5 mm / s. The fault alarm condition is set as issuing a first-level alarm when the bearing temperature exceeds 90℃. The emergency shutdown logic is set as immediately triggering an emergency shutdown when the vibration acceleration is not less than 10 times the gravitational acceleration. The fifth step is model fusion, which integrates and associates the geometric model, physical model, behavioral model, and business rule model to establish an explicit data mapping relationship based on the unique device identifier of the centrifugal pump. This enables real-time interaction of data between different models, thus forming a complete initial digital twin model.

[0026] After the initial model was built, full runtime sequence data of the centrifugal pump was collected in real time, and the PC algorithm was used to mine the implicit causal relationships between different parameters. The PC algorithm gradually eliminated spurious associations between variables through conditional independence tests, and finally determined the causal direction and strength between variables. Among them, the conditional independence test adopted the chi-square test method, with the significance level set at 0.05. When the p value was greater than 0.05, the two variables were considered to be conditionally independent and the edge between them was deleted. Through continuous data collection and analysis, multiple implicit causal relationships were uncovered, leading to the construction of an implicit causal relationship network for equipment parameters, which includes causal direction, time delay, and influence weight. Among these, a key relationship exists where increased motor current leads to increased bearing temperature, which in turn causes increased vibration acceleration. The time delay from motor current to bearing temperature is 5 minutes, with an influence weight of 0.7; the time delay from bearing temperature to vibration acceleration is 3 minutes, with an influence weight of 0.8. This relationship, not mentioned in the equipment design manual, is the main reason for the recent frequent vibration alarms of the centrifugal pump. Figure 3 As shown.

[0027] To record the changes of each causal relationship over time, a causal chain time decay memory model is established for each causal edge in the implicit causal network. Specifically, this includes: assigning a unique 128-bit UUID to each causal edge and accurately recording its first discovery timestamp, accurate to the millisecond level; then, counting the number of triggers for each causal edge in real time, saving the timestamp of each trigger and the corresponding influence weight value, and generating an influence weight time series. When the relationship between two parameters satisfies the causal condition, the causal edge is considered to have been triggered once; next, the intensity decay coefficient is calculated based on the influence weight time series. The calculation method is the difference between the current time and the first discovery time, divided by the product of the average influence weight and the historical trigger count. The average influence weight is the average of all historical trigger influence weights for this causal edge. Next, a dynamic aging threshold is set, with an initial value of 0.6. This threshold will be dynamically adjusted according to the equipment's operating status and maintenance records, and will be adjusted once a quarter. For each additional year of equipment operation, the threshold will be reduced by 5%. This is because the longer the equipment operates, the more easily the causal relationship between parameters will change. That is, agingThreshold=0.6×(1−0.05×runningYears), where agingThreshold is the dynamic aging threshold and runningYears is the cumulative operating years of the equipment. If the equipment undergoes major maintenance or component replacement, the aging threshold will be reset to 0.6. Finally, when the intensity attenuation coefficient exceeds the dynamic aging threshold, the causal relationship edge is marked as a state to be verified. The conditional independence test is performed on the causal relationship edge marked as a state to be verified every week. If the test result shows that the relationship no longer exists, it is removed from the causal relationship network and the implicit causal relationship network of the device parameters is updated.

[0028] The real-time operating data of the equipment is compared with the simulated output of the initial digital twin model every moment to calculate the real-time deviation value of the target parameter. When the deviation value exceeds the preset threshold, the multi-path root cause tracing process with time weight correction is immediately triggered.

[0029] Suppose that at 14:32 on May 12, 2024, the real-time deviation of the vibration acceleration at the non-drive end of a centrifugal pump is detected to exceed a preset threshold of 0.5 mm / s. Root cause analysis is immediately initiated. First, the node position of the vibration acceleration parameter in the implicit causal network is determined. A breadth-first search with a depth of 5 layers is performed along the reverse path of the causal relationship to generate all possible causal chain paths that could lead to the vibration acceleration deviation, resulting in 8 candidate paths. Then, the intensity attenuation coefficient of all causal edges in each path is obtained, and the time correction weight of that path is calculated. The time correction weight is equal to the product of the intensity attenuation coefficients of all causal edges in the path. In the formula, timeWeight is the time-adjusted weight of the causal chain path, with a value range of (0, 1], and n is the total number of causal edges contained in the causal chain path; decayCoefficient i Let be the intensity decay coefficient of the i-th causal relationship edge in the path, where 1-decayCoefficient i The value range is (0, 1]; it is mainly constructed based on the probability multiplication rule. The overall reliability of the causal chain is equal to the product of the reliability of all causal edges in the chain; the entire causal chain is reliable only when all edges in the chain are reliable; it is used to assign different weights to causal chains with different aging degrees when tracing the root cause, effectively eliminating the interference of weakened or invalid causal associations and improving the accuracy of root cause location. Next, the correction contribution of each path to the vibration acceleration deviation was calculated. The correction contribution was equal to the sum of the deviation values ​​of all upstream parameter nodes in the path multiplied by the time correction weight of the corresponding path. Then, paths with a correction contribution exceeding 0.1 were marked as valid causal chains, resulting in 3 valid paths. Finally, the path with the largest correction contribution, i.e., from motor current to bearing temperature to vibration acceleration, was selected from all valid causal chains. The motor current was marked as the core root cause parameter, and the complete causal chain of this path was extracted. After on-site inspection, it was confirmed that the increase in motor current was indeed caused by the deterioration of bearing grease, leading to increased friction, which was completely consistent with the system's tracing results. The maintenance personnel replaced the grease in time, and the vibration abnormality problem was quickly resolved.

[0030] Based on the root cause analysis results, corresponding model update instructions are generated. First, based on the type of core root cause parameters and the hierarchical structure of the associated causal chain, the hierarchical parameter update is determined. When the equipment undergoes component replacement, structural modification, or major maintenance, a local structural update process is triggered. The geometric parameters of the changed parts are re-scanned using a 3D laser scanner, and the corresponding local geometric structure and component composition in the initial digital twin model are modified. The physical parameters are recalibrated and the six-dimensional validity is verified. After the verification is passed, the digital twin models at the edge and in the cloud are updated synchronously. Then, the aging degree of two edges in the associated causal chain is calculated. The aging degree is equal to the ratio of the intensity attenuation coefficient to the dynamic aging threshold. Assuming that the calculated aging degree of the motor current to bearing temperature edge is 0.72 and the aging degree of the bearing temperature to vibration acceleration edge is 0.65, both belong to medium priority update tasks. The aging degree between 0.5 and 0.8 is medium priority. Then, according to the update priority from high to low, starting from the level where the motor current is located, the adjustment range of the correlation parameters of each level is determined in sequence along the causal forward transmission direction. The adjustment range is calculated by multiplying the deviation value of the core root cause parameter by the original influence weight of the corresponding causal correlation edge to obtain the first product. Then, the first product is multiplied by the difference between 1 and the intensity attenuation coefficient to obtain the adjustment range. Therefore, only the parameter values ​​on the three causal chain paths—motor current, bearing temperature, and vibration acceleration—are updated, while the remaining non-related parameters remain unchanged, and the model is updated incrementally. Then, the adjusted parameters are checked for logical consistency, including checking the parameter value range to ensure that the adjusted parameters are within a physically reasonable range; checking the causal relationship direction to ensure that the parameter adjustment conforms to the direction of the causal network; and checking the conformity with physical laws to ensure that the adjustment results do not violate basic physical laws. The parameter adjustment is considered valid only when all checks pass. Finally, the parameter adjustment results that have passed the checks are synchronized to the initial digital twin model to generate a real-time digital twin model.

[0031] When the equipment undergoes component replacement, structural modification, or major maintenance, a local structural update is triggered; the equipment's geometric parameters are rescanned, the corresponding 3D model structure is modified, the physical property parameters are updated, and the model is re-fused and verified in six dimensions. Installation is only performed after the verification is passed.

[0032] After the model update is completed, the model verification and installation unit conducts six-dimensional validity verification of the real-time digital twin model, including: (1) geometric accuracy verification, comparing the geometric dimensions of the model with the three-dimensional laser scanning data to verify that the geometric error between the model and the physical equipment does not exceed 1 mm; (2) physical accuracy verification, comparing the physical parameters such as temperature and pressure output by the model with the actual measured values ​​to verify that the error does not exceed 5%; (3) behavioral consistency verification, comparing the equipment operation rules simulated by the model with the actual operation data to verify that the consistency is not less than 95%; (4) business compliance verification, checking whether the model complies with the enterprise's production management specifications and equipment operation procedures to ensure that the alarm and shutdown logic is correct; (5) causal chain consistency verification, checking whether the causal relationship inside the model is consistent with the implicit causal relationship network of the equipment parameters; (6) time decay characteristic verification, verifying that the model can correctly reflect the time decay characteristics of the causal relationship and making appropriate weight adjustments to the aging causal relationship; after the verification results show that all dimensions meet the requirements, a verification report containing the verification scores of each dimension and improvement suggestions is generated.

[0033] The validated real-time digital twin model is synchronously installed on the edge computing nodes and cloud servers of the industrial internet platform. The edge computing nodes are equipped with a lightweight real-time digital twin model to perform real-time simulation of equipment operation status, anomaly detection, and rapid response. The cloud servers are equipped with a complete digital twin model to be responsible for long-term trend analysis, fault prediction, and process optimization of the equipment, generating an equipment operation analysis report once a day. Through the edge-cloud collaborative installation method, both the real-time requirements of the industrial site and the computational needs for complex analysis are met.

[0034] The real-time digital twin model after installation is continuously monitored, and model operation data, actual equipment operation data, and environmental change data are collected every 30 minutes. These data are fed back to the implicit causal association mining process and the causal chain time decay memory model update process. A fixed model collaborative update cycle of 7 days is set. In each update cycle, new samples are added to the incremental learning sample library, and the entire link model is updated in a backward order. That is, the time-weighted root cause tracing model is updated first, then the causal chain time decay memory model is updated, and finally the implicit causal association mining model is updated. During the update process, the transfer learning algorithm is used to retain the causal structure knowledge that the model has learned, so as to avoid error accumulation. Based on newly collected data, the implicit causal network is continuously improved, new causal relationships are discovered, and the causal chain time decay memory model is updated. The strength decay coefficient and aging threshold of the causal relationship edges are adjusted. In this way, the model can be updated synchronously with the aging, maintenance and transformation of the equipment, and always maintain a high degree of consistency with the physical equipment.

[0035] It has a complete built-in anomaly handling mechanism, which automatically switches to the backup channel at the same location and generates a fault warning when the sensor fails; enables edge local caching when the network is interrupted and resumes the transmission after the network is restored; when the model inference is abnormal, if the output deviation of 3 consecutive frames exceeds 15%, it automatically loads the most recently verified model version and triggers temporary calibration.

[0036] Example 2: This embodiment introduces an implicit causal association mining algorithm, a causal chain time decay memory model, a time-weighted root cause tracing method, and a hierarchical incremental update strategy. Implicit causal association mining employs a constraint-based PC algorithm, which gradually eliminates spurious associations between variables through conditional independence tests to determine the causal structure between variables; details are as follows: Initialize a completely undirected graph where each node represents a device operating parameter. Then, check the unconditional independence between each pair of variables; if they are independent, delete the edge between them. Next, check the conditional independence of each pair of connected variables given one other variable; if they are independent, delete the edge between them. Gradually increase the number of conditional variables and repeat the above process until no edge can be deleted. Finally, determine the direction of the undirected edges based on the V-structure and other causal inference rules, ultimately obtaining a directed acyclic graph, i.e., the implicit causal relationship network of device parameters. In practical applications, it was found that the PC algorithm has a large computational load when processing high-dimensional data, so some optimizations were made to the algorithm. First, the mutual information method was used to filter out features related to key state parameters, reducing the number of variables, which can greatly improve the running efficiency of the algorithm.

[0037] The causal chain time decay memory model is one of the core innovations of this invention. It can accurately record the changes of each causal relationship over time; specifically as follows: Design a data structure for each causal link, including a unique identifier, first discovery timestamp, historical trigger time series, impact weight of each trigger, average impact weight, intensity decay coefficient, dynamic aging threshold, and a status identifier to be verified. Then calculate the intensity decay coefficient, adjust the dynamic aging threshold, and finally process the edges to be verified. Perform conditional independence tests on causal links marked as statuses to be verified weekly. If the test results show that the association no longer exists (i.e., p-value is greater than 0.05), remove it from the causal network.

[0038] The time-weighted root cause analysis method effectively eliminates interference from weakened or invalid causal relationships by introducing time-adjusted weights, thus improving the accuracy of root cause localization. Specifically: The node position of the target parameter causing the deviation in the implicit causal relationship network is determined. Then, a breadth-first traversal with a preset depth is performed along the reverse path of the causal relationship to generate all causal chain paths that may cause the target parameter deviation. The traversal depth is set to 5 layers, which can ensure that all possible paths are covered while avoiding excessive computation. Next, the time correction weight of each causal chain path is calculated, and then the correction contribution of each path to the target parameter deviation is calculated. Finally, the path with a correction contribution exceeding a preset threshold of 0.1 is marked as a valid causal chain. From all valid causal chains, the path with the largest correction contribution is selected, and its starting node is marked as the core root cause parameter.

[0039] The hierarchical incremental update strategy prioritizes updates based on the aging of causal chains, making targeted adjustments only to the correlation parameters on valid causal chains, thereby improving the efficiency of model updates; specifically as follows: First, based on the type of the core root cause parameters and the hierarchical structure of the causal chain, determine whether to perform a hierarchical parameter update or a local structure update. If the model error is caused by parameter deviation, perform a hierarchical parameter update; if the model error is caused by changes in equipment structure or replacement of components, perform a local structure update. Next, calculate the aging degree of each edge in the causal chain and divide the update task into three priorities based on the aging degree: aging degree greater than 0.8 is high priority, between 0.5 and 0.8 is medium priority, and less than 0.5 is low priority. Following the order of update priority from high to low, starting from the level where the core root cause parameters are located, determine the adjustment range of the associated parameters at each level in sequence along the forward causal propagation direction. Then, only update the parameter values ​​on the causal chain path, leaving non-associated parameters unchanged. After the update is completed, perform a logical consistency check on the adjusted parameters to ensure that the causal relationship inside the model is consistent with the updated implicit causal relationship network.

[0040] The overall approach adopts a collaborative incremental update mechanism across the entire model chain, with the update order proceeding from back to front. Specifically, it first updates the hierarchical incremental update, then updates the time-weighted root cause tracing, followed by the causal chain time decay memory, and finally updates the implicit causal association mining. Furthermore, it employs an incremental learning algorithm, updating model parameters only with new samples while retaining existing knowledge and avoiding full retraining. Simultaneously, it uses the causal constraints of subsequent models as regularization terms to correct the prediction errors of preceding models and prevent error accumulation.

[0041] Example 3: This embodiment adopts a three-layer architecture of edge acquisition, edge computing, and cloud backup to introduce a dynamic modeling system for digital twins of equipment based on the Industrial Internet. The system mainly consists of eight modules, which communicate with each other through a standardized OPC UA industrial interface to ensure the reliability and compatibility of data transmission; details are as follows: The data acquisition and preprocessing module is the data foundation of the entire system. It consists of an edge gateway, various sensors, PLC, industrial cameras, and synchronous trigger controllers. It is mainly responsible for the drive docking and real-time data processing of the detection equipment, and completes the synchronous acquisition and standardized preprocessing of multi-scale multi-source data. This provides reliable basic data for subsequent analysis. It can upload the preprocessed standardized data to the full-cycle performance evolution prediction module and the detection sample set construction module in real time. At the same time, it receives differentiated detection cycle information issued by the full-cycle performance evolution prediction module, adjusts the acquisition parameters in real time, and dynamically adjusts the acquisition strategy. The initial model building module, installed inside the cloud server, is mainly used to build an initial digital twin model based on the preprocessed dataset. This module communicates in real time with the data acquisition and preprocessing module and the model verification and installation module. It receives a unified format dataset from the data acquisition and preprocessing module and sends the completed initial model to the model verification and installation module for verification. The implicit causal relationship mining module is used to mine implicit causal relationships in the timing data of equipment operation in real time and construct an implicit causal relationship network of equipment parameters. This module communicates in real time with the data acquisition and preprocessing module and the causal chain time decay memory module, receives real-time operation data from the data acquisition and preprocessing module, and sends the constructed causal relationship network to the causal chain time decay memory module. The causal chain time decay memory module is used to build a time decay memory model for each causal association edge, record its time characteristics and intensity changes, and dynamically update the causal association network. This module communicates in real time with the implicit causal association mining module and the time weight root cause tracing module, receives the causal association network from the implicit causal association mining module, and provides the time characteristic data of the causal association edges to the time weight root cause tracing module. The time-weighted root cause tracing module is used to compare the deviation values ​​calculated by the real-time operation data of the equipment with the simulation output of the initial model, and to perform multi-path root cause tracing based on the causal chain time decay memory model to locate the core root cause parameters and related causal chains. This module communicates in real time with the causal chain time decay memory module and the hierarchical incremental update module, receives the time characteristic data from the causal chain time decay memory module, and sends the root cause tracing results to the hierarchical incremental update module. The hierarchical incremental update module is used to generate hierarchical incremental model update instructions based on the root cause tracing results, adjust the parameters and optimize the structure of the initial model, and generate a real-time digital twin model. This module communicates in real time with the time-weighted root cause tracing module and the model verification and installation module, receives the root cause tracing results from the time-weighted root cause tracing module, and sends the updated real-time model to the model verification and installation module for verification. The modules mentioned above are all installed on edge computing nodes, namely the implicit causal association mining module, the causal chain time decay memory module, the time weight root cause tracing module, and the hierarchical incremental update module.

[0042] The model verification and installation unit, installed on the cloud server, is used to perform six-dimensional validity verification of real-time digital twin models and generate verification reports. It synchronously installs qualified models to edge computing nodes and the cloud server. This module communicates in real time with the hierarchical incremental update module and the full lifecycle monitoring and optimization module. It receives real-time models from the hierarchical incremental update module and sends the installed model information to the full lifecycle monitoring and optimization module.

[0043] The full lifecycle monitoring and optimization module is also installed on the cloud server. It is used to continuously monitor the installed model, collect various data and feed them back to the front-end to form a closed-loop optimization. This module communicates with all other modules in real time, receives and stores all modules' running data, detection data, model parameters and detection reports, and provides a data foundation for model updates.

[0044] In the specific implementation process, the data acquisition and preprocessing module initiates synchronous acquisition to obtain multi-source detection data and scene parameters during equipment operation, and uploads them to the edge computing terminal after preprocessing; the implicit causal association mining module receives the data and performs implicit causal association mining and causal association network construction; the causal chain time decay memory module establishes a time decay memory model for each causal association edge; when a deviation in equipment operation is detected, the time weight root cause tracing module initiates root cause tracing; the hierarchical incremental update module generates model update instructions based on the root cause tracing results and updates the model; the model verification and installation module verifies and installs the updated model; and the full lifecycle monitoring and optimization module continuously monitors the operating status of the model and equipment, collects data, and feeds it back to the front-end to facilitate optimization.

[0045] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0046] The above description is only a specific embodiment of this application, but the protection scope of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the protection scope of this application.

Claims

1. A dynamic modeling method for equipment digital twins based on the Industrial Internet, characterized in that, The steps of this method are as follows: After acquiring multi-source heterogeneous operational data of target equipment in industrial sites and obtaining a unified format dataset through standardized preprocessing, an initial digital twin model containing a four-layer structure of geometry, physics, behavior, and business rules is constructed. Real-time acquisition of full runtime sequence data of the target device, mining of implicit causal relationships between different parameters, construction of implicit causal relationship network of device parameters, and comparison of real-time running data of the device with the simulation output of the initial digital twin model at time intervals to calculate the real-time deviation value of the target parameters; Establish a causal chain time decay memory model for each causal link in the network, and record the first discovery time, historical trigger time series, single trigger impact weight and intensity fluctuation characteristics; Multi-path root cause tracing based on the causal chain time decay memory model performs time weight correction on real-time deviation values, locates core root cause parameters and all related causal chains, and quantifies the deviation contribution ratio of each causal chain. Based on the root cause tracing results, a hierarchical incremental model update instruction based on the aging degree of the causal chain is generated. The initial model is then adjusted layer by layer and its local structure is optimized according to the causal transmission direction and update priority to generate a real-time digital twin model. The effectiveness of the real-time digital twin model is then verified, and a verification report is obtained. The validated model is synchronously installed on the edge computing nodes and cloud servers of the industrial internet platform, and the installed model is monitored throughout its entire lifecycle. Data on model operation, actual equipment operation, and environmental changes are continuously collected at preset time intervals and updated accordingly.

2. The method for dynamic modeling of equipment digital twins based on the Industrial Internet according to claim 1, characterized in that, When acquiring multi-source heterogeneous operational data, an encrypted transmission channel is established between the industrial internet platform edge gateway and various data acquisition terminals of the target device to synchronously collect the following data: A 3D laser scanner acquires the equipment's external dimensions, component locations, assembly gaps, and surface defect geometric parameters; the equipment's built-in temperature, pressure, vibration, speed, and current sensors collect physical property data. The PLC controller outputs real-time operating status, including start / stop status, runtime, operation commands, valve opening degree, and motor power; industrial cameras capture equipment operation images and component movement processes; and environmental monitoring devices collect environmental perception data such as temperature, humidity, dust concentration, noise intensity, and vibration intensity.

3. The method for dynamic modeling of equipment digital twins based on the Industrial Internet according to claim 1, characterized in that, The standardization preprocessing steps are as follows: Statistical outlier detection methods are used to remove erroneous data that exceeds a reasonable threshold, as well as incomplete data with a missing rate exceeding a preset missing threshold. The raw data of different formats are uniformly converted into a columnar storage format compatible with the industrial internet platform, and all data are mapped to the [0, 1] interval through the linear normalization method; A feature selection method is used to extract key features related to equipment operating status and model construction, and redundant information is removed. Interpolation is used to align key features in the time and space, ensuring that the temporal error of data from different sources does not exceed a preset time threshold and the spatial error does not exceed a preset spatial threshold, thus forming a dataset in a unified format.

4. The method for dynamic modeling of equipment digital twins based on the Industrial Internet according to claim 1, characterized in that, When constructing the initial digital twin model, a three-dimensional geometric model is built based on the geometric parameter data to restore the equipment's external structure, component composition, assembly relationship, and surface texture; A physical model is constructed by combining physical property data with equipment design parameters, and the material properties, mechanical properties and thermodynamic properties of each component are defined. Based on historical operational data and process data, a behavioral model is trained using machine learning algorithms to simulate the operating patterns and response characteristics of equipment under different working conditions. In accordance with the enterprise's production management standards and equipment operating procedures, a business rule model is constructed through a rule engine to set equipment safe operation thresholds, fault alarm conditions, and emergency shutdown logic. By integrating and associating the four-layer model, an explicit data mapping relationship based on the unique identifier of the device is established, forming an initial digital twin model.

5. The method for dynamic modeling of equipment digital twins based on the Industrial Internet according to claim 1, characterized in that, A constraint-based causal discovery algorithm is employed to mine implicit causal relationships in runtime sequence data, construct an implicit causal relationship network for device parameters, and establish a causal chain time decay memory model for each causal relationship edge, as detailed below: Assign a unique identifier to each edge and record the first discovery timestamp; count the number of times each edge is triggered in real time, save the timestamp of each trigger and the corresponding impact weight value, generate an impact weight time series, and calculate the intensity decay coefficient; Set an aging threshold that is dynamically adjusted based on the equipment's operating status and maintenance records. When the intensity attenuation coefficient exceeds the dynamic aging threshold, the edge is marked as a state to be verified. Perform conditional independence checks on the edges to be verified, remove associated edges whose strength has decayed to an invalid level, and update the implicit causal relationship network of the device parameters.

6. The method for dynamic modeling of equipment digital twins based on the Industrial Internet according to claim 1, characterized in that, When performing multi-path root cause tracing with time weight correction, first determine the node position of the target parameter that caused the deviation in the implicit causal relationship network; then perform a breadth-first traversal of a preset depth along the reverse causal relationship path to generate all causal chain paths that may cause deviation of the target parameter. Obtain the strength attenuation coefficient of all causal edges in each path, and calculate the time correction weight of the path and the correction contribution of each path to the deviation of the target parameters. Paths whose correction contribution exceeds a preset threshold are marked as valid causal chains. The path with the largest correction contribution is selected from all valid causal chains, its starting node is marked as the core root cause parameter, and the complete causal chain of the path is extracted.

7. The method for dynamic modeling of equipment digital twins based on the Industrial Internet according to claim 1, characterized in that, After generating a hierarchical incremental model update instruction based on the aging degree of the causal chain, hierarchical parameter update or local structure update is performed according to the type of the core root cause parameter and the hierarchical structure of the associated causal chain. The aging degree of each edge in the causal chain is calculated, and the update task is divided into three priority levels: high, medium and low. The aging degree is the ratio of the intensity decay coefficient to the dynamic aging threshold. Following the order of priority from high to low, starting from the level where the core root cause parameter is located, the adjustment range of each level of related parameters is set sequentially along the causal forward transmission direction, and the parameter values ​​on the causal chain path are updated, while non-related parameters remain unchanged. When performing local structural updates, the corresponding local geometry and component composition in the initial model are modified according to the actual structural changes of the equipment or the replacement of components. The adjusted parameters and optimized structure are logically consistent to ensure that the causal relationships within the model are consistent with the updated implicit causal relationship network; and the results of the successful verification are synchronized to the initial model to generate a real-time digital twin model.

8. A dynamic modeling system for equipment digital twins based on the Industrial Internet, characterized in that, The modeling system includes: The data acquisition and preprocessing module acquires multi-source heterogeneous operating data of target equipment in the industrial field and generates a unified format dataset through standardized preprocessing. The initial model building module constructs an initial digital twin model with a four-layer structure, including geometry, physics, behavior, and business rules, based on a unified format dataset. The implicit causal association mining module collects all runtime sequence data of the target device in real time, and uses a constraint-based causal discovery algorithm to mine the implicit causal relationships between parameters and construct an implicit causal association network of device parameters. The causal chain time decay memory module establishes a causal chain time decay memory model for each causal connection edge in the network, records its first discovery time, historical trigger time sequence, single trigger impact weight and intensity fluctuation characteristics, and dynamically updates the causal connection network. The time-weighted root cause tracing module compares the real-time operating data of the equipment with the initial model simulation output to calculate the deviation value, and performs multi-path root cause tracing based on the causal chain time decay memory model to perform time-weight correction, and locate the core root cause parameters and all related causal chains. The hierarchical incremental update module generates hierarchical incremental model update instructions based on the aging degree of the causal chain according to the root cause tracing results. It adjusts the parameters and optimizes the local structure of the initial model layer by layer according to the causal transmission direction and update priority, and generates a real-time digital twin model. The model verification and installation module verifies the effectiveness of real-time digital twin models and generates verification reports. It then synchronously installs compliant models to the edge computing nodes and cloud servers of the industrial internet platform. The full lifecycle monitoring and optimization module monitors the installed model throughout its entire lifecycle, continuously collecting various types of data at preset time intervals and providing feedback updates.

9. The equipment digital twin dynamic modeling system based on the Industrial Internet according to claim 8, characterized in that, The causal chain time decay memory module includes: The causal edge identification unit assigns a unique identifier to each causal edge and records the first discovery timestamp; Trigger statistics unit to count the number of triggers and their corresponding impact weights in real time, and generate time series of impact weights; The attenuation coefficient calculation unit calculates the intensity attenuation coefficient based on the influence weight time series and dynamically adjusts the aging threshold based on the equipment operating status and maintenance records. When the intensity attenuation coefficient exceeds the aging threshold, the edge is marked as a state to be verified. The causal network update unit periodically performs conditional independence checks on the edges to be verified, removes invalid association edges, and updates the implicit causal network of device parameters.

10. The equipment digital twin dynamic modeling system based on the Industrial Internet according to claim 8, characterized in that, The hierarchical incremental update module includes: The update instruction generation unit generates corresponding update instructions based on the type of the core root cause parameters and the hierarchical structure of the associated causal chain. Priority partitioning unit: calculates the aging degree of each edge in the associated causal chain and partitions the update priority; The layer-by-layer parameter adjustment unit, upon receiving a layer-by-layer parameter update instruction, determines the adjustment range of the associated parameters at each level in sequence according to priority, starting from the level where the core root cause parameter is located, and updates the parameters on the causal chain path. The local structure optimization unit, upon receiving a local structure update instruction, modifies the local geometry and component composition of the initial model according to the actual changes in the equipment, and performs logical consistency verification between the adjusted parameters and the optimized structure. The model synchronization unit synchronizes the verified results to the initial model, generating a real-time digital twin model.

Images