A wind power ring forging forming control system and method

Abstract

The application discloses a wind power ring forging forming control system and method, and belongs to the technical field of ring forging forming control. The system comprises a data synchronization module, a topology twin module, a trust evaluation module, a dynamic transmission module, a twin prediction module and a control execution module. Through physical event time correction and control cycle alignment of multi-source process data, a ring topology digital twin state is constructed, a variable level trustworthiness is generated, and a trust matrix is formed. According to the influence degree of industrial data flow on the forming state and the organization quality potential variable, 5G-TSN deterministic communication link resources are configured. In the unified forming state space, the trust matrix, network residence time, organization quality potential variable and ring space non-uniformity characteristics are fused to generate a next control cycle prediction forming state and determine the actual control amount. The application is suitable for the prediction control of the wind power ring forging forming process.

Description

Technical Field

[0001] This invention belongs to the field of ring forging forming control technology, and particularly relates to a wind power ring forging forming control system and method. Background Technology

[0002] Wind turbine ring forgings are crucial load-bearing components in wind power equipment. During the forming process, the workpiece temperature, geometry, deformation state, and material microstructure interact with each other, and changes in control parameters directly affect the dimensional accuracy, shape accuracy, and microstructure quality of the ring. Existing forming control methods typically rely on equipment detection data, sensor data, and process experience for parameter adjustments. However, technologies such as digital twins, industrial communication, and predictive control are increasingly being applied to large forging or ring forming processes.

[0003] The forming process of wind turbine ring forgings involves issues such as inconsistent sampling times from multiple data sources, feedback lag, transmission delays, and data quality fluctuations. Existing control methods struggle to simultaneously ensure the temporal consistency, data reliability, and communication status changes of process data. Furthermore, the temperature, deformation, and geometric dimensions of the ring forging process vary across different spatial locations, and existing control methods do not adequately consider these spatial distribution differences and their impact on subsequent forming processes.

[0004] The forming quality of wind turbine ring forgings includes not only geometric qualities such as diameter, wall thickness, and ellipticity, but also microstructural quality factors such as grain coarsening, overheating, deformation uniformity, and residual stress. Existing control methods do not adequately consider the correlation between microstructural quality risks and control parameters; when communication status changes, key data feedback lags, or data reliability declines, the judgment of forming status and the adjustment of control parameters are easily affected. Summary of the Invention

[0005] The purpose of this invention is to provide a wind turbine ring forging forming control system and method to solve the problems of insufficient accuracy in predicting forming state and insufficient matching between control quantity generation and actual working conditions when existing technologies have multiple sources of data timing differences, data reliability fluctuations, communication status changes, ring spatial state differences and organizational quality risks coexist.

[0006] To achieve the above objectives, the present invention employs the following technical solution:

[0007] This invention provides a wind turbine ring forging forming control system, comprising:

[0008] The data synchronization module is used to acquire multi-source process data during the production of wind turbine ring forgings. Based on the 5G-TSN deterministic communication link, it performs physical event timing correction and control cycle alignment to obtain synchronized process data.

[0009] The topology twin module is used to construct a ring topology digital twin state based on the ring spatial structure and synchronization process data of the wind turbine ring forging, and to generate ring spatial non-uniformity characteristics.

[0010] The credibility assessment module is used to generate variable-level credibility based on the deviation between the measured variables and the digital twin predictive variables, and to form a credibility matrix.

[0011] The dynamic transmission module is used to configure the data flow priority and bearer resources in the 5G-TSN deterministic communication link based on the degree of influence of industrial data flow on the forming state and latent variables of organizational quality in the next control cycle, as well as the variable-level reliability, and output the network dwell time.

[0012] The twin prediction module is used to integrate the current process stage, the ring topology digital twin state, control variables, reliability matrix, network dwell time, latent variables of organizational quality, and ring space non-uniformity characteristics within a unified forming state space to generate the predicted forming state for the next control cycle.

[0013] The control execution module is used to generate the control quantity for the next control cycle based on the target forming state, the predicted forming state for the next control cycle, the network dwell time, the variable-level credibility, and the organizational quality latent variables, and to determine the actual control quantity to be issued according to the security trigger conditions.

[0014] The data synchronization module, topology twin module, trust evaluation module, dynamic transmission module, twin prediction module, and control execution module are connected sequentially according to the data flow direction.

[0015] Preferably, the data synchronization module includes:

[0016] The data acquisition unit is used to acquire process status data, equipment execution data, and communication status data. The process status data includes ring temperature data, ring geometry data, forming force data, and motion status data. The equipment execution data includes heating power execution feedback, radial feed execution feedback, axial feed execution feedback, ring rotation speed execution feedback, and equipment operating status. The communication status data includes network dwell time, synchronization error, and communication bearer status in the 5G-TSN deterministic communication link.

[0017] The physical event timing correction unit is used to uniformly correct the sampling timing of process status data, the feedback timing of equipment execution data, and the recording timing of communication status data based on the time synchronization message, boundary clock correction information, transparent clock correction information, network dwell time, and synchronization error carried by the 5G-TSN deterministic communication link, so as to obtain the physical event timing corresponding to each data source.

[0018] The synchronization data frame generation unit is used to merge data whose physical event times fall within the same control cycle into synchronization process data frames according to a preset control cycle.

[0019] The synchronization quality identifier generation unit is used to generate synchronization quality identifiers based on the time correction deviation and data missing status of each data source.

[0020] The synchronization process data includes synchronization process data frames and synchronization quality identifiers.

[0021] Preferably, the topological twin module includes:

[0022] The ring node partitioning unit is used to divide the wind turbine ring forging into multiple circumferential sectors, multiple radial layers and multiple axial layers, and the discrete region jointly defined by the circumferential sector, radial layer and axial layer is used as the ring space node.

[0023] The node state generation unit is used to generate the node state of each ring space node based on the synchronization process data, and to normalize the node state to obtain the dimensionless space node state.

[0024] The topology matrix construction unit is used to construct a node-level ring topology matrix based on the circumferential closed adjacency relationship between adjacent circumferential sectors, the radial adjacency relationship between adjacent radial layers, and the axial adjacency relationship between adjacent axial layers. The node-level ring topology matrix is ​​then extended according to the characteristic dimensions of the spatial node states to obtain an extended ring topology matrix. The node-level ring topology matrix is ​​used to record the adjacency weights and difference calculation relationships between adjacent ring spatial nodes.

[0025] The non-uniformity feature generation unit is used to extract neighborhood differences of the dimensionless spatial node states based on the extended annular topology matrix, and generate annular spatial non-uniformity features.

[0026] The non-uniformity characteristics of the annular space include the differences in the distribution of annular temperature, geometry, and forming deformation in the circumferential, radial, and axial directions, and are used as inputs to the twin prediction module.

[0027] Preferably, the reliability assessment module includes:

[0028] The variable deviation calculation unit is used to calculate the variable-level deviation between the measured values ​​and the digital twin predicted values ​​of the monitoring variables in the synchronization process data.

[0029] The deviation normalization unit is used to normalize the variable-level deviation based on the deviation scale parameter of the corresponding monitoring variable under healthy operating conditions, so as to obtain the normalized variable-level deviation.

[0030] The variable confidence generation unit is used to perform weighted accumulation of normalized variable-level deviations within a preset scrolling window and generate variable-level confidence through a monotonically decreasing mapping; the variable-level confidence decreases as the accumulation of normalized variable-level deviations increases.

[0031] The credibility matrix generation unit is used to assemble the variable-level credibility corresponding to the equipment execution data into a control quantity credibility matrix, and to assemble the variable-level credibility corresponding to the process status data into a measurement output credibility matrix, and to assemble the variable-level credibility corresponding to the communication status data into a network data credibility matrix.

[0032] The comprehensive credibility generation unit is used to perform weighted aggregation of variable-level credibility to obtain a comprehensive credibility, with the sum of the aggregation weights being 1.

[0033] Preferably, the dynamic transmission module includes:

[0034] Sensitivity calculation unit, used for the first Industrial data streams, for the first The network dwell time of industrial data streams is subjected to a unit normalized perturbation, and the impact of the unit normalized perturbation on the latent variables of organizational quality and forming state in the next control cycle is calculated based on the twin prediction module, so as to obtain the organizational quality sensitivity and forming state sensitivity.

[0035] Priority scoring unit, used in the first The calculation of the first control cycle Process sensitivity priority scoring for industrial data streams:

[0036] ;

[0037] In the formula, Indicates the first Industrial data streams in the first Process sensitivity priority score for each control cycle Indicates the first The non-negative dimensionless sensitivity of network dwell perturbations in industrial-like data streams to predict latent variables of organizational quality. Indicates the first The non-negative dimensionless sensitivity of network dwell perturbations in industrial-like data streams to the prediction of forming states. Indicates the first The overall reliability of each control cycle Indicates the first Does the industrial data flow belong to the indication of the critical flow in closed-loop control? What is the corresponding critical flow in closed-loop control? Non-closed-loop control critical flow corresponding , , , and This represents the non-negative dimensionless scoring weight;

[0038] The priority determination unit is used to prioritize data based on the process sensitivity priority score and the score distribution of each industrial data stream within the current control cycle. Industrial data streams are mapped to integer TSN priority code points in the range of 0 to 7;

[0039] The bearer matching unit is used to match the corresponding bearer resources from the pre-configured differential service code points, 5G service flow identifiers and data radio bearer resource sets based on the integer TSN priority code points.

[0040] The dwell time output unit is used to acquire the network dwell time of various industrial data streams in the 5G-TSN deterministic communication link, and outputs it to the twin prediction module after forming a network dwell time vector according to the preset data stream order.

[0041] Preferably, the twin prediction module includes:

[0042] The process stage identification unit is used to determine the process stage based on synchronous process data. The process stage corresponding to each predictive recursive control cycle;

[0043] The reliable input assembly unit is used to acquire dimensionless forming state, dimensionless control quantity, control quantity reliability matrix, network data reliability matrix, measurement output reliability matrix, normalized network residence time vector, dimensionless tissue quality latent variable and annular space non-uniformity characteristics.

[0044] Prediction model building unit, used to construct a network trusted collaborative prediction model, which satisfies the following:

[0045] ;

[0046] In the formula, Indicates the first Dimensionless forming state of a predictive recursive control cycle Indicates the first The predicted forming state of each prediction recursion cycle, Indicates the first The current process stage corresponding to each predictive recursive cycle. This represents the state evolution matrix corresponding to the current process stage. This represents the control input matrix corresponding to the current process stage. Represents the confidence matrix of the control quantity. This represents a dimensionless control quantity. This represents the network dwell time compensation matrix corresponding to the current process stage. Represents the network data credibility matrix. This represents the normalized network dwell time vector. This represents the measurement reliability perturbation compensation matrix. This represents the measurement output confidence matrix. Represents a vector consisting entirely of 1s. This represents the organizational quality impact matrix corresponding to the current process stage. This represents the dimensionless latent variable of organizational quality. This represents the annular topology compensation matrix corresponding to the current process stage. This indicates the non-uniformity of the annular space;

[0047] The microstructure quality prediction unit is used to predict the forming state, discrete unit temperature, equivalent plastic strain increment, cooling rate and historical microstructure quality latent variables based on the next control cycle, and update the microstructure quality latent variables for the next control cycle.

[0048] The prediction state output unit is used to predict the forming state and the latent variables of the microstructure quality in the next control cycle based on the next control cycle, and outputs the predicted temperature, predicted ring diameter, predicted ring wall thickness, predicted ellipticity and predicted microstructure quality risk.

[0049] The model calibration unit is used to recursively update the control input matrix in the network trusted collaborative prediction model based on the measured forming state, the predicted forming state, and the trusted weighted control input data, and to recursively update the compensation matrix in the network trusted collaborative prediction model when there is a compensation error.

[0050] Dimensionless forming state, dimensionless control quantity, normalized network residence time vector, dimensionless tissue quality latent variable and annular space non-uniformity feature are all mapped to the same forming state space, and each input item has the same dimension as the forming state predicted in the next control cycle.

[0051] Preferably, the organization quality prediction unit is used to update the organization quality latent variables for the next control period based on the organization quality latent variables;

[0052] The microstructure quality latent variables are generated from multiple discrete unit states of the wind turbine ring forging. The microstructure quality latent variables include grain coarsening risk index, overheating risk index, deformation uniformity index, and residual stress risk index.

[0053] The grain coarsening risk index is determined based on the normalized exceedance degree of the discrete unit temperature relative to the grain coarsening risk temperature threshold, the duration of high temperature, and the equivalent plastic strain increment.

[0054] The overheating risk index is determined based on the normalized degree of exceeding the limit of the discrete unit temperature relative to the overheating risk temperature threshold and the duration of exceeding the limit.

[0055] The deformation uniformity index is determined based on the deviation of the equivalent plastic strain increment of each discrete element from the weighted average equivalent plastic strain increment.

[0056] The residual stress risk index is determined based on the annular topological temperature gradient, cooling rate, and equivalent plastic strain increment deviation.

[0057] The grain coarsening risk index, overheating risk index, deformation uniformity index, and residual stress risk index are all dimensionless risk indices, and are obtained by weighting the weights of each discrete unit, with the sum of the weights of each discrete unit being 1.

[0058] Preferably, the control execution module includes:

[0059] The target building unit is used to construct a rolling optimization problem based on a network-trusted collaborative prediction model. The rolling optimization problem takes the deviation of the predicted forming state from the target forming state, the deviation of the predicted tissue quality latent variable from the target tissue quality latent variable, the non-uniformity characteristics of the annular space, and the cost of changes in the control quantity as the objective terms.

[0060] The constraint generation unit is used to generate constraints for the rolling optimization problem. These constraints include: the predicted forming state satisfies the network's reliable collaborative prediction model; the control quantity is within preset upper and lower control limits; the rate of change of the control quantity is within a preset rate of change range; the predicted temperature obtained by inverse normalization of the predicted forming state is within the allowable temperature range; the predicted ring diameter and predicted ring wall thickness are within the allowable size range; the predicted ellipticity does not exceed the allowable ellipticity upper limit; each risk indicator in the predicted tissue quality latent variables does not exceed the corresponding allowable tissue quality risk upper limit; both the predicted network dwell time and the measured network dwell time do not exceed the allowable network dwell time upper limit; and a safety constraint is triggered when either the predicted or measured network dwell time exceeds the allowable network dwell time upper limit.

[0061] The rolling solver unit is used to solve the control sequence for multiple future control cycles based on the objective term and constraints.

[0062] The control output unit is used to output the current control quantity in the control sequence in each control cycle, and to re-trigger the solution of the rolling optimization problem in the next control cycle based on the updated synchronization process data.

[0063] Preferably, the control execution module further includes a safety triggering unit, a safety control quantity generation unit, and a control quantity fusion unit;

[0064] The security trigger unit is used to determine whether a security trigger condition is met when one of the following security trigger conditions is satisfied: the minimum value of the variable-level credibility is lower than the preset credibility threshold; the network dwell time is higher than the upper limit of the allowed network dwell time; or any risk indicator of the organizational quality potential variables is higher than the upper limit of the allowed network dwell time.

[0065] The safety control quantity generation unit is used to solve for the safety envelope control quantity when the safety triggering conditions are met, with the goals of reducing organizational quality risks, minimizing control quantity mutations, and maintaining the temperature of the components within a safe insulation range.

[0066] The control quantity fusion unit is used to determine the security fusion coefficient based on the comprehensive credibility, the degree of network dwell time exceeding the limit, and the degree of organizational quality risk exceeding the limit, and to fuse the current control quantity and the security envelope control quantity obtained by rolling optimization according to the security fusion coefficient to generate the actual issued control quantity;

[0067] When the network dwell time and organizational quality risk exceedances remain constant, the security integration coefficient increases as the overall credibility decreases; when the overall credibility and organizational quality risk exceedances remain constant, the security integration coefficient increases as the network dwell time exceeds the limit; when the overall credibility and network dwell time exceedances remain constant, the security integration coefficient increases as the organizational quality risk exceeds the limit.

[0068] The current control quantity and the safety envelope control quantity are represented in the same order of control variables, which include multi-zone heating power, radial feed control quantity, axial feed control quantity, and ring speed control quantity.

[0069] This invention also provides a method for controlling the forming of wind turbine ring forgings, comprising the following steps:

[0070] S1. Acquire multi-source process data during the production of wind turbine ring forgings, and perform physical event timing correction and control cycle alignment on the multi-source process data based on the 5G-TSN deterministic communication link to obtain synchronous process data.

[0071] S2. Construct a ring topology digital twin state based on the ring space structure and synchronous process data of the wind turbine ring forging, and generate ring space non-uniformity characteristics.

[0072] S3. Generate variable-level confidence levels based on the deviation between the measured variables and the digital twin predicted variables, and form a confidence matrix;

[0073] S4. Based on the degree of influence of industrial data flow on the forming state and organizational quality latent variables of the next control cycle, as well as the variable-level credibility, configure the data flow priority and bearer resources in the 5G-TSN deterministic communication link, and obtain the network dwell time.

[0074] S5. In a unified forming state space, integrate the current process stage, the annular topology digital twin state, control variables, reliability matrix, network dwell time, latent variables of organizational quality, and annular space non-uniformity characteristics to generate the predicted forming state for the next control cycle.

[0075] S6. Based on the target forming state, the predicted forming state for the next control cycle, the network dwell time, the variable-level confidence level, and the latent variables of organizational quality, generate the control quantity for the next control cycle.

[0076] S7. Determine whether the safety triggering condition is met; if the safety triggering condition is met, determine the actual control quantity to be issued based on the safety envelope control quantity; if the safety triggering condition is not met, use the control quantity of the next control cycle as the actual control quantity to be issued.

[0077] The beneficial effects of this invention are as follows: In the forming control process of wind turbine ring forgings, this invention coordinates the timing correction of multi-source process data, annular space state characterization, data reliability assessment, communication resource configuration, twin prediction, and control execution. By correcting the timing of physical events and aligning the control cycle of multi-source process data, the impact of timing deviations generated during sampling, feedback, and transmission from different data sources on the judgment of forming state is reduced. By constructing an annular topology digital twin state with the annular space structure of the wind turbine ring forging and generating annular space non-uniformity features, the forming state prediction can reflect the state differences at different spatial locations of the ring. Variable-level reliability is generated by the deviation between measured variables and digital twin predicted variables, and the reliability matrix, network dwell time, latent variables of organizational quality, and annular space non-uniformity features are used together in the next step. The generation of the forming state prediction in the control cycle enables the prediction process to simultaneously consider the data reliability state, communication state, spatial non-uniformity state, and microstructure quality state. By configuring data stream priority and carrying resources according to the degree of influence of industrial data stream on the forming state and microstructure quality latent variables, the communication resource configuration is adapted to the key data requirements in forming control. By comprehensively considering the target forming state, the predicted forming state of the next control cycle, network dwell time, variable-level reliability, and microstructure quality latent variables, the control quantity of the next control cycle is generated, and the actual control quantity is determined according to the safety triggering conditions. This ensures that the generated control quantity matches the current data state, communication state, and microstructure quality state, thereby improving the accuracy of the forming state prediction of wind turbine ring forgings and the rationality of the generated control quantity, improving the stability of the forming process, and facilitating the consideration of both the geometric quality and microstructure quality of the ring. Attached Figure Description

[0078] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a schematic diagram of the modular structure of the system of the present invention; Figure 2 This is a flowchart of the method of the present invention. Detailed Implementation

[0079] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention are within the scope of protection of the present invention.

[0080] In this embodiment, the wind turbine ring forging forming control system based on digital twins includes a data synchronization module, a topology twin module, a reliable evaluation module, a dynamic transmission module, a twin prediction module, and a control execution module. The output of the data synchronization module is connected to the topology twin module, the reliable evaluation module, and the twin prediction module, respectively; the output of the topology twin module is connected to the twin prediction module; the output of the reliable evaluation module is connected to the dynamic transmission module, the twin prediction module, and the control execution module, respectively; the output of the dynamic transmission module is connected to the twin prediction module; the output of the twin prediction module is connected to the control execution module; and the output of the control execution module is used to output the actual issued control quantity.

[0081] like Figure 1 As shown, this is an embodiment of the present invention, which provides a wind turbine ring forging forming control system, including:

[0082] (1) Data synchronization module

[0083] In this embodiment, the data synchronization module is located in the edge control equipment or industrial server of the wind turbine ring forging production line and is connected to the field sensors, equipment controllers, and 5G-TSN communication gateway. The data synchronization module is used to acquire multi-source process data during the wind turbine ring forging process and unify the data from different data sources to the same time base and the same control cycle, forming synchronized process data for subsequent modules to call.

[0084] The data acquisition unit is used to acquire process status data, equipment execution data, and communication status data. Process status data reflects the actual state of the wind turbine ring forging during the forming process, including ring temperature data, ring geometry data, forming force data, and motion status data. Ring temperature data is obtained from infrared thermography, thermal imaging, or temperature sensors; ring geometry data includes ring diameter, wall thickness, and ellipticity data; forming force data is obtained from pressure sensors, rolling force sensors, or feedback from the hydraulic system; motion status data includes ring rotation speed, radial feed displacement, and axial feed displacement. Equipment execution data reflects the actual execution of control commands on the equipment side, including heating power execution feedback, radial feed execution feedback, axial feed execution feedback, ring rotation speed execution feedback, and equipment operating status. Communication status data reflects the data transmission status of the 5G-TSN deterministic communication link, including network dwell time, synchronization error, and communication bearer status. When the communication bearer status is a categorical state quantity, it is converted into a numerical state quantity according to a preset coding rule, facilitating variable deviation calculation by the subsequent reliability assessment module.

[0085] Because process status data, equipment execution data, and communication status data come from different acquisition nodes, the data sampling time, feedback time, recording time, and arrival time are not entirely consistent. The physical event timing correction unit is used to convert various types of data from their original timestamps to physical event timestamps under a unified time base. The physical event timing refers to the actual time when the corresponding production event occurs, not the time when the data arrives at the edge control device or industrial server. Specifically, the physical event timing correction unit corrects the sampling time of process status data, the feedback time of equipment execution data, and the recording time of communication status data based on the time synchronization messages carried by the 5G-TSN deterministic communication link, boundary clock correction information, transparent clock correction information, network dwell time, and synchronization error. This allows data from different data sources to be compared and merged under a unified time base.

[0086] Within a control cycle, the synchronization data frame generation unit merges data corrected for physical event moments according to a preset control cycle. The preset control cycle is set based on the control frequency and digital twin prediction cycle of the wind turbine ring forging forming equipment. The synchronization data frame generation unit groups process status data, equipment execution data, and communication status data whose physical event moments fall within the same control cycle into the same synchronization process data frame. For cases where the same variable is sampled multiple times within the same control cycle, the synchronization data frame generation unit selects a representative value according to preset rules. These preset rules include taking the most recent value at the end of the control cycle, taking the average value of the control cycle, and taking the maximum value of the control cycle. Preset rules for different variables are pre-configured according to the variable type. For example, temperature data uses the average or maximum value of the control cycle, geometric data uses the most recent value at the end of the control cycle, forming force data uses the average or peak value of the control cycle, equipment execution data uses the execution feedback value within the control cycle, and communication status data uses the network dwell time record value within the corresponding control cycle.

[0087] The synchronization quality identifier generation unit generates synchronization quality identifiers based on the time correction deviation and data missing status of each data source. The time correction deviation characterizes the time difference before and after data correction, as well as the residual synchronization error; the data missing status characterizes whether valid data exists in the corresponding data source within the control cycle. The synchronization quality identifier is output along with the synchronization process data frame. For missing data within a control cycle, the synchronization data frame generation unit retains the acquired data and records the missing status in the synchronization quality identifier; for data requiring temporary supplementation, the synchronization quality identifier records the supplementation status, enabling the subsequent reliability assessment module to reduce the reliability of the corresponding variable.

[0088] Therefore, the synchronization process data output by the data synchronization module includes synchronization process data frames and synchronization quality identifiers. The synchronization process data frames provide process status data, equipment execution data, and communication status data within the same control cycle after physical event timing correction and control cycle alignment; the synchronization quality identifier provides information on the temporal consistency and integrity of the above data. The topology twin module constructs a ring topology digital twin state based on the synchronization process data, the reliability assessment module generates variable-level reliability based on the synchronization process data and synchronization quality identifiers, the dynamic transmission module configures data stream priorities based on communication status data and variable-level reliability, and the twin prediction module generates a predicted forming state for the next control cycle based on the synchronization process data.

[0089] The data synchronization module in this embodiment does not involve artificial intelligence model training. Input data, time correction criteria, control cycle merging rules, and output data are all determined by production site data and preset control parameters. Those skilled in the art, based on the aforementioned data sources, correction methods, and merging methods, can implement the data synchronization module in the edge control equipment, industrial server, or industrial control system of the wind turbine ring forging production line.

[0090] (2) Topological twin module

[0091] In this embodiment, the topology twin module receives synchronization process data output by the data synchronization module and constructs a ring topology digital twin state based on the ring spatial structure of the wind turbine ring forging. The topology twin module includes a ring node partitioning unit, a node state generation unit, a topology matrix construction unit, and a non-uniformity feature generation unit. The ring topology digital twin state and ring spatial non-uniformity features output by the topology twin module are used by the subsequent twin prediction module.

[0092] The annular node partitioning unit is used to discretize the wind turbine ring forging in space. Specifically, the ring is divided into multiple circumferential sectors, multiple radial layers, and multiple axial layers along its circumference. A region defined by one circumferential sector, one radial layer, and one axial layer is considered a single annular spatial node. The number of circumferential sectors, radial layers, and axial layers are pre-set based on the ring size, detection point arrangement, control cycle, and computational resources. Through this partitioning, the spatial state of the wind turbine ring forging can be expressed using multiple annular spatial nodes.

[0093] The node state generation unit generates the node state of each annular spatial node based on the synchronization process data. For data with a clear spatial location, the node state generation unit maps the data to the corresponding annular spatial node according to the correspondence between the acquisition location and the annular spatial node. For overall state data such as forming force, ring rotation speed, and equipment execution feedback, the node state generation unit writes the overall state as a global state into the annular topology digital twin state, or maps it to the relevant annular spatial nodes according to a preset allocation relationship. The node state of each annular spatial node includes node temperature, node geometry, and node forming deformation. For annular spatial nodes lacking direct measurement data, the node state generation unit generates node state estimates based on adjacent node data, historical cycle data, or process model calculation results, and sets data quality tags for the node states generated by the estimation method. The subsequent reliability assessment module adjusts the corresponding variable-level reliability based on the data quality tags. The node state generation unit normalizes the node states to obtain dimensionless spatial node states, enabling data with different dimensions to participate in the same subsequent calculation process.

[0094] The topology matrix construction unit is used to construct a node-level ring topology matrix based on the adjacency relationships between ring-shaped spatial nodes. The node-level ring topology matrix is ​​established based on the circumferential closed adjacency relationships between adjacent circumferential sectors, the radial adjacency relationships between adjacent radial layers, and the axial adjacency relationships between adjacent axial layers. It is used to record the adjacency weights and difference calculation relationships between adjacent ring-shaped spatial nodes. The first and last sectors in the circumferential direction are established as adjacencies according to the ring closure relationship. The adjacency weights are pre-set based on spatial distance, node volume, or the degree of process influence. The topology matrix construction unit further extends the node-level ring topology matrix according to the characteristic dimensions of the spatial node states, obtaining an extended ring topology matrix, enabling the topological relationships to act on multiple state characteristics such as node temperature, node geometry, and node forming deformation.

[0095] The non-uniformity feature generation unit is used to extract neighborhood differences from the dimensionless spatial node states based on the extended annular topology matrix, generating annular spatial non-uniformity features. Neighborhood difference extraction is used to calculate the state differences between adjacent annular spatial nodes. The annular spatial non-uniformity features include the distribution differences of ring component temperature, geometric dimensions, and forming deformation in the circumferential, radial, and axial directions. Since the spatial node states involved in the calculation have been normalized, the generated annular spatial non-uniformity features are dimensionless and can be used as input to the twin prediction module.

[0096] The annular topology digital twin state includes dimensionless spatial node states, node-level annular topology matrices, extended annular topology matrices, and global states related to the overall state of the wind turbine ring forging. Global states include the current process stage, average ring temperature, average wall thickness, average diameter, ellipticity, total forming force, ring rotation speed, and equipment execution status. Annular spatial non-uniformity features, extracted from the annular topology digital twin state, are input into the twin prediction module to participate in the generation of the forming state in the next control cycle.

[0097] The topological twin module in this embodiment does not involve training an artificial intelligence model. The number of ring nodes, node state characteristics, normalization scale, adjacency weight, and node arrangement order are all configured as preset parameters. Those skilled in the art can realize the above-mentioned ring topological digital twin state construction and ring spatial non-uniformity feature generation process based on the specifications, dimensions, detection points, and control cycle of the wind turbine ring forging.

[0098] (3) Credibility Assessment Module

[0099] In this embodiment, the reliability assessment module receives the synchronization process data output by the data synchronization module and obtains the digital twin predictive variables, and performs a reliability assessment on the monitoring variables in the synchronization process data. The inputs of the reliability assessment module include measured variables, digital twin predictive variables, deviation scale parameters of the corresponding monitoring variables under healthy operating conditions, and synchronization quality identifiers; the outputs include variable-level reliability, control quantity reliability matrix, measurement output reliability matrix, network data reliability matrix, and overall reliability.

[0100] The variable deviation calculation unit is used to calculate variable-level deviations for monitored variables in the synchronization process data. Monitored variables include ring temperature, ring geometry, forming force, and motion state variables in the process status data; heating power execution feedback, radial feed execution feedback, axial feed execution feedback, and ring rotation speed execution feedback in the equipment execution data; and network dwell time, synchronization error, and communication bearer status variables in the communication status data. The variable deviation calculation unit obtains the measured values ​​and corresponding digital twin predicted values ​​of the monitored variables within the current control cycle, and uses the difference or absolute difference between the measured values ​​and the digital twin predicted values ​​as the variable-level deviation. The digital twin predicted values ​​are generated by the twin prediction module based on the previous control cycle or historical process data. For discrete monitored variables such as communication bearer status, the variable-level deviation is obtained based on preset state code differences or state consistency judgment results.

[0101] The deviation normalization unit normalizes variable-level deviations based on the deviation scale parameters of the corresponding monitored variables under healthy operating conditions, resulting in normalized variable-level deviations. The deviation scale parameters are determined based on normal operating data, historical data of qualified products, process tolerance ranges, or trial production data. Through normalization, deviations of variables with different dimensions, such as temperature, size, forming force, execution feedback, and network dwell time, are converted into dimensionless deviations, avoiding direct comparisons of data with different dimensions. For monitored variables where synchronous quality indicators show missing data, data timeouts, or incomplete data, the deviation normalization unit applies a preset penalty to the corresponding normalized variable-level deviation, ensuring that data integrity anomalies are reflected in subsequent reliability assessments.

[0102] The variable confidence level generation unit is used to perform weighted accumulation of normalized variable-level deviations within a preset rolling window and generate variable-level confidence levels through a monotonically decreasing mapping. The preset rolling window includes the current control cycle and several control cycles preceding it. The window length is preset based on the forming equipment control cycle, data sampling frequency, and process response speed. The sum of the weights corresponding to each cycle within the window is 1, with data closer to the current control cycle having a higher weight than data from earlier control cycles. The larger the weighted accumulation value of the normalized variable-level deviation, the lower the variable-level confidence level. The variable-level confidence level ranges from 0 to 1; a larger value indicates a higher level of confidence in the monitored variable. The monotonically decreasing mapping is implemented using an exponential decay function, a piecewise linear function, or a preset lookup table function. The specific mapping relationship is pre-configured according to the equipment type and process requirements.

[0103] The credibility matrix generation unit generates credibility matrices based on variable-level credibility. Variable-level credibility corresponding to equipment execution data is assembled into a control quantity credibility matrix according to the order of control variables, including heating power, radial feed, axial feed, and ring rotation speed. Variable-level credibility corresponding to process status data is assembled into a measurement output credibility matrix according to the order of measurement output variables, including ring temperature, ring geometry, forming force, and motion state. Variable-level credibility corresponding to communication status data is assembled into a network data credibility matrix according to the order of communication variables, including network dwell time, synchronization error, and communication bearer status. The control quantity credibility matrix, measurement output credibility matrix, and network data credibility matrix are in diagonal matrix form, with diagonal elements representing the variable-level credibility of the corresponding variables.

[0104] The comprehensive credibility generation unit is used to perform weighted aggregation of variable-level credibility to obtain a comprehensive credibility score. The aggregation weights are pre-set according to the importance of each monitored variable to the prediction of the forming state and the generation of control quantities, and the sum of the aggregation weights is 1. The comprehensive credibility score is used to represent the overall credibility status of the synchronization process data within the current control cycle and is output to the dynamic transmission module, the twin prediction module, and the control execution module.

[0105] The credibility assessment module in this embodiment does not involve training an artificial intelligence model. The bias scale parameter, rolling window length, window weight, monotonically decreasing mapping relationship, penalty amount, and aggregation weight are all preset parameters. Those skilled in the art, based on synchronization process data, digital twin predictor variables, and the aforementioned preset parameters, can realize variable-level bias calculation, bias normalization, variable-level credibility generation, credibility matrix assembly, and comprehensive credibility generation processes.

[0106] (4) Dynamic transmission module

[0107] In this embodiment, the dynamic transmission module receives the variable-level credibility and overall credibility output by the credibility assessment module, and calls the prediction results of the twin prediction module to evaluate the process sensitivity of the industrial data stream during the wind turbine ring forging process. Based on the process sensitivity evaluation results, the dynamic transmission module configures the data stream priority and bearer resources in the 5G-TSN deterministic communication link, and outputs the network dwell time to the twin prediction module. The industrial data stream includes process status data stream, equipment execution data stream, and communication status data stream, specifically including temperature data stream, geometric dimension data stream, forming force data stream, motion status data stream, equipment execution feedback data stream, and communication status data stream.

[0108] The sensitivity calculation unit is used to calculate the first The degree of influence of industrial data streams on the latent variables of organizational quality and the forming state in the next control cycle is assessed. In practice, the sensitivity calculation unit first uses the network dwell time of each current industrial data stream as a baseline input, and calls the twin prediction module to obtain the baseline prediction result; subsequently, it calculates the... A unit normalized perturbation is applied to the normalized network dwell time corresponding to the industrial data stream. With the network dwell time of other industrial data streams remaining unchanged, the twin prediction module is called again to obtain the perturbation prediction result. The sensitivity calculation unit obtains the tissue quality sensitivity based on the change in the predicted latent variable of tissue quality before and after the perturbation, and obtains the forming state sensitivity based on the change in the predicted forming state before and after the perturbation.

[0109] At the computational level, the normalized network dwell time is obtained by the ratio between the actual network dwell time and the upper limit of the allowed network dwell time, and is limited to a preset value range. The upper limit of the allowed network dwell time is preset based on the forming control cycle, the 5G-TSN deterministic communication link configuration, and the process-allowed latency. The unit normalized perturbation is a preset small positive number, or a unit step size under the normalization scale. The tissue quality sensitivity is determined based on the norm of the difference between the predicted tissue quality latent variable after the perturbation and the predicted tissue quality latent variable before the perturbation, and is normalized by the sum of the norm of the predicted tissue quality latent variable before the perturbation and the positive stable term. The forming state sensitivity is determined based on the norm of the difference between the predicted forming state after the perturbation and the predicted forming state before the perturbation, and is normalized by the sum of the norm of the predicted forming state before the perturbation and the positive stable term. The tissue quality sensitivity and forming state sensitivity obtained through the above processing are both non-negative dimensionless values.

[0110] Priority scoring unit, used in the first The calculation of the first control cycle Process sensitivity priority scoring for industrial data streams:

[0111] ;

[0112] In the formula, Indicates the first Industrial data streams in the first Process sensitivity priority score for each control cycle Indicates the first The non-negative dimensionless sensitivity of network dwell perturbations in industrial-like data streams to predict latent variables of organizational quality. Indicates the first The non-negative dimensionless sensitivity of network dwell perturbations in industrial-like data streams to the prediction of forming states. Indicates the first The overall reliability of each control cycle Indicates the first Does the industrial data flow belong to the indication of the critical flow in closed-loop control? What is the corresponding critical flow in closed-loop control? Non-closed-loop control critical flow corresponding , , , and This represents the non-negative dimensionless scoring weight; all the above input quantities are dimensionless, which can avoid the inconsistency caused by directly adding parameters with different dimensions.

[0113] The overall reliability score is output by the reliability assessment module, ranging from 0 to 1. A lower overall reliability score indicates a worse data reliability status within the current control cycle, and a larger reliability compensation term in the process sensitivity priority score. The closed-loop control critical flow indication is determined based on whether the industrial data flow directly participates in the generation of control quantities for the next control cycle. Data flows used for transmitting control commands, equipment execution feedback, critical process status measurements, and necessary inputs to twin prediction models are pre-marked as closed-loop control critical flows, with a corresponding indication of 1; data flows not directly involved in closed-loop control calculations are pre-marked as non-closed-loop control critical flows, with a corresponding indication of 0. The closed-loop control critical flow marking table is pre-configured based on the production line control object and data flow type.

[0114] Rating weight , , and These are non-negative, dimensionless parameters used to adjust the influence of microstructure quality sensitivity, forming state sensitivity, overall reliability, and closed-loop control critical flow indicators on the process sensitivity priority scoring. The scoring weights can be normalized or non-normalized, with each scoring item maintaining a dimensionless scale. In implementations where different scoring weights correspond to different process stages, the weight corresponding to forming state sensitivity is increased during the hole-expanding stage, the weight corresponding to microstructure quality sensitivity is increased during the heat preservation stage, and the weight corresponding to overall reliability is increased when communication quality fluctuates significantly. The corresponding weight table is pre-stored in the dynamic transmission module and retrieved according to the current process stage.

[0115] The priority determination unit is used to determine the priority of the process based on the process sensitivity priority score and the score distribution of each industrial data stream in the current control cycle. Industrial data streams are mapped to integer TSN priority codes ranging from 0 to 7. In practice, the priority determination unit determines the TSN priority codes based on both the score ranking and the score range. Industrial data streams with higher scores falling within the high-score range are assigned higher integer TSN priority codes, while those with lower scores falling within the low-score range are assigned lower integer TSN priority codes. Since the integer TSN priority codes are determined by both the score ranking and the score range, changes in overall reliability can affect the priority level and resource carrying capacity of the industrial data stream. For closed-loop control critical flows, the priority determination unit protects them according to a preset lower priority limit to prevent them from being mapped to excessively low priorities.

[0116] The bearer matching unit is used to match corresponding bearer resources based on integer TSN priority code points. The bearer matching unit pre-stores the mapping relationship between TSN priority code points, differential service code points, 5G service flow identifiers, and data radio bearer resources. After priority determination, the bearer matching unit matches the corresponding bearer resources based on the integer TSN priority code points. For industrial data streams, corresponding TSN priority code points are identified and configured with the corresponding differential service code points, 5G service flow identifiers, and data radio bearer resources. This enables data streams of different priorities to be transmitted in the 5G-TSN deterministic communication link according to a preset bearer strategy. Higher-priority data streams are configured with low-latency and high-reliability bearer resources, while lower-priority data streams are configured with ordinary bearer resources.

[0117] The dwell time output unit is used to acquire the network dwell time of various industrial data streams in the 5G-TSN deterministic communication link. The network dwell time is recorded by the 5G-TSN gateway, network controller, or edge control device, representing the transmission, queuing, and scheduling time experienced by the corresponding industrial data stream in the communication link. The dwell time output unit, according to a preset data stream order, assembles the network dwell times corresponding to various industrial data streams into a network dwell time vector, and outputs the network dwell time vector to the twin prediction module. The preset data stream order is consistent with the industrial data stream number in sensitivity calculation, priority scoring, and bearer matching, enabling the twin prediction module to correctly retrieve the network dwell time of the corresponding data stream.

[0118] The dynamic transmission module in this embodiment does not involve artificial intelligence model training. Industrial data stream classification, unit normalized disturbance magnitude, maximum allowable network dwell time, scoring weights, closed-loop control key flow labeling table, priority mapping rules, bearer resource mapping relationships, and preset data stream order are all configurable parameters. Those skilled in the art, based on 5G-TSN link configuration, industrial data stream types, reliable assessment results, and twin prediction results, can implement the processes of industrial data stream sensitivity calculation, priority scoring, TSN priority code point determination, bearer resource matching, and network dwell time output.

[0119] (5) Twin prediction module

[0120] In this embodiment, the twin prediction module receives data output from the data synchronization module, the topology twin module, the reliability assessment module, and the dynamic transmission module. This data is used to generate a predicted forming state for the next control cycle and to provide a prediction basis for the control execution module. The inputs to the twin prediction module include synchronization process data, the ring topology digital twin state, control variables, a control variable reliability matrix, a network data reliability matrix, a measurement output reliability matrix, a normalized network dwell time vector, dimensionless tissue quality latent variables, and ring spatial inhomogeneity characteristics. The outputs include the predicted forming state for the next control cycle, the tissue quality latent variables for the next control cycle, as well as predicted temperature, predicted ring diameter, predicted ring wall thickness, predicted ellipticity, and predicted tissue quality risk.

[0121] The process stage identification unit determines the process stage corresponding to the current control cycle based on synchronous process data. The process stage is determined based on the process formula, equipment operating status, ring temperature range, radial feed status, axial feed status, ring rotation speed, and forming force changes. In practice, a process stage determination table is pre-established, associating and storing the heating, holding, expansion, rolling, shaping, and cooling stages with their corresponding process parameter ranges. Within each control cycle, the process stage identification unit reads the equipment operating status, control command status, and process status data from the synchronous process data, matches the current production status to the corresponding process stage, and outputs the current process stage identifier. This identifier is used to call the corresponding stage's state evolution matrix, control input matrix, and compensation matrix.

[0122] The trusted input assembly unit is used to organize data from different sources into a unified input required by the network trusted collaborative prediction model. Specifically, the trusted input assembly unit renders the current control cycle's formed state dimensionless to obtain a dimensionless formed state; renders the current control quantity dimensionless according to the control variable order to obtain a dimensionless control quantity; reads the trusted matrix of the control quantity, the trusted matrix of the network data, and the trusted matrix of the measurement output from the trusted evaluation module; reads the network residence time vector from the dynamic transmission module and normalizes it according to the upper limit of the allowed network residence time to obtain a normalized network residence time vector; reads the dimensionless organization quality latent variable from the organization quality prediction unit; and reads the annular spatial inhomogeneity characteristics from the topological twin module. The above inputs are assembled according to a preset vector order to ensure that the same variable has a fixed position in different control cycles.

[0123] The prediction model building unit is used to construct a network-based trusted collaborative prediction model. This model describes, within a unified forming state space, the impact of the current forming state, control variables, data trustworthiness, network dwell time, latent variables of organizational quality, and annular spatial inhomogeneities on the predicted forming state for the next control cycle. The network-based trusted collaborative prediction model satisfies the following:

[0124] ;

[0125] In the formula, Indicates the first Dimensionless forming state of a predictive recursive control cycle Indicates the first The predicted forming state of each prediction recursion cycle, Indicates the first The current process stage corresponding to each predictive recursive cycle. This represents the state evolution matrix corresponding to the current process stage. This represents the control input matrix corresponding to the current process stage. Represents the confidence matrix of the control quantity. This represents a dimensionless control quantity. This represents the network dwell time compensation matrix corresponding to the current process stage. Represents the network data credibility matrix. This represents the normalized network dwell time vector. This represents the measurement reliability perturbation compensation matrix. This represents the measurement output confidence matrix. Represents a vector consisting entirely of 1s. This represents the organizational quality impact matrix corresponding to the current process stage. This represents the dimensionless latent variable of organizational quality. This represents the annular topology compensation matrix corresponding to the current process stage. This indicates the non-uniformity of the annular space;

[0126] At the computational level, , , , and All are dimensionless quantities, matrices , , , , and This is used to map all input items to the same forming state space. The forming state space includes temperature state, diameter state, wall thickness state, ellipticity state, and deformation state related to the forming process. The number of rows in each matrix is ​​consistent with the dimension of the forming state vector. Each input item, after being mapped by the corresponding matrix, is... It maintains a consistent dimension, avoiding inconsistencies in dimensions caused by directly adding temperature, size, time, reliability, and tissue quality indicators. The reliability matrix for control quantities, network data, and measurement outputs is used to apply reliability weights to the corresponding input items; when the reliability of a variable decreases, the corresponding variable's effect on the prediction result weakens, and the measurement reliability disturbance compensation term increases accordingly.

[0127] The aforementioned matrix parameters were determined using a process mechanism model, historical production data, trial production data, and offline identification results. The offline identification process for the matrix parameters included: collecting dimensionless forming states, dimensionless control quantities, normalized network residence time vectors, dimensionless microstructure quality latent variables, and annular spatial inhomogeneity characteristics within historical control cycles; grouping samples according to process stages; using the measured dimensionless forming states of the next control cycle as output samples, and obtaining the matrix parameters corresponding to each process stage through least-squares identification. For process stages lacking historical data, initial matrix parameters were set based on finite element simulation results, equipment process parameters, and expert process rules; during system operation, the model correction unit recursively updated some matrices. The above model is a state prediction model and does not require neural network training as a necessary condition.

[0128] The microstructure quality prediction unit is used to predict the forming state, discrete element temperature, equivalent plastic strain increment, cooling rate, and historical microstructure quality latent variables for the next control cycle, and update the microstructure quality latent variables for the next control cycle. Discrete element temperature, equivalent plastic strain increment, and cooling rate are obtained by mapping synchronous process data, the toroidal topology digital twin state, and the predicted forming state for the next control cycle. Microstructure quality latent variables include grain coarsening risk indicators, overheating risk indicators, deformation uniformity indicators, and residual stress risk indicators. The grain coarsening risk indicator is determined based on the normalized exceedance degree of the discrete element temperature relative to the grain coarsening risk temperature threshold, the duration of the high temperature, and the equivalent plastic strain increment; the overheating risk indicator is determined based on the normalized exceedance degree of the discrete element temperature relative to the overheating risk temperature threshold and the duration of the exceedance; the deformation uniformity indicator is determined based on the deviation of the equivalent plastic strain increment of each discrete element relative to the weighted average equivalent plastic strain increment; and the residual stress risk indicator is determined based on the deviation of the toroidal topology temperature gradient, cooling rate, and equivalent plastic strain increment. Each risk indicator is normalized to form dimensionless microstructure quality latent variables.

[0129] In one implementation, the microstructure quality prediction unit calculates the risks of grain coarsening, overheating, deformation uniformity, and residual stress for each discrete unit. These risks are then weighted and summed according to the discrete unit weights to obtain the overall microstructure quality latent variables for the ring component. The discrete unit weights are pre-set based on the discrete unit's volume, mass, or importance to the target quality region, with a sum of weights equal to 1. The temperature threshold, allowable high-temperature duration, overheating risk temperature threshold, cooling rate scale, and equivalent plastic strain increment scale are pre-determined based on the material grade, process specifications, experimental data, or historical qualified product data. These microstructure quality latent variables do not directly replace material testing results but serve as risk characterization quantities for predictive control during the forming process.

[0130] The prediction state output unit is used to predict the forming state and the latent variables of the microstructure quality in the next control cycle, and output the prediction results. Specifically, the prediction state output unit predicts the forming state for the next control cycle. Inverse normalization is performed to obtain the predicted temperature, predicted ring diameter, predicted ring wall thickness, and predicted ellipticity. Based on the latent variables of the organization quality in the next control cycle, the predicted organization quality risk is obtained. The predicted temperature, predicted ring diameter, predicted ring wall thickness, predicted ellipticity, and predicted organization quality risk are output to the control execution module for rolling optimization, constraint judgment, and safety trigger judgment.

[0131] The model calibration unit recursively updates the network reliable collaborative prediction model based on the measured forming state, the predicted forming state, and the reliable weighted control input data. Specifically, after obtaining the measured forming state in the next control cycle, the model calibration unit calculates the prediction error between the measured forming state and the predicted forming state of the previous cycle, and applies reliable weighting to the control input data in conjunction with the reliable control matrix. The model calibration unit recursively updates the control input matrix based on the prediction error and the reliable weighted control input data. When the prediction error contains compensation errors caused by network dwell time, measurement reliable state, or ring space inhomogeneity, the network dwell time compensation matrix, measurement reliable disturbance compensation matrix, or ring topology compensation matrix is ​​recursively updated. The recursive update is implemented using recursive least squares, least squares with a forgetting factor, or gain limiting update methods. The update step size, forgetting factor, and parameter limiting range are stored as preset parameters in the twin prediction module to avoid abrupt changes in model parameters under the influence of abnormal data.

[0132] The key parameters of the twin prediction module in this embodiment include the dimensions of the forming state vector, control quantity vector, network residence time vector, latent microstructure quality variables, annular spatial inhomogeneity characteristics, normalization scale, process stage judgment table, matrix parameters corresponding to each process stage, microstructure quality risk threshold, discrete unit weight, recursive update step size, and parameter limit range. All of these parameters are predetermined based on the specifications of the wind turbine ring forging, material grade, equipment capacity, control cycle, historical production data, and trial production data. Those skilled in the art, based on the above input, model structure, parameter determination method, and output rules, can enable the twin prediction module to generate predictions of the forming state and latent microstructure quality variables for the next control cycle.

[0133] (6) Control Execution Module

[0134] In this embodiment, the control execution module receives the predicted forming state, predicted latent variables of tissue quality, and predicted tissue quality risk for the next control cycle from the twin prediction module, and receives the network dwell time from the dynamic transmission module and the variable-level credibility and overall credibility from the credibility assessment module. The control execution module generates the control quantity for the next control cycle based on the target forming state and the current production state, and determines the actual control quantity to be issued when the data credibility state, communication state, or tissue quality state is abnormal. The actual control quantity output by the control execution module is used to drive the heating equipment, radial feed actuator, axial feed actuator, and ring rotation speed actuator.

[0135] The target building unit is used to construct a rolling optimization problem based on a network-based trusted collaborative prediction model. The target forming state is determined by the process specifications, product size requirements, and the current process stage. The target forming state includes the target temperature, target ring diameter, target ring wall thickness, target ellipticity, and the target deformation state related to the forming process. The target microstructure quality latent variable is determined by the material grade, heat treatment window, and quality control requirements, representing the target control levels for grain coarsening risk, overheating risk, deformation uniformity risk, and residual stress risk. The target building unit uses the deviation of the predicted forming state from the target forming state, the deviation of the predicted microstructure quality latent variable from the target microstructure quality latent variable, the annular spatial inhomogeneity characteristics, and the cost of control quantity changes as the objective terms of the rolling optimization problem. The cost of control quantity changes is used to limit excessive changes in the control quantity between adjacent control cycles, ensuring that the control output matches the equipment's execution capability.

[0136] The constraint generation unit generates constraints for the rolling optimization problem. These constraints include model constraints, equipment constraints, process constraints, quality constraints, and communication constraints. Model constraints ensure that the predicted forming state satisfies the network-reliable collaborative prediction model in the twin prediction module. Equipment constraints limit the control quantity to preset upper and lower limits and the rate of change of the control quantity to a preset range. Process constraints limit the predicted temperature obtained by inverse normalization of the predicted forming state to an allowable temperature range. Quality constraints limit the predicted ring diameter and predicted ring wall thickness to allowable dimensions, the predicted ellipticity to not exceed the upper limit of allowable ellipticity, and each risk indicator in the predicted microstructure quality latent variables to not exceed the corresponding upper limit of allowable microstructure quality risk. Communication constraints limit both the predicted network dwell time and the measured network dwell time to the upper limit of allowable network dwell time. If either the predicted or measured network dwell time exceeds the upper limit of allowable network dwell time, the constraint generation unit generates a safety constraint and transmits it to the safety triggering unit. The predicted network dwell time is estimated by the dynamic transmission module based on the current bearer resource configuration and historical link status, while the measured network dwell time is obtained from the 5G-TSN deterministic communication link record.

[0137] The preset control upper and lower limits, preset rate of change range, allowable temperature range, allowable size range, allowable ellipticity upper limit, allowable microstructure quality risk upper limit, and allowable network dwell time upper limit are all preset based on the specifications of the wind turbine ring forging, material grade, equipment capacity, process procedures, and 5G-TSN deterministic communication link configuration. The upper and lower limits of multi-zone heating power are determined by the rated power of the heating equipment and the temperature process window; the upper and lower limits of radial feed control and axial feed control are determined by the stroke of the forming equipment and the allowable load; and the upper and lower limits of the ring rotation speed control are determined by the equipment rotation speed range and process stability requirements.

[0138] The rolling solver unit is used to solve for the control sequence of multiple future control cycles based on the objective term and constraints. These multiple future control cycles constitute the prediction time domain, and the rolling solver unit solves for the control sequence within the prediction time domain within the current control cycle. The control variables in the control sequence are arranged in a unified control variable order, which includes multi-zone heating power, radial feed control, axial feed control, and ring rotation speed control. In one implementation, when the network reliable collaborative prediction model, after linearization, forms a quadratic objective term and linear constraints, the rolling solver unit uses quadratic programming to solve the control sequence; when the constraints or prediction model contain nonlinear relationships, the rolling solver unit uses sequential quadratic programming or the interior-point method to solve the control sequence. During the solution process, the rolling solver unit uses the current synchronization process data, the current reliability matrix, the current network dwell time, and the current organizational quality latent variables as initial conditions to obtain a control sequence that satisfies the constraints.

[0139] The control output unit outputs the current control quantity in the control sequence for each control cycle. Specifically, after the rolling solution unit obtains the control sequence for multiple future control cycles, the control output unit only outputs the control quantity corresponding to the first control cycle as the control quantity for the next control cycle. Upon entering the next control cycle, the control execution module reconstructs and solves the rolling optimization problem based on the updated synchronization process data, predicted forming state, variable-level confidence, network dwell time, and organizational quality latent variables. Through this rolling approach, the control execution module can continuously update the control quantity according to changes in the production site status.

[0140] The security triggering unit determines whether security triggering conditions are met. These conditions include: the minimum value of the variable-level credibility being lower than a preset credibility threshold; network dwell time exceeding the allowed upper limit of network dwell time; and any risk indicator among the organizational quality latent variables exceeding the allowed upper limit of organizational quality risk. The minimum value of the variable-level credibility reflects the state of the least credible monitored variable within the current control cycle; network dwell time reflects the impact of the 5G-TSN deterministic communication link on the timeliness of closed-loop control; and the risk indicator among the organizational quality latent variables reflects the organizational quality risk in the current formation process. When any of the above security triggering conditions is met, the security triggering unit determines that it has entered a security control state and sends a security control request to the security control quantity generation unit.

[0141] The safety control quantity generation unit is used to solve for the safety envelope control quantity when the safety triggering conditions are met. The safety envelope control quantity is generated with the goals of reducing tissue quality risk, minimizing control quantity mutations, and maintaining the ring component temperature within a safe insulation range. In practice, the safety control quantity generation unit limits the variation range of radial and axial feed control quantities based on the current temperature, tissue quality latent variables, network dwell time, and equipment execution status. It maintains or slowly adjusts the multi-zone heating power to keep the ring component temperature within a safe insulation range and controls the ring component rotation speed within a preset stable range. The safety envelope control quantity does not primarily aim to quickly reach the target size, but rather to prevent further risk accumulation and maintain stable equipment execution. The safety envelope control quantity is output in the order of multi-zone heating power, radial feed control quantity, axial feed control quantity, and ring component rotation speed control quantity.

[0142] The control quantity fusion unit determines the security fusion coefficient based on the overall reliability, network dwell time exceeding limits, and organizational quality risk exceeding limits. It then fuses the current control quantity obtained from rolling optimization and the security envelope control quantity according to the security fusion coefficient to generate the actual control quantity issued. A lower overall reliability indicates a worse current data reliability status; a higher network dwell time exceeding limits indicates a greater impact of the communication link on closed-loop control; and a higher organizational quality risk exceeding limits indicates a higher potential quality risk from continuing to execute the original control quantity. The control quantity fusion unit determines the security fusion coefficient based on these three states, with the coefficient ranging from 0 to 1. When the security fusion coefficient is close to 0, the actual control quantity issued primarily uses the current control quantity obtained from rolling optimization; when the security fusion coefficient is close to 1, the actual control quantity issued primarily uses the security envelope control quantity.

[0143] At the computational level, the control quantity fusion unit first calculates the credibility risk, network dwell time exceeding the limit, and organizational quality risk exceeding the limit, respectively. The credibility risk is determined based on the difference between the overall credibility and a preset credibility threshold; the network dwell time exceeding the limit is determined based on the proportion of network dwell time exceeding the preset dwell time threshold; and the organizational quality risk exceeding the limit is determined based on the maximum proportion of each risk indicator in the organizational quality latent variables exceeding its corresponding organizational risk threshold. The control quantity fusion unit weights and sums these risk quantities and limits them to the range of 0 to 1 to obtain the security fusion coefficient. When the degree of network dwell time exceeding the limit and the degree of organizational quality risk exceeding the limit remain unchanged, the security fusion coefficient increases as the overall credibility decreases; when the overall credibility and the degree of organizational quality risk exceeding the limit remain unchanged, the security fusion coefficient increases as the degree of network dwell time exceeding the limit increases; and when the overall credibility and the degree of network dwell time exceeding the limit remain unchanged, the security fusion coefficient increases as the degree of organizational quality risk exceeding the limit increases.

[0144] The current control quantity and the safety envelope control quantity are represented in the same control variable order. This control variable order includes multi-zone heating power, radial feed control quantity, axial feed control quantity, and ring component speed control quantity. The control quantity fusion unit fuses the current control quantity and the safety envelope control quantity according to the position of the same variable, ensuring that each control variable in the actual issued control quantity corresponds one-to-one with the corresponding actuator. After the actual issued control quantity is generated, it is sent by the control execution module to the heating controller, radial feed controller, axial feed controller, and speed controller.

[0145] The control execution module in this embodiment does not involve artificial intelligence model training. The target forming state, target tissue quality latent variables, upper and lower limits of control quantities, range of control quantity change rates, allowable temperature range, allowable size range, upper limit of allowable ellipticity, upper limit of allowable tissue quality risk, upper limit of allowable network dwell time, preset credibility threshold, preset dwell time threshold, preset tissue risk threshold, prediction time domain, control time domain, and security fusion weights are all pre-configurable parameters. Those skilled in the art, based on the prediction results output by the twin prediction module, the credibility output by the credibility assessment module, the network dwell time output by the dynamic transmission module, and the aforementioned preset parameters, can achieve the processes of rolling optimization solution, security trigger judgment, security envelope control quantity generation, and actual control quantity determination.

[0146] like Figure 2 As shown, another embodiment of the present invention provides a method for controlling the forming of wind turbine ring forgings, comprising the following steps:

[0147] S1. Acquire multi-source process data during the production of wind turbine ring forgings, and perform physical event timing correction and control cycle alignment on the multi-source process data based on the 5G-TSN deterministic communication link to obtain synchronous process data.

[0148] S2. Construct a ring topology digital twin state based on the ring space structure and synchronous process data of the wind turbine ring forging, and generate ring space non-uniformity characteristics.

[0149] S3. Generate variable-level confidence levels based on the deviation between the measured variables and the digital twin predicted variables, and form a confidence matrix;

[0150] S4. Based on the degree of influence of industrial data flow on the forming state and organizational quality latent variables of the next control cycle, as well as the variable-level credibility, configure the data flow priority and bearer resources in the 5G-TSN link, and obtain the network dwell time.

[0151] S5. In a unified forming state space, integrate the current process stage, the annular topology digital twin state, control variables, reliability matrix, network dwell time, latent variables of organizational quality, and annular space non-uniformity characteristics to generate the predicted forming state for the next control cycle.

[0152] S6. Based on the target forming state, the predicted forming state for the next control cycle, the network dwell time, the variable-level confidence level, and the latent variables of organizational quality, generate the control quantity for the next control cycle.

[0153] S7. Determine whether the safety triggering condition is met; if the safety triggering condition is met, determine the actual control quantity to be issued based on the safety envelope control quantity; if the safety triggering condition is not met, use the control quantity of the next control cycle as the actual control quantity to be issued.

[0154] Through the above steps, the method can continuously update the synchronous process data, predict the forming state, and issue actual control quantities according to the control cycle, thereby completing the closed-loop control of the wind power ring forging process.

[0155] In summary, this invention provides a complete system structure and methodology for the integration of data acquisition, state modeling, communication coordination, and control execution in the wind turbine ring forging process. It also defines the calling relationships between multi-source process data, ring topology digital twin states, variable-level reliability, network dwell time, latent organizational quality variables, and control quantities. These features enable this invention to serve as an implementation scheme for digital forming control of wind turbine ring forgings, applicable to process monitoring, state prediction, communication resource coordination, and control execution in large ring forging production lines. Furthermore, it provides a scalable technical implementation path for the intelligent manufacturing control of similar large ring components.

[0156] In the description of this specification, the terms "an embodiment," "specific embodiment," etc., refer to specific technical features, structures, parameters, or processing methods described in connection with that embodiment, which are included in at least one embodiment of the present invention. Technical features from different embodiments may be combined without contradicting each other.

[0157] The modules, units, and steps described in this specification are used to illustrate the implementation of the present invention and are not intended to limit the implementation of the modules, units, and steps to the specific forms listed. Without changing the technical concept of the present invention, those skilled in the art can make adaptive adjustments to the relevant parameters, data acquisition methods, calculation rules, and execution methods according to the specifications of wind turbine ring forgings, production line equipment configuration, control cycle, and communication link conditions.

[0158] The above description is merely a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any equivalent substitutions, modifications, or variations made within the scope of the technical concept and claims of the present invention should be included within the scope of protection of the present invention. The scope of protection of the present invention is determined by the claims.

Claims

1. A wind turbine ring forging forming control system, characterized in that, include: The data synchronization module is used to acquire multi-source process data during the production of wind turbine ring forgings. Based on the 5G-TSN deterministic communication link, it performs physical event timing correction and control cycle alignment to obtain synchronized process data. The topology twin module is used to construct a ring topology digital twin state based on the ring spatial structure and synchronization process data of the wind turbine ring forging, and to generate ring spatial non-uniformity characteristics. The credibility assessment module is used to generate variable-level credibility based on the deviation between the measured variables and the digital twin predictive variables, and to form a credibility matrix. The credibility matrix generation unit is used to assemble the variable-level credibility corresponding to the equipment execution data into a control quantity credibility matrix, and to assemble the variable-level credibility corresponding to the process status data into a measurement output credibility matrix, and to assemble the variable-level credibility corresponding to the communication status data into a network data credibility matrix. The comprehensive credibility generation unit is used to perform weighted aggregation of variable-level credibility to obtain the comprehensive credibility, and the sum of the aggregation weights is 1. The dynamic transmission module is used to configure the data flow priority and bearer resources in the 5G-TSN deterministic communication link based on the degree of influence of industrial data flow on the forming state and latent variables of organizational quality in the next control cycle, as well as the variable-level reliability, and output the network dwell time. The dynamic transmission module includes: Sensitivity calculation unit, used for the first Industrial data streams, for the first The network dwell time of industrial data streams is subjected to a unit normalized perturbation, and the impact of the unit normalized perturbation on the latent variables of organizational quality and forming state in the next control cycle is calculated based on the twin prediction module, so as to obtain the organizational quality sensitivity and forming state sensitivity. Priority scoring unit, used in the first The calculation of the first control cycle Process sensitivity priority scoring for industrial data streams: ； In the formula, Indicates the first Industrial data streams in the first Process sensitivity priority score for each control cycle Indicates the first The non-negative dimensionless sensitivity of network dwell perturbations in industrial-like data streams to predict latent variables of organizational quality. Indicates the first The non-negative dimensionless sensitivity of network dwell perturbations in industrial-like data streams to the prediction of forming states. Indicates the first The overall reliability of each control cycle Indicates the first Does the industrial data flow belong to the indication of the critical flow in closed-loop control? What is the corresponding critical flow in closed-loop control? Non-closed-loop control critical flow corresponding , , , and This represents the non-negative dimensionless scoring weight; The priority determination unit is used to prioritize data based on the process sensitivity priority score and the score distribution of each industrial data stream within the current control cycle. Industrial-like data streams are mapped to integer TSN priority code points in the range of 0 to 7; The bearer matching unit is used to match the corresponding bearer resources from the pre-configured differential service code points, 5G service flow identifiers and data radio bearer resource sets based on the integer TSN priority code points. The dwell time output unit is used to acquire the network dwell time of various industrial data streams in the 5G-TSN deterministic communication link, and output the network dwell time vector according to the preset data stream order to the twin prediction module. The twin prediction module is used to integrate the current process stage, the ring topology digital twin state, control variables, reliability matrix, network dwell time, latent variables of organizational quality, and ring space non-uniformity characteristics within a unified forming state space to generate the predicted forming state for the next control cycle. The twin prediction module includes: The process stage identification unit is used to determine the first process stage based on synchronous process data. The process stage corresponding to each predicted recursive cycle; The reliable input assembly unit is used to acquire dimensionless forming state, dimensionless control quantity, control quantity reliability matrix, network data reliability matrix, measurement output reliability matrix, normalized network residence time vector, dimensionless tissue quality latent variable and annular space non-uniformity characteristics. Prediction model building unit, used to construct a network trusted collaborative prediction model, which satisfies the following: ； In the formula, Indicates the first The dimensionless forming state of a predictive recursive cycle Indicates the first The predicted forming state of each prediction recursion cycle, Indicates the first The current process stage corresponding to each predictive recursive cycle. This represents the state evolution matrix corresponding to the current process stage. This represents the control input matrix corresponding to the current process stage. Represents the confidence matrix of the control quantity. This represents a dimensionless control quantity. This represents the network dwell time compensation matrix corresponding to the current process stage. Represents the network data credibility matrix. This represents the normalized network dwell time vector. This represents the measurement reliability perturbation compensation matrix corresponding to the current process stage. This represents the measurement output confidence matrix. Represents a vector consisting entirely of 1s. This represents the organizational quality impact matrix corresponding to the current process stage. This represents the dimensionless latent variable of organizational quality. This represents the annular topology compensation matrix corresponding to the current process stage. This indicates the non-uniformity of the annular space; The microstructure quality prediction unit is used to predict the forming state, discrete unit temperature, equivalent plastic strain increment, cooling rate and historical microstructure quality latent variables based on the next control cycle, and update the microstructure quality latent variables for the next control cycle. The prediction state output unit is used to predict the forming state and the latent variables of the microstructure quality in the next control cycle based on the next control cycle, and outputs the predicted temperature, predicted ring diameter, predicted ring wall thickness, predicted ellipticity and predicted microstructure quality risk. The model calibration unit is used to recursively update the control input matrix in the network trusted collaborative prediction model based on the measured forming state, the predicted forming state, and the trusted weighted control input data, and to recursively update the compensation matrix in the network trusted collaborative prediction model when there is a compensation error. Dimensionless forming state, dimensionless control quantity, normalized network residence time vector, dimensionless tissue quality latent variable and annular space non-uniformity feature are all mapped to the same forming state space, and each input item has the same dimension as the forming state predicted in the next control cycle. The control execution module is used to generate the control quantity for the next control cycle based on the target forming state, the predicted forming state for the next control cycle, the network dwell time, the variable-level credibility, and the organizational quality latent variables, and to determine the actual control quantity to be issued according to the security trigger conditions. The data synchronization module, topology twin module, trust evaluation module, dynamic transmission module, twin prediction module, and control execution module are connected sequentially according to the data flow direction.

2. The wind turbine ring forging forming control system according to claim 1, characterized in that, The data synchronization module includes: The data acquisition unit is used to acquire process status data, equipment execution data, and communication status data. The process status data includes ring temperature data, ring geometry data, forming force data, and motion status data. The equipment execution data includes heating power execution feedback, radial feed execution feedback, axial feed execution feedback, ring rotation speed execution feedback, and equipment operating status. The communication status data includes network dwell time, synchronization error, and communication bearer status in the 5G-TSN deterministic communication link. The physical event timing correction unit is used to uniformly correct the sampling timing of process status data, the feedback timing of equipment execution data, and the recording timing of communication status data based on the time synchronization message, boundary clock correction information, transparent clock correction information, network dwell time, and synchronization error carried by the 5G-TSN deterministic communication link, so as to obtain the physical event timing corresponding to each data source. The synchronization data frame generation unit is used to merge data whose physical event times fall within the same control cycle into synchronization process data frames according to a preset control cycle. The synchronization quality identifier generation unit is used to generate synchronization quality identifiers based on the time correction deviation and data missing status of each data source. The synchronization process data includes synchronization process data frames and synchronization quality identifiers.

3. The wind turbine ring forging forming control system according to claim 1, characterized in that, The topology twin module includes: The ring node partitioning unit is used to divide the wind turbine ring forging into multiple circumferential sectors, multiple radial layers and multiple axial layers, and the discrete region jointly defined by the circumferential sector, radial layer and axial layer is used as the ring space node. The node state generation unit is used to generate the node state of each ring space node based on the synchronization process data, and to normalize the node state to obtain the dimensionless space node state. The topology matrix construction unit is used to construct a node-level ring topology matrix based on the circumferential closed adjacency relationship between adjacent circumferential sectors, the radial adjacency relationship between adjacent radial layers, and the axial adjacency relationship between adjacent axial layers. The node-level ring topology matrix is ​​then extended according to the characteristic dimensions of the spatial node states to obtain an extended ring topology matrix. The node-level ring topology matrix is ​​used to record the adjacency weights and difference calculation relationships between adjacent ring spatial nodes. The non-uniformity feature generation unit is used to extract neighborhood differences of the dimensionless spatial node states based on the extended annular topology matrix, and generate annular spatial non-uniformity features. The non-uniformity characteristics of the annular space include the differences in the distribution of annular temperature, geometry, and forming deformation in the circumferential, radial, and axial directions, and are used as inputs to the twin prediction module.

4. The wind turbine ring forging forming control system according to claim 1, characterized in that, The reliability assessment module includes: The variable deviation calculation unit is used to calculate the variable-level deviation between the measured values ​​and the digital twin predicted values ​​of the monitoring variables in the synchronization process data. The deviation normalization unit is used to normalize the variable-level deviation based on the deviation scale parameter of the corresponding monitoring variable under healthy operating conditions, so as to obtain the normalized variable-level deviation. The variable confidence generation unit is used to perform weighted accumulation of normalized variable-level deviations within a preset scrolling window and generate variable-level confidence through a monotonically decreasing mapping; the variable-level confidence decreases as the accumulation of normalized variable-level deviations increases.

5. The wind turbine ring forging forming control system according to claim 1, characterized in that, The organization quality prediction unit is used to update the organization quality latent variables for the next control period based on the organization quality latent variables. The microstructure quality latent variables are generated from multiple discrete unit states of the wind turbine ring forging. The microstructure quality latent variables include grain coarsening risk index, overheating risk index, deformation uniformity index, and residual stress risk index. The grain coarsening risk index is determined based on the normalized exceedance degree of the discrete unit temperature relative to the grain coarsening risk temperature threshold, the duration of high temperature, and the equivalent plastic strain increment. The overheating risk index is determined based on the normalized degree of exceeding the limit of the discrete unit temperature relative to the overheating risk temperature threshold and the duration of exceeding the limit. The deformation uniformity index is determined based on the deviation of the equivalent plastic strain increment of each discrete element from the weighted average equivalent plastic strain increment. The residual stress risk index is determined based on the annular topological temperature gradient, cooling rate, and equivalent plastic strain increment deviation. The grain coarsening risk index, overheating risk index, deformation uniformity index, and residual stress risk index are all dimensionless risk indices, and are obtained by weighting the weights of each discrete unit, with the sum of the weights of each discrete unit being 1.

6. The wind turbine ring forging forming control system according to claim 1, characterized in that, The control execution module includes: The target building unit is used to construct a rolling optimization problem based on a network-trusted collaborative prediction model. The rolling optimization problem takes the deviation of the predicted forming state from the target forming state, the deviation of the predicted tissue quality latent variable from the target tissue quality latent variable, the non-uniformity characteristics of the annular space, and the cost of changes in the control quantity as the objective terms. The constraint generation unit is used to generate constraints for the rolling optimization problem. These constraints include: the predicted forming state satisfies the network's reliable collaborative prediction model; the control quantity is within preset upper and lower control limits; the rate of change of the control quantity is within a preset rate of change range; the predicted temperature obtained by inverse normalization of the predicted forming state is within the allowable temperature range; the predicted ring diameter and predicted ring wall thickness are within the allowable size range; the predicted ellipticity does not exceed the allowable ellipticity upper limit; each risk indicator in the predicted tissue quality latent variables does not exceed the corresponding allowable tissue quality risk upper limit; both the predicted network dwell time and the measured network dwell time do not exceed the allowable network dwell time upper limit; and a safety constraint is triggered when either the predicted or measured network dwell time exceeds the allowable network dwell time upper limit. The rolling solver unit is used to solve the control sequence for multiple future control cycles based on the objective term and constraints. The control output unit is used to output the current control quantity in the control sequence in each control cycle, and to re-trigger the solution of the rolling optimization problem in the next control cycle based on the updated synchronization process data.

7. A wind turbine ring forging forming control system according to claim 6, characterized in that, The control execution module further includes a security triggering unit, a security control quantity generation unit, and a control quantity fusion unit; The security trigger unit is used to determine that a security trigger condition is met when one of the following security trigger conditions is satisfied: the minimum value of the variable-level confidence level is lower than a preset confidence threshold; The network dwell time exceeds the maximum allowed network dwell time limit; Any risk indicator among the organization's quality potential variables exceeds the permissible upper limit for organizational quality risk; The safety control quantity generation unit is used to solve for the safety envelope control quantity when the safety triggering conditions are met, with the goals of reducing organizational quality risks, minimizing control quantity mutations, and maintaining the temperature of the components within a safe insulation range. The control quantity fusion unit is used to determine the security fusion coefficient based on the comprehensive credibility, the degree of network dwell time exceeding the limit, and the degree of organizational quality risk exceeding the limit, and to fuse the current control quantity and the security envelope control quantity obtained by rolling optimization according to the security fusion coefficient to generate the actual issued control quantity; When the network dwell time and organizational quality risk exceedances remain constant, the security integration coefficient increases as the overall credibility decreases; when the overall credibility and organizational quality risk exceedances remain constant, the security integration coefficient increases as the network dwell time exceeds the limit; when the overall credibility and network dwell time exceedances remain constant, the security integration coefficient increases as the organizational quality risk exceeds the limit. The current control quantity and the safety envelope control quantity are represented in the same order of control variables, which include multi-zone heating power, radial feed control quantity, axial feed control quantity, and ring speed control quantity.

8. A method for controlling the forming of wind turbine ring forgings, implemented using a wind turbine ring forging control system as described in any one of claims 1-7, characterized in that, Includes the following steps: S1. Acquire multi-source process data during the production of wind turbine ring forgings, and perform physical event timing correction and control cycle alignment on the multi-source process data based on the 5G-TSN deterministic communication link to obtain synchronous process data. S2. Construct a ring topology digital twin state based on the ring space structure and synchronous process data of the wind turbine ring forging, and generate ring space non-uniformity characteristics. S3. Generate variable-level confidence levels based on the deviation between the measured variables and the digital twin predicted variables, and form a confidence matrix; S4. Based on the degree of influence of industrial data flow on the forming state and organizational quality latent variables of the next control cycle, as well as the variable-level credibility, configure the data flow priority and bearer resources in the 5G-TSN deterministic communication link, and obtain the network dwell time. S5. In a unified forming state space, integrate the current process stage, the annular topology digital twin state, control variables, reliability matrix, network dwell time, latent variables of organizational quality, and annular space non-uniformity characteristics to generate the predicted forming state for the next control cycle. S6. Based on the target forming state, the predicted forming state for the next control cycle, the network dwell time, the variable-level confidence level, and the latent variables of organizational quality, generate the control quantity for the next control cycle. S7. Determine whether the safety triggering conditions are met; When the safety triggering conditions are met, the actual control quantity to be issued is determined based on the safety envelope control quantity; If the safety triggering conditions are not met, the control quantity of the next control cycle will be used as the actual control quantity issued.

Images