Equipment digital twin model updating method and system, terminal and medium

Abstract

The invention discloses an equipment digital twin model updating method and system, a terminal and a medium. The method comprises the following steps: collecting data of a plurality of positions of equipment; for any data acquisition point on the equipment, the monitoring frequency is dynamically adjusted according to the change sensitivity index of the monitoring data; preprocessing the collected data; according to the processed data, updating a digital twin model of the equipment; when the model is updated, the allocation of computing resources is dynamically adjusted according to the change trend of the data; and updating the part, influenced by the physical entity change, in the model by adopting an incremental learning algorithm. The method has the advantages that the updating efficiency of the equipment digital twin model can be improved, the real-time performance and reliability of a digital twin system are remarkably improved, and powerful support is provided for safe operation, maintenance and management of large equipment.

Description

Technical Field

[0001] The present application relates to the field of digital twin technology, and in particular to a method, system, terminal and medium for updating a digital twin model of a device. Background Art

[0002] In the application of digital twin technology, a key issue is the real-time update of the model. As the physical entity operates, its state will continue to change, such as equipment wear, structural deformation, etc. The digital twin model needs to be updated accordingly to accurately reflect these changes. However, there are many limitations in existing technologies, such as the limitations of sensor monitoring frequency and coverage, which make it impossible to fully and timely capture the change information of physical entities. At the same time, the algorithms and computing resources required to update the model are complex, which restricts the real-time performance of the digital twin system. Especially for large equipment, such as large cranes, structural changes under the action of wind loads, loads, etc. are difficult to reflect in real time in the digital twin model, which greatly affects the effectiveness and application value of digital twin technology. Summary of the invention

[0003] The present application provides a method, system, terminal and medium for updating the digital twin model of a device, which has the advantage of improving the update efficiency of the digital twin model of the device, significantly improving the real-time performance and reliability of the digital twin system, and providing strong support for the safe operation, maintenance and management of large-scale equipment.

[0004] The technical solution of this application is as follows: On the one hand, the present application provides a method for updating a device digital twin model, comprising the following steps: S1: collects data from several locations of the device; For any data collection point on the equipment, the monitoring frequency is dynamically adjusted according to the sensitivity index of the monitoring data change; S2: Preprocess the collected data, including data cleaning, outlier detection and noise reduction; S3: Update the digital twin model of the device based on the processed data; when the model is updated: Select the update algorithm based on the data type. For the strain data of structural deformation, the update algorithm based on finite element analysis is adopted. For the equipment wear data, the algorithm based on historical wear data and machine learning is used to predict the wear trend and update the wear status of the components in the model. Dynamically adjust the allocation of computing resources according to the changing trend of data, and select the model update algorithm according to the allocation ratio of computing resources; An incremental learning algorithm is used to update the parts of the model that are affected by changes in physical entities.

[0005] Furthermore, in step S1, the collected equipment data includes strain data, displacement data and vibration data of different parts of the equipment.

[0006] Furthermore, in step S1, a distributed sensor network is used to collect data from several locations of the device, and the collection frequency of the sensor is obtained by the following formula:

[0007] in, is the initial monitoring frequency of the sensor, k is the adjustment coefficient, and S is the sensitivity index of the change of the monitoring data; The change sensitivity index S of the monitoring data is obtained by any of the following methods: (1) For structural strain data monitoring:

[0008] : The variance of the data in the current time window; Time window length T: set according to the response time of the physical entity; : Historical maximum variance, historical variance update cycle: daily or weekly update; (2) For displacement data and vibration data:

[0009] : The time gradient of the data; the gradient calculation cycle : consistent with the sensor sampling frequency; : Maximum historical gradient; When the time gradient of the data exceeds the preset threshold for multiple times in a row, it is judged as abnormal and an abnormal warning is issued; (3) For complex working conditions of equipment:

[0010] : Weight coefficient; through historical data regression analysis, select the model that minimizes the prediction error , value.

[0011] Furthermore, in step S1, a distributed sensor network is used to collect data from a plurality of locations of the device, and sensors in the sensor network communicate with each other, and: (1) Sensors communicate with each other using a dynamic routing protocol based on ant colony optimization: The dynamic routing mechanism is designed by using the ant colony optimization algorithm. Each sensor node releases virtual pheromones to mark the quality of the communication path. When some nodes fail or data transmission is blocked, the neighboring nodes dynamically select alternative paths based on the pheromone concentration. The path selection probability formula is:

[0012] in: : pheromone concentration of path itoj at time t; : Heuristic factor, is the path length; : pheromone weight coefficient; : Heuristic factor weight coefficient; The pheromone update rules are:

[0013] : Pheromone volatilization rate; :The kth ant is on the path The increment of pheromone released, Q=1 is a constant, L k is the total path length of ant k; (2) Each sensor is embedded with a lightweight reinforcement learning model, which makes autonomous decisions on monitoring strategies through local observation. When a neighboring sensor fails, the reinforcement learning model triggers the "compensation mode", automatically increases its own monitoring frequency, and optimizes resource allocation through the Qlearning algorithm. The Q value update formula is:

[0014] in: : learning rate; : discount factor; Reward function design:

[0015] W1=0.7,W2=0.3: weight coefficient; (3) Adaptive energy-data trade-off mechanism: The sensor node evaluates its own energy level and data importance in real time, and dynamically adjusts the transmission mode. When energy is insufficient, it gives priority to transmitting key data, including abnormal signals, and reduces the sampling frequency of non-critical data. When data transmission is blocked, it switches to low-power broadcast mode and transmits data through multi-hop relays. (4) Fountain Code redundant coding technology is used to divide the data into blocks and store them in adjacent sensors. Fountain Code redundant coding: The data is divided into k original packets, generating n coded packets, n ≥ k; Each coded packet is the XOR of d randomly selected original packets; Decode Threshold: Receive The encoded packets were successfully decoded. is the failure probability; (5) Collaborative Compressed Sensing and Data Fusion Neighboring sensors share raw data and use compressed sensing algorithms for joint sparse representation to reduce redundant data transmission. At the same time, multi-sensor data is integrated through Kalman filtering or Bayesian fusion algorithms to improve data quality; when a sensor fails, neighboring nodes reconstruct missing signals based on historical data and current observations; collaborative processing reduces communication load and improves fault tolerance; moving average filtering filter formula:

[0016] Parameter selection: N=5: window size; Data entropy calculation: Entropy formula: ; Binning rules: adaptive binning; (6) Trust management based on blockchain: A lightweight blockchain protocol is introduced to record the status and data interaction history of sensor nodes. When a node fails or the data is abnormal, neighboring nodes verify the data credibility through a consensus mechanism and isolate malicious or faulty nodes.

[0017] Furthermore, in step S2, a statistical analysis method is used to identify and remove abnormal data points. The 3σ principle is used. If the data x satisfies , is the mean, σ is the standard deviation, then the data is judged as an outlier and excluded; Moving average filtering is used to reduce the noise of the data. The formula is:

[0018] in is the filtered value of the nth data point, are the original data points and N is the size of the moving average window.

[0019] Furthermore, in step S3, the allocation of computing resources is dynamically adjusted according to the change trend of the data as follows: When a set of data is used to update the device digital twin model, the computing resource allocation ratio R of the set of data is:

[0020] Where a and b are coefficients determined based on system performance and equipment characteristics; Entropy H can be calculated by Shannon entropy formula:

[0021] is the probability distribution of the data.

[0022] Furthermore, in step S3, an incremental learning algorithm is used to update the part of the model affected by the change of the physical entity; The incremental update of the finite element model is based on the variational principle and the virtual work principle; assuming that the structural deformation , which satisfies the following variational equation:

[0023] in: is the domain of the structure, σ is the stress tensor, is the strain tensor, It's physical strength. is the surface force on the boundary, is the virtual displacement; by solving this equation, the deformation state of the structure is updated, and then the structural information in the digital twin model is updated.

[0024] On the other hand, the present application provides a device digital twin model update system, comprising: A data acquisition unit, wherein the data acquisition unit includes a plurality of data acquisition devices, which are distributed and deployed at a plurality of locations of the equipment to collect data of the equipment; the data acquisition device includes a sensor module, a data processing module and a communication module, the sensor module includes one or more of a strain sensor, a displacement sensor and a vibration sensor, and the data processing module is used to pre-process the data collected by the sensor module; A data center is connected to the data acquisition device for receiving data from the data acquisition device, and the data center updates the digital twin model of the equipment according to the received data; when updating the model: an update algorithm is selected according to the data type, and an update algorithm based on finite element analysis is used for strain data of structural deformation; for equipment wear data, an algorithm based on historical wear data and machine learning is used to predict the wear trend and update the wear status of components in the model; the allocation of computing resources is dynamically adjusted according to the change trend of the data; an incremental learning algorithm is used to update the part of the model affected by the change of the physical entity; and a storage and display unit for storing and displaying the digital twin model.

[0025] On the other hand, the present application provides a digital twin model terminal, including a memory, a processor and a display, the display is used to display the digital twin model, the memory stores a computer program, and when the computer program is called and executed by the processor, the device digital twin model update method as described above is implemented.

[0026] On the other hand, the present application provides a computer-readable medium, characterized in that the computer-readable medium memory stores a computer program, and when the computer program is called and executed by a computer, it implements the device digital twin model updating method as described above.

[0027] In summary, the beneficial effects of this application are: 1. Dynamically adjust the monitoring frequency according to the operating status and environment of the physical entity, realize the optimal allocation of resources, collect less data when the data is small, and reduce the amount of data processing; calculate the monitoring frequency by using the sensitivity index of the change of the monitoring data, so that the sensor can intelligently adapt to the data collection needs under different working conditions; 2. Use the 3σ principle and moving average filtering to accurately detect outliers and reduce data noise, providing high-quality data for subsequent model updates; 3. The update algorithm is flexibly selected according to different data types, and computing resources are dynamically allocated according to data changes, which avoids the limitations of fixed algorithms and fixed resource allocation in traditional methods and improves the utilization efficiency of computing resources; computing resources are dynamically allocated based on data entropy to achieve adaptive adjustment of resources; the introduction of incremental learning algorithms only updates the affected model parts, significantly reducing the amount of calculation and time required for the update, making the model update more real-time. The incremental update of the finite element model is realized by using the variational principle and the virtual work principle, which ensures the physical rationality and accuracy of the model update. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 It is a schematic diagram of the device digital twin model update system in this application. DETAILED DESCRIPTION

[0029] The specific implementation of the present application is described in detail below with reference to the accompanying drawings.

[0030] A specific embodiment of the present application provides a method for updating a digital twin model of a device, comprising the following steps: S1: collects data from several locations of the device; For any data collection point on the equipment, the monitoring frequency is dynamically adjusted according to the sensitivity index of the monitoring data change; The collected equipment data include strain data, displacement data and vibration data of different parts of the equipment.

[0031] A distributed sensor network is used to collect data from several locations of the device. The collection frequency of the sensor is obtained by the following formula:

[0032] in, is the initial monitoring frequency of the sensor, k is the adjustment coefficient, and S is the sensitivity index of the change of the monitoring data; The change sensitivity index S of the monitoring data is obtained by any of the following methods: (1) Based on data variance calculation, it is applicable to scenarios where the data changes significantly but slowly, typically for structural strain data monitoring:

[0033] : The variance of the data in the current time window; Time window length T: set according to the response time of the physical entity (e.g. crane strain monitoring, T=10s); : Historical maximum variance (used for normalization), historical variance update cycle: daily or weekly update to prevent long-term drift.

[0034] (2) Based on data gradient calculation, it is applicable to scenarios where the data change rate is sensitive, typically for vibration monitoring and displacement data detection:

[0035] : The time gradient of the data; the gradient calculation cycle : consistent with the sensor sampling frequency (for example = 0.1s); : Maximum historical gradient; When the time gradient of the data exceeds the preset threshold multiple times in a row, it is judged as abnormal and an abnormal warning is issued.

[0036] (3) Based on the calculation of data mixed indicators (variance + gradient), it is applicable to complex working conditions with comprehensive change amplitude and rate, typically for crane swing monitoring under wind load:

[0037] : Weight coefficient, which needs to be optimized according to the scenario (e.g. = 0.6, = 0.4); through historical data regression analysis, select the one that minimizes the model prediction error , value.

[0038] Distributed sensor networks consist of multiple smart sensors that are deployed at key locations and important components of physical entities. Smart sensors are not only highly accurate and sensitive, but also have certain intelligent processing capabilities and can perform preliminary processing on the collected data. For example, for large cranes, different types of sensors, including strain sensors, displacement sensors, vibration sensors, etc., are installed at their main force-bearing structures, key connecting components, transmission devices, etc.

[0039] The sensors used are shown in Table 1: Table 1: Sensor types in distributed sensor networks Serial number type brand model applicability 1 Strain Sensors HottingerBaldwinMesstechnikGmbH C10 High precision, high stability and good linearity can accurately measure the slight strain changes of the structure. The accuracy can reach ±0.1%FS (full scale), which can meet the accuracy requirements of structural stress and strain monitoring of large cranes and other equipment. 2 Displacement Sensors Micro-Epsilon ILD1402 Using the principle of laser triangulation, high-precision displacement measurement can be achieved, with a resolution of up to sub-micron level. The measurement range can be selected from a few millimeters to a few meters according to different models. It is very suitable for measuring the displacement changes of crane hoisting mechanisms, booms, etc. 3 Vibration Sensor Bruel&Kjaer 4507 It has a wide frequency response range (from a few Hz to tens of kHz), can accurately measure the vibration of the crane during operation, and plays an important role in detecting abnormal vibration and fault warning of equipment. Its measurement accuracy can maintain a high level in different frequency bands, and can effectively capture tiny vibration changes. The deployment of sensors adopts a distributed network structure, which can automatically adjust its monitoring frequency and monitoring range according to the complexity and importance of the physical entity. For example, when abnormal changes (such as stress concentration) are detected in key parts of the equipment, the sensors in that part will automatically increase the monitoring frequency to obtain more detailed data; while for relatively stable parts, the monitoring frequency can be appropriately reduced to save energy and data transmission resources.

[0040] In some other embodiments, the communication connection between sensors realizes self-organizing communication to form an intelligent network. When some sensors fail or data transmission is blocked, adjacent sensors can adjust monitoring strategies through the self-organizing communication mechanism to ensure data continuity and integrity.

[0041] Specifically, sensors in a sensor network communicate with each other and: (1) Sensors communicate with each other using a dynamic routing protocol based on ant colony optimization: The dynamic routing mechanism is designed using the ant colony optimization algorithm. Each sensor node simulates the behavior of "ants". Each sensor node releases virtual pheromones to mark the quality of the communication path. When some nodes fail or data transmission is blocked, the neighboring nodes dynamically select alternative paths based on the pheromone concentration to bypass the faulty area. The pheromone concentration is updated in real time, and low-latency, high-stability paths are given priority. Without relying on central nodes, sensors can independently complete path discovery and maintenance, significantly improving the network's fault tolerance. Path selection probability formula:

[0042] in: : pheromone concentration of path itoj at time t; : Heuristic factor, is the path length; : Pheromone weight coefficient. Experiments show that the convergence speed is optimal when α=1; : Heuristic factor weight coefficient, focusing on path length optimization; The pheromone update rules are:

[0043] : The pheromone volatilization rate is determined through simulation experiments to balance exploration and utilization; :The kth ant is on the path The increment of pheromone released, Q=1 is a constant, L k is the total path length of ant k; (2) Each sensor is embedded with a lightweight reinforcement learning model, which makes autonomous monitoring decisions based on local observations (such as the status of neighboring nodes, its own remaining energy, and sensitivity to data changes). When a neighboring sensor fails, the reinforcement learning model triggers the "compensation mode", automatically increases its own monitoring frequency, and optimizes resource allocation through the Qlearning algorithm. This approach combines edge computing and reinforcement learning to achieve decentralized intelligent decision-making and reduce reliance on cloud computing power. Q value update formula:

[0044] in: : Learning rate, a low learning rate is conducive to stable convergence; : Discount factor, focusing on long-term returns; Reward function design:

[0045] W1=0.7,W2=0.3: weight coefficients, (optimized through A / B testing, prioritizing data integrity); (3) Adaptive energy-data trade-off mechanism: Sensor nodes evaluate their own energy levels and data importance in real time, and dynamically adjust transmission modes. When energy is insufficient, they give priority to transmitting critical data, which includes abnormal signals, and reduce the sampling frequency of non-critical data. When data transmission is blocked, they switch to low-power broadcast mode and transmit data through multi-hop relays. By dynamically weighing energy consumption and data value, the network life is extended while ensuring the integrity of critical data.

[0046] (4) Fountain Code redundant coding technology is used to divide data into blocks and store them in neighboring sensors. When some nodes fail, the system restores the original information through any sufficient number of data blocks to avoid data loss. Combining erasure codes with distributed storage, highly robust data backup and recovery can be achieved. Fountain Code (LT Code) redundant coding: The data is divided into k original packets, generating n coded packets, n ≥ k; Each coded packet is the XOR of d randomly selected original packets (d follows the degree distribution Ω(d), such as the robust solitary wave distribution); Decode Threshold: Receive The encoded packets were successfully decoded. , is the failure probability; (5) Collaborative Compressed Sensing and Data Fusion Neighboring sensors share raw data and use compressed sensing algorithms for joint sparse representation to reduce redundant data transmission. At the same time, multi-sensor data is integrated through Kalman filtering or Bayesian fusion algorithms to improve data quality; when a sensor fails, neighboring nodes reconstruct missing signals based on historical data and current observations; collaborative processing reduces communication load and improves fault tolerance; moving average filtering filter formula:

[0047] Parameter selection: N=5: window size (determined by frequency domain analysis, balancing smoothness and real-time performance); Data entropy calculation (Shannon entropy): Entropy formula: ; Binning rules: adaptive binning; (6) Trust management based on blockchain: A lightweight blockchain protocol is introduced to record the status of sensor nodes and the data interaction history. When a node fails or the data is abnormal, neighboring nodes verify the data credibility through a consensus mechanism and isolate malicious or faulty nodes to enhance network security and prevent data pollution caused by single point failures or malicious attacks.

[0048] In this embodiment, in the sensor network, the parameters are as follows: (1) Sensor nodes: Core chip: Microprocessor: ARMCortex-M7 (main frequency 300MHz, support floating-point operations, meet local filtering and decision-making calculations); Communication module: SemtechSX1276LoRa chip (frequency band 868MHz, transmission power 14dBm, communication distance 1km).

[0049] Energy Management: Power supply: 3.7V lithium thionyl chloride battery (capacity 19Ah, low self-discharge rate); Dynamic power consumption adjustment: sleep mode current ≤ 1μA, active mode current ≤ 20mA.

[0050] (2) Edge computing unit: Hardware configuration: Processor: NVIDIA Jetson Nano (quad-core ARM A57, 128-core GPU, supports TensorFlow Lite); Storage: 64GB eMMC (read and write speed 100MB / s).

[0051] (3) Communication network: 5G module: QuectelRM500Q-GL (supports Sub-6GHz, peak rate 2.5Gbps); Protocol configuration: CoAPoverUDP (small header overhead, suitable for low-bandwidth transmission).

[0052] Example of implementing a sensor network that self-organizes communication and dynamically adjusts monitoring strategies (taking a large crane scenario as an example): Failure scenario: A strain sensor on the crane boom fails due to mechanical shock.

[0053] Dynamic adjustment process: 1. Neighboring sensors detect failed nodes through the ant colony routing protocol and re-plan the data transmission path; 2. The displacement sensors near the failed nodes trigger the reinforcement learning model, increasing the monitoring frequency from 5 Hz to 10 Hz to compensate for the missing strain data; 3. The data preprocessing module recovers the historical data of the failed node from other nodes through redundant coding; 4. The adaptive algorithm module infers the local stress changes of the boom based on the fused displacement and vibration data and updates the digital twin model.

[0054] The above technologies can achieve the following technical effects: Data continuity: The data loss rate of faulty nodes is reduced to less than 5%; Real-time: Network self-healing time is controlled within 200ms; Energy efficiency: Dynamic energy management extends the life of sensor networks by 30%.

[0055] Through distributed intelligence and adaptive mechanisms, high robustness, low latency and high energy efficiency of the sensor network are achieved, providing reliable guarantee for the real-time update of the digital twin model.

[0056] S2: Preprocess the collected data, including data cleaning, outlier detection and noise reduction; In step S2, the statistical analysis method is used to identify and remove abnormal data points. The 3σ principle is used. If the data x satisfies , If is the mean and σ is the standard deviation, the data is considered to be an outlier and is excluded; Moving average filtering is used to reduce the noise of the data. The formula is:

[0057] in is the filtered value of the nth data point, are the original data points and N is the size of the moving average window.

[0058] S3: Update the digital twin model of the device based on the processed data; when the model is updated: The update algorithm is selected according to the data type. For the strain data of structural deformation, the update algorithm based on finite element analysis is adopted. For the equipment wear data, the algorithm based on historical wear data and machine learning is used to predict the wear trend and update the wear status of the components in the model.

[0059] According to the trend of data changes, dynamically adjust the allocation of computing resources, and select the model update algorithm according to the allocation ratio of computing resources; when data changes are relatively stable, allocate fewer computing resources; when sudden changes or anomalies occur, automatically call more computing resources for accurate calculations and model updates. Specifically: When a set of data is used to update the device digital twin model, the computing resource allocation ratio R of the set of data is:

[0060] Where a and b are coefficients determined based on system performance and equipment characteristics; Entropy H can be calculated by Shannon entropy formula:

[0061] is the probability distribution of the data.

[0062] When the data change rate is large, it indicates that the physical changes of the corresponding part of the equipment are also large. At this time, more computing resources are called to process data to update the model. When the data change rate reaches the set threshold, a high-precision finite element model is used to update the calculation; in normal operation, the data change rate is small, fewer resources are allocated to process data, and a simplified model update algorithm is used to update the model.

[0063] The simplified model updating algorithm is: a fast response method based on dynamic order reduction.

[0064] The dynamic reduced-order model (DROM) is a simplified algorithm designed for low-resource scenarios. It significantly reduces computational complexity through data-driven main mode selection and reduced-order space construction, while ensuring the available accuracy of model updates. Its core innovation lies in the dynamic adjustment mechanism and resource perception capabilities, filling the gap in the simplified update algorithm of the digital twin model.

[0065] When the computing resource allocation ratio R is low, the system uses the Dynamic Reduced Order Model (DROM) as a simplified algorithm to achieve efficient and real-time digital twin model updates. The following are the specific implementation steps and technical details: Under the premise of ensuring a certain accuracy, the computational complexity is significantly reduced, which is suitable for low-resource scenarios (such as R<0.3). Driven by real-time data, key modes (such as principal strain and principal displacement direction) are dynamically selected to construct a reduced-order space, and the model is updated only in this subspace.

[0066] The specific implementation steps are: 1. Data-driven modality selection Input: Current sensor data (strain, displacement, etc.) ---- PCA results of historical data ---- Standardize real-time data to eliminate dimension effects ---- Calculate the projection coefficients of real-time data and historical principal components:

[0067] U: principal component matrix of historical PCA; : The mean of historical data.

[0068] The first k modes with the largest absolute values ​​of projection coefficients are selected (for example, k = 3, experiments show that the first three principal components can explain more than 90% of the data variance, taking into account both efficiency and accuracy) to form the reduced-order subspace.

[0069] 2. Reduced-order model construction In the reduced-order subspace, the original finite element model is simplified to:

[0070] : Reduced stiffness matrix (U k are the first k principal components); : Reduced-order load vector.

[0071] 3. Incremental Updates and Refactoring Directly solve the reduced-order equation to obtain the reduced-order displacement increment , reconstruct the global displacement field:

[0072] Update model: Superimpose it on the current displacement field and update the stress and strain synchronously.

[0073] 4. Dynamic mode adjustment. Adaptive mechanism: If the residual error of the reduced-order model exceeds a threshold (such as , determined through A / B testing, a value higher than this will result in a model error of >5%, triggering modal expansion), automatically expanding the number of modalities k (e.g. ); When resources are sufficient, the principal component matrix U is periodically recalculated. The principal component update cycle is 24 hours, which is based on the equipment operation cycle to avoid frequent calculation of principal components and consumption of resources.

[0074] Adopting an incremental learning algorithm, the part of the model affected by the change of physical entities is updated. Only the part of the model affected by the change of physical entities is updated, avoiding large-scale recalculation of the entire model, greatly improving the update efficiency.

[0075] The incremental update of the finite element model is based on the variational principle and the virtual work principle; assuming that the structural deformation , which satisfies the following variational equation:

[0076] in: is the domain of the structure, σ is the stress tensor, is the strain tensor, It's physical strength. is the surface force on the boundary, is the virtual displacement; by solving this equation, the deformation state of the structure is updated, and then the structural information in the digital twin model is updated.

[0077] The above method is further explained below using the digital twin application of a large crane as an example.

[0078] 1. Deployment of Distributed Intelligent Sensor Network For the digital twin application of large cranes, intelligent sensors are deployed in key parts such as the boom, tower, hoisting mechanism, slewing mechanism, and walking mechanism. For example, strain sensors are installed in different sections and key connection points of the boom, and displacement and vibration sensors are installed on the ropes and transmission components of the hoisting mechanism.

[0079] Each intelligent sensor is equipped with a microprocessor and a communication module. The microprocessor is responsible for preliminary processing of the collected data, and the communication module realizes self-organizing communication between sensors. The sensor determines the initial monitoring frequency through a built-in decision algorithm based on its location and monitoring data. For example, in the initial state, for the root of the boom that is subjected to greater stress, the monitoring frequency of the strain sensor is set to 10 times per second, while for the walking mechanism, the monitoring frequency of the displacement sensor is set to 5 times per second. Specifically, assuming that the adjustment coefficient of the strain sensor, the sensitivity index of the change of the monitoring data is calculated based on the initial strain data variance to be 0.2, then the adjusted monitoring frequency is:

[0080] 2. Data transmission and preprocessing The data collected by the sensors is transmitted to the data center through low-latency wireless communication protocols (such as 5G, LoRa, etc.). During the transmission process, the data is encrypted and integrity verified to ensure data security and accuracy.

[0081] In the data center, sliding window technology is used to clean the data in real time to remove noise and abnormal data caused by environmental interference or sensor abnormalities. For abnormal data, by comparing it with adjacent data and analyzing historical data, it is determined whether it is a real abnormality. For example, if the strain data at a certain moment differs from the data before and after by more than a certain threshold and does not match the trend of historical data, it is determined to be abnormal data and excluded. When using moving average filtering, if the window size N=5, the strain data of the nth data point is filtered to obtain:

[0082] 3. Adaptive model update algorithm The system automatically determines the update algorithm based on the type of data passed in. When receiving strain data of structural deformation, an update algorithm based on finite element analysis is used to update the structural mechanical properties in the digital twin model based on the strain data. For equipment wear data, an algorithm based on historical wear data and machine learning is used to predict wear trends and update the wear status of components in the model.

[0083] The system dynamically allocates computing resources based on the magnitude and rate of data change. For example, when the rate of change of strain data exceeds the set threshold, more computing resources are called to enable high-precision finite element model update calculations; in normal operation, a simplified model update algorithm is used. Assuming that the probability distribution of strain data is known, calculate its entropy

[0084] According to the coefficient , calculate the computing resource allocation ratio:

[0085] For model updating, an incremental learning algorithm is used. Assume that when a local deformation occurs in the crane boom, only the finite element mesh and structural mechanics parameters related to the local deformation are updated instead of recalculating the finite element model of the entire boom. According to the variational equation

[0086] The updated deformation information is substituted into the equation and solved by the finite element solver to obtain the updated structural deformation field and stress field, and then the corresponding parts in the digital twin model are updated.

[0087] 4. Storage and display of digital twin models The updated digital twin model is stored in a distributed storage system, using a distributed file system or distributed database to ensure data redundancy and scalability.

[0088] The digital twin model is displayed to users through a three-dimensional visualization engine. Users can view the status of the crane under different working conditions through the interactive interface, such as observing the deformation of the boom under wind load, and rotate and scale the model through mouse operations to view detailed information and parameters of different components.

[0089] Another specific embodiment of the present application provides a device digital twin model update system, including: A data acquisition unit, wherein the data acquisition unit includes a plurality of data acquisition devices, which are distributed and deployed at a plurality of locations of the equipment to collect data of the equipment; the data acquisition device includes a sensor module, a data processing module and a communication module, the sensor module includes one or more of a strain sensor, a displacement sensor and a vibration sensor, and the data processing module is used to pre-process the data collected by the sensor module; The data center is connected to the data acquisition device for receiving data from the data acquisition device and adopts a low-latency, high-bandwidth data transmission protocol to ensure that the data can reach the data center quickly and accurately.

[0090] The data center updates the digital twin model of the equipment according to the received data; when updating the model: an update algorithm is selected according to the data type, and for strain data of structural deformation, an update algorithm based on finite element analysis is adopted; for equipment wear data, an algorithm based on historical wear data and machine learning is used to predict the wear trend, and the wear status of components in the model is updated; according to the change trend of the data, the allocation of computing resources is dynamically adjusted; and an incremental learning algorithm is adopted to update the part of the model affected by the change of the physical entity; and a storage and display unit for storing and displaying the digital twin model.

[0091] Another specific embodiment of the present application provides a digital twin model terminal, including a memory, a processor and a display, the display is used to display the digital twin model, the memory stores a computer program, and when the computer program is called and executed by the processor, the device digital twin model update method as described above is implemented.

[0092] Another specific embodiment of the present application provides a computer-readable medium, characterized in that the computer-readable medium memory stores a computer program, and when the computer program is called and executed by a computer, it implements the device digital twin model updating method as described above.

[0093] The above is only a preferred implementation of the present application. It should be pointed out that a person skilled in the art can make several modifications and improvements without departing from the inventive concept of the present application, and these all fall within the scope of protection of the present application.

Claims

1. A method for updating a device digital twin model, characterized in that: The following steps are involved: S1: collects data from several locations of the device; For any data collection point on the equipment, the monitoring frequency is dynamically adjusted according to the sensitivity index of the monitoring data change; S2: Preprocess the collected data, including data cleaning, outlier detection and noise reduction; S3: Update the digital twin model of the device based on the processed data; when the model is updated: Select the update algorithm based on the data type. For the strain data of structural deformation, the update algorithm based on finite element analysis is adopted. For the equipment wear data, the algorithm based on historical wear data and machine learning is used to predict the wear trend and update the wear status of the components in the model. Dynamically adjust the allocation of computing resources according to the changing trend of data, and select the model update algorithm according to the allocation ratio of computing resources; An incremental learning algorithm is used to update the parts of the model that are affected by changes in physical entities.

2. The device digital twin model updating method according to claim 1, characterized in that: In step S1, the collected equipment data includes strain data, displacement data and vibration data of different parts of the equipment.

3. The device digital twin model updating method according to claim 2, characterized in that: In step S1, a distributed sensor network is used to collect data from several locations of the device, and the collection frequency of the sensor is obtained by the following formula: ； in, is the initial monitoring frequency of the sensor, k is the adjustment coefficient, and S is the sensitivity index of the change of the monitoring data; The change sensitivity index S of the monitoring data is obtained by any of the following methods: (1) For structural strain data monitoring: ； : The variance of the data in the current time window; Time window length T: set according to the response time of the physical entity; : Historical maximum variance, historical variance update cycle: daily or weekly update; (2) For displacement data and vibration data: ； : The time gradient of the data; the gradient calculation cycle : consistent with the sensor sampling frequency; : Maximum historical gradient; When the time gradient of the data exceeds the preset threshold for multiple times in a row, it is judged as abnormal and an abnormal warning is issued; (3) For complex working conditions of equipment: ； : Weight coefficient; through historical data regression analysis, select the model that minimizes the prediction error , value.

4. The device digital twin model updating method according to claim 2, characterized in that: In step S1, a distributed sensor network is used to collect data from several locations of the device. The sensors in the sensor network communicate with each other, and: (1) Sensors communicate with each other using a dynamic routing protocol based on ant colony optimization: The dynamic routing mechanism is designed by using the ant colony optimization algorithm. Each sensor node releases virtual pheromones to mark the quality of the communication path. When some nodes fail or data transmission is blocked, the neighboring nodes dynamically select alternative paths based on the pheromone concentration. The path selection probability formula is: ； in: : pheromone concentration of path itoj at time t; : Heuristic factor, is the path length; : pheromone weight coefficient; : Heuristic factor weight coefficient; The pheromone update rules are: ； : Pheromone volatilization rate; :The kth ant is on the path The increment of pheromone released, Q=1 is a constant, L k is the total path length of ant k; (2) Each sensor is embedded with a lightweight reinforcement learning model, which makes autonomous decisions on monitoring strategies through local observation. When a neighboring sensor fails, the reinforcement learning model triggers the "compensation mode", automatically increases its own monitoring frequency, and optimizes resource allocation through the Qlearning algorithm; Q value update formula: ； in: : learning rate; : discount factor; Reward function design: ； W1=0.7,W2=0.3: weight coefficient; (3) Adaptive energy-data trade-off mechanism: The sensor node evaluates its own energy level and data importance in real time, and dynamically adjusts the transmission mode. When energy is insufficient, it gives priority to transmitting key data, including abnormal signals, and reduces the sampling frequency of non-critical data. When data transmission is blocked, it switches to low-power broadcast mode and transmits data through multi-hop relays. (4) Fountain Code redundant coding technology is used to divide the data into blocks and store them in adjacent sensors. Fountain Code redundant coding: The data is divided into k original packets, generating n coded packets, n ≥ k; Each coded packet is the XOR of d randomly selected original packets; Decode Threshold: Receive The encoded packets were successfully decoded. is the failure probability; (5) Collaborative compressed sensing and data fusion: Neighboring sensors share raw data, use compressed sensing algorithms for joint sparse representation, and integrate multi-sensor data through Kalman filtering or Bayesian fusion algorithms; when a sensor fails, neighboring nodes reconstruct missing signals based on historical data and current observations; moving average filtering formula: ； Parameter selection: N=5: window size; Data entropy calculation: Entropy formula: ; Binning rules: adaptive binning; (6) Trust management based on blockchain: A lightweight blockchain protocol is introduced to record the status of sensor nodes and the data interaction history. When a node fails or the data is abnormal, neighboring nodes verify the data credibility through a consensus mechanism and isolate the faulty node.

5. The device digital twin model updating method according to claim 1, characterized in that: In step S2, the statistical analysis method is used to identify and remove abnormal data points. The 3σ principle is used. x satisfy , If is the mean and σ is the standard deviation, the data is considered to be an outlier and is excluded; Moving average filtering is used to reduce the noise of the data. The formula is: ； in is the filtered value of the nth data point, are the original data points and N is the size of the moving average window.

6. The device digital twin model updating method according to claim 1, characterized in that: In step S3, according to the change trend of the data, the allocation of computing resources is dynamically adjusted as follows: When a set of data is used to update the device digital twin model, the computing resource allocation ratio R of the set of data is: ； Where a and b are coefficients determined based on system performance and equipment characteristics; Entropy H can be calculated by Shannon entropy formula: ； is the probability distribution of the data.

7. The device digital twin model updating method according to claim 1, characterized in that: In step S3, an incremental learning algorithm is used to update the part of the model affected by the change of the physical entity; The incremental update of the finite element model is based on the variational principle and the virtual work principle; assuming that the structural deformation , which satisfies the following variational equation: ； in: is the domain of the structure, σ is the stress tensor, is the strain tensor, It's physical strength. is the surface force on the boundary, is the virtual displacement; by solving this equation, the deformation state of the structure is updated, and then the structural information in the digital twin model is updated.

8. A device digital twin model update system, characterized in that: include: A data collection unit, the data collection unit includes a plurality of data collection devices, the data collection devices are distributedly deployed at a plurality of locations of the device to collect data of the device; The data acquisition device includes a sensor module, a data processing module and a communication module. The sensor module includes one or more of a strain sensor, a displacement sensor and a vibration sensor. The data processing module is used to pre-process the data collected by the sensor module. A data center is connected to the data acquisition device for receiving data from the data acquisition device, and the data center updates the digital twin model of the equipment according to the received data; when updating the model: an update algorithm is selected according to the data type, and an update algorithm based on finite element analysis is used for strain data of structural deformation; for equipment wear data, an algorithm based on historical wear data and machine learning is used to predict the wear trend and update the wear status of components in the model; the allocation of computing resources is dynamically adjusted according to the change trend of the data; an incremental learning algorithm is used to update the part of the model affected by the change of the physical entity; and a storage and display unit for storing and displaying the digital twin model.

9. A digital twin model terminal, characterized in that: It includes a memory, a processor and a display, the display is used to display the digital twin model, the memory stores a computer program, and when the computer program is called and executed by the processor, the device digital twin model updating method as described in any one of claims 1 to 7 is implemented.

10. A computer-readable medium, characterized in that The computer-readable medium memory stores a computer program, and when the computer program is called and executed by a computer, it implements the device digital twin model updating method as described in any one of claims 1 to 7.

Images