Intelligent manufacturing flexible practical training management and control system and method based on digital twinning

Abstract

This invention relates to the field of virtual simulation teaching technology, specifically to a smart manufacturing flexible training and management system and method based on digital twins. By constructing a device twin state evolution analysis mechanism and combining it with technologies such as device state update records, operation attribution analysis, and dynamic permission control, this invention achieves precise tracking of device state changes and effective correlation of operational behaviors during smart manufacturing flexible training and management. It can achieve more accurate training evaluation and permission adjustments based on actual state feedback and operational results, thereby improving the accuracy and reliability of training process management. Simultaneously, this solution enhances the information synchronization capability between the digital twin model and physical equipment, making the training management process more intelligent and contributing to improved device interaction security and management efficiency in smart manufacturing flexible training scenarios.

Description

Technical Field

[0001] This invention relates to the field of virtual simulation teaching technology, and in particular to a flexible training and management system and method for intelligent manufacturing based on digital twins. Background Technology

[0002] The field of virtual simulation teaching technology mainly involves using computer hardware and software technology to generate realistic audiovisual environments for learners to interact with. Its main purpose is to construct virtual operation scenarios and guide learners to carry out internship and training tasks within the environment.

[0003] Traditional intelligent manufacturing flexible training and management processes typically rely on data collection from training equipment operation status, personnel operation records, and data aggregation from management platforms to monitor and evaluate changes in equipment status and personnel operational behaviors during the training process. While this approach can achieve basic training process management, in intelligent manufacturing flexible training, where the actual equipment status and digital management status are constantly changing, problems can easily arise such as untimely updates to equipment status information and insufficient correlation between abnormal states and actual operational behaviors. This makes it difficult for managers to accurately determine the causes of equipment status changes in a timely manner, affecting subsequent adjustments to training permissions and the effectiveness of process control. Summary of the Invention

[0004] To address the technical problems existing in the prior art, embodiments of the present invention provide an intelligent manufacturing flexible training and control system and method based on digital twins, the technical solution of which is as follows: On the one hand, a smart manufacturing flexible training and management system based on digital twins is provided, which includes: The sequence construction module acquires the equipment twin status observation sequence of the intelligent manufacturing flexible training and control production line; The state evolution module constructs the current state sequence matrix and the next state sequence matrix of the device twin state observation sequence, calculates the corresponding predicted device twin state matrix, extracts the pre-stored uncorrected device twin state matrix, calculates the matrix difference between the predicted device twin state matrix and the uncorrected device twin state matrix, and obtains the initial state deviation matrix. The matrix update module obtains a pre-stored topology mapping table containing physical device location information and workstation device twin nodes, searches for workstation device twin nodes in the topology mapping table with reference to the initial state deviation matrix, updates the workstation device twin nodes, and obtains a device twin state update record table. The operation attribution module collects the actual status feedback values ​​of the current training station and the expected status values ​​of the operation instructions, compares them with the equipment twin status update record table, and generates a multi-dimensional operation attribution identifier record set based on the comparison results. The access control module determines the current training performance update record of the trainee based on the multi-dimensional operation attribution identifier record set, uses the training performance update record as the performance-side access control quantity, and combines the device twin status update record table to determine the access enable signal and access lock blocking signal of the physical device location information. Based on the access enable signal and access lock blocking signal, the module performs the device interaction access control switching operation, which is used as the training control result.

[0005] As a further aspect of the present invention, the equipment twin state observation sequence includes normalized conveyor line running speed, normalized robot end-effector spatial coordinates, normalized processing station instruction execution time, normalized robot load weight, and normalized robot actual arrival time. The initial state deviation matrix includes the difference between each matrix term. The equipment twin state update record table includes station equipment twin nodes and the updated predicted equipment twin state matrix. The multidimensional operation attribution identifier record set includes a first anomaly exemption identifier, a second anomaly exemption identifier, and a training operation deduction identifier. The training control results include a training score update record sheet, permission opening enable signal, and permission locking block signal.

[0006] As a further aspect of the present invention, when the sequence construction module obtains the device twin state observation sequence, it performs the following processing in sequence: The system acquires the conveyor running speed, robot end-effector spatial coordinates, and processing station instruction execution time of the intelligent manufacturing flexible training and control production line. It also collects the robot end-effector load weight and the actual arrival time of the robot affected by the robot end-effector load weight, and integrates them into the initial observation parameters of the training operation. The extreme value normalization mapping is performed on the initial observation parameters of the training operation, including the conveyor running speed, the spatial coordinates of the robot end, the execution time of the processing station command, the load weight of the robot, and the actual arrival time of the robot, to obtain the corresponding normalized conveyor running speed, normalized robot end spatial coordinates, normalized processing station command execution time, normalized robot load weight, and normalized robot actual arrival time, and to establish normalized training equipment mapping parameters; The normalized training equipment mapping parameters are concatenated based on the chronological order to generate a twin state observation sequence for the equipment.

[0007] As a further aspect of the present invention, when the state evolution module constructs the current state sequence matrix and the next state sequence matrix and obtains the initial state deviation matrix, it performs the following processes in sequence: Extract the normalized conveyor running speed, normalized robot end-effector spatial coordinates, normalized processing station instruction execution time, normalized robot load weight, and normalized robot actual arrival time at the same acquisition time from the equipment twin state observation sequence. Arrange and combine them vertically to form a single-step state column vector. Obtain multiple single-step state column vectors of the equipment twin state observation sequence in a continuous time period. Remove the last item in the time sorting and horizontally concatenate the multiple single-step state column vectors in time order to form the current time state sequence matrix. Remove the first item in the time sorting and horizontally concatenate the multiple single-step state column vectors in time order to form the associated next time state sequence matrix. Perform pseudo-inverse matrix processing on the current time state sequence matrix to obtain the corresponding current time pseudo-inverse matrix, calculate the product between the next time state sequence matrix and the current time pseudo-inverse matrix, and obtain the linear mapping matrix of the physical device state evolution. Calculate the product between the linear mapping matrix and the current time-state sequence matrix to obtain the predicted device twin state matrix. Extract the pre-stored uncorrected device twin state matrix with the same row and column dimensions as the predicted device twin state matrix. Calculate the matrix difference between the predicted device twin state matrix and the uncorrected device twin state matrix to obtain the initial state deviation matrix.

[0008] As a further aspect of the present invention, when the matrix update module updates the workstation equipment twin node with reference to the initial state deviation matrix and obtains the equipment twin state update record table, it performs the following processing in sequence: Extract the matrix difference of each item in the initial state deviation matrix, calculate the absolute value of each matrix difference to obtain the initial deviation value, compare the initial deviation value with the preset deviation judgment threshold, and filter out the initial deviation exceeding the deviation judgment threshold. Obtain a pre-stored topology mapping table containing physical device location information and workstation device twin nodes; extract the matrix row and column coordinates of the initial deviation exceeding the limit value in the initial state deviation matrix; and query and match the workstation device twin node corresponding to the initial deviation exceeding the limit value in the topology mapping table based on the matrix row and column coordinates. The predicted equipment twin state matrix is ​​invoked to replace and update the uncorrected equipment twin state matrix associated with the workstation equipment twin node. The replaced and updated predicted equipment twin state matrix is ​​then assembled to generate an equipment twin state update record table.

[0009] As a further aspect of the present invention, when the operation attribution module generates a multidimensional operation attribution identifier record set based on the actual state feedback value, the expected state value of the operation instruction, and the equipment twin state update record table, it performs the following processing in sequence: Collect the actual status feedback values ​​and expected status values ​​of the operation instructions at the current training station, perform extreme value normalization mapping on the actual status feedback values ​​and expected status values ​​of the operation instructions, and generate normalized actual feedback values ​​and normalized expected status values. Extract the predicted device twin state matrix value corresponding to the normalized actual feedback value in the updated predicted device twin state matrix recorded in the device twin state update record table, calculate the difference between the normalized actual feedback value and the predicted device twin state matrix value, and obtain the first state offset difference. Extract the uncorrected device twin state matrix value corresponding to the normalized actual feedback value in the uncorrected device twin state matrix, calculate the difference between the normalized actual feedback value and the uncorrected device twin state matrix value, and obtain the second state offset difference. Calculate the difference between the predicted device twin state matrix value and the normalized expected state value, and obtain the third state offset difference. A first anomaly exemption flag for physical state changes is generated for the first state offset difference; a second anomaly exemption flag for equipment twin deviation is generated for the second state offset difference; and a training operation deduction flag for operational errors is generated for the third state offset difference. The first anomaly exemption flag, the second anomaly exemption flag, and the training operation deduction flag are merged to generate a multi-dimensional operation attribution flag record set.

[0010] As a further aspect of the present invention, when the access control module generates the training control result based on the multi-dimensional operation attribution identifier record set and the device twin status update record table, it performs the following processing in sequence: Obtain the initial behavior evaluation score of the current trainee, extract the score deduction compensation coefficient corresponding to the first abnormality exemption identifier, the second abnormality exemption identifier and the training operation deduction identifier recorded in the multidimensional operation attribution identifier record set, calculate the product between the score deduction compensation coefficient and the initial behavior evaluation score, and generate a training performance update record sheet. The system calls the physical device location information recorded in the topology mapping table, and generates permission enabling signal and permission locking blocking signal for the physical device location information based on the training performance update record sheet and the device twin status update record sheet. The permission enable signal and the permission lock block signal are sent to the intelligent manufacturing flexible training and control production line according to the physical device location information to perform the device interaction permission switching operation and generate training and control results.

[0011] On the other hand, the intelligent manufacturing flexible training and management method based on digital twins, which is executed based on the aforementioned intelligent manufacturing flexible training and management system based on digital twins, includes the following steps: S1: Obtain the equipment twin status observation sequence of the intelligent manufacturing flexible training and control production line; S2: Construct the current time state sequence matrix and the next time state sequence matrix of the device twin state observation sequence, calculate the corresponding predicted device twin state matrix, extract the pre-stored uncorrected device twin state matrix, calculate the matrix difference between the predicted device twin state matrix and the uncorrected device twin state matrix, and obtain the initial state deviation matrix. S3: Obtain a pre-stored topology mapping table containing physical device location information and workstation device twin nodes; refer to the initial state deviation matrix to find the workstation device twin nodes from the topology mapping table; update the workstation device twin nodes; and obtain the device twin state update record table. S4: Collect the actual status feedback value and expected status value of the operation instruction of the current training station, compare them with the equipment twin status update record table, and generate a multi-dimensional operation attribution identifier record set based on the comparison result; S5: Based on the multi-dimensional operation attribution identifier record set, determine the current training performance update record sheet of the trainee, use the training performance update record sheet as the performance side permission judgment quantity, combine with the device twin status update record table, determine the permission opening enable signal and permission locking block signal of the physical device location information, and perform the device interaction permission switching operation according to the permission opening enable signal and permission locking block signal, as the training control result.

[0012] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: This invention constructs a device twin state evolution analysis mechanism and combines it with technologies such as device state update records, operation attribution analysis, and dynamic permission control. This enables precise tracking of device state changes and effective correlation of operational behaviors during the flexible training and management of intelligent manufacturing. Based on actual state feedback and operational results, it allows for more accurate training evaluation and permission adjustments, thereby improving the accuracy and reliability of training process management. Simultaneously, this solution enhances the information synchronization capability between the digital twin model and physical equipment, making the training management process more intelligent and contributing to improved device interaction security and management efficiency in flexible intelligent manufacturing training scenarios. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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.

[0014] Figure 1A schematic diagram of the closed-loop management and control process for intelligent manufacturing flexible training based on digital twins; Figure 2 A schematic diagram illustrating the data flow for constructing the equipment twin state observation sequence; Figure 3 A schematic diagram illustrating the evolution of the equipment twin state and the formation of the initial state deviation matrix; Figure 4 This is a schematic diagram of the twin node update for workstation equipment based on topology mapping. Figure 5 A diagram illustrating the relationship formation of multidimensional operational attribution identifiers; Figure 6 This is a diagram illustrating the linkage and control between practical training results and equipment interaction permissions. Detailed Implementation

[0015] This embodiment provides a smart manufacturing flexible training and control system based on digital twins. In practical applications, for example, when the smart manufacturing flexible training and control production line is in a continuous teaching and training operation state, the conveyor line, robot arm, processing station and workstation equipment twin nodes are within the same training and control boundary. The trainees' operation instructions, production line physical feedback, equipment twin status and performance-side permission judgment quantities are transmitted in a closed loop within the control platform. The system includes a sequence construction module, a state evolution module, a matrix update module, an operation attribution module and a permission control module.

[0016] Please see Figures 1-6 The sequence construction module is used to acquire the equipment twin state observation sequence of the intelligent manufacturing flexible training and control production line. In this embodiment, the equipment twin state observation sequence is a time-series data object that carries the state of the production line equipment according to the acquisition time. It consists of normalized conveyor line running speed, normalized robot end space coordinates, normalized processing station instruction execution time, normalized robot load weight, and normalized robot actual arrival time. Its input sources are the production line controller, robot control feedback terminal, processing station instruction execution record terminal, and robot load acquisition terminal. Its output destination is the state evolution module.

[0017] When the sequence construction module forms the equipment twin state observation sequence, it continuously processes the production line state data according to the order of initial observation parameter acquisition, extreme value normalization mapping, and time sequence splicing during actual training operation. The data objects formed in the previous processing stage are used as the input for the next processing stage, and the traceability of the processing chain is maintained by using the acquisition source identifier and time sequence identifier.

[0018] The sequence construction module acquires the conveyor speed, robotic arm end-effector spatial coordinates, and processing station instruction execution time of the intelligent manufacturing flexible training and control production line. It also collects the robotic arm's load weight and the actual arrival time of the robotic arm affected by the load weight, integrating these into the initial observation parameters for the training operation. In this embodiment, the initial observation parameters for the training operation are a set of physical observation data of the production line that has not yet entered the same-scale mapping, encompassing the conveyor motion state, robotic arm spatial position state, processing station instruction execution state, robotic arm load state, and robotic arm arrival feedback state. After collection, a collection source identifier and a time sequence identifier are added.

[0019] The conveyor line's operating speed is derived from the conveyor line controller's operating feedback channel; the robot arm's end-effector spatial coordinates are derived from the robot arm's pose feedback channel; the processing station's instruction execution time is derived from the station-side record chain between the processing station's instruction issuance record and execution completion record; the robot arm's load weight is derived from the robot arm's end-effector load acquisition terminal; and the robot arm's actual arrival time is derived from the robot arm's arrival confirmation feedback terminal. The sequence construction module performs consistency checks on the field categories, acquisition sources, and time sequences of data from various sources. When an acquisition item is missing, repeatedly reported, has a field structure that does not conform to the preset acquisition format, or its source channel is not confirmed, the abnormal acquisition item is marked as pending verification and written to the acquisition trace record. The trace record includes the data source category, access channel, acquisition trigger condition, verification status, and subsequent traceability purpose. Acquisition items that fail verification do not proceed to subsequent extreme value normalization processing.

[0020] The sequence construction module performs extreme value normalization mapping on the initial observation parameters of the training operation, including the conveyor line running speed, robot end-effector spatial coordinates, processing station instruction execution time, robot load weight, and robot actual arrival time. This yields the corresponding normalized conveyor line running speed, normalized robot end-effector spatial coordinates, normalized processing station instruction execution time, normalized robot load weight, and normalized robot actual arrival time, establishing normalized training equipment mapping parameters. In this embodiment, extreme value normalization mapping is a preprocessing rule that maps initial observation parameters of the training operation from different sources and with different engineering meanings to a unified and comparable scale. The mapping is based on pre-stored records from the rated state of the production line equipment, station control boundaries, and historical calibration benchmarks. The mapping output does not change the physical meaning of each parameter.

[0021] The normalized conveyor line operating speed represents the current operating status of the conveyor line on a unified scale; the normalized robot end-effector spatial coordinates represent the spatial position of the robot end-effector on a unified scale; the normalized machining station instruction execution time represents the progress of machining station instruction execution on a unified scale; the normalized robot end-effector load weight represents the load status of the robot end-effector on a unified scale; and the normalized robot end-effector actual arrival time represents the robot end-effector arrival response status on a unified scale. For parameters that cannot match the pre-stored mapping boundary, the sequence construction module outputs a mapping unavailable status and writes the unavailability reason, source channel, and parameter category into the preprocessing log; available mapping results are entered into subsequent time sequence splicing processing according to parameter category.

[0022] After completing the unified scale mapping, the sequence construction module concatenates the normalized training equipment mapping parameters based on their temporal order to generate a twin state observation sequence for the equipment. In this embodiment, the normalized training equipment mapping parameters are a set of state parameters at the same scale after extreme value normalization mapping. Internally, they maintain the correspondence between the same acquisition time among the conveyor line, robot arm, processing station, and load feedback. The concatenation process uses the order of acquisition time and the consistency of parameter sources as the arrangement rules.

[0023] Specifically, the sequence construction module binds the normalized conveyor line running speed, normalized robot end-effector spatial coordinates, normalized processing station instruction execution time, normalized robot load weight, and normalized robot actual arrival time, all generated at the same acquisition time, into a state record for the same moment. These records are then connected according to their chronological order to form a twin state observation sequence for the equipment. If time sequence conflicts, incomplete parameters at the same acquisition time, misalignment of source records, or cached state invalidation occur, the sequence construction module maintains the integrity of the verified timing sequence portion, marks the abnormal position as a state to be supplemented, and passes it to the state evolution module. This ensures that the state evolution module only performs subsequent matrix processing on observation sequence segments that meet the timing alignment conditions.

[0024] State Evolution Module: The state evolution module is used to construct the current-moment state sequence matrix and the next-moment state sequence matrix of the equipment twin state observation sequence, forming the corresponding predicted equipment twin state matrix. It extracts the pre-stored uncorrected equipment twin state matrix and performs matrix difference processing on the predicted and uncorrected equipment twin state matrices to obtain the initial state deviation matrix. In this embodiment, the current-moment state sequence matrix is ​​a matrix data object formed by horizontally arranging the single-step state column vectors (excluding the last moment) within a continuous time period. The next-moment state sequence matrix is ​​also a matrix data object formed by horizontally arranging the single-step state column vectors (excluding the first moment) within a continuous time period. Both are used to express the evolution relationship of the same production line state between adjacent moments.

[0025] In this embodiment, the predicted device twin state matrix is ​​a twin-side predicted state data object formed by applying a linear mapping matrix of the physical device state evolution to the current state sequence matrix. The uncorrected device twin state matrix is ​​a pre-stored twin-side historical state data object that has not yet been replaced by the current predicted state. The initial state deviation matrix is ​​a data object carrying the matrix differences between the predicted device twin state matrix and the uncorrected device twin state matrix at each position, and its output goes to the matrix update module. The state evolution module completes the state evolution deduction according to the sequential relationship of state sequence assembly, dynamic feature extraction, and matrix difference processing.

[0026] The state evolution module extracts the normalized conveyor running speed, normalized robot end-effector spatial coordinates, normalized processing station instruction execution time, normalized robot load weight, and normalized robot actual arrival time at the same acquisition time from the equipment twin state observation sequence. These are arranged vertically as a single-step state column vector. The module then obtains the single-step state column vectors within a continuous time period from the equipment twin state observation sequence. The single-step state column vectors, excluding the last item in the time sorting, are horizontally concatenated in chronological order to form the current time-based state sequence matrix. Similarly, the single-step state column vectors, excluding the first item in the time sorting, are horizontally concatenated in chronological order to form the associated next-time-based state sequence matrix. In this embodiment, the single-step state column vector is a state carrier formed by the vertical arrangement of normalized state parameters within the same acquisition time. The X-axis, Y-axis, and Z-axis coordinates of the robot end-effector spatial coordinates are each used as independent elements of the single-step state column vector.

[0027] After receiving the equipment twin state observation sequence output by the sequence construction module, the state evolution module first verifies whether the state records at each time moment simultaneously contain the corresponding parameters of the conveyor line, robot arm, processing station, load, and position feedback. Then, it constructs the current time moment state sequence matrix and the next time moment state sequence matrix according to the time sequence. If there are unresolvable state records, repeated acquisition times, or discontinuous states between consecutive times in the time sequence, the state evolution module isolates the corresponding state records into time sequence segments that cannot participate in matrix processing, while retaining their acquisition source and verification status. Matrixes that meet the assembly conditions proceed to subsequent dynamic feature extraction processing.

[0028] The state evolution module performs pseudo-inverse matrix processing on the current state sequence matrix to obtain the corresponding current-time pseudo-inverse matrix. Based on the product relationship between the next-time state sequence matrix and the current-time pseudo-inverse matrix, it obtains the linear mapping matrix for the state evolution of the physical device. In this embodiment, the pseudo-inverse matrix processing is a matrix transformation process performed on the current-time state sequence matrix that can be used to solve the linear mapping. Its function is to convert the current-time state sequence matrix into a matrix object that can establish an evolutionary relationship with the next-time state sequence matrix. The current-time pseudo-inverse matrix only participates in the formation of the linear mapping matrix within the state evolution module.

[0029] In this embodiment, the linear mapping matrix for the state evolution of physical equipment is a matrix data object describing the state transition relationships between adjacent moments, including the normalized conveyor line running speed, the normalized spatial coordinates of the robot's end effector, the normalized processing station instruction execution time, the normalized robot's load weight, and the normalized robot's actual arrival time. Before outputting the linear mapping matrix, the state evolution module verifies the dimensional matching relationship, temporal correspondence, and source continuity between the current moment's state sequence matrix and the next moment's state sequence matrix. If the pseudo-inverse matrix at the current moment cannot be formed from a valid state sequence, the state evolution module outputs a state evolution unusable flag and accordingly blocks the transmission of erroneous states to the matrix update module.

[0030] The state evolution module forms a predicted device twin state matrix based on the product relationship between the linear mapping matrix and the current time-sequence matrix. It then extracts a pre-stored uncorrected device twin state matrix with the same row and column dimensions as the predicted twin state matrix. A matrix difference is performed between the predicted and uncorrected twin state matrices to obtain the initial state deviation matrix. In this embodiment, the matrix difference records the deviation between the predicted and uncorrected twin state matrices at the same row and column positions. Each item in the matrix difference retains its corresponding state parameter category, time position, and matrix row and column coordinates.

[0031] When extracting the uncalibrated twin state matrix, the state evolution module searches for pre-stored objects according to the row and column structure of the predicted twin state matrix, ensuring that the two types of matrices correspond in terms of state parameter categories and temporal positions. If a pre-stored object is missing, has a dimension mismatch, or the stored record cannot be parsed, the state evolution module marks the stored object as a twin state to be verified and prevents it from participating in bias formation as a valid uncalibrated twin state matrix. After completing the matrix interpolation process, the initial state bias matrix is ​​passed to the matrix update module, while the predicted twin state matrix is ​​retained as the data source for subsequent replacement updates.

[0032] Matrix Update Module: The matrix update module is used to obtain a pre-stored topology mapping table containing physical equipment location information and workstation equipment twin nodes. Referring to the initial state deviation matrix, it searches the topology mapping table for workstation equipment twin nodes, updates the workstation equipment twin nodes, and obtains an equipment twin state update record table. In this embodiment, the topology mapping table is a pre-stored table of correspondence between workstation spatial relationships and twin nodes, carrying the mapping relationship between physical equipment location information, workstation equipment twin nodes, and matrix row and column coordinates. The workstation equipment twin nodes are the twin-side state-carrying nodes corresponding to the training workstation, conveyor line segment, robot end effector, and processing workstation.

[0033] In this embodiment, the equipment twin state update record table is a data table that records the workstation equipment twin nodes and the updated predicted equipment twin state matrix. Its input sources are the initial state deviation matrix, the predicted equipment twin state matrix, the uncorrected equipment twin state matrix, and the topology mapping table. Its output destinations are the operation attribution module and the access control module. The matrix update module completes the location and update of the workstation equipment twin nodes according to the sequence of deviation determination, topology coordinate query, and update record compilation.

[0034] The matrix update module extracts the matrix difference value of each item in the initial state deviation matrix, performs an absolute deviation representation on each matrix difference value to obtain the initial deviation value, compares the initial deviation value with the preset deviation judgment benchmark, and filters out the initial deviation exceeding the limit value that is greater than the deviation judgment benchmark. In this embodiment, the preset deviation judgment benchmark is an engineering judgment rule used to distinguish between twin differences that can be accommodated and twin node differences that need to be updated. Its source is the pre-configuration of production line equipment calibration records, twin state synchronization strategies, and training control rules, and it is not disclosed in the embodiment in the form of a specific value.

[0035] In this embodiment, the initial deviation value is the deviation level object after absolute deviation representation of each matrix difference in the initial state deviation matrix. The initial deviation exceeding the limit value is the deviation object that is determined to need to enter the subsequent topology coordinate query processing after the initial deviation value is compared with the preset deviation judgment benchmark. The matrix update module outputs the pending judgment status for cases where the matrix difference is missing, the symbol marking is abnormal, the source matrix is ​​untraceable, or the judgment benchmark is not loaded, and writes the status and the corresponding matrix row and column coordinates into the deviation judgment record; the initial deviation exceeding the limit value is used for subsequent topology coordinate query processing.

[0036] The matrix update module obtains a pre-stored topology mapping table containing physical device location information and workstation twin nodes. It extracts the matrix row and column coordinates of the initial deviation exceeding the limit value in the initial state deviation matrix. Based on these matrix row and column coordinates, it queries the topology mapping table to obtain the workstation twin node corresponding to the initial deviation exceeding the limit value. In this embodiment, the physical device location information consists of spatial location records of conveyor lines, robotic arms, and processing workstations in the intelligent manufacturing flexible training and control production line, along with their workstation affiliation records. This information is used to point the matrix row and column coordinates to the controllable physical device locations.

[0037] In this embodiment, the matrix row and column coordinates serve as the position index of the initial deviation exceeding the limit value in the initial state deviation matrix, carrying the correspondence between the state parameter category and the time position. During the matching process, the matrix update module simultaneously verifies the correspondence between the matrix row and column coordinates, physical device location information, and the workstation device twin node. If the topology mapping table is missing, the coordinates cannot be matched, the physical device location is inconsistent with the twin node, or there is a conflict in the mapping record, the matrix update module outputs the matching status to be reviewed and retains the original initial deviation exceeding the limit value to avoid unconfirmed positions entering subsequent replacement updates and record assembly processing.

[0038] After location matching is completed, the matrix update module calls the predicted equipment twin state matrix to perform a replacement update on the uncorrected equipment twin state matrix associated with the workstation equipment twin node, assembles the replaced and updated predicted equipment twin state matrix, and generates an equipment twin state update record table. In this embodiment, the replacement update is a twin-side state refresh action that replaces the position of the uncorrected equipment twin state matrix corresponding to the matched workstation equipment twin node with the corresponding position state in the predicted equipment twin state matrix. The replacement range is jointly limited by the matrix row and column coordinates corresponding to the initial deviation exceeding the limit value and the topology mapping table.

[0039] The equipment twin status update record table retains the workstation equipment twin nodes, physical equipment location information, the updated predicted equipment twin status matrix, the source of the initial deviation exceeding the limit, and the update status. When writing to the record table, the matrix update module verifies whether the workstation equipment twin nodes are in an updateable state. If a node is locked, mapping pending review, or storage write unconfirmed, the replacement update is marked as pending, and the original uncorrected equipment twin status matrix remains traceable. After write confirmation, the equipment twin status update record table is sent to the operation attribution module as a basis for status comparison and to the access control module as a basis for access control determination.

[0040] Operation Attribution Module: The operation attribution module collects the actual status feedback values ​​and expected status values ​​of the operation commands at the current training station, compares them with the equipment twin status update record table, and generates a multi-dimensional operation attribution identifier record set based on the comparison results. In this embodiment, the actual status feedback values ​​specifically include the actual robot arm spatial coordinates, the actual conveyor line running speed, and the actual processing command execution time at the current training station. The expected status values ​​of the operation commands specifically include the target robot arm spatial coordinates, the set conveyor line running speed, and the expected processing command execution time. The actual robot arm spatial coordinates include the actual X-axis coordinates, actual Y-axis coordinates, and actual Z-axis coordinates, while the target robot arm spatial coordinates include the target X-axis coordinates, target Y-axis coordinates, and target Z-axis coordinates, each serving as an independent element of the single-step status column vector.

[0041] In this embodiment, the multi-dimensional operation attribution identifier record set is a data record set for the division of responsibility for practical training operations, including a first anomaly exemption identifier, a second anomaly exemption identifier, and a practical training operation deduction identifier. The first anomaly exemption identifier is an attribution identifier formed for changes in physical state, the second anomaly exemption identifier is an attribution identifier formed for equipment twin deviations, and the practical training operation deduction identifier is an attribution identifier formed for operational errors. The operation attribution module completes the division of responsibility for practical training operations according to the order of numerical normalization, deviation conversion, and attribution identifier aggregation.

[0042] The operation attribution module collects the actual status feedback values ​​of the current training workstation and the expected status values ​​of the operation instructions. It then performs extreme value normalization mapping on the actual status feedback values ​​and the expected status values ​​of the operation instructions to generate normalized actual feedback values ​​and normalized expected status values. In this embodiment, the normalized actual feedback value is the status representation of the current training workstation's physical feedback state after being mapped to a unified scale, and the normalized expected status value is the status representation of the workstation state expected to be achieved by the training operation instructions after being mapped to a unified scale.

[0043] The operation attribution module obtains the actual robot's spatial coordinates, actual conveyor line speed, and actual processing instruction execution time from the physical feedback channel of the current training station. It also obtains the target robot's spatial coordinates, set conveyor line speed, and expected processing instruction execution time from the operation instruction recording channel. Data from both sources is mapped using source identifiers, station identifiers, and instruction association identifiers. Feedback values ​​or expected state values ​​that fail to correspond are marked as incomplete attribution input and written to the station's attribution log. The normalized actual feedback values ​​and normalized expected state values, after completing extreme value normalization mapping, are then processed for subsequent multi-dimensional deviation conversion.

[0044] The operation attribution module extracts the predicted device twin state matrix values ​​corresponding to the normalized actual feedback values ​​in the updated predicted device twin state matrix recorded in the device twin state update record table. A first state offset difference is formed based on the difference between the normalized actual feedback values ​​and the predicted device twin state matrix values. The module also extracts the uncorrected device twin state matrix values ​​corresponding to the normalized actual feedback values ​​in the uncorrected device twin state matrix. A second state offset difference is formed based on the difference between the normalized actual feedback values ​​and the uncorrected device twin state matrix values. Finally, a third state offset difference is formed based on the difference between the predicted device twin state matrix values ​​and the normalized expected state values.

[0045] In this embodiment, the first state offset difference is used to express the offset relationship between the actual feedback state of the physical training station and the updated twin prediction state. The second state offset difference is used to express the offset relationship between the actual feedback state of the physical training station and the uncorrected twin state. The third state offset difference is used to express the offset relationship between the updated twin prediction state and the expected state of the training operation instruction. The operation attribution module performs position correspondence verification on the normalized actual feedback value, the normalized expected state value, the predicted equipment twin state matrix value, and the uncorrected equipment twin state matrix value. If the corresponding position does not exist, the update record table status is not confirmed, or the uncorrected equipment twin state matrix is ​​untraceable, the corresponding offset difference is marked as a state to be attributed and does not enter the valid range of subsequent attribution identifier summarization.

[0046] The operation attribution module generates a first anomaly exemption flag for physical state changes based on the first state offset difference, a second anomaly exemption flag for equipment twin deviations based on the second state offset difference, and a training operation deduction flag for operational errors based on the third state offset difference. It then merges the first anomaly exemption flag, the second anomaly exemption flag, and the training operation deduction flag to generate a multi-dimensional operation attribution flag record set. The inputs to this process are the first state offset difference, the second state offset difference, and the third state offset difference, and the output directly enters the access control module.

[0047] When generating attribution identifiers, the operation attribution module categorizes attribution objects according to the state source corresponding to the offset difference. When the first state offset difference corresponds to a change in physical state, the first anomaly exemption identifier carries the source of the difference between the physical equipment feedback state and the twin predicted state. When the second state offset difference corresponds to a twin deviation in equipment, the second anomaly exemption identifier carries the impact of the uncorrected twin state on the current attribution judgment. When the third state offset difference corresponds to the difference between the expected state of the operation instruction and the predicted twin state of the equipment, the training operation deduction identifier carries the category of performance impact corresponding to the trainee's operational error. If a state to be attributed exists, the operation attribution module writes this state, along with the corresponding workstation, state source, and offset category, into a multi-dimensional operation attribution identifier record set, which is then used by the access control module to identify attribution availability in performance conversion and access determination.

[0048] Access Control Module: This module determines the current trainee's training performance update record based on a multi-dimensional operation attribution identifier record set. The training performance update record is used as the performance-side access control metric. Combined with the device twin status update record table, it determines the access enable signal and access lock blocking signal for physical device location information. Based on these signals, it executes a device interaction access switching operation, which serves as the training control result. In this embodiment, the training performance update record is a performance-side data object recording the current trainee's behavior evaluation score after attribution identifier correction. The performance-side access control metric is the input object used by the access control module to determine the device interaction access status.

[0049] In this embodiment, the permission opening enable signal is a control signal that allows the corresponding physical device location information to enter an interactive open state, while the permission locking block signal is a control signal that keeps or locks the corresponding physical device location information to an interactive locked state. The training management results include the training grade update record sheet, the permission opening enable signal, and the permission locking block signal. The permission management module completes the linkage management between the grade side and the device side according to the sequence of grade conversion, enabling or blocking judgment, and permission switching.

[0050] The access control module obtains the initial behavior evaluation score of the current trainee, extracts the first anomaly exemption identifier, the second anomaly exemption identifier, and the corresponding score deduction compensation coefficient from the multi-dimensional operation attribution identifier record set, and generates a training performance update record based on the product relationship between the score deduction compensation coefficient and the initial behavior evaluation score. In this embodiment, the initial behavior evaluation score is the basic behavior evaluation data provided by the performance management record before the trainee operates at the current training station. The score deduction compensation coefficient is a performance correction rule object set by the preset physical equipment anomaly tolerance rate and the trainee's operation error deduction rate.

[0051] In this embodiment, the physical equipment fault tolerance rate is used to characterize the source of grade correction when anomalies occur due to changes in the physical equipment's state, while the trainee's operational error deduction rate is used to characterize the source of grade deduction when the training operation deviates from the expected state. Both are derived from the training control rule configuration, and their specific values ​​are not disclosed. Before generating the training grade update record sheet, the access control module verifies the availability of the attribution identifier, the current trainee's identity association status, and the source of the initial behavior evaluation score. If the attribution identifier is in a pending attribution state or the trainee's identity cannot correspond to the workstation operation record, the training grade update record sheet is marked as a pending confirmation grade status and is not used as a direct basis for determining open permissions.

[0052] The access control module calls upon the physical device location information recorded in the topology mapping table, and generates access enable signals and access lock blocking signals based on the training performance update record sheet and the device twin status update record sheet. The inputs to this process are the training performance update record sheet, the device twin status update record sheet, and the topology mapping table, and the output is an access enable signal or an access lock blocking signal pointing to the physical device location information.

[0053] The access control module first reads the update status of the workstation equipment twin nodes in the equipment twin status update record table, and then reads the status category of the grade-side access control quantity in the training grade update record sheet, matching the two with the physical equipment location information in the topology mapping table. If the workstation equipment twin node update has been confirmed and the grade-side access control quantity meets the opening conditions, an access control enable signal is generated for the corresponding physical equipment location information; if the workstation equipment twin node is in a state of delayed update, pending review and matching, pending grade confirmation, or attributed to unavailable, an access control block signal is generated for the corresponding physical equipment location information. The control basis, input source, judgment status, and output signal in the judgment process are all written into the access control traceability record for subsequent auditing and traceability.

[0054] After completing the permission determination, the permission control module sends the permission enable signal and the permission lock blocking signal to the intelligent manufacturing flexible training and control production line based on the physical device location information to perform the device interaction permission switching operation and generate the training and control results. In this embodiment, the device interaction permission switching operation is a control action to open or lock the conveyor line interaction entry, the robot operation entry, and the processing station instruction entry. The controlled objects are limited by the physical device location information, and the control basis comes from the aforementioned permission enable signal or permission lock blocking signal.

[0055] When switching device interaction permissions, the permission control module sends a permission enable signal or a permission lock blocking signal to the control interface of the corresponding physical device location and waits for the production line to return a control confirmation status. If the control interface returns an error, the transmission is not confirmed, or the target device location is inconsistent with the topology mapping table, the permission control module maintains the locked state of the corresponding device interaction permission and writes the control error status, target physical device location information, output signal type, and production line feedback status into the permission switching log. If the production line confirms the completion of the permission switch, the training score update record, the permission enable signal, and the permission lock blocking signal are all incorporated into the training control results, so that the update of trainee scores, the update of twin node status, and the physical device interaction permissions form a closed loop within the same training control boundary.

[0056] A digital twin-based intelligent manufacturing flexible training and management method includes the following steps: S1: Obtain the equipment twin status observation sequence of the intelligent manufacturing flexible training and control production line; S2: Construct the current time state sequence matrix and the next time state sequence matrix of the equipment twin state observation sequence, calculate the corresponding predicted equipment twin state matrix, extract the pre-stored uncorrected equipment twin state matrix, calculate the matrix difference between the predicted equipment twin state matrix and the uncorrected equipment twin state matrix, and obtain the initial state deviation matrix. S3: Obtain a pre-stored topology mapping table containing physical device location information and workstation device twin nodes; refer to the initial state deviation matrix to find the workstation device twin nodes from the topology mapping table; update the workstation device twin nodes; and obtain the device twin state update record table. S4: Collect the actual status feedback values ​​of the current training station and the expected status values ​​of the operation instructions, compare them with the equipment twin status update record table, and generate a multi-dimensional operation attribution identifier record set based on the comparison results; S5: Based on the multi-dimensional operation attribution identifier record set, determine the current training performance update record sheet of the trainee, use the training performance update record sheet as the performance side permission judgment quantity, combine with the equipment twin status update record table, determine the permission opening enable signal and permission locking block signal of physical equipment location information, and perform equipment interaction permission switching operation according to the permission opening enable signal and permission locking block signal, as the training control result.

[0057] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A flexible training and control system for intelligent manufacturing based on digital twins, characterized in that, The system includes: The sequence construction module acquires the equipment twin status observation sequence of the intelligent manufacturing flexible training and control production line; The state evolution module constructs the current state sequence matrix and the next state sequence matrix of the device twin state observation sequence, calculates the corresponding predicted device twin state matrix, extracts the pre-stored uncorrected device twin state matrix, calculates the matrix difference between the predicted device twin state matrix and the uncorrected device twin state matrix, and obtains the initial state deviation matrix. The matrix update module obtains a pre-stored topology mapping table containing physical device location information and workstation device twin nodes, searches for workstation device twin nodes in the topology mapping table with reference to the initial state deviation matrix, updates the workstation device twin nodes, and obtains a device twin state update record table. The operation attribution module collects the actual status feedback values ​​of the current training station and the expected status values ​​of the operation instructions, compares them with the equipment twin status update record table, and generates a multi-dimensional operation attribution identifier record set based on the comparison results. The access control module determines the current training performance update record of the trainee based on the multi-dimensional operation attribution identifier record set, uses the training performance update record as the performance-side access control quantity, and combines the device twin status update record table to determine the access enable signal and access lock blocking signal of the physical device location information. Based on the access enable signal and access lock blocking signal, the module performs the device interaction access control switching operation, which is used as the training control result.

2. The intelligent manufacturing flexible training and control system based on digital twins according to claim 1, characterized in that: The equipment twin state observation sequence includes normalized conveyor line running speed, normalized robot end-effector spatial coordinates, normalized processing station instruction execution time, normalized robot load weight, and normalized robot actual arrival time. The initial state deviation matrix includes the difference between each matrix item. The equipment twin state update record table includes the workstation equipment twin node and the updated predicted equipment twin state matrix. The multi-dimensional operation attribution identifier record set includes the first anomaly exemption identifier, the second anomaly exemption identifier, and the training operation deduction identifier. The training control results include the training score update record sheet, permission opening enable signal, and permission locking block signal.

3. The intelligent manufacturing flexible training and control system based on digital twins according to claim 1, characterized in that: When the sequence construction module obtains the device twin state observation sequence, it performs the following processing in sequence: The system acquires the conveyor running speed, robot end-effector spatial coordinates, and processing station instruction execution time of the intelligent manufacturing flexible training and control production line. It also collects the robot end-effector load weight and the actual arrival time of the robot affected by the robot end-effector load weight, and integrates them into the initial observation parameters of the training operation. The extreme value normalization mapping is performed on the initial observation parameters of the training operation, including the conveyor running speed, the spatial coordinates of the robot end, the execution time of the processing station command, the load weight of the robot, and the actual arrival time of the robot, to obtain the corresponding normalized conveyor running speed, normalized robot end spatial coordinates, normalized processing station command execution time, normalized robot load weight, and normalized robot actual arrival time, and to establish normalized training equipment mapping parameters; The normalized training equipment mapping parameters are concatenated based on the chronological order to generate a twin state observation sequence for the equipment.

4. The intelligent manufacturing flexible training and control system based on digital twins according to claim 3, characterized in that: When the state evolution module constructs the current state sequence matrix and the next state sequence matrix and obtains the initial state deviation matrix, it performs the following processes in sequence: Extract the normalized conveyor running speed, normalized robot end-effector spatial coordinates, normalized processing station instruction execution time, normalized robot load weight, and normalized robot actual arrival time at the same acquisition time from the equipment twin state observation sequence. Arrange and combine them vertically to form a single-step state column vector. Obtain multiple single-step state column vectors of the equipment twin state observation sequence in a continuous time period. Remove the last item in the time sorting and horizontally concatenate the multiple single-step state column vectors in time order to form the current time state sequence matrix. Remove the first item in the time sorting and horizontally concatenate the multiple single-step state column vectors in time order to form the associated next time state sequence matrix. Perform pseudo-inverse matrix processing on the current time state sequence matrix to obtain the corresponding current time pseudo-inverse matrix, calculate the product between the next time state sequence matrix and the current time pseudo-inverse matrix, and obtain the linear mapping matrix of the physical device state evolution. Calculate the product between the linear mapping matrix and the current time-state sequence matrix to obtain the predicted device twin state matrix. Extract the pre-stored uncorrected device twin state matrix with the same row and column dimensions as the predicted device twin state matrix. Calculate the matrix difference between the predicted device twin state matrix and the uncorrected device twin state matrix to obtain the initial state deviation matrix.

5. The intelligent manufacturing flexible training and control system based on digital twins according to claim 4, characterized in that: When the matrix update module updates the workstation equipment twin node with reference to the initial state deviation matrix and obtains the equipment twin state update record table, it performs the following processes in sequence: Extract the matrix difference of each item in the initial state deviation matrix, calculate the absolute value of each matrix difference to obtain the initial deviation value, compare the initial deviation value with the preset deviation judgment threshold, and filter out the initial deviation exceeding the deviation judgment threshold. Obtain a pre-stored topology mapping table containing physical device location information and workstation device twin nodes; extract the matrix row and column coordinates of the initial deviation exceeding the limit value in the initial state deviation matrix; and query and match the workstation device twin node corresponding to the initial deviation exceeding the limit value in the topology mapping table based on the matrix row and column coordinates. The predicted equipment twin state matrix is ​​invoked to replace and update the uncorrected equipment twin state matrix associated with the workstation equipment twin node. The replaced and updated predicted equipment twin state matrix is ​​then assembled to generate an equipment twin state update record table.

6. The intelligent manufacturing flexible training and control system based on digital twins according to claim 5, characterized in that: When the operation attribution module generates the multidimensional operation attribution identifier record set based on the actual state feedback value, the expected state value of the operation command, and the device twin state update record table, it performs the following processing in sequence: Collect the actual status feedback values ​​and expected status values ​​of the operation instructions at the current training station, perform extreme value normalization mapping on the actual status feedback values ​​and expected status values ​​of the operation instructions, and generate normalized actual feedback values ​​and normalized expected status values. Extract the predicted device twin state matrix value corresponding to the normalized actual feedback value in the updated predicted device twin state matrix recorded in the device twin state update record table, calculate the difference between the normalized actual feedback value and the predicted device twin state matrix value, and obtain the first state offset difference. Extract the uncorrected device twin state matrix value corresponding to the normalized actual feedback value in the uncorrected device twin state matrix, calculate the difference between the normalized actual feedback value and the uncorrected device twin state matrix value, and obtain the second state offset difference. Calculate the difference between the predicted device twin state matrix value and the normalized expected state value, and obtain the third state offset difference. A first anomaly exemption flag for physical state changes is generated for the first state offset difference; a second anomaly exemption flag for equipment twin deviation is generated for the second state offset difference; and a training operation deduction flag for operational errors is generated for the third state offset difference. The first anomaly exemption flag, the second anomaly exemption flag, and the training operation deduction flag are merged to generate a multi-dimensional operation attribution flag record set.

7. The intelligent manufacturing flexible training and control system based on digital twins according to claim 6, characterized in that: When the access control module generates the training control result based on the multi-dimensional operation attribution identifier record set and the device twin status update record table, it performs the following processing in sequence: Obtain the initial behavior evaluation score of the current trainee, extract the score deduction compensation coefficient corresponding to the first abnormality exemption identifier, the second abnormality exemption identifier and the training operation deduction identifier recorded in the multidimensional operation attribution identifier record set, calculate the product between the score deduction compensation coefficient and the initial behavior evaluation score, and generate a training performance update record sheet. The system calls the physical device location information recorded in the topology mapping table, and generates permission enabling signal and permission locking blocking signal for the physical device location information based on the training performance update record sheet and the device twin status update record sheet. The permission enable signal and the permission lock block signal are sent to the intelligent manufacturing flexible training and control production line according to the physical device location information to perform the device interaction permission switching operation and generate training and control results.

8. The intelligent manufacturing flexible training and control system based on digital twins according to claim 6, characterized in that: The actual state feedback values ​​are specifically the actual robot space coordinates of the current training station, the actual conveyor line running speed, and the actual processing instruction execution time. The expected state values ​​of the operation instruction are specifically the target robot space coordinates, the set conveyor line running speed, and the expected processing instruction execution time. The actual robot space coordinates include the actual X-axis coordinates, the actual Y-axis coordinates, and the actual Z-axis coordinates. The target robot space coordinates include the target X-axis coordinates, the target Y-axis coordinates, and the target Z-axis coordinates, which are each independent elements of the single-step state column vector.

9. The intelligent manufacturing flexible training and control system based on digital twins according to claim 7, characterized in that: The score deduction compensation coefficient is set based on the preset physical equipment fault tolerance rate and the training personnel's operational error deduction rate.

10. A flexible training and management method for intelligent manufacturing based on digital twins, characterized in that, The execution of the intelligent manufacturing flexible training and control system based on digital twins according to any one of claims 1-9 includes the following steps: S1: Obtain the equipment twin status observation sequence of the intelligent manufacturing flexible training and control production line; S2: Construct the current time state sequence matrix and the next time state sequence matrix of the device twin state observation sequence, calculate the corresponding predicted device twin state matrix, extract the pre-stored uncorrected device twin state matrix, calculate the matrix difference between the predicted device twin state matrix and the uncorrected device twin state matrix, and obtain the initial state deviation matrix. S3: Obtain a pre-stored topology mapping table containing physical device location information and workstation device twin nodes; refer to the initial state deviation matrix to find the workstation device twin nodes from the topology mapping table; update the workstation device twin nodes; and obtain the device twin state update record table. S4: Collect the actual status feedback value and expected status value of the operation instruction of the current training station, compare them with the equipment twin status update record table, and generate a multi-dimensional operation attribution identifier record set based on the comparison result; S5: Based on the multi-dimensional operation attribution identifier record set, determine the current training performance update record sheet of the trainee, use the training performance update record sheet as the performance side permission judgment quantity, combine with the device twin status update record table, determine the permission opening enable signal and permission locking block signal of the physical device location information, and perform the device interaction permission switching operation according to the permission opening enable signal and permission locking block signal, as the training control result.

Images