A digital twin workshop management and control method and system based on multi-layer data fusion

Abstract

This invention relates to a digital twin workshop management and control method and system based on multi-layer data fusion, belonging to the field of service-oriented industrial system data science. The method includes: acquiring multi-source heterogeneous data across the entire digital twin workshop; classifying, processing, and storing the multi-source heterogeneous data; performing data-level fusion to obtain standard data for the digital twin workshop; extracting the standard data; performing semantic association, feature extraction, and feature analysis on the standard data; performing feature-level fusion to obtain key state features of the digital twin workshop; constructing a decision model for the digital twin workshop; inputting key operational features of the digital twin workshop; solving based on the decision model, decision optimization, and decision methods; performing decision-level fusion; and outputting workshop production management and control decisions. This invention can provide support for the governance and analysis of digital twin workshop operational data and accurate production management and control decisions.

Description

Technical Field

[0001] This invention belongs to the field of service-oriented industrial system data science, specifically relating to a digital twin workshop management method and system based on multi-layer data fusion. Background Technology

[0002] With the continuous development of modern manufacturing, workshop production environments are becoming increasingly complex, especially in discrete manufacturing workshops. The uncertainty and dynamism of the production process require enterprises to have greater production flexibility and control precision. Traditional workshop management models suffer from data silos, with equipment status, production plans, and material flow data scattered across independent systems. Format differences make data fusion difficult. Simultaneously, the production patterns hidden within massive amounts of data are not fully explored, making it difficult to support accurate decision-making. Static scheduling strategies based on experience rules cannot adapt to real-time dynamic production changes. To address these issues, in recent years, academia and industry have proposed various intelligent manufacturing and industrial internet solutions. These solutions introduce technologies such as big data, cloud computing, and the Internet of Things (IoT) to attempt to achieve data fusion and intelligent control in workshop management. Digital twins represent a new paradigm for workshop data fusion and control. By constructing a virtual mirror of the physical workshop, digital twins enable real-time mapping and optimized decision-making of the production process. However, existing digital twin workshop data fusion systems still have the following limitations: First, a data fusion paradigm is lacking; sensor data at the physical layer and simulation data at the virtual layer lack unified representation and processing methods, resulting in low efficiency in digital twin-physical collaboration. Second, they focus primarily on physical domain feature extraction, neglecting multi-domain virtual-physical data mining, making it difficult to extract high-value feature information. Third, optimization schemes generated in the virtual environment lack feedback on physical execution effects, making it difficult to form a continuous improvement twin optimization mechanism and flexibly adjust according to the production situation in the physical workshop. These problems collectively lead to bottlenecks in data control and decision-making in digital twin workshops. Summary of the Invention

[0003] To address the aforementioned issues, this invention proposes a digital twin workshop management and control method and system based on multi-layer data fusion. The aim is to enhance data management and control capabilities during the digital twin workshop production process through multi-level data fusion. Effective data-level, feature-level, and decision-level fusion methods are designed to uncover the application value of multi-source heterogeneous data in the digital twin workshop, achieving accurate and reliable decision support. Through these technical means, this invention realizes full-process, full-domain data management and control in the digital twin workshop, providing effective technical support for multi-source heterogeneous data management and intelligent dynamic production decision-making in the digital twin workshop.

[0004] The technical solution adopted by the present invention to solve the above technical problems is as follows:

[0005] A digital twin workshop management and control method based on multi-layer data fusion includes the following steps:

[0006] Step 1: Acquire multi-source heterogeneous data across the entire digital twin workshop, classify, process, and store the multi-source heterogeneous data, and perform data-level fusion to obtain standard data for the digital twin workshop;

[0007] Step 2: Extract standard data for the digital twin workshop, perform semantic association, feature extraction, and feature analysis on the standard data for the digital twin workshop, and perform feature-level fusion to obtain the key state features of the digital twin workshop;

[0008] Step 3: Construct a decision model for the digital twin workshop, input the key state features of the digital twin workshop, solve the problem based on the decision model, decision optimization, and decision method, perform decision-level fusion, and output the workshop production control decision.

[0009] A digital twin workshop control system based on multi-layer data fusion includes:

[0010] The data-level fusion module takes multi-source heterogeneous data from the digital twin workshop as input and outputs standard data for the digital twin workshop, including:

[0011] The data classification module is used to globally acquire and classify entries of the multi-source heterogeneous dataset of the digital twin workshop;

[0012] The data processing module is used to perform preliminary preprocessing of the data, filtering out noise and duplicate values.

[0013] The data storage module is used to standardize and hierarchically store data to obtain standardized digital twin workshop data;

[0014] The feature-level fusion module takes standard data from the digital twin workshop as input and outputs key state features of the digital twin workshop, including:

[0015] The data association module is used to perform semantic association of virtual and real data based on ontology models on heterogeneous models of digital twin workshops, and to build a global view of digital twin workshop data;

[0016] The feature extraction module is used to extract and represent multi-domain features from the global data of the digital twin workshop.

[0017] The feature analysis module is used to mine key features of the digital twin workshop and analyze and characterize the workshop's operating status.

[0018] The decision-level fusion module takes as input the key state characteristics of the digital twin workshop and outputs workshop production control decisions, including:

[0019] Decision Model Module: Used to build a model of decision-making relationships in the digital twin workshop, serving as input for the decision-making object;

[0020] The decision optimization module is used to define decision constraints, decision conditions, and decision objectives, and to optimize the decision model.

[0021] The decision-making method solution module is used to select appropriate decision-making methods, analyze and solve decision problems, and obtain production control decisions.

[0022] The present invention has the following beneficial effects:

[0023] 1. Data control level: By designing a standardized modeling and representation method for the entire workshop data, data-level fusion is achieved, enabling data control of the entire process and all areas of the digital twin workshop, providing a reliable data foundation for subsequent analysis and decision-making.

[0024] 2. Analysis and Decision-Making Level: Design a multi-dimensional data feature mining method to achieve feature-level fusion and further explore the value of data; and design a decision optimization method for data enhancement of digital twin workshop for effective features to achieve decision-level fusion, provide accurate decision-making for production control, and help meet the status prediction needs and production scheduling goals of 3C digital twin workshop.

[0025] 3. Production efficiency level: It provides a theoretical basis for enhancing the production control efficiency of digital twin workshops, helps to improve the flexibility and accuracy of production, improve the production efficiency of workshops, meet the needs of discrete production mode, has high practical application value, and promotes the application of digital twin technology in the 3C industry and the development of new quality productivity. Attached Figure Description

[0026] Figure 1 This is a flowchart illustrating the steps of a digital twin workshop management method based on multi-layer data fusion according to the present invention.

[0027] Figure 2 This is a module relationship diagram of a digital twin workshop control system based on multi-layer data fusion according to the present invention. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other. To achieve the above objectives, this invention adopts the following technical solution.

[0029] This invention uses a 3C workshop manufacturing system as an example to illustrate the method of this invention. The 3C workshop manufacturing system faces uncertainties in the production process and requires higher control precision. The 3C workshop model is multi-layered and coupled, discrete and flexible, with internal data from multiple sources and heterogeneous structures.

[0030] like Figure 1 As shown, the technical solution adopted by this invention is: a digital twin workshop management method based on multi-layer data fusion, comprising the following steps:

[0031] Step 1: Acquire multi-source heterogeneous data from the entire digital twin workshop and synchronize it to the virtual model of the digital twin workshop to establish a physical-virtual data mapping relationship. Based on the characteristics of the multi-source heterogeneous data and the complex system hierarchy of the digital twin workshop, sort out the internal elements and data content of the digital twin workshop and classify the data; then perform preliminary preprocessing of the data to remove noise and duplicate values; finally, store the data in a standard computer format to obtain standard digital twin workshop data, and perform data-level fusion based on data classification, data processing, and data storage.

[0032] Step 2: Extract standard data from the digital twin workshop, perform heterogeneous data semantic association on the physical entities, virtual entities, and interaction relationships of the digital twin workshop, and construct global data for the digital twin workshop; extract multi-domain features from the global data of the digital twin workshop; combine deep learning algorithms to perform feature analysis on the equipment's temperature, speed, energy consumption, and product processing time, accuracy, and material consumption features in the multi-domain features, and obtain key state features of the digital twin workshop through time series prediction, such as equipment failure warning time, equipment maintenance time, and key process parameter deviations for product quality fluctuations. Based on data semantic association, feature extraction, and feature analysis, perform feature-level fusion;

[0033] Step 3: Construct a decision model for the digital twin workshop. This model includes parameterized decision indicators for various variables required for decision-making. These variables are the key state characteristics of the digital twin workshop obtained in Step 2, such as equipment failure warning time and equipment maintenance time. Select specific decision-making methods to solve problems, define physical constraints, twin constraints, decision conditions, and decision objectives for decision optimization. Select appropriate decision-making methods to achieve enhanced decision results from the digital twin. Based on the decision model, decision optimization, and decision-making method solutions, perform decision-level fusion to output workshop production control decisions.

[0034] The detailed process of data-level fusion is as follows:

[0035] Acquire multi-source heterogeneous data across the entire digital twin workshop, including: acquiring all production data from the digital twin workshop. This analysis examines the data composition of the digital twin workshop. The data sources for the 3C physical workshop include Equipment Management System (DMS), Material Management System (WMS), Environmental Management System (EMS), Public Network (PN), Enterprise Resource Planning (ERP), and Manufacturing Execution System (MES). Among these, production data... Including data objects, data virtual and real, data status, and data structures, the above data are classified in detail according to their corresponding attributes to obtain the global data composition. ,in Indicates the twin data field type, The domain represents twin-workshop data. The field represents physical workshop data. The number of data source system categories. For the first Data entries within an information system;

[0036] Data preprocessing includes: performing deduplication and noise reduction operations on the collected digital twin workshop data to obtain high-quality twin data. Specific steps include:

[0037] Data imputation: Regression imputation was used, with the missing attribute as the dependent variable and the other attributes as independent variables. A multiple linear regression model was established between the missing attribute and the other attributes. The model is expressed as follows:

[0038] ,

[0039] in, For estimated missing attribute values, such as product defect rate, These are estimates of the multiple regression coefficients. For attributes related to missing attributes, such as product size, product temperature, etc.;

[0040] Data deduplication: Checking data Does the primary key ID contain duplicates? In the workshop, the primary key ID is often the product number. A latest timestamp priority strategy is adopted: find records where the update time field is a timestamp, retain the record with the largest timestamp value, and delete duplicate records, keeping only the latest version.

[0041] Data denoising: For single time-series data points with attributes such as equipment vibration and temperature, a moving window averaging denoising method is used. A window size of 9 is used, and the mean value of the data within the window replaces the original value at the center of the window, resulting in smoothed data. The formula is as follows:

[0042] ,

[0043] in, for Time data point in time, This is the size of the sliding window.

[0044] Data storage includes: addressing the heterogeneous data from physical entity systems (such as sensor groups, AGV logistics systems), virtual model systems (such as process simulation modules, digital twins), and business management systems (such as MES, ERP) in the digital twin workshop; organizing the data-feature-decision information processing chain in the digital twin workshop; and using XML text language to express data entity relationships. The data mapping relationships are constructed in four layers: conceptual layer... Direct object layer Indirect object layer and attribute layer The data acquisition process is structured as follows: The conceptual layer integrates abstract concepts of workshop resources, inputs and outputs, and production events within the 3C manufacturing process. The direct object layer is a general model derived from the integration of production process objects, representing common elements of the workshop and forming a general object description of the themes carried by the production process data. The indirect object layer is a refined extension of the direct objects; these objects are actual objects existing in the physical workshop and are instantiations of general production process objects. The attribute layer uses indirect objects as specific clustering attributes and serves as the basic unit for data collection from the physical workshop. Finally, for this basic data collection unit, the data is stored in XML text format using the XMI format, serving as a computer-readable data language.

[0045] The detailed process of feature-level fusion is as follows:

[0046] Data semantic association: Define the multi-dimensional context of heterogeneous data, dividing workshop data into three categories: physical domain (e.g., real-time vibration data), virtual domain (e.g., simulation prediction values), and interactive domain (e.g., OPC UA communication process data). Label the source domain for data in different field domains. Collection timestamp Spatial location Semantic tags Data information, such as acquisition timestamps, spatial locations, and semantic tags, is retrieved from the physical and virtual domains. An OPC UA communication protocol is used to establish a unified metric for the data across different domains, enabling semantic association between them. Based on this unified metric, the physical and virtual domains are scaled to ensure comparability under the same dimensions. For example, real-time vibration data from a mobile phone production line and vibration data from a virtual workshop simulation. Finally, the conceptual layer from step 1 is... Direct object layer Indirect object layer and attribute layer Update all XML text data to ensure the correlation and consistency of data in different fields of the digital twin workshop;

[0047] Feature extraction: Multi-domain feature extraction is performed on the global data of the digital twin workshop, including:

[0048] ①Time-domain feature extraction: Taking the moving parts of the placement machine in the SMT production line as the extraction object, extract the conventional indicators of the time series data, including 16 time-domain features such as maximum value, maximum absolute value, minimum value, mean, peak-to-peak value, absolute mean, root mean square value, root mean square amplitude, standard deviation, kurtosis, skewness, margin index, waveform index, impulse index, peak index, and kurtosis index.

[0049] ② Frequency Domain Feature Extraction: The equipment vibration signal is converted into a frequency domain representation using a Fast Fourier Transform (FFT), including the amplitude and phase information of all frequency components. The FFT formula is as follows:

[0050] ,

[0051] in, for Time data point in time, The number of sampling points. For the first Complex values ​​of each frequency domain component For a complex exponential rotation on the unit circle. From the calculated... The frequency domain conventional indicators are extracted, including four frequency domain features: centroid frequency, root mean square frequency, average frequency, and frequency variance.

[0052] Feature analysis:

[0053] ① Data Feature Training: For the extracted 16 time-domain features and 4 frequency-domain features of equipment vibration, with the remaining lifespan of the equipment as the dependent variable, a data feature analysis model based on CNN-BiGRU-Attention was constructed. First, the original feature set was divided into a training set and a test set. Next, the feature analysis model was constructed, including a one-dimensional convolution module, a bidirectional gated recurrent network module, and an attention module. The one-dimensional convolution module performs convolution and pooling on the time-domain features of the one-dimensional signal itself and the transformed frequency-domain features to extract global features. The bidirectional gated recurrent network module simultaneously processes the positive and negative relationships of the above features in the time domain, fully utilizing the sequential information in the features to capture long-term trends in the data, such as periodic fluctuations and degradation trends. The attention module identifies and strengthens the learning of features with a greater impact on the dependent variable among the 16 time-domain features and 4 frequency-domain features, ensuring effective feature selection and ensuring the accuracy of extracting key state features. Model initialization parameters such as convolution kernel size, convolution function, number of hidden layers, virtual-real collaborative loss function, and number of iterations were set, and the feature analysis model was trained to obtain the initial training model.

[0054] ② Data Feature Prediction: Select a test set, perform predictions on the initial trained model, output the feature prediction results, and predict the loss function based on the deviation between the predicted and actual values. As shown below.

[0055] ,

[0056] in, for The actual value of the data at any given time point. for Predicted values ​​for time points based on real-time data. The number of sampling points is specified. Based on the calculated loss function and fine-tuning of the original model parameters, real-time feedback from physical sensors corrects the prediction error of the virtual model, resulting in prediction correction. This training calculation and parameter update process is repeated until the preset number of iterations is reached, yielding a trained feature analysis model. Finally, the trained feature analysis model is used to select a test set and output the test results.

[0057] The detailed process of decision-level fusion is as follows:

[0058] Decision Model Construction: Considering various variables in the production process, including equipment status, production tasks, resource consumption, personnel scheduling, and simulation model parameters of the virtual workshop, as well as virtual-physical synchronization delays, this paper describes the operational status, virtual-physical operational deviations, and decision objectives within the digital twin workshop, constructing a decision model for digital twin workshop production. It is known that the 3C physical workshop has 10 production tasks. The equipment is awaiting processing; the workshop includes 6 processing machines. Each task has a variety of processing steps. , for the first The task processing needs to go through the first step Each process involves multiple steps; each step takes different processing times on different equipment. ,in This refers to the equipment number. Based on the above physical workshop, virtual workshop production parameters are defined, and a corresponding virtual workshop decision model is constructed. The decision-making content of this model is to allocate different machines to different task processes according to the needs of production tasks, thereby forming a production plan.

[0059] Decision optimization: Constraints, decision objectives, and decision conditions are defined based on the decision model. Constraints are the limitations that the workshop operation must meet, including the finiteness of physical workshop production resources and production capacity limitations, as well as the limitations of the virtual workshop simulation model, such as the limit on the number of placement machine heads and the error in placement pressure prediction. The decision objective is to minimize the total production time of the production plan. Decision conditions are the conditions that the workshop decisions must meet, such as the ability of the same equipment to process only one workpiece at a time. Parameter optimization and improvement are then performed based on the decision model.

[0060] Decision-making method: The genetic algorithm from the metaheuristic algorithm family is selected for solving the decision-making method in the digital twin workshop. First, virtual workshop simulation data is introduced, including equipment status and processing time, to determine multiple processes in the digital twin workshop, the technology of each process, and the corresponding processing equipment. In the genetic algorithm, chromosomes are defined as double-gene strings encoding both process codes and equipment processing time codes. The population is the set of decision schemes. An initial decision solution set with a population size of 100 is generated first, and the completion time is calculated for these solution sets. The calculation formula is as follows:

[0061] ,

[0062] For each decision-making individual in the parent population, perform crossover and mutation operations using a genetic algorithm to generate a new offspring population, and calculate the completion time in the same way. Select from the parent-child population The smallest 100 individuals form a new generation. This constitutes one iteration; repeat this process for 100 generations to obtain the final population. Then, select the individuals from the population... The solution with the smallest minimum value is selected as the final decision. This decision is then distributed to the physical workshop for execution, yielding the actual completion time of the physical workshop. The sensors provide real-time feedback of decision-making results to the virtual workshop, which is then compared with the actual execution results in the physical workshop. Fine-tune the virtual workshop model until the physical workshop is operational. Equal to the virtual workshop calculation The above steps enable the management and control of the digital twin workshop.

[0063] like Figure 2 As shown, this invention provides a digital twin workshop control system based on multi-layer data fusion, comprising:

[0064] (1) Data-level fusion module: Inputs multi-source heterogeneous data from the digital twin workshop and outputs clean, standardized, and mineable data from the workshop. This includes:

[0065] The data classification module is used to globally acquire and classify entries of the multi-source heterogeneous dataset of the digital twin workshop;

[0066] The data processing module is used to perform preliminary preprocessing of the data, filtering out noise and duplicate values.

[0067] The data storage module is used to standardize and hierarchically store data to obtain standardized digital twin workshop data;

[0068] (2) Feature-level fusion module: Inputs the value data to be mined and outputs key state features of the digital twin workshop. These include:

[0069] The data association module is used to perform semantic association of virtual and real data based on ontology models on heterogeneous models of digital twin workshops, and to build a global view of digital twin workshop data;

[0070] The feature extraction module is used to extract and represent multi-domain features from the global data of the digital twin workshop.

[0071] The feature analysis module is used to mine key features of the digital twin workshop and analyze and characterize the workshop's operating status.

[0072] (3) Decision-level fusion module: Input key characteristics of the digital twin workshop operation, output workshop production control decisions. This includes:

[0073] The decision model module is used to build a model of decision-making associations in the digital twin workshop, which serves as the input for the decision object;

[0074] The decision optimization module is used to define decision constraints, decision conditions, and decision objectives, and to optimize the decision model.

[0075] The decision-making method solution module is used to select appropriate decision-making methods, analyze and solve decision problems, and obtain production control decisions.

Claims

1. A digital twin workshop management and control method based on multi-layer data fusion, characterized in that, Includes the following steps: Step 1: Acquire multi-source heterogeneous data across the entire digital twin workshop, classify, process, and store the multi-source heterogeneous data, and perform data-level fusion to obtain standard data for the digital twin workshop; Step 2: Extract standard data for the digital twin workshop, perform semantic association, feature extraction, and feature analysis on the standard data for the digital twin workshop, and perform feature-level fusion to obtain the key state features of the digital twin workshop; Step 3: Construct a decision model for the digital twin workshop, input the key state characteristics of the digital twin workshop, solve the problem based on the decision model, decision optimization, and decision method, perform decision-level fusion, and output the workshop production control decision; In step 2, semantic association includes: defining the multi-dimensional context of heterogeneous data, dividing workshop data into three categories: physical domain, virtual domain, and interaction domain, and labeling the source domain and collection timestamp for data in different domains. Spatial location Semantic tags For each data information, read the collection timestamp, spatial location, and semantic tags of data from the physical and virtual domains, and use communication protocols to establish unified data metrics under different domains to achieve semantic association of data in different domains; Step 2, feature analysis, includes: ① Data Feature Training: Based on the extracted time-domain and frequency-domain features, and taking the dependent variable, a feature analysis model is constructed. First, the original feature set is divided into a training set and a test set. Next, the feature analysis model is constructed, which includes a one-dimensional convolution module, a bidirectional gated recurrent network module, and an attention module. The one-dimensional convolution module performs convolution and pooling on the time-domain features of the one-dimensional signal itself and the transformed frequency-domain features to extract global features. The bidirectional gated recurrent network module simultaneously processes the positive and negative relationships of the above features in the time domain, using the sequential information in the features to capture the long-term trend of the data. The attention module identifies and strengthens the learning of features in the time-domain and frequency-domain features that have a high degree of influence on the dependent variable. The initialization parameters of each model, such as the convolution kernel size, convolution function, number of hidden layers, virtual-real collaborative loss function, and number of iterations, are set to train the feature analysis model to obtain the initial training model. ② Data Feature Prediction: Select a test set, perform predictions on the initial trained model, output the feature prediction results, and predict the loss function based on the deviation between the predicted and actual values. As shown below: ， in, for The actual value of the data at any given time point. for Predicted values ​​for time points based on real-time data. The number of sampling points is used; based on the calculation results of the loss function and the fine-tuning of the original model parameters, and based on real-time feedback to correct the prediction error of the virtual model, prediction correction is performed; The decision-making model construction includes: considering various variables in the production process, describing the operating status, virtual-real operation deviation and decision-making objectives in the digital twin workshop, and constructing a decision-making model for digital twin workshop production; Decision optimization includes: defining constraints, decision objectives, and decision conditions based on the decision model. The constraints are the constraints that the workshop operation itself must meet, the decision objective is to minimize the total production time of the production plan, and the decision conditions are the conditions that the workshop decision must meet. Parameter optimization and improvement are then carried out based on the decision model. Step 3 involves solving the decision-making method, including: selecting the genetic algorithm from metaheuristic algorithms to solve the decision-making method for the digital twin workshop. First, virtual workshop simulation data is introduced, including workshop equipment status and processing time. Multiple processes in the digital twin workshop, the technology of each process, and the corresponding processing equipment are determined. Chromosomes in the genetic algorithm are defined as double-gene strings encoding both process codes and equipment processing time codes. The population is the set of decision schemes. An initial decision solution set is first established, and the completion time is calculated for these solution sets. ; For each decision-making individual in the parent population, perform crossover and mutation operations using a genetic algorithm to generate a new offspring population, and calculate the completion time in the same way. .

2. The digital twin workshop management method based on multi-layer data fusion according to claim 1, characterized in that, Step 1 involves acquiring multi-source heterogeneous data across the entire digital twin workshop, including: acquiring all production data from the digital twin workshop. Regarding the above production data Detailed classification based on corresponding attributes yields multi-source heterogeneous data across the entire domain. ,in Indicates the twin data field type, The domain represents twin-workshop data. The field represents physical workshop data. The number of data source system categories. For the first Data entries within an information system.

3. The digital twin workshop management method based on multi-layer data fusion according to claim 1, characterized in that, Step 1, data processing, includes: deduplication and noise reduction of the collected multi-source heterogeneous data from the digital twin workshop; specific steps include: Data imputation: Regression imputation was used, with the missing attribute as the dependent variable and the other attributes as independent variables. A multiple linear regression model was established between the missing attribute and the other attributes. The model is expressed as follows: ， in, For the estimated missing attribute values, These are estimates of the multiple regression coefficients. For attributes related to missing attributes, including product size and product temperature; Data deduplication: Checking data To check if there are duplicate primary key IDs, in the workshop, the primary key ID is the product number. The latest timestamp priority strategy is adopted. Find the record in the data whose update time field is a timestamp, keep the record with the largest timestamp value, delete duplicate records and keep only the latest version. Data denoising: For a single time-series data point, a moving window averaging denoising method is used. The mean value of the data within the window is taken and replaced with the original value at the center of the window to obtain smoothed data; the formula is as follows: ， in, for Time data point in time, This is the size of the sliding window.

4. The digital twin workshop management method based on multi-layer data fusion according to claim 1, characterized in that, In step 1, data storage includes: for the heterogeneous data of the physical entity system, virtual model system, and business management system in the digital twin workshop, the data-feature-decision information processing chain of the digital twin workshop is sorted out, and XML text language is used to express the data entity relationship to construct the data mapping relationship, which is divided into four levels: conceptual layer. Direct object layer Indirect object layer and attribute layer .

5. The digital twin workshop management method based on multi-layer data fusion according to claim 1, characterized in that, Step 2, feature extraction, includes: ①Time-domain feature extraction: Extract indicators of time-series data, including maximum value, maximum absolute value, minimum value, mean, peak-to-peak value, absolute mean, root mean square value, root mean square amplitude, standard deviation, kurtosis, skewness, margin index, waveform index, impulse index, peak index, and kurtosis index; ② Frequency Domain Feature Extraction: The signal is converted into a frequency domain representation using Fast Fourier Transform (FFT), including the amplitude and phase information of all frequency components. The FFT formula is as follows: ， in, for Time data point in time, The number of sampling points. For the first Complex values ​​of each frequency domain component For a complex exponential rotation on the unit circle, from the calculated Frequency domain metrics are extracted, including centroid frequency, root mean square frequency, average frequency, and frequency variance.

6. A digital twin workshop management and control method system based on multi-layer data fusion, used to implement the method of claim 1, characterized in that, include: The data-level fusion module takes multi-source heterogeneous data from the digital twin workshop as input and outputs standard data for the digital twin workshop, including: The data classification module is used to globally acquire and classify entries of the multi-source heterogeneous dataset of the digital twin workshop; The data processing module is used to perform preliminary preprocessing of the data, filtering out noise and duplicate values. The data storage module is used to standardize and hierarchically store data to obtain standardized digital twin workshop data; The feature-level fusion module takes standard data from the digital twin workshop as input and outputs key state features of the digital twin workshop, including: The data association module is used to perform semantic association of virtual and real data based on ontology models on heterogeneous models of digital twin workshops, and to build a global view of digital twin workshop data; The feature extraction module is used to extract and represent multi-domain features from the global data of the digital twin workshop. The feature analysis module is used to mine key features of the digital twin workshop and analyze and characterize the workshop's operating status. The decision-level fusion module takes as input the key state characteristics of the digital twin workshop and outputs workshop production control decisions, including: Decision Model Module: Used to build a model of decision-making relationships in the digital twin workshop, serving as input for the decision-making object; The decision optimization module is used to define decision constraints, decision conditions, and decision objectives, and to optimize the decision model. The decision-making method solution module is used to select appropriate decision-making methods, analyze and solve decision problems, and obtain production control decisions.

Images