Intelligent manufacturing collaborative decision-making method based on digital twinning

Abstract

The application discloses an intelligent manufacturing collaborative decision-making method based on digital twinning, which comprises the following steps: establishing a global digital twinning model and a local digital twinning model according to the process of intelligent manufacturing; collecting production data in the intelligent manufacturing process in real time, determining a regulation and control index data set under a sampling number, constructing a time series augmented matrix, constructing a dynamic weight matrix by using a Pearson correlation coefficient, adjusting the time series augmented matrix into a weighted time series augmented matrix, performing singular value decomposition, introducing production process constraints to modify the singular value matrix, and reducing the dimension of the weighted time series augmented matrix by using the rank of the modified singular value matrix; standardizing the reduced weighted time series augmented matrix, dividing the standardized matrix by using a dynamic sliding window, inputting the matrix into a double-branch attention network in sequence, combining a current actual production decision scheme, and predicting a production decision scheme. The application improves the response speed of intelligent manufacturing through collaborative local self-collaboration and global scheduling.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing technology, and more specifically, to an intelligent manufacturing collaborative decision-making method based on digital twins. Background Technology

[0002] In the context of the digital economy era, fierce market competition and continuously evolving customer personalization demands have created a dual driving force, jointly propelling the rise of the Make-to-Order (MTO) model. This model, characterized by high customization and agile manufacturing, not only imposes stringent requirements on order delivery cycles but also necessitates the construction of a production system with real-time responsiveness—a closed-loop decision-making mechanism capable of millisecond-level anomaly diagnosis and minute-level strategy adjustment in the event of sudden disruptions such as equipment failures or order changes.

[0003] To adapt to these dynamic production demands, distributed architectures based on multi-agent negotiation have become the mainstream technology in production scheduling due to their autonomous and collaborative characteristics. It is worth noting that with the accelerated implementation of Cyber-Physical Production Systems (CPPS) under the Industry 4.0 framework, the construction of a virtual-physical integrated production environment through digital twins may bring disruptive changes to the make-to-order production model. This technological integration not only enhances the real-time perception capabilities of the production system but also enables millisecond-level responses to distributed decisions through edge computing, opening up new practical dimensions for intelligent manufacturing.

[0004] Digital twin technology typically involves continuously synchronizing the states of physical entities to build dynamic mapping relationships, forming a virtual mirror system that supports holistic insights. This virtual-physical interaction technology not only enables millimeter-level precision modeling and visualization of manufacturing processes but also achieves process optimization, anomaly warning, and full lifecycle management through real-time data-driven approaches. However, the traditional global digital twin paradigm based on full-domain modeling focuses on descriptive tasks such as production traceability and status monitoring, lacking self-optimization of process parameters and dynamic scheduling decisions, which severely diminishes the technological potential of digital twins.

[0005] Current mainstream research on production process anomaly management focuses primarily on the global restructuring of upper-level production plans, such as order prioritization and equipment load balancing, but fails to effectively activate the distributed intelligent potential of edge computing nodes. This technical deficiency directly leads to a cyber-physical integration gap in the "edge-cloud" collaborative architecture: when encountering sudden equipment failures, centralized decision-making in the cloud still needs to be relied upon. Existing frameworks lack device-level autonomous negotiation and decision-making mechanisms and have failed to build a distributed anomaly handling network based on digital threads, resulting in the production line's self-healing capabilities being limited to the traditional linear process of "monitoring-reporting-response," and failing to achieve automatic triggering of smart contracts and dynamic resource reorganization between production line nodes.

[0006] Regarding digital twin-driven production management methods, there is a significant discrepancy between the "global" advantage of digital twins in continuously synchronizing the state of physical entities to build dynamic mapping relationships and their precise real-time response. Although there have been reports on the concept of local digital twins, these carriers should ideally facilitate data exchange and intelligent analysis between local equipment and the global manufacturing system. However, a standardized architecture for local digital twin carriers has yet to be established. In most current application scenarios, local digital twin service functions are forced to be directly loaded onto the physical production equipment itself. This "hard-coupling" deployment mode not only violates the principle of decoupling between the virtual and physical worlds in digital twins but also leads to a surge in the load on edge computing nodes, resulting in a chain reaction of data transmission delays and decreased system stability. Summary of the Invention

[0007] To address the problems existing in the prior art, this invention provides a collaborative decision-making method for intelligent manufacturing based on digital twins. By constructing a two-level system that integrates global and local digital twins, it coordinates local self-cooperation and global scheduling based on intelligent edge industrial control computers, thereby improving the response speed of intelligent manufacturing.

[0008] To achieve the above technical objectives, the present invention adopts the following technical solution:

[0009] A collaborative decision-making method for intelligent manufacturing based on digital twins includes the following steps:

[0010] Step S1: Establish a global digital twin model and a local digital twin model based on the intelligent manufacturing process;

[0011] Step S2: Collect production data in real time during the intelligent manufacturing process to determine the control index dataset under the sampling number;

[0012] Step S3: Construct a time-series augmented matrix for the control indicator datasets from all sampling times, and calculate the first... s -1 sampling number and the first s The Pearson correlation coefficients at each sampling number are used to construct a dynamic weight matrix, and the time-series augmented matrix is ​​adjusted to a weighted time-series augmented matrix.

[0013] Step S4: Perform singular value decomposition on the weighted time series augmented matrix to obtain the singular value matrix. Introduce the production process constraints in the intelligent manufacturing process to correct the singular value matrix. Use the rank of the corrected singular value matrix to reduce the dimensionality of the weighted time series augmented matrix.

[0014] Step S5: Standardize the dimensionality-reduced weighted temporal augmentation matrix, divide the standardized matrix using a dynamic sliding window, and input it sequentially into the dual-branch attention network. Combined with the current actual production decision scheme, predict the production decision scheme.

[0015] Step S6: If the production decision scheme is local self-cooperation, use the corresponding local digital twin model for scheduling; if the production decision scheme is global scheduling, use the global digital twin model for scheduling.

[0016] Furthermore, the dataset of control indicators under the sampling number is represented as follows:

[0017]

[0018] in, Indicates the number of samples t The following dataset of regulatory indicators z This represents the total number of tasks in the intelligent manufacturing process. i express z index, Indicates the number of samples t Next i The average quality error of each task Indicates the number of samples t Next i The cumulative deviation of each task's cycle. Indicates the number of samples t Next i The cumulative cost deviation for each task.

[0019] Furthermore,

[0020] Number of samples t Next i The average quality error of the task is based on the first t The first data collection during the intelligent manufacturing process i Actual production quality of each task and expected production quality The calculation yielded:

[0021]

[0022] Number of samples t Next i The cumulative deviation of the cycle of the first task is based on the first... t The first data collection during the intelligent manufacturing process i The actual end time of each task and expected end time The calculation yielded:

[0023]

[0024] Number of samples t Next i The cumulative cost deviation of the task is based on the first t The first data collection during the intelligent manufacturing process i The actual production cost of each task and expected production costs The calculation yielded:

[0025]

[0026] in, t Indicates the number of samples t The index.

[0027] Furthermore, step S3 includes the following sub-steps:

[0028] Step S3.1: Construct a time-series augmented matrix from the datasets of control indicators across all sampling periods. ,in, s Indicates the total number of samples;

[0029] Step S3.2: Calculate the first data point of the control index dataset for each sampling number in the time-series augmented matrix. j The mean of the column elements, combined with each element in the time-series augmented matrix, is used to calculate the first... s -1 sampling number and the first s Pearson correlation coefficient for each sampling number:

[0030]

[0031] in, k Indicates the preceding s -1 index for sampling count, Indicates the first s The dataset of control indicators under the sampling number is the first... j Column elements and the first k The dataset of control indicators under the sampling number is the first... j Pearson correlation coefficients of column elements, Indicates the first s The dataset of control indicators under the sampling number is the first... j The mean of the column elements. Indicates the first k The dataset of control indicators under the sampling number is the first... j The mean of the column elements. Indicates the first s The first data point in the control indicator dataset under the sampling number . j Liede i Data corresponding to each task Indicates the first k The dataset of control indicators under the sampling number is the first... j Liede i Data corresponding to each task;

[0032] Step S3.3: [The text appears to be incomplete and contains several grammatical errors. A more accurate translation would require the full context.] Normalization is performed as the time decay weight. A dynamic weight matrix is ​​constructed based on the time decay weight. ,in, diagnosis ( ) denotes a diagonal matrix;

[0033] Step S3.4: Introduce a dynamic weight matrix into the time-series augmented matrix to obtain the weighted time-series augmented matrix. .

[0034] Furthermore, the corrected singular value matrix Represented as:

[0035]

[0036] in, Represents a singular value matrix. The diag() represents the weighting coefficients of the production process constraints. s Indicates the total number of samples. t express s index, j The index represents the control indicators. 1 indicates the average quality error between the actual and expected production quality of all tasks in intelligent manufacturing; 2 indicates the cumulative deviation of the actual and expected completion times of all tasks in intelligent manufacturing; and 3 indicates the cumulative cost deviation between the actual and expected production costs of all tasks in intelligent manufacturing. Indicates the number of samples t Next j The production process constraints corresponding to each control indicator, when j When =1, if the number of samples t The average quality error is greater than the average quality error threshold. ,otherwise, ;when j When =2, if the number of samples t The cumulative period deviation is greater than the cumulative period deviation threshold. ,otherwise, ;when j When =3, if the number of samples t The cumulative cost deviation is greater than the cumulative cost deviation threshold. ,otherwise, .

[0037] Furthermore, the dimensionality reduction process of the weighted temporal augmented matrix is ​​as follows:

[0038]

[0039] in, Y This represents the weighted temporal augmentation matrix after dimensionality reduction. UThis represents an orthogonal matrix obtained by performing singular value decomposition on a weighted time-series augmented matrix. M r express Obtain the rank through row and column swapping. r Equal non-zero column rows in a matrix.

[0040] Furthermore, step S5 includes the following sub-steps:

[0041] Step S5.1: Perform Z-score normalization on the dimensionality-reduced weighted time-series augmented matrix and calculate the first-order difference variance of the Z-score normalized matrix;

[0042] Step S5.2: Adjust the sliding window according to the first-order difference variance to divide the Z-score standardized matrix;

[0043] Step S5.3: Input the partitioned matrix into a one-dimensional convolutional layer to extract local features. Input the extracted local features into an LSTM to capture temporal dependencies and output a hidden state sequence. The first self-attention layer uses the softmax function to calculate the attention weights of the hidden states at each time step, and weights the hidden state sequence to obtain the feature output. ;

[0044] Step S5.4: Encode the current actual production decision scheme into a one-hot vector through the encoding layer to form a decision feature matrix. Obtain the decision feature attention matrix through the second self-attention layer, compress it through the global average pooling layer, and then map it into the decision feature output through the first fully connected layer. ;

[0045] Step S5.5: Output the features and decision feature output The fusion is performed through a fusion layer to obtain fused features. ,in, Indicates feature output Weighting coefficients;

[0046] Step S5.6: Merge features The global scheduling probability is predicted by inputting into the second fully connected layer. ,in, This indicates the bias of the second fully connected layer. This represents the weight matrix of the second fully connected layer;

[0047] Step S5.7: If Adjust the production decision-making plan to global scheduling, if Adjust the production decision-making process to a local self-cooperation model; if =0, continue using the current actual production decision plan.

[0048] Furthermore, in step S5.2, if the first-order difference variance Set the sliding window size to 3×3 and the step size to 1; if the first difference variance Set the sliding window size to 7×7 with a step size of 1; otherwise, set the sliding window size to 5×5 with a step size of 1.

[0049] Furthermore, if the production decision-making scheme is a local self-cooperative one, the specific process of scheduling using the corresponding local digital twin model is as follows:

[0050] Identify candidate intelligent edge industrial control machines for local self-cooperation in each local digital twin model, obtain the expected quality, expected completion time and expected cost of each candidate intelligent edge industrial control machine's control task in the local digital twin model, combine the quality, completion time and cost of the original intelligent manufacturing plan, score the task control performance of each candidate intelligent edge industrial control machine, and use the candidate intelligent edge industrial control machine with the highest score in each task for local scheduling.

[0051] The process of scoring the task control performance of each candidate intelligent edge industrial control computer is as follows:

[0052]

[0053] in, In the local digital twin model, the first... l The first candidate intelligent edge industrial control computer control i The score for each task. m Indicates the index of evaluation indicators. q Indicates quality indicators, d Indicates the end time indicator. c Indicates cost indicators, In the local digital twin model, the first... l The first candidate intelligent edge industrial control computer control i The first task m The expected score of each evaluation indicator In the local digital twin model, the first... l The first candidate intelligent edge industrial control computer control i The first task m The original planned scores of each evaluation indicator, In the local digital twin model, the first... l The first candidate intelligent edge industrial control computer control i The first task m The deviation between the expected score and the original planned score for each evaluation indicator. In the local digital twin model, the first... mThe weighting coefficients of each evaluation indicator. Furthermore, if the production decision-making scheme is global scheduling, the specific process of scheduling using a global digital twin model is as follows:

[0054] The expected quality, expected end time, and expected cost of each candidate intelligent edge industrial control machine's control task are obtained in the global scheduling model. Combined with the quality, end time, and cost of the original intelligent manufacturing plan, the task control performance of each candidate intelligent edge industrial control machine is scored. The candidate intelligent edge industrial control machine with the highest score under each task is used for global scheduling.

[0055] The process of scoring the task control performance of each candidate intelligent edge industrial control computer is as follows:

[0056]

[0057] in, In the global digital twin model, the first... L The first candidate intelligent edge industrial control computer control i The score for each task. m Indicates the index of evaluation indicators. q Indicates quality indicators, d Indicates the end time indicator. c Indicates cost indicators, In the global digital twin model, the first... L The first candidate intelligent edge industrial control computer control i The first task m The expected score of each evaluation indicator In the global digital twin model, the first... L The first candidate intelligent edge industrial control computer control i The first task m The original planned scores of each evaluation indicator, In the global digital twin model, the first... L The first candidate intelligent edge industrial control computer control i The first task m The deviation between the expected score and the original planned score for each evaluation indicator. In the global digital twin model, the first... m The weighting coefficients of each evaluation indicator.

[0058] Compared with the prior art, the present invention has the following beneficial effects:

[0059] (1) The intelligent manufacturing collaborative decision-making method based on digital twins in this invention calculates the pre-processing timeline before... s -1 sampling number and the first sThe Pearson correlation coefficients under each sampling number are used to construct a dynamic weight matrix. The time-series augmented matrix is ​​then adjusted to a weighted time-series augmented matrix, effectively capturing the time-series correlation and dynamic evolution of the control indicators. By quantifying the correlation of the control indicators under each sampling number, the weights of the control indicator data are dynamically adjusted to strengthen key time-series characteristics. This provides data support for collaborative scheduling decisions that reflects the dynamic evolution of the intelligent manufacturing production process, avoiding information lag caused by static weighting.

[0060] (2) The intelligent manufacturing collaborative decision-making method based on digital twins of the present invention corrects the singular value matrix by introducing production process constraints in the intelligent manufacturing process, so as to ensure that the dimensionality reduction of the subsequent weighted time-series augmented matrix can both retain the core process features and conform to the feasibility boundary of actual production, thereby improving the response accuracy and realizing the precision and efficiency of digital twin collaborative decision-making.

[0061] (3) This invention extracts local high-frequency disturbance features and global temporal trend features in the intelligent manufacturing process by standardizing the weighted time-series augmented matrix of the dimension reduction. This improves the accuracy of feature extraction and the dynamic adaptation of the dual-branch attention network, reduces computational redundancy and energy consumption, optimizes collaborative decision-making efficiency, and uses a dynamic sliding window to divide the standardized matrix to adaptively match the complexity features of production data, thereby improving the flexibility and pertinence of feature extraction. Combined with historical decision-making experience, the dual-branch attention network is used to predict production decision schemes, which can improve the accuracy and dynamic adaptability of collaborative scheduling, thereby realizing accurate and rapid decision-making in the intelligent manufacturing process.

[0062] In summary, the intelligent manufacturing collaborative decision-making method based on digital twins of this invention can effectively reduce data processing energy consumption in the intelligent manufacturing process, greatly shorten production time, and reduce data transmission latency. Attached Figure Description

[0063] Figure 1 This is a flowchart of the intelligent manufacturing collaborative decision-making method based on digital twins according to the present invention;

[0064] Figure 2 This is a schematic diagram of the dual-branch attention network in this invention. Detailed Implementation

[0065] The technical solution of the present invention will be further explained and described below with reference to the accompanying drawings.

[0066] like Figure 1 This is a flowchart of the intelligent manufacturing collaborative decision-making method based on digital twins according to the present invention. The intelligent manufacturing collaborative decision-making method includes the following steps:

[0067] Step S1: Establish a global digital twin model and a local digital twin model based on the intelligent manufacturing process; the local digital twin model is used for decision-making and scheduling of local intelligent edge industrial control computers to achieve local self-cooperation; the global digital twin model is used to realize the global scheduling of all intelligent edge industrial control computers in the intelligent manufacturing process.

[0068] Step S2: Real-time collection of production data during the intelligent manufacturing process to determine the control index dataset for the number of samplings. The control index dataset includes: average quality error, cumulative cycle deviation, and cumulative cost deviation. This allows for coordinated consideration of "quality-efficiency-cost" in the intelligent manufacturing process, improving the accuracy of local self-cooperation and global scheduling collaborative decision-making, shortening production response time, and reducing data transmission latency. Specifically:

[0069] According to the t The first data collection during the intelligent manufacturing process i Actual production quality of each task and expected production quality Calculate the number of samples t Next i Average quality error of each task ;

[0070] According to the t The first data collection during the intelligent manufacturing process i The actual end time of each task and expected end time Calculate the number of samples t Next i Cumulative deviation of each task's cycle ;

[0071] According to the t The first data collection during the intelligent manufacturing process i The actual production cost of each task and expected production costs Calculate the number of samples t Next i Cumulative cost deviation for each task ;

[0072] Number of samples t All tasks , , Composition of regulatory indicator dataset ,in, t Indicates the number of samples t The index.

[0073] Step S3: Construct a time-series augmented matrix for the control indicator datasets from all sampling times, and calculate the first... s-1 sampling number and the first s The Pearson correlation coefficients for each sampling number are used to construct a dynamic weight matrix. The time-series augmented matrix is ​​then adjusted to a weighted time-series augmented matrix, effectively capturing the time-series correlation and dynamic evolution of regulatory indicators. By quantifying the linear correlation strength of regulatory indicators under different sampling numbers, capturing the time-varying characteristics of the correlation, and reducing noise interference through centering, the accuracy of capturing the dynamic patterns of regulatory indicators is effectively improved. This provides support for extracting time-series evolution patterns, reduces redundant parameters to improve computational efficiency, identifies coupling risks of regulatory indicators in advance, enhances decision robustness, and provides accurate, efficient, and robust technical support for collaborative decision-making. Specifically, the following sub-steps are included: Step S3.1: Construct a time-series augmented matrix for the regulatory indicator datasets under all sampling numbers. ,in, s Indicates the total number of samples;

[0074] Step S3.2: Calculate the first data point of the control index dataset for each sampling number in the time-series augmented matrix. j The mean of the column elements, combined with each element in the time-series augmented matrix, is used to calculate the first... s -1 sampling number and the first s Pearson correlation coefficient for each sampling number:

[0075]

[0076] in, k Indicates the preceding s -1 index for sampling count, Indicates the first s The dataset of control indicators under the sampling number is the first... j Column elements and the first k The dataset of control indicators under the sampling number is the first... j Pearson correlation coefficients of column elements, Indicates the first s The dataset of control indicators under the sampling number is the first... j The mean of the column elements. Indicates the first k The dataset of control indicators under the sampling number is the first... j The mean of the column elements. Indicates the first s The first data point in the control indicator dataset under the sampling number . j Liede i Data corresponding to each task Indicates the first k The dataset of control indicators under the sampling number is the first... j Liede i Data corresponding to each task;

[0077] Step S3.3: [The text appears to be incomplete and contains several grammatical errors. A more accurate translation would require the full context.] Normalization is performed as the time decay weight. A dynamic weight matrix is ​​constructed based on the time decay weight. ,in, diagnosis ( ) denotes a diagonal matrix;

[0078] Step S3.4: Introduce a dynamic weight matrix into the time-series augmented matrix to obtain the weighted time-series augmented matrix. .

[0079] Step S4: To address the computational complexity, overfitting, and ambiguity of physical meaning caused by high-dimensional data redundancy in intelligent manufacturing, this invention performs singular value decomposition (SVD) on the weighted time-series augmented matrix to obtain a singular value matrix (SVM). Production process constraints from intelligent manufacturing are then introduced to correct the SVM, ensuring that subsequent dimensionality reduction of the weighted time-series augmented matrix is ​​both mathematically optimized and physically feasible. This preserves key process characteristics in intelligent manufacturing, improving response accuracy and achieving precise and efficient digital twin collaborative decision-making. The rank of the corrected SVM is used to reduce the dimensionality of the weighted time-series augmented matrix, enabling the reduced matrix to support collaborative decision-making for both global scheduling and local self-cooperation, thereby reducing energy consumption and data transmission latency in intelligent manufacturing. Specifically, this includes the following sub-steps: Step S4.1: Perform SVD on the weighted time-series augmented matrix to obtain the orthogonal matrix and the SVM:

[0080]

[0081] in, U and V They represent orthogonal matrices, M Represents a singular value matrix. T Indicates transpose;

[0082] Step S4.2: Introduce production process constraints from the intelligent manufacturing process to correct the singular value matrix. The corrected singular value matrix... Represented as:

[0083]

[0084] in, Represents a singular value matrix. The weighting coefficients represent the production process constraints, with values ​​ranging from (0,1], and are determined by domain experts; diag() represents a diagonal matrix. s Indicates the total number of samples. t express s index, jThe index represents the control indicators. 1 indicates the average quality error between the actual and expected production quality of all tasks in intelligent manufacturing; 2 indicates the cumulative deviation of the actual and expected completion times of all tasks in intelligent manufacturing; and 3 indicates the cumulative cost deviation between the actual and expected production costs of all tasks in intelligent manufacturing. Indicates the number of samples t Next j The production process constraints corresponding to each control indicator, when j When =1, if the number of samples t The average quality error is greater than the average quality error threshold. This is to increase the modified singular value matrix and strengthen quality control; otherwise, ;when j When =2, if the number of samples t The cumulative period deviation is greater than the cumulative period deviation threshold. Reduce the modified singular value matrix to trigger scheduling to accelerate production; otherwise, ;when j When =3, if the number of samples t The cumulative cost deviation is greater than the cumulative cost deviation threshold. Reduce the modified singular value matrix to trigger scheduling to accelerate production; otherwise, ;

[0085] Step S4.3: Dimensionality reduction is performed using the rank-pair weighted time-series augmented matrix of the corrected singular value matrix:

[0086]

[0087] in, Y This represents the weighted temporal augmentation matrix after dimensionality reduction. U This represents an orthogonal matrix obtained by performing singular value decomposition on a weighted time-series augmented matrix. M r express Obtain the rank through row and column swapping. r Equal non-zero column rows in a matrix.

[0088] Step S5: To facilitate the extraction of local high-frequency disturbance features and global temporal trend features that characterize the intelligent manufacturing process, improve the accuracy of feature extraction and the dynamic adaptation of the dual-branch attention network, reduce computational redundancy and energy consumption, and optimize collaborative decision-making efficiency, this invention standardizes the dimensionality-reduced weighted temporal augmented matrix, uses a dynamic sliding window to divide the standardized matrix, adaptively matches the complexity features of production data, improves the flexibility and targeting of feature extraction, and sequentially inputs it into the dual-branch attention network. Combined with the current actual production decision-making scheme, it predicts the production decision-making scheme. By extracting the features of control indicators and using historical decision-making experience for collaborative modeling, it can improve the accuracy and dynamic adaptability of collaborative scheduling, thereby achieving accurate and rapid decision-making in the intelligent manufacturing process; including the following sub-steps:

[0089] Step S5.1: Perform Z-score normalization on the dimensionality-reduced weighted time-series augmented matrix and calculate the first-order difference variance of the Z-score normalized matrix. :

[0090]

[0091] in, d This indicates the row number of an element in the Z-score normalized matrix. h express d index, g express r index, Represents the Z-score normalized matrix. h Line 1 g Column elements, This represents the mean of the indexed elements in the Z-score normalized matrix;

[0092] Step S5.2: Using the first-order difference variance as a feature complexity index, adjust the sliding window to divide the Z-score-normalized matrix. This invention adaptively adjusts the window size based on the complexity index, balancing the extraction of local details and global trends, enhancing the adaptability of features to dynamic production scenarios, and providing high-quality, low-redundancy feature input for scheduling decisions; specifically, if the first-order difference variance... This indicates that the data in the Z-score standardized matrix fluctuates greatly and has high complexity. Setting the sliding window size to 3×3 with a step size of 1 allows for focusing on local details. If the first-order difference variance... If the result is positive, it indicates that the data in the Z-score-normalized matrix has small fluctuations and low complexity. Setting the sliding window size to 7×7 with a step size of 1 will expand the window to capture more information. Otherwise, it indicates that the data complexity in the Z-score-normalized matrix is ​​moderate. Setting the sliding window size to 5×5 with a step size of 1 will balance local and global information.

[0093] Step S5.3: As Figure 2 The partitioned matrix is ​​sequentially input into a one-dimensional convolutional layer to extract local features. These extracted local features are then input into an LSTM to capture temporal dependencies, outputting a hidden state sequence. The first self-attention layer uses the softmax function to calculate the attention weights of the hidden states at each time step, weighting the hidden state sequence to obtain the feature output. ;

[0094] Step S5.4: Encode the current actual production decision scheme into a one-hot vector through the encoding layer to form a decision feature matrix. Obtain the decision feature attention matrix through the second self-attention layer to capture the correlation of historical decisions. Compress the matrix through the global average pooling layer and then map it to the decision feature output through the first fully connected layer. ;

[0095] Step S5.5: Output the features and decision feature output The fusion is performed through a fusion layer to obtain fused features. ,in, Indicates feature output Weighting coefficients;

[0096] Step S5.6: Merge features The global scheduling probability is predicted by inputting into the second fully connected layer. ,in, This indicates the bias of the second fully connected layer. This represents the weight matrix of the second fully connected layer;

[0097] Step S5.7: If Adjust the production decision-making plan to global scheduling, if Adjust the production decision-making process to a local self-cooperation model; if =0, continue using the current actual production decision plan.

[0098] Step S6: If the production decision scheme is local self-cooperation, use the corresponding local digital twin model for scheduling; if the production decision scheme is global scheduling, use the global digital twin model for scheduling.

[0099] The scheduling process using the corresponding local digital twin model in this invention is as follows: Candidate intelligent edge control machines for local self-cooperation in each local digital twin model are determined. The expected quality, expected completion time, and expected cost of each candidate intelligent edge control machine's control task are obtained. Combined with the quality, completion time, and cost of the original intelligent manufacturing plan, the task control performance of each candidate intelligent edge control machine is scored. The candidate intelligent edge control machine with the highest score in each task is used for local scheduling. The candidate intelligent edge control machines are scored using quality, time, and cost to achieve dynamic balance and collaborative optimization among the three factors, thereby improving the objectivity, scientific nature, and adaptability to intelligent manufacturing scenarios of the scheduling decision.

[0100] The process of scoring the task control performance of each candidate intelligent edge industrial control computer is as follows:

[0101]

[0102] in, In the local digital twin model, the first... l The first candidate intelligent edge industrial control computer control i The score for each task. m Indicates the index of evaluation indicators. q Indicates quality indicators, d Indicates the end time indicator. c Indicates cost indicators, In the local digital twin model, the first... l The first candidate intelligent edge industrial control computer control i The first task m The expected score of each evaluation indicator In the local digital twin model, the first... l The first candidate intelligent edge industrial control computer control i The first task m The original planned scores of each evaluation indicator, In the local digital twin model, the first... l The first candidate intelligent edge industrial control computer control i The first task m The deviation between the expected score and the original planned score for each evaluation indicator. In the local digital twin model, the first... m The weighting coefficients of each evaluation indicator.

[0103] The scheduling process using a global digital twin model in this invention is as follows:

[0104] The expected quality, expected end time, and expected cost of each candidate intelligent edge industrial control machine's control task are obtained in the global scheduling model. Combined with the quality, end time, and cost of the original intelligent manufacturing plan, the task control performance of each candidate intelligent edge industrial control machine is scored. The candidate intelligent edge industrial control machine with the highest score under each task is used for global scheduling.

[0105] The process of scoring the task control performance of each candidate intelligent edge industrial control computer is as follows:

[0106]

[0107] in, In the global digital twin model, the first... L The first candidate intelligent edge industrial control computer control i The score for each task. m Indicates the index of evaluation indicators. q Indicates quality indicators, d Indicates the end time indicator. c Indicates cost indicators, In the global digital twin model, the first... L The first candidate intelligent edge industrial control computer control i The first task m The expected score of each evaluation indicator In the global digital twin model, the first... L The first candidate intelligent edge industrial control computer control i The first task m The original planned scores of each evaluation indicator, In the global digital twin model, the first... L The first candidate intelligent edge industrial control computer control i The first task m The deviation between the expected score and the original planned score for each evaluation indicator. In the global digital twin model, the first... m The weighting coefficients of each evaluation indicator.

[0108] In one technical solution of this invention, the intelligent edge industrial control unit forms a digital twin hub, requiring service functions such as communication, adaptation, analysis, and control. Furthermore, the intelligent edge industrial control unit supports mainstream communication protocols such as UDP, TCP, OPC UA, and MQTT, and can run high-level programming software program instructions such as Matlab and Python. ms Level-one real-time response capability.

[0109] This invention addresses the problems of "imbalance in global modeling functionality, delays in cloud-centralized decision-making, and lack of local carrier architecture" in traditional digital twin decision-making. It constructs a two-tiered collaborative architecture that integrates global and local digital twins, combining lightweight data processing and intelligent decision-making to achieve collaborative decision-making and scheduling in intelligent manufacturing processes.

[0110] The global digital twin model achieves macro-level scheduling through the virtual-physical mapping of the virtual production line and the physical production line. The local digital twin model relies on the intelligent edge industrial control computer to build a distributed decision-making unit to support device-level autonomous collaboration. The two achieve real-time data interaction through a digital twin hub with integrated communication and adaptation functions, solving the edge node load problem caused by "hard coupling" deployment.

[0111] At the data processing level, this invention reduces the dimensionality of the weighted temporal augmented matrix by introducing a dynamic weight matrix and a singular value matrix modified by production process constraints. Combined with historical decision-making experience, it uses a dual-branch attention network to quickly output the probability of global scheduling in order to determine whether to adopt a global scheduling or local self-cooperation decision scheme, which can shorten the response time to the millisecond level.

[0112] In the decision-making scheme of global scheduling and local self-cooperation, a scoring model for candidate intelligent edge industrial control machines is constructed using multiple indicators such as quality, time and cost. The machine with the highest score is selected as the scheduler, which effectively reduces the processing energy consumption in the intelligent manufacturing process and solves the problems of decision lag, excessive energy consumption and transmission delay in the traditional architecture.

[0113] In one technical solution of the present invention, a computer-readable storage medium is also provided, storing a computer program that enables a computer to execute the intelligent manufacturing collaborative decision-making method based on digital twins of the present invention.

[0114] In one technical solution of the present invention, an electronic device is also provided, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the intelligent manufacturing collaborative decision-making method based on digital twins of the present invention.

[0115] In the embodiments disclosed in this application, a computer storage medium may be a tangible medium that may contain or store programs for use by or in conjunction with an instruction execution system, apparatus, or device. The computer storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of computer storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0116] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0117] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should be considered within the scope of protection of the present invention.

Claims

1. An intelligent manufacturing collaborative decision-making method based on digital twinning, characterized in that, The method comprises the following steps: Step S1: establishing a global digital twin model and a local digital twin model according to a smart manufacturing process; Step S2: collecting production data in the smart manufacturing process in real time, determining a control index data set under a sampling number, and representing as diag in, Indicates the number of samples t The following dataset of regulatory indicators z This represents the total number of tasks in the intelligent manufacturing process. i express z index, Indicates the number of samples t Next i The average quality error of each task Indicates the number of samples t Next i The cumulative deviation of each task's cycle. Indicates the number of samples t Next i Cumulative cost deviation for each task; Step S3: Construct a time series augmented matrix for the regulation index data set under all sampling times, respectively calculate the Pearson correlation coefficients under the first s -1 sampling times and the first s -1 sampling times, construct a dynamic weight matrix, and adjust the time series augmented matrix to a weighted time series augmented matrix; Step S3 includes the following substeps: Step S3.1: Constructing the time series augmented matrix of the regulation index dataset under all sampling times wherein, s denotes the total number of sampling times; Step S3.2: Calculate the mean of the column elements of the time- augmented matrix, combine each element of the time-augmented matrix with the mean of the column elements, and calculate the Pearson correlation coefficient between the first j sample and the first s sample, respectively. s sample, respectively. in, k Indicates the preceding s -1 index for sampling count, Indicates the first s The dataset of control indicators under the sampling number is the first... j Column elements and the first k The dataset of control indicators under the sampling number is the first... j Pearson correlation coefficients of column elements Indicates the first s The dataset of control indicators under the sampling number is the first... j The mean of the column elements. Indicates the first k The dataset of control indicators under the sampling number is the first... j The mean of the column elements. Indicates the first s The first data point in the control index dataset under the sampling number . j Liede i Data corresponding to each task Indicates the first k The dataset of control indicators under the sampling number is the first... j Liede i Data corresponding to each task; Step S3.3: Constructing the dynamic weight matrix Wt Performing normalization as time decay weights and constructing the dynamic weight matrix Wt according to the time decay weights wherein Step S4: performing singular value decomposition on the weighted time sequence augmented matrix to obtain a singular value matrix, introducing production process constraints in the smart manufacturing process to modify the singular value matrix, and performing dimension reduction on the weighted time sequence augmented matrix by using a rank of the modified singular value matrix; ( ) denotes a diagonal matrix; Step S3.4: Introducing a dynamic weight matrix in the timing augmented matrix to obtain a weighted timing augmented matrix ; Step S5: standardizing the dimension-reduced weighted time sequence augmented matrix, dividing the standardized matrix by using a dynamic sliding window, inputting the matrix into a double-branch attention network in sequence, combining a current actual production decision scheme, and predicting a production decision scheme; Step S6: if the production decision scheme is local self-collaboration, performing scheduling by using a corresponding local digital twin model; or if the production decision scheme is global scheduling, performing scheduling by using a global digital twin model.

2. The smart manufacturing collaborative decision method based on a digital twin according to claim 1, characterized in that, τ Sampling times t The average quality error of the next task is calculated according to the actual production quality and the expected production quality of the first task in the intelligent manufacturing process collected the first time. i τ i ​​​​​ Number of samples t Next i The cumulative deviation of the cycle of the first task is based on the first... τ The first data collection during the intelligent manufacturing process i The actual end time of each task and expected end time The calculation yielded: Sampling times t The next task cost cumulative deviation is calculated according to the actual production cost and the expected production cost of the first task in the intelligent manufacturing process collected the first time. i τ i ​​​​​ wherein The dimension reduction process of the weighted time sequence augmented matrix is as follows: represents the number of samples t index of the sample.

3. The intelligent manufacturing collaborative decision-making method based on digital twinning according to claim 1, characterized in that, the modified singular value matrix is expressed as: wherein, represents a singular value matrix, represents a weight coefficient of a production process constraint, diag() represents a diagonal matrix, s represents a total number of sampling, t represents s an index of, j represents an index of a control index, 1 represents an average quality error of actual production quality and expected production quality of all tasks in intelligent manufacturing, 2 represents a period cumulative deviation of actual end time and expected end time of all tasks in intelligent manufacturing, 3 represents a cost cumulative deviation of actual production cost and expected production cost of all tasks in intelligent manufacturing, represents a sampling number t The production process constraint corresponding to the first j control index under the condition that j =1, if the average quality error under the sampling number t is greater than the average quality error threshold, , otherwise, ; when j =2, if the period cumulative deviation under the sampling number t is greater than the period cumulative deviation threshold, , otherwise, ; when j =3, if the cost cumulative deviation under the sampling number t is greater than the cost cumulative deviation threshold, , otherwise, .

4. The intelligent manufacturing collaborative decision-making method based on digital twinning according to claim 3, characterized in that, Step S5 comprises the following sub-steps: wherein, Y denotes a reduced dimension weighted time-augmented matrix, U denotes an orthogonal matrix of singular value decomposition of the weighted time-augmented matrix, M r denotes a column non-zero row matrix with rank r equal to the rank 5. The intelligent manufacturing collaborative decision-making method based on digital twinning according to claim 1, characterized in that, Step S5.1: performing Z-score standardization on the dimension-reduced weighted time sequence augmented matrix, and calculating a first-order difference variance of the Z-score standardized matrix; Step S5.2: adjusting a sliding window according to the first-order difference variance, and dividing the Z-score standardized matrix; If the production decision scheme is local self-collaboration, the specific process of performing scheduling by using a corresponding local digital twin model is as follows: Step S5.3: input the divided matrix into a one-dimensional convolution layer in sequence to extract local features, input the extracted local features into an LSTM to capture time sequence dependency, output a hidden state sequence, and a first self-attention layer uses a softmax function to calculate attention weights of hidden states at each time step, weights the hidden state sequence, and obtains a feature output ; Step S5.4: encode the current actual production decision scheme into a one-hot vector through the encoding layer to form a decision feature matrix, obtain a decision feature attention matrix through a second self-attention layer, compress through a global average pooling layer, and then map to a decision feature output through a first fully connected layer ; Step S5.5: output the feature and the decision feature fusion through the fusion layer to obtain a fused feature wherein, denotes a weight coefficient of the feature output ​ Step S5.6: fusing the features In the input second fully connected layer, the global dispatch probability is predicted wherein, denotes a bias of the second fully connected layer, denotes a weight matrix of the second fully connected layer; Step S5.7: If the production decision scheme is adjusted to a global dispatch, if the production decision scheme is adjusted to a local self-coordination; and if =0, the current actual production decision scheme is continued to be used.

6. The intelligent manufacturing collaborative decision-making method based on digital twinning according to claim 5, characterized in that, If the first-order difference variance in step S5.2 , the size of the sliding window is set to 3x3 and the step is 1; if the first-order difference variance , the size of the sliding window is set to 7x7 and the step is 1; otherwise, the size of the sliding window is set to 5x5 and the step is 1.

7. The intelligent manufacturing collaborative decision-making method based on digital twinning according to claim 1, characterized in that, determining a candidate intelligent edge industrial computer for local self-collaboration of each local digital twin model, obtaining expected quality, expected end time and expected cost of a control task of each candidate intelligent edge industrial computer in the local digital twin model, combining quality, end time and cost of an original smart manufacturing plan, scoring a task control situation of each candidate intelligent edge industrial computer, and using a candidate intelligent edge industrial computer with the highest score under each task for local scheduling; The process of scoring the task control situation of each candidate intelligent edge industrial computer is as follows: If the production decision scheme is global scheduling, the specific process of performing scheduling by using a global digital twin model is as follows: in, In the local digital twin model, the first... l The first candidate intelligent edge industrial control computer control i The score for each task. m Indicates the index of evaluation indicators. q Indicates quality indicators, d Indicates the end time indicator. c Indicates cost indicators, In the local digital twin model, the first... l The first candidate intelligent edge industrial control computer control i The first task m The expected scores for each evaluation indicator In the local digital twin model, the first... l The first candidate intelligent edge industrial control computer control i The first task m The original planned scores of each evaluation indicator, In the local digital twin model, the first... l The first candidate intelligent edge industrial control computer control i The first task m The deviation between the expected score and the original planned score for each evaluation indicator. In the local digital twin model, the first... m The weighting coefficients of each evaluation indicator.

8. The intelligent manufacturing collaborative decision-making method based on digital twinning according to claim 1, characterized in that, obtaining expected quality, expected end time and expected cost of a control task of each candidate intelligent edge industrial computer in the global scheduling model, combining quality, end time and cost of an original smart manufacturing plan, scoring a task control situation of each candidate intelligent edge industrial computer, and using a candidate intelligent edge industrial computer with the highest score under each task for global scheduling; The process of scoring the task control situation of each candidate intelligent edge industrial computer is as follows: ​ in, In the global digital twin model, the first... L The first candidate intelligent edge industrial control computer control i The score for each task. m Indicates the index of evaluation indicators. q Indicates quality indicators, d Indicates the end time indicator. c Indicates cost indicators, In the global digital twin model, the first... L The first candidate intelligent edge industrial control computer control i The first task m The expected scores for each evaluation indicator In the global digital twin model, the first... L The first candidate intelligent edge industrial control computer control i The first task m The original planned scores of each evaluation indicator, In the global digital twin model, the first... L The first candidate intelligent edge industrial control computer control i The first task m The deviation between the expected score and the original planned score for each evaluation indicator. In the global digital twin model, the first... m The weighting coefficients of each evaluation indicator.

Images