Full-life-cycle digital twin management system and modeling method

Abstract

The invention relates to the technical field of digital twinborn management, and discloses a full-life-cycle digital twinborn management system and a modeling method. A model driving unit; a decision optimization unit; and a credible evidence storage unit. The modeling method comprises the following steps: S1, analyzing a product design bill of materials; s2, establishing a topological mapping relation between the sensor nodes and the virtual model area; s3, running Monte Carlo simulation to predict the defect probability; and S4, binding and uploading the new virtual model version and the physical product serial number. Aiming at the problem of data dispersion in the prior art, the system automatically converges three types of heterogeneous data of geometric dimension design, working condition manufacturing and maintenance monitoring into a shared cache pool through a main data table of a unified metadata structure, so that cross-stage semantic alignment and time sequence synchronization are realized, data format difference is eliminated, and data consistency is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of digital twin management, and specifically to a full life cycle digital twin management system and modeling method. Background Art

[0002] With the advancement of intelligent manufacturing, digital twin technology, as the core tool for the interaction between physical entities and virtual space, has been initially applied in product design, production and manufacturing, equipment operation and maintenance, etc. Its core is to build a dynamic virtual mapping of physical objects through IoT sensors, three-dimensional modeling and real-time data analysis to achieve status monitoring and decision optimization.

[0003] After searching, the Chinese patent number CN119624082A discloses a product management system and management method for the entire life cycle, involving the field of product management technology, including: dividing the product life cycle and clarifying the product data management content of each stage; using AI technology to perform preliminary analysis of the product data in the cache; recursively obtaining the product data of each child node based on the product's root node number and version number combined with the preliminary analysis report; this invention divides the product life cycle, clarifies the data management content of each stage, and introduces AI technology and digital twin technology to realize the automatic collection, anomaly detection and simulation modeling of product data.

[0004] In the above patent, although the life cycle is divided into design, trial production, procurement, maintenance and mass production stages, the data of each stage are still scattered in independent systems, such as PDM, ERP and MES systems. This dispersion leads to the lack of data standards, heterogeneous data formats of different systems, and lack of unified metadata specifications, which hinders cross-stage data integration. Design changes cannot be synchronized to manufacturing or maintenance links in real time and need to be manually transmitted, causing version dislocation. For example, the problem of using expired design drawings in the trial production stage occurs. Based on this, the present invention designs a full life cycle digital twin management system and modeling method to solve the above problems. Summary of the Invention

[0005] The purpose of the present invention is to provide a full life cycle digital twin management system and modeling method, which solves the problem of serious system dispersion in the background technology.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] A full life cycle digital twin management system and modeling method, including:

[0008] The data fusion unit is configured to: Based on a master data table with a unified metadata structure, integrate geometric dimension and material property data from the product design phase, equipment operating condition data from the manufacturing phase, and sensor monitoring data from the maintenance phase, thereby establishing a shared cache pool for multi-source heterogeneous data;

[0009] A model driving unit is in communication with the data fusion unit and is configured to: perform dynamic incremental updates on the parameterized virtual model according to the cross-stage data flow in the shared cache pool, where the update trigger condition is that the deviation of the real-time physical quantity exceeds a set threshold m;

[0010] a decision optimization unit, in communication with the model driving unit, configured to: generate, based on the updated virtual model, an optimization instruction including a process parameter adjustment range of ±5% and a work order instruction including a maintenance item list, and send the result to an execution terminal;

[0011] The trusted evidence storage unit is connected to each unit and is configured to bind and store the unique serial number of the physical product and the virtual model version number at three stages: when the design is completed, when the manufacturing process is changed, and when the maintenance event ends, to form an unalterable twin entity association chain.

[0012] Preferably, the data fusion unit includes:

[0013] Design data interface, used to extract design master data from product data management system;

[0014] Manufacturing data interface, which collects real-time current, pressure, and displacement data from the manufacturing execution system through industrial communication protocols;

[0015] Maintenance data interface, receiving data streams from vibration sensors and temperature sensors deployed on the equipment body.

[0016] Preferably, the model driving unit comprises:

[0017] Multi-physics coupling engine for simultaneous simulation of structural stress and temperature fields;

[0018] Deviation response module, when the real-time sensor measurement value y and the virtual model output value y0 meet:

[0019] , activates model status correction.

[0020] Preferably, the state correction adopts the Kalman filter algorithm to update the model parameters according to the state equation:

[0021] ;

[0022] x is the updated virtual model state vector, which contains key variables of position, stress, and temperature;

[0023] x - is the prior state estimate before updating;

[0024] K is the Kalman gain matrix, which is used to dynamically adjust the observation weight;

[0025] y is the physical quantity measurement value collected by the sensor in real time;

[0026] H is the measurement matrix, which is used to map the state vector to the sensor measurement space;

[0027] After each revision, a new version identifier with a timestamp is generated and the incremental change point is marked.

[0028] Preferably, the decision optimization unit includes:

[0029] The life prediction module trains a time series model based on the equipment's historical operating data and outputs the remaining service life (RUL) and a 90%-95% confidence interval.

[0030] The instruction decision module generates a load rate adjustment instruction of ±5% when the lower limit of the RUL confidence interval is lower than the preset safety threshold n.

[0031] Preferably, the optimization instruction includes:

[0032] Processing parameter instructions, spindle speed adjustment range;

[0033] Maintenance work order instructions, a list of preventive maintenance items and execution time windows.

[0034] Preferably, the trusted evidence storage unit performs a first binding operation at the end of the design phase, and stores the following data in an associated evidence:

[0035] Unique product serial number;

[0036] Initial virtual model version number;

[0037] Design review digital signature;

[0038] Binding operation timestamp;

[0039] Phase end flag code.

[0040] Preferably, the trusted evidence storage unit adopts a dual blockchain architecture:

[0041] Model version chain: stores full model snapshots and adjacent version difference data, with a storage interval of ≤24 hours;

[0042] Product event chain: stores manufacturing process parameter change records and maintenance work order execution result hash values.

[0043] According to a second aspect of the present invention, a modeling method for full lifecycle digital twin management is also proposed, the modeling method comprising the following steps:

[0044] Step S1: parse the product design bill of materials, extract ≥8 parameters including geometric dimensions, tolerances, and material properties to construct a master data table;

[0045] Step S2, establishing a topological mapping relationship between the sensor nodes and the virtual model area;

[0046] Step S3: Run Monte Carlo simulation to predict the defect probability P. When P exceeds the preset defect safety threshold λ, the parameter optimization algorithm is triggered to generate an adjustment instruction.

[0047] Step S4: Within 24 hours after the manufacturing process change event occurs, the new virtual model version is bound to the physical product serial number on the chain.

[0048] Preferably, the execution process of the parameter optimization algorithm in step S3 includes:

[0049] a) Constructing a response surface model of process parameters and defect probability:

[0050] ;

[0051] Where S is the spindle speed, F is the feed rate, and C is the cutting depth;

[0052] b) Solve the objective function based on the gradient descent method:

[0053] , where S is the spindle speed, F is the feed rate, C is the cutting depth, λ is the preset defect safety threshold, and α is the weighting coefficient;

[0054] Constraints: , where S min The minimum spindle speed, S max is the maximum spindle speed, ΔF is the feed rate change, C 安全 Safe cutting depth.

[0055] Compared with the prior art, the present invention has the following beneficial effects:

[0056] 1. In response to the data dispersion problem in the prior art, the present invention uses a master data table with a unified metadata structure to automatically merge three types of heterogeneous data, namely design geometry, manufacturing conditions, and maintenance monitoring, into a shared cache pool, achieving cross-stage semantic alignment and time synchronization, eliminating data format differences, and significantly improving data consistency.

[0057] 2. To address the problem of "design changes cannot be synchronized in real time and require manual transmission, causing version misalignment", the present invention uses a model-driven unit that uses real-time sensor deviation as a trigger condition, utilizes Kalman filtering to perform millisecond-level incremental updates on the virtual model, and automatically generates a version identifier with a timestamp. The design, manufacturing, and maintenance parties work collaboratively on the same model version, eliminating the risk of using outdated drawings during the trial production stage.

[0058] 3. In order to solve the problem that the lack of unified metadata specifications hinders cross-stage integration, the present invention adopts a dual blockchain architecture for trusted evidence units. The model version chain stores full model snapshots and differences, and the product event chain stores process changes and maintenance result hashes. The two are bound by the entity serial number and the virtual version number to form an unalterable twin entity association chain, thereby achieving traceability and auditability of data throughout the entire life cycle. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 This is the content diagram of the full life cycle digital twin management system of the present invention;

[0060] Figure 2 This is a data fusion unit diagram of the present invention;

[0061] Figure 3 is a flow chart of the modeling method of the present invention;

[0062] Figure 4 This is a flow chart of the trusted evidence chain mechanism of the present invention. DETAILED DESCRIPTION

[0063] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0064] Example 1;

[0065] See also Figures 1-4In an embodiment of the present invention, a full lifecycle digital twin management system includes: a data fusion unit, configured to integrate geometric dimension and material property data of the product design stage, equipment operating condition data of the manufacturing stage, and sensor monitoring data of the maintenance stage based on a master data table with a unified metadata structure, and establish a shared cache pool of multi-source heterogeneous data; a model driving unit, communicatively connected to the data fusion unit, configured to perform dynamic incremental updates on the parameterized virtual model according to the cross-stage data flow in the shared cache pool, where the update trigger condition is that the real-time physical quantity deviation exceeds a set threshold m; a decision optimization unit, communicatively connected to the model driving unit, configured to generate optimization instructions including a process parameter adjustment range of ±5% and a work order instruction including a maintenance item list based on the updated virtual model, and issue them to an execution terminal; a trusted evidence unit, communicatively connected to each unit, configured to bind and store the unique serial number of the physical product with the virtual model version number at three stages: when the design is completed, when the manufacturing process is changed, and when the maintenance event ends, to form an unalterable twin entity association chain.

[0066] The data fusion unit includes: a design data interface for extracting design master data from the product data management system; a manufacturing data interface for collecting real-time current, pressure, and displacement data from the manufacturing execution system through industrial communication protocols; and a maintenance data interface for receiving data streams from vibration sensors and temperature sensors deployed on the equipment itself.

[0067] The model driving unit includes: a multi-physics field coupling engine for synchronously executing structural stress field and temperature field simulation calculations; a deviation response module, when the real-time sensor measurement value y and the virtual model output value y0 meet:

[0068] , activates model status correction.

[0069] The state correction adopts the Kalman filter algorithm to update the model parameters according to the state equation:

[0070] ;

[0071] x is the updated virtual model state vector, which contains key variables of position, stress, and temperature;

[0072] x - is the prior state estimate before updating;

[0073] K is the Kalman gain matrix, which is used to dynamically adjust the observation weight;

[0074] y is the physical quantity measurement value collected by the sensor in real time;

[0075] H is the measurement matrix, which is used to map the state vector to the sensor measurement space;

[0076] After each revision, a new version identifier with a timestamp is generated and the incremental change point is marked.

[0077] The decision optimization unit includes: a life prediction module, which trains a time series model based on the historical operation data of the equipment and outputs the remaining service life RUL and a 90%-95% confidence interval; an instruction decision module, which generates a load rate adjustment instruction of ±5% when the lower limit of the RUL confidence interval is lower than the preset safety threshold n.

[0078] Optimization instructions include: processing parameter instructions, spindle speed adjustment range; maintenance work order instructions, preventive maintenance item list and execution time window.

[0079] The trusted evidence storage unit performs the first binding operation at the end of the design phase, and associates the following data for evidence storage: the product's unique serial number; the initial virtual model version number; the design review digital signature; the binding operation timestamp; and the phase end flag code.

[0080] The trusted evidence storage unit adopts a dual blockchain architecture: model version chain: stores full model snapshots and adjacent version difference data, with an evidence storage interval of ≤24 hours; product event chain: stores manufacturing process parameter change records and maintenance work order execution result hash values.

[0081] The working principle of the embodiment of the present invention is as follows: the operating mechanism is centered on the "data-model-decision-evidence" closed loop. When the system is started, the data fusion unit first performs semantic alignment and time synchronization on multi-source heterogeneous data such as three-dimensional geometric dimensions, tolerance chains, material stress-strain curves, and heat treatment parameters in the product design stage, CNC spindle current, hydraulic system pressure, servo motor displacement in the manufacturing stage, and accelerometers, infrared thermocouples, and acoustic emission sensors in the maintenance stage based on a master data table with a unified metadata structure, thereby building a shared cache pool with a version control mechanism.

[0082] After receiving the cross-stage data stream pushed by the data fusion unit, the model driver activates the multiphysics coupling engine and simultaneously solves the structural stress field and transient temperature field based on the finite volume method (FVM), forming a high-fidelity parameterized virtual model. This model incorporates a state estimator based on a Kalman filter. Its observation equation uses the Jacobian matrix H to map the state vector (node ​​displacement, von Mises stress, and temperature gradient) to the sensor measurement space. When the Euclidean distance between the real-time sensor measurement value y and the model output value y0 exceeds a set threshold m (default 0.05 times the range), an incremental update mechanism is triggered. The update process dynamically calculates the Kalman gain matrix K by solving the Riccati equation, achieving Bayesian optimal estimation of the model parameters.

[0083] Based on the updated virtual model, the decision-making optimization unit calls the Bi-LSTM time series model trained in the life prediction module, taking as input 12-dimensional features including vibration spectrum entropy, temperature cycle amplitude, and load power spectrum density. It then outputs the remaining useful life (RUL) and its 90%-95% confidence interval. When the lower limit of the RUL confidence interval falls below a preset safety threshold n (e.g., 100 hours), the instruction decision module uses a reinforcement learning-based policy gradient algorithm to search for the optimal action within the feasible domain of process parameters. It generates optimized instructions with adjustment ranges for spindle speed (±5%), feed rate (±3%), and depth of cut (±2%), as well as preventive maintenance work orders covering items such as bearing lubrication, tool replacement, and thermal balance calibration. These instructions are then sent to the CNC controller and CMMS system via the MQTT protocol for execution.

[0084] The trusted evidence storage unit utilizes a dual-blockchain architecture to achieve tamper-proof evidence storage for the twin entity association chain. The model version chain, built on Hyperledger Fabric, stores full model snapshots and differences between adjacent versions. It utilizes the IPFS distributed file system to store large-capacity geometric models, with an evidence storage interval of ≤24 hours. The product event chain, built on an Ethereum private chain, stores Keccak-256 hash values ​​and digital signatures of manufacturing process parameter changes and maintenance work order execution results. The initial binding operation is performed at the end of the design phase. A smart contract links the product's unique serial number (GTIN+serial number based on the GS1 standard), the initial virtual model version number, the design review digital signature, the binding operation timestamp (ISO8601 format, with millisecond accuracy), and the phase end marker (UUIDv4) for evidence storage. Subsequent on-chain updates are automatically triggered by event-driven mechanisms upon completion of manufacturing process changes and maintenance events.

[0085] Example 2;

[0086] See also Figures 1-4 In an embodiment of the present invention, the modeling method includes the following steps: step S1, parsing the product design material list, extracting geometric dimensions, tolerances, material properties ≥ 8 parameters to construct a master data table; step S2, establishing a topological mapping relationship between sensor nodes and virtual model areas; step S3, running Monte Carlo simulation to predict the defect probability P, when P exceeds the preset defect safety threshold λ, triggering the parameter optimization algorithm to generate an adjustment instruction; step S4, within 24 hours after the manufacturing process change event occurs, binding the new virtual model version to the physical product serial number on the chain.

[0087] The execution process of the parameter optimization algorithm in step S3 includes:

[0088] a) Constructing a response surface model of process parameters and defect probability:

[0089] ;

[0090] Where S is the spindle speed, F is the feed rate, and C is the cutting depth;

[0091] b) Solve the objective function based on the gradient descent method:

[0092] , where S is the spindle speed, F is the feed rate, C is the cutting depth, λ is the preset defect safety threshold, and α is the weighting coefficient;

[0093] Constraints: , where S min The minimum spindle speed, S max is the maximum spindle speed, ΔF is the feed rate change, C 安全 Safe cutting depth.

[0094] The working principle of this embodiment of the present invention is as follows: the operating mechanism is centered on a four-step closed loop of "data parsing - topology mapping - defect prediction - version binding." After the system is started, step S1 is first executed. The product design bill of materials (BOM) is parsed through the CAD / CAE software API interface, extracting no fewer than eight key parameters, including geometric dimensions (such as aperture, wall thickness, and profile), dimensional tolerances (such as ISO 2768-m grade), and material properties (such as elastic modulus, Poisson's ratio, thermal expansion coefficient, yield strength, hardness, density, thermal conductivity, and specific heat capacity), ensuring machine readability and scalability of the parameters.

[0095] In step S2, the system discretizes the virtual model into finite element units using a three-dimensional meshing algorithm (such as Delaunay tetrahedron meshing) and establishes a topological mapping relationship between sensor nodes and the model area based on the sensor deployment diagram. The mapping process uses a KD-Tree spatial index structure to accelerate queries, associating each sensor node (such as a triaxial accelerometer or infrared thermocouple) with the nearest finite element node. The mapping weight matrix is ​​recorded and used for spatial interpolation calculations during subsequent data assimilation.

[0096] Step S3 initiates the Monte Carlo simulation engine. Based on Latin hypercube sampling (LHS), a random sample set of process parameters (spindle speed S, feed rate F, and depth of cut C) is generated. The sample size is ≥10,000 groups, and the defect probability P (defined as the volume fraction of regions with a stress concentration factor greater than 3) is statistically analyzed. When P exceeds a preset defect safety threshold λ (e.g., 0.1%), a parameter optimization algorithm is triggered. This algorithm first constructs a process parameter-defect probability response surface model based on a radial basis function (RBF). The kernel function uses a multi-quadratic surface form, and the optimal shape parameters are determined through cross-validation. Constrained gradient descent is then used to solve the objective function, which is defined as a weighted combination of the defect probability P and the process time T (weight coefficients α = 0.7 and β = 0.3). Constraints include the upper limit of machine power, the lower limit of tool life, and the upper limit of surface roughness.

[0097] In step S4, within 24 hours of a manufacturing process change, the system automatically invokes a smart contract to bind the new virtual model version (including geometry updates, material parameter corrections, and boundary condition adjustments) to the physical product serial number and upload it to the blockchain. This is written into a new block via the Ethereum private blockchain's PoA consensus mechanism, ensuring the unalterable association between the model version and the physical state. This also triggers the incremental synchronization mechanism within the model version chain, completing the closed-loop update of the digital twin model throughout its lifecycle.

[0098] Example 3;

[0099] See also Figures 1-4 , provides a specific implementation example of full life cycle digital twin management for aircraft engine high-pressure compressor blades.

[0100] The data fusion unit creates a master data table with a unified metadata structure during system initialization. The fields include: blade section chord length 82.4 mm, maximum thickness 3.2 mm, leading edge radius 0.8 mm, material TC4 titanium alloy elastic modulus 114 GPa, Poisson's ratio 0.342, density 4.43 g cm -3 , yield strength 880 MPa. During the manufacturing phase, real-time data from the five-axis machining center was collected via the OPC UA protocol at a 250 Hz sampling rate: spindle current 45.7A, hydraulic pressure 7.3 MPa, and X-axis displacement -12.4 μm. During the maintenance phase, eight triaxial MEMS accelerometers placed on the blade surface transmitted vibration signals at a 1 kHz sampling rate, with a peak-to-peak error of 12.3 ms. -2 At the same time, the surface temperature was measured by infrared thermocouple at 412℃.

[0101] The model-driven unit uses a tetrahedral mesh with 1,874,652 elements, a cell size of 0.15 mm, and a coupled solver time step of 0.1 ms. The Kalman filter observation matrix H is 8 × 12 in dimension, mapping the state vector to the acceleration and temperature measurement points. The deviation threshold m is set to 5% of the range, meaning that when the difference between the sensor measurement value of any channel and the model output exceeds 0.615 ms, the error is detected. -2 The incremental update is triggered when the temperature reaches 20.6°C; the updated state vector x contains the blade tip displacement of 0.34 mm, the maximum von Mises stress of 615 MPa, and the leading edge temperature gradient of 38 °C mm -1 .

[0102] The decision optimization unit uses a Bi-LSTM model trained on a dataset consisting of 3.2 TB of operational data from the past 36 months. The model uses an input feature dimension of 12, a hidden layer of 256 units, a learning rate of 0.001, and a batch size of 512. The model outputs a RUL of 217 hours for the current blade, with a 90% confidence interval of 196 hours to 238 hours. The safety threshold n is set to 200 hours. Because the lower limit of 196 hours is below the threshold, the instruction decision module generates a process parameter adjustment instruction: the spindle speed is increased from 8000 rpm to 1000 rpm. -1 Reduce to 7600 rpm -1 , feed rate is 0.25 mm rev -1 Downgraded to 0.238 mm rev -1 The cutting depth was reduced from 0.8mm to 0.76mm. A maintenance work order was also generated: complete three tasks before the next shift: fluorescent penetrant testing of the blade surface, ultrasonic flaw detection of the blade root groove, and random inspection of the coating porosity.

[0103] The trusted evidence unit performs the first binding at the end of the design phase. The product's unique serial number is recorded as GTIN095060001234567890, serial number S / N AE2025-0718-0001, initial virtual model version number v1.0.0, and the design review digital signature uses the ECDSA P-256 algorithm with a signature value of 3046022100... The timestamp is 2025-07-18T09:17:42.123Z, and the phase end identifier UUID is 3ce3f8b0-... The manufacturing process change occurred at 2025-07-18T14:32:05Z. The change records include spindle speed adjustment of -5%, feed rate adjustment of -5%, and cutting depth adjustment of -5%. At the end of the maintenance event, it is bound again, and the execution results are recorded as no cracks detected, coating porosity 2.1%, RUL updated to 278 hours, version number upgraded to v1.1.3, binding timestamp 2025-07-18T22:45:10Z, completing the full life cycle closed loop.

[0104] Working principle: This patent is centered on the closed loop of "data fusion-model drive-decision optimization-trusted evidence storage" and runs through the entire life cycle of the product. First, the data fusion unit uses a unified metadata structure to merge the key data of the three stages of design, manufacturing, and maintenance into a shared cache pool; then, the model drive unit uses a multi-physics field coupling engine and a Kalman filter to dynamically update the model based on the deviation between the real-time sensor and the virtual model, and generates a new version with a timestamp. The decision optimization unit calculates the remaining life based on the updated model. When the lower limit of the life confidence interval is lower than the set threshold, it automatically generates a ±5% adjustment instruction for the process parameters and a preventive maintenance work order, and sends it to the execution terminal. The trusted evidence storage unit binds the physical product serial number and the virtual model version number to the dual blockchain at the three nodes of design completion, process change, and maintenance completion, forming an unalterable twin entity association chain. The entire system realizes real-time synchronization of data across stages, adaptive evolution of models, closed-loop optimization of decisions, and traceability of the entire process.

[0105] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and alterations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A full life cycle digital twin management system, characterized by: include: The data fusion unit is configured to: Based on a master data table with a unified metadata structure, integrate geometric dimension and material property data from the product design phase, equipment operating condition data from the manufacturing phase, and sensor monitoring data from the maintenance phase, thereby establishing a shared cache pool for multi-source heterogeneous data; A model driving unit is in communication with the data fusion unit and is configured to: perform dynamic incremental updates on the parameterized virtual model according to the cross-stage data flow in the shared cache pool, where the update trigger condition is that the deviation of the real-time physical quantity exceeds a set threshold m; a decision optimization unit, in communication with the model driving unit, configured to: generate, based on the updated virtual model, an optimization instruction including a process parameter adjustment range of ±5% and a work order instruction including a maintenance item list, and send the result to an execution terminal; The trusted evidence storage unit is connected to each unit and is configured to bind and store the unique serial number of the physical product and the virtual model version number at three stages: when the design is completed, when the manufacturing process is changed, and when the maintenance event ends, to form an unalterable twin entity association chain.

2. A full life cycle digital twin management system according to claim 1, characterized in that: The data fusion unit includes: Design data interface, used to extract design master data from product data management system; Manufacturing data interface, which collects real-time current, pressure, and displacement data from the manufacturing execution system through industrial communication protocols; Maintenance data interface, receiving data streams from vibration sensors and temperature sensors deployed on the equipment body.

3. A full life cycle digital twin management system according to claim 1, characterized in that: The model driving unit includes: Multi-physics coupling engine for simultaneous simulation of structural stress and temperature fields; Deviation response module, when the real-time sensor measurement value y and the virtual model output value y0 meet , activates model status correction.

4. A full life cycle digital twin management system according to claim 3, characterized in that: The state correction updates the model parameters according to the state equation: ； x is the updated virtual model state vector, which contains the key variables of position, stress and temperature; x - is the prior state estimate before updating; K is the Kalman gain matrix, which is used to dynamically adjust the observation weight; y is the physical quantity measurement value collected by the sensor in real time; H is the measurement matrix, which is used to map the state vector to the sensor measurement space; After each revision, a new version identifier with a timestamp is generated and the incremental change point is marked.

5. A full life cycle digital twin management system according to claim 1, characterized in that: The decision optimization unit includes: The life prediction module trains a time series model based on the equipment's historical operating data and outputs the remaining service life (RUL) and a 90%-95% confidence interval. The instruction decision module generates a load rate adjustment instruction of ±5% when the lower limit of the RUL confidence interval is lower than the preset safety threshold n.

6. A full life cycle digital twin management system according to claim 1, characterized in that: The optimization instructions include: Processing parameter instructions, spindle speed adjustment range; Maintenance work order instructions, a list of preventive maintenance items and execution time windows.

7. A full life cycle digital twin management system according to claim 1, characterized in that: The trusted evidence storage unit performs the first binding operation at the end of the design phase, associating the following data for evidence storage: Unique product serial number; Initial virtual model version number; Design review digital signature; Binding operation timestamp; Phase end flag code.

8. A full life cycle digital twin management system according to claim 1, characterized in that: The trusted evidence storage unit adopts a dual blockchain architecture: Model version chain: stores full model snapshots and adjacent version difference data, with a storage interval of ≤24 hours; Product event chain: stores manufacturing process parameter change records and maintenance work order execution result hash values.

9. A modeling method for full life cycle digital twin management, implemented using the full life cycle digital twin management system according to any one of claims 1 to 8, characterized in that: The modeling method comprises the following steps: Step S1: Parse the product design bill of materials, extract geometric dimensions, tolerances, and material property parameters to construct a master data table; Step S2, establishing a topological mapping relationship between the sensor nodes and the virtual model area; Step S3: Run Monte Carlo simulation to predict the defect probability P. When P exceeds the preset defect safety threshold λ, the parameter optimization algorithm is triggered to generate an adjustment instruction. Step S4: Within 24 hours after the manufacturing process change event occurs, the new virtual model version is bound to the physical product serial number on the chain.

10. The modeling method for full life cycle digital twin management according to claim 9, characterized in that: The execution process of the parameter optimization algorithm in step S3 includes: a) Constructing a response surface model of process parameters and defect probability: ； Where S is the spindle speed, F is the feed rate, and C is the cutting depth; b) Solve the objective function based on the gradient descent method: ； Constraints: .

Images