A smart factory management platform and management method based on digital twins

Abstract

The present invention discloses a smart factory management platform and management method based on digital twins, which specifically relates to the field of digital twin technology, and includes the following steps: dividing and preliminarily analyzing the digital twin smart factory, then extracting features and analyzing anomalies of components with hidden dangers, calculating a data source quality index and a predicted performance quality index, and evaluating the comprehensive operation quality using a pre-trained machine learning model; based on the evaluation results, dividing the components into inefficient and efficient operation components; determining the optimization method for the inefficient components; and performing intelligent regulation on the efficient components based on the initial sampling frequency; the present invention can comprehensively evaluate the operation status of digital twin model components, evaluate the degree of adaptation of each component to the actual factory components, perform intelligent regulation on the efficient operation components, and optimize the inefficient operation components, thereby significantly improving resource utilization, ensuring the stability and reliability of the system under various operating conditions, and preventing system failures and shutdowns.

Description

Technical Field

[0001] The present invention relates to the field of digital twin technology, and more specifically, to a smart factory management platform and management method based on digital twin. Background Art

[0002] Smart factory management based on digital twins is an advanced manufacturing management method that uses a combination of virtual and real technologies to achieve comprehensive monitoring and optimization of various equipment, processes and systems in the factory. Digital twin technology creates a virtual model that is completely corresponding to the actual factory. Through real-time data collection and analysis, this virtual model can reflect the actual operating status of the factory and simulate various operations and changes, thereby helping managers predict problems, optimize production processes, and improve production efficiency and resource utilization.

[0003] The core of this management model is to achieve all-round perception and intelligent control of the factory through digital twin technology. Smart factories can collect a large amount of real-time data through sensors and Internet of Things technology, and then process and interpret these data through big data analysis and artificial intelligence technology, so as to optimize various links such as equipment maintenance, production planning, and quality control. Ultimately, smart factory management based on digital twins can significantly improve the factory's production efficiency and reduce costs, while enhancing its flexibility and responsiveness to adapt to rapidly changing market demands.

[0004] Existing technical problems: The digital twin model needs to accurately reflect the status of the actual factory, but the factory's production equipment and processes will change over time. Ensuring the timely update and accuracy of the model is an ongoing challenge. If the model cannot accurately reflect the actual situation, it will lead to wrong decisions. Therefore, a smart factory management platform and management method based on digital twins are proposed here. Summary of the invention

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A smart factory management method based on digital twins includes the following steps:

[0007] Divide the digital twin smart factory into multiple digital twin model components, then collect the twin operation information of multiple digital twin model components respectively, and then conduct a preliminary analysis on each digital twin model component to determine whether there is a hidden danger of incompatibility between the digital twin model component and the current actual factory components;

[0008] Perform feature extraction and anomaly analysis on digital twin model components that have hidden dangers of being incompatible with current actual factory components, and obtain the data source quality index and prediction performance quality index respectively. Then substitute the data source quality index and prediction performance quality index into the pre-trained machine learning model to evaluate the comprehensive operation quality of the digital twin model components.

[0009] Based on the output data of the machine learning model, the digital twin model components are divided into inefficient operating components and efficient operating components;

[0010] For inefficiently running components, the optimization method is determined based on the data source quality index, the predicted performance quality index, and the output data of the machine learning model. For efficiently running components, the actual sampling frequency is intelligently regulated based on its initial sampling frequency.

[0011] In a preferred embodiment, each digital twin model component is preliminarily analyzed to determine whether the digital twin model component has a hidden danger of being incompatible with the current actual factory component as follows:

[0012] Within a fixed time window, the model data lag time of the digital twin model component is obtained: when the actual data is generated, each data is timestamped. When the digital twin model receives and applies the data, the update timestamp is recorded. The lag time is obtained by subtracting the data timestamp from the model update timestamp to obtain the model data lag time series data of the digital twin model component. Then, the mean and standard deviation of the data lag time series data are calculated. Then, the standard deviation is divided by the mean to obtain the coefficient of variation. Then, the coefficient of variation is compared with the preset standard data distribution reference threshold. If the coefficient of variation is less than the preset standard data distribution reference threshold, a normal signal is generated. If the coefficient of variation is not less than the preset standard data distribution reference threshold, an abnormal signal is generated. When an abnormal signal is generated, it is judged that the digital twin model component has a hidden danger of being incompatible with the current actual factory component.

[0013] In a preferred embodiment, the logic for obtaining the data source quality index is:

[0014] Extract the lag time feature data Ti from different data sources, and calculate the average value of the lag time feature data. The calculation formula is: ; represents the average value of the lag time feature data, N is the total number of lag time feature data, and then calculates the average value of each lag time feature data Ti and the lag time feature data The absolute value of the difference is , and then calculate the consistency coefficient, the calculation formula is: ; YZ is the consistency coefficient;

[0015] Extract the actual physical state feature data Ai and the corresponding twin state feature data Mi in the digital twin model component, calculate the error Ei between the actual data and the model data, the error Ei is the absolute value of the difference between the actual physical state feature data Ai and the corresponding twin state feature data Mi in the digital twin model component, and use the Bayesian update method to combine the historical error and the current error to update the error estimate, specifically: calculate the error posterior probability under given data conditions , ; represents the likelihood of the data given the error Ei, represents the prior probability of error Ei, Represents the marginal probability of the data, which is a normalized constant, and then the accuracy coefficient is calculated by combining the comprehensive error and the posterior probability: ; represents the accuracy coefficient, N is the total number of data points, represents the error of the ith data point, represents the error posterior probability of the i-th data point;

[0016] The consistency coefficient and the accuracy coefficient are summarized to obtain vector one, a first standard vector is preset, and the Euclidean distance between vector one and the preset first standard vector is calculated to obtain a data source quality index.

[0017] In a preferred embodiment, the acquisition logic of the predicted performance quality index is:

[0018] Extract the historical feature data of the actual plant state and the predicted feature data of the corresponding digital twin model components, calculate the difference between the historical feature data of the actual plant state and the predicted feature data of the corresponding digital twin model components, and obtain the prediction error. Then, calculate the root mean square error and the mean absolute percentage error based on the prediction error. Calculate the coefficient of variation of the sequence data through the sequence data composed of the obtained prediction errors. Summarize the root mean square error, the mean absolute percentage error, and the coefficient of variation of the sequence data to obtain vector two. A second standard vector is preset. Calculate the Euclidean distance between vector two and the preset second standard vector to obtain the prediction performance quality index.

[0019] In a preferred embodiment, after the data source quality index and the predicted performance quality index are substituted into a pre-trained machine learning model, the machine learning model performs a comprehensive analysis to generate a comprehensive operation quality coefficient.

[0020] In a preferred embodiment, dividing the digital twin model components into inefficient operation components and efficient operation components means:

[0021] Obtain the comprehensive operation quality coefficient generated by the target digital twin model component in the machine learning model, and then compare the comprehensive operation quality coefficient with the preset operation reference threshold. If the comprehensive operation quality coefficient is greater than or equal to the preset operation reference threshold, the target digital twin model component is classified as an efficient operation component; if the comprehensive operation quality coefficient is less than the preset operation reference threshold, the target digital twin model component is classified as an inefficient operation component.

[0022] In a preferred embodiment, determining the optimization method for the inefficiently running component refers to:

[0023] The data source quality index, predicted performance quality index and comprehensive operation quality coefficient generated by the machine learning model of the inefficient running component are obtained, the data source quality index, predicted performance quality index and comprehensive operation quality coefficient are taken as input variables, the optimization method is taken as the output variable, the input variables are fuzzified and the values ​​of the input variables are converted into fuzzy sets, the output variables are fuzzified and the output variables are converted into fuzzy sets, fuzzy rules are formulated to describe the optimization requirements under different data type combinations, the fuzzified input variables are inferred through fuzzy rules, and the type of optimization that needs to be performed on the target inefficient running component is determined.

[0024] In a preferred embodiment, for the efficient operation component, the actual sampling frequency is intelligently regulated based on its initial sampling frequency, and the intelligent regulation formula is: ; represents the actual sampling frequency after regulation, represents the initial sampling frequency before regulation, k is the preset proportional coefficient, and F is the ratio of the preset operation reference threshold to the comprehensive operation quality coefficient.

[0025] In a preferred embodiment, a smart factory management platform based on digital twins includes an information collection module, a digital twin module, a preliminary analysis module, an evaluation module, and an optimization module;

[0026] The digital twin module is used to build a smart factory that is a digital twin of the actual factory;

[0027] The information collection module is used to divide the digital twin smart factory into multiple digital twin model components, and then collect the twin operation information of multiple digital twin model components respectively;

[0028] The preliminary analysis module is used to conduct preliminary analysis on each digital twin model component to determine whether there is a hidden danger of incompatibility between the digital twin model component and the current actual factory component;

[0029] The evaluation module is used to extract features and perform anomaly analysis on digital twin model components that have hidden dangers of being incompatible with current actual factory components, and obtain the data source quality index and the prediction performance quality index respectively. The data source quality index and the prediction performance quality index are then substituted into the pre-trained machine learning model to evaluate the comprehensive operation quality of the digital twin model components.

[0030] The optimization module is used to divide the digital twin model components into inefficient operating components and efficient operating components based on the output data of the machine learning model. Then, for the inefficient operating components, the optimization method is judged according to the data source quality index, the prediction performance quality index, and the output data of the machine learning model. For the efficient operating components, the actual sampling frequency is intelligently regulated based on its initial sampling frequency.

[0031] Technical effects and advantages of the present invention:

[0032] The present invention introduces a comprehensive operation quality coefficient (CQI), which can comprehensively evaluate the operating status of digital twin model components and the degree of adaptation of each component to the actual factory components. The comprehensive operation quality coefficient combines the data source quality index (DSQI) and the prediction performance quality index (PPQI), which can accurately reflect the overall operation quality of the system. The present invention improves the dynamic response capability of the system by adjusting the actual sampling frequency. The sampling frequency can be automatically adjusted according to the real-time operating status of the system, thereby optimizing the data acquisition process and ensuring data quality and system stability.

[0033] The present invention uses the Bayesian update method to combine historical errors and current errors to obtain dynamic errors in dynamically changing application environments, and to understand the error behavior of the system more objectively and accurately. For inefficient operating components, the present invention uses fuzzy logic methods to fuzzify the data source quality index, the prediction performance quality index, and the comprehensive operating quality coefficient, and determines the optimization method through reasoning through fuzzy rules. The fuzzy logic method can handle complex multivariate decision-making problems and improve the flexibility and accuracy of optimization decisions.

[0034] The present invention can significantly improve resource utilization by intelligently regulating efficient operating components and optimizing inefficient operating components. The optimized system can reduce unnecessary resource consumption and improve overall production efficiency while ensuring high-quality operation. The present invention ensures the stability and reliability of the system under various operating conditions by comprehensively analyzing and optimizing various model components. The dynamic adjustment and intelligent regulation methods can timely discover and correct deviations and problems in system operation, prevent system failures and shutdowns, and improve the stability and reliability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to facilitate understanding by those skilled in the art, the present invention is further described below in conjunction with the accompanying drawings;

[0036] Figure 1 This is a schematic diagram of a smart factory management method based on digital twins in the present invention.

[0037] Figure 2 This is a schematic diagram of a smart factory management platform based on digital twins in the present invention. DETAILED DESCRIPTION

[0038] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0039] Reference Figure 1 - Figure 2 The following embodiments are obtained:

[0040] Embodiment 1: A smart factory management method based on digital twins, comprising the following steps:

[0041] Divide the digital twin smart factory into multiple digital twin model components, then collect the twin operation information of multiple digital twin model components respectively, and then conduct a preliminary analysis on each digital twin model component to determine whether there is a hidden danger of incompatibility between the digital twin model component and the current actual factory components;

[0042] Perform feature extraction and anomaly analysis on digital twin model components that have hidden dangers of being incompatible with current actual factory components, and obtain the data source quality index and prediction performance quality index respectively. Then substitute the data source quality index and prediction performance quality index into the pre-trained machine learning model to evaluate the comprehensive operation quality of the digital twin model components.

[0043] Based on the output data of the machine learning model, the digital twin model components are divided into inefficient operating components and efficient operating components;

[0044] For inefficiently running components, the optimization method is determined based on the data source quality index, the predicted performance quality index, and the output data of the machine learning model. For efficiently running components, the actual sampling frequency is intelligently regulated based on its initial sampling frequency.

[0045] The digital twin model can be divided into multiple model components, each of which focuses on a different factory system or function. In this way, each part of the factory can be managed and optimized more meticulously, and accuracy prediction and update can also be performed separately. The following is an example of the present invention, giving a feasible division method, which is described in detail as follows:

[0046] Equipment-level components: Each key equipment (such as robots, conveyor belts, CNC machine tools, etc.) has its own digital twin model, which can monitor the equipment’s operating status, performance data, failure rate, etc., helping to predict maintenance needs and optimize operations.

[0047] Production line-level components: Each production line has an independent digital twin model. The model integrates the status data of all equipment on the production line and simulates the production process, output, efficiency, etc. It can be used to optimize production scheduling, reduce bottlenecks, and improve the overall efficiency of the production line.

[0048] Environmental Monitoring Component: This model component monitors various parameters in the factory environment, such as temperature, humidity, air quality, etc. Through this data, it can ensure that the production environment meets process requirements and optimize energy use and environmental control systems.

[0049] Logistics and supply chain components: These models focus on the flow of raw materials and products in and out of the factory. They can help optimize inventory management, supply chain scheduling and transportation routes, improve logistics efficiency, and reduce inventory costs and delivery times.

[0050] Quality Control Component: This model component is used to monitor and analyze product quality data, identify quality issues that may occur during the production process, predict and prevent defects, and optimize the quality inspection process.

[0051] Human Resources and Safety Component: This component tracks the work status, skill level, and safety status of factory employees. By analyzing this data, it is possible to optimize personnel scheduling, improve production safety, and improve employee satisfaction.

[0052] Dividing the digital twin model into multiple model components can manage various aspects of the factory more finely and perform accurate predictions and optimizations separately, which improves the flexibility and reliability of the system, ensures that each part can operate efficiently and respond to changes and adjustments in a timely manner, and improves overall production efficiency and benefits.

[0053] The steps to conduct a preliminary analysis on each digital twin model component and determine whether the digital twin model component has hidden dangers of incompatibility with the current actual factory components are as follows:

[0054] Within a fixed time window, the model data lag time of the digital twin model component is obtained: data timestamp record: when the actual data is generated, each data is timestamped; model update timestamp record: when the digital twin model receives and applies the data, the update timestamp is recorded. The data timestamp and model update timestamp are recorded to calculate the lag time of each data, so as to evaluate the delay in data transmission and processing. The lag time is obtained by subtracting the data timestamp from the model update timestamp to obtain the model data lag time series data of the digital twin model component. The lag time series data reflects the time difference from the generation of each data to its reception and application by the model, representing the delay in data transmission and processing. The mean and standard deviation of the data lag time series data are then calculated. The mean and standard deviation respectively measure the central tendency and dispersion of the data lag, providing a basis for the calculation of the coefficient of variation. The standard deviation is then divided by the mean to obtain the coefficient of variation. The coefficient of variation is a dimensionless value that reflects the lag time relative to the model. The degree of dispersion of its average value provides a standardized volatility measure, and then the coefficient of variation is compared with the preset standard data distribution reference threshold. By comparing with the preset threshold, it is possible to determine whether the volatility of the data lag is within an acceptable range, thereby evaluating the stability and adaptability of the model. If the coefficient of variation is less than the preset standard data distribution reference threshold, a normal signal is generated. If the coefficient of variation is not less than the preset standard data distribution reference threshold, an abnormal signal is generated. When the abnormal signal is generated, it is determined that the digital twin model component has a hidden danger of incompatibility with the current actual factory components; normal signal: the coefficient of variation is less than the preset standard data distribution reference threshold; abnormal signal: the coefficient of variation is greater than or equal to the preset standard data distribution reference threshold. The abnormal signal indicates that there are potential problems in the system, which requires further analysis and adjustment to ensure the efficient adaptation of the digital twin model to the actual factory components, and to ensure that the model can reflect the actual status of the factory in real time and accurately, thereby improving production efficiency and resource utilization.

[0055] The logic for obtaining the data source quality index is:

[0056] Extract the lag time feature data Ti of different data sources, represent the lag time feature data of the i-th data source, and calculate the average value of the lag time feature data. The calculation formula is: ; represents the average value of the lag time feature data, N is the total number of lag time feature data, and then calculates the average value of each lag time feature data Ti and the lag time feature data The absolute value of the difference is , and then calculate the consistency coefficient, the calculation formula is: ; YZ is the consistency coefficient, which is used to measure the consistency of the data source lag time;

[0057] Extract the actual physical state feature data Ai and the corresponding twin state feature data Mi in the digital twin model component, calculate the error Ei between the actual data and the model data, the error Ei is the absolute value of the difference between the actual physical state feature data Ai and the corresponding twin state feature data Mi in the digital twin model component, and use the Bayesian update method to combine the historical error and the current error to update the error estimate, specifically: calculate the error posterior probability under given data conditions , ; represents the likelihood of the data given the error Ei, represents the prior probability of error Ei, Represents the marginal probability of the data, which is a normalized constant, and then the accuracy coefficient is calculated by combining the comprehensive error and the posterior probability: ; represents the accuracy coefficient, N is the total number of data points, represents the error of the ith data point, represents the error posterior probability of the i-th data point;

[0058] The consistency coefficient and the accuracy coefficient are summed up to obtain vector 1. A first standard vector is preset. The Euclidean distance between vector 1 and the preset first standard vector is calculated to obtain the data source quality index. Specifically, the consistency coefficient YZ and the accuracy coefficient AC are summed up. The accuracy coefficient AC is For the convenience of expression, a vector V is formed: ; Preset the first standard vector : ; Calculate the Euclidean distance between vector 1 and the preset first standard vector: DSQI is the data source quality index, which indicates the distance between the actual data source and the ideal data source. The smaller the distance, the higher the quality of the data source. The Bayesian update method is combined with historical errors and current errors to provide a method for dynamically adjusting error estimates, thereby improving the accuracy and reliability of data source quality assessment. The data source quality index comprehensively reflects the overall quality of the data source through a comprehensive analysis of the consistency coefficient and the accuracy coefficient, providing a reliable foundation for subsequent optimization and improvement.

[0059] It is necessary to elaborate that the Bayesian update method provides a method for dynamically adjusting the error estimate by combining prior information (historical error) and new observation information (current error). Historical error (prior probability): represents the model's previous understanding of the error. This part of information is based on past observations and experience and can provide a preliminary framework for the current estimate. Current error (likelihood function): reflects the latest data observation results. By combining current observations with historical information, the current system state can be more accurately reflected. Real systems are often dynamic, and the environment and conditions may change at any time. The Bayesian update method can continuously update the error estimate with each new observation. The Bayesian update method allows multiple sources of information to be integrated together to obtain a more comprehensive and accurate error estimate: historical data provides a long-term perspective on the stability of the system, while current data reflects the latest changes in the system. Combining the information of these two can provide a more comprehensive understanding of the error behavior of the system. The prior and likelihood functions in the Bayesian method can be adjusted by weight to flexibly weigh the impact of historical data and current observations.

[0060] The logic of obtaining the prediction performance quality index is as follows: extract the historical feature data of the actual plant state and the prediction feature data of the corresponding digital twin model component, calculate the difference between the historical feature data of the actual plant state and the prediction feature data of the corresponding digital twin model component, and obtain the prediction error. Then, according to the prediction error, calculate the root mean square error and the mean absolute percentage error. Through the obtained sequence data composed of the prediction errors, calculate the coefficient of variation of the sequence data. The coefficient of variation is used to measure the degree of dispersion of the error sequence relative to its average value. The root mean square error, the mean absolute percentage error, and the coefficient of variation of the sequence data are summarized to obtain vector 2. A second standard vector is preset to represent the root mean square error, the mean absolute percentage error, and the coefficient of variation under ideal conditions. The Euclidean distance between vector 2 and the preset second standard vector is calculated to obtain the prediction performance quality index. The root mean square error, the mean absolute percentage error, and the coefficient of variation are all obtained by calculation formulas well known to those skilled in the art, and will not be elaborated here. Root mean square error (RMSE): reflects the average size of the prediction error and is an important indicator for measuring the prediction performance of the model. The smaller the RMSE, the higher the accuracy of the model prediction. Mean absolute percentage error (MAPE): reflects the relative size of the prediction error and is an important indicator for measuring the prediction performance of the model. The smaller the MAPE, the higher the accuracy of the model prediction. Coefficient of variation (CV): reflects the degree of dispersion of the prediction error and is an important indicator for measuring the stability of the model prediction. The smaller the CV, the better the stability of the model prediction. Prediction performance quality index (PPQI): comprehensively measures the prediction accuracy, accuracy and stability of the model, and can fully reflect the prediction performance of the model. The smaller the PPQI, the better the prediction performance of the model. The prediction performance quality index PPQI combines the root mean square error, mean absolute percentage error and coefficient of variation to provide a comprehensive performance evaluation indicator to help identify the prediction ability of the model.

[0061] After the data source quality index and the prediction performance quality index are substituted into the pre-trained machine learning model, the machine learning model performs a comprehensive analysis to generate a comprehensive operation quality coefficient. The machine learning model is not specifically specified. The present invention provides a specific implementation method: the data source quality index and the prediction performance quality index are standardized respectively, and the processing method is exemplified by Z-score standardization, and then the standardized ones are substituted into the formula: ; represents the standardized data source quality index, represents the standardized prediction performance quality index, w1, w2, and w3 are weight coefficients used to adjust the importance of each indicator in the comprehensive operation quality, k1 and k2 are scaling coefficients used to control the sensitivity of nonlinear transformation, tanh represents the hyperbolic tangent function, which is used to introduce nonlinear effects and make the model more sensitive to extreme values. The hyperbolic tangent function maps the input value to the range of [-1,1], which enhances the ability to distinguish high and low values. The interaction term :Considering the interaction between the data source quality index and the prediction performance quality index, we can capture the situation where the two jointly affect the comprehensive operation quality. Comprehensive operation quality coefficient (CQI): The smaller the CQI is, the better the comprehensive operation quality is, the more the digital twin model matches the actual factory operation status, and the data source quality and prediction performance are both excellent. The larger the CQI is, the worse the comprehensive operation quality is, and there is a large difference between the digital twin model and the actual factory operation status, and the data source quality and prediction performance need to be further optimized.

[0062] Dividing digital twin model components into inefficiently functioning components and efficiently functioning components means:

[0063] Obtain the comprehensive operation quality coefficient generated by the target digital twin model component in the machine learning model, and then compare the comprehensive operation quality coefficient with the preset operation reference threshold. If the comprehensive operation quality coefficient is greater than or equal to the preset operation reference threshold, the target digital twin model component is classified as an efficient operation component; if the comprehensive operation quality coefficient is less than the preset operation reference threshold, the target digital twin model component is classified as an inefficient operation component.

[0064] The way to optimize inefficient components is to:

[0065] Obtain the data source quality index, prediction performance quality index, and comprehensive operation quality coefficient of the inefficient operation component, use the data source quality index, prediction performance quality index, and comprehensive operation quality coefficient as input variables, use the optimization method as the output variable, fuzzify the input variables, convert the values ​​of the input variables into fuzzy sets, fuzzify the output variables, convert the output variables into fuzzy sets, formulate fuzzy rules, describe the optimization requirements under different data type combinations, and reason the fuzzified input variables through fuzzy rules to determine the type of optimization that needs to be performed on the target inefficient operation component. Assumed input variable range: DSQI range: 0 to 100; PPQI range: 0 to 100; CQI range: -1 to 1; define the fuzzy set of input variables: DSQI: low, medium, high; PPQI: low, medium, high; CQI: poor, general, good; define the fuzzy set of output variables (optimization type): optimize data source, optimize prediction model, comprehensive optimization, and formulate rules to describe the optimization requirements under different data combinations. Examples are as follows: If DSQI is high and PPQI is low and CQI is poor, then optimize the data source. If DSQI is low and PPQI is high and CQI is poor, then optimize the prediction model. If DSQI is medium and PPQI is medium and CQI is average, then perform comprehensive optimization, reasoning on the fuzzified input variables through fuzzy rules, and finally provide reasonable optimization suggestions for inefficiently running components.

[0066] For efficient operation components, the actual sampling frequency is intelligently regulated based on its initial sampling frequency. The intelligent regulation formula is: ; represents the actual sampling frequency after regulation, represents the initial sampling frequency before regulation, k is the preset proportional coefficient, F is the ratio of the preset operation reference threshold to the comprehensive operation quality coefficient, represents the accuracy coefficient. The initial sampling frequency is the basic frequency set by the system during design to ensure basic data collection requirements. The preset proportional coefficient is used to adjust the sensitivity of the sampling frequency adjustment. The size of k can be adjusted according to the needs of the specific system. The operation reference threshold is the operation quality standard that the system expects to achieve, indicating the operation quality that the system should have under ideal conditions. The comprehensive operation quality coefficient reflects the actual operation quality of the current system. By comparing with the reference threshold, the sampling frequency can be adjusted to optimize the system performance, and the actual sampling frequency of the efficient operation component can be intelligently regulated. The larger the ratio F, the lower the current operation quality of the system, and the more it is necessary to increase the sampling frequency to obtain more accurate data. The initial sampling frequency, the preset proportional coefficient, and the ratio of the operation reference threshold to the comprehensive operation quality coefficient are taken into account during regulation, so as to achieve dynamic adjustment of the sampling frequency to optimize the system performance.

[0067] Embodiment 2: A smart factory management platform based on digital twins, comprising an information collection module, a digital twin module, a preliminary analysis module, an evaluation module, and an optimization module;

[0068] The digital twin module is used to build a smart factory that is a digital twin of the actual factory;

[0069] The information collection module is used to divide the digital twin smart factory into multiple digital twin model components, and then collect the twin operation information of multiple digital twin model components respectively;

[0070] The preliminary analysis module is used to conduct preliminary analysis on each digital twin model component to determine whether there is a hidden danger of incompatibility between the digital twin model component and the current actual factory component;

[0071] The evaluation module is used to extract features and perform anomaly analysis on digital twin model components that have hidden dangers of being incompatible with current actual factory components, and obtain the data source quality index and the prediction performance quality index respectively. The data source quality index and the prediction performance quality index are then substituted into the pre-trained machine learning model to evaluate the comprehensive operation quality of the digital twin model components.

[0072] The optimization module is used to divide the digital twin model components into inefficient operating components and efficient operating components based on the output data of the machine learning model. Then, for the inefficient operating components, the optimization method is judged according to the data source quality index, the prediction performance quality index, and the output data of the machine learning model. For the efficient operating components, the actual sampling frequency is intelligently regulated based on its initial sampling frequency.

[0073] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0074] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0075] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0076] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0077] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A smart factory management method based on digital twins, characterized in that: The following steps are involved: Divide the digital twin smart factory into multiple digital twin model components, then collect the twin operation information of multiple digital twin model components respectively, and then conduct a preliminary analysis on each digital twin model component to determine whether there is a hidden danger of incompatibility between the digital twin model component and the current actual factory components; Perform feature extraction and anomaly analysis on digital twin model components that have hidden dangers of being incompatible with current actual factory components, and obtain the data source quality index and prediction performance quality index respectively. Then substitute the data source quality index and prediction performance quality index into the pre-trained machine learning model to evaluate the comprehensive operation quality of the digital twin model components. Based on the output data of the machine learning model, the digital twin model components are divided into inefficient operating components and efficient operating components; For inefficiently running components, the optimization method is determined based on the data source quality index, the predicted performance quality index, and the output data of the machine learning model. For efficiently running components, the actual sampling frequency is intelligently regulated based on its initial sampling frequency. The logic for obtaining the data source quality index is: Extract the lag time feature data Ti from different data sources, and calculate the average value of the lag time feature data. The calculation formula is: ; represents the average value of the lag time feature data, N is the total number of lag time feature data, and then calculates the average value of each lag time feature data Ti and the lag time feature data The absolute value of the difference is , and then calculate the consistency coefficient, the calculation formula is: ; YZ is the consistency coefficient; Extract the actual physical state feature data Ai and the corresponding twin state feature data Mi in the digital twin model component, calculate the error Ei between the actual data and the model data, the error Ei is the absolute value of the difference between the actual physical state feature data Ai and the corresponding twin state feature data Mi in the digital twin model component, and use the Bayesian update method to combine the historical error and the current error to update the error estimate, specifically: calculate the error posterior probability under given data conditions , ; represents the likelihood of the data given the error Ei, represents the prior probability of error Ei, Represents the marginal probability of the data, which is a normalized constant, and then the accuracy coefficient is calculated by combining the comprehensive error and the posterior probability: ; represents the accuracy coefficient, N is the total number of data points, represents the error of the ith data point, represents the error posterior probability of the i-th data point; The consistency coefficient and the accuracy coefficient are summarized to obtain a vector 1, a first standard vector is preset, and the Euclidean distance between the vector 1 and the preset first standard vector is calculated to obtain a data source quality index; The logic for obtaining the predicted performance quality index is: Extract the historical feature data of the actual plant state and the predicted feature data of the corresponding digital twin model components, calculate the difference between the historical feature data of the actual plant state and the predicted feature data of the corresponding digital twin model components, and obtain the prediction error. Then, calculate the root mean square error and the mean absolute percentage error based on the prediction error. Calculate the coefficient of variation of the sequence data through the sequence data composed of the obtained prediction errors. Summarize the root mean square error, the mean absolute percentage error, and the coefficient of variation of the sequence data to obtain vector two. A second standard vector is preset. Calculate the Euclidean distance between vector two and the preset second standard vector to obtain the prediction performance quality index.

2. According to the digital twin-based smart factory management method of claim 1, it is characterized in that: The steps to conduct a preliminary analysis on each digital twin model component and determine whether the digital twin model component has hidden dangers of incompatibility with the current actual factory components are as follows: Within a fixed time window, the model data lag time of the digital twin model component is obtained: when the actual data is generated, each data is timestamped. When the digital twin model receives and applies the data, the update timestamp is recorded. The lag time is obtained by subtracting the data timestamp from the model update timestamp to obtain the model data lag time series data of the digital twin model component. Then, the mean and standard deviation of the data lag time series data are calculated. Then, the standard deviation is divided by the mean to obtain the coefficient of variation. Then, the coefficient of variation is compared with the preset standard data distribution reference threshold. If the coefficient of variation is less than the preset standard data distribution reference threshold, a normal signal is generated. If the coefficient of variation is not less than the preset standard data distribution reference threshold, an abnormal signal is generated. When an abnormal signal is generated, it is judged that the digital twin model component has a hidden danger of being incompatible with the current actual factory component.

3. According to the digital twin-based smart factory management method of claim 2, it is characterized in that: After the data source quality index and the prediction performance quality index are substituted into the pre-trained machine learning model, the machine learning model performs a comprehensive analysis to generate a comprehensive operation quality coefficient.

4. The method for managing a smart factory based on digital twins according to claim 3 is characterized in that: Dividing digital twin model components into inefficiently functioning components and efficiently functioning components means: Obtain the comprehensive operation quality coefficient generated by the target digital twin model component in the machine learning model, and then compare the comprehensive operation quality coefficient with the preset operation reference threshold. If the comprehensive operation quality coefficient is greater than or equal to the preset operation reference threshold, the target digital twin model component is classified as an efficient operation component; if the comprehensive operation quality coefficient is less than the preset operation reference threshold, the target digital twin model component is classified as an inefficient operation component.

5. The smart factory management method based on digital twin according to claim 4 is characterized in that: The way to optimize inefficient components is to: The data source quality index, predicted performance quality index and comprehensive operation quality coefficient generated by the machine learning model of the inefficient running component are obtained, the data source quality index, predicted performance quality index and comprehensive operation quality coefficient are taken as input variables, the optimization method is taken as the output variable, the input variables are fuzzified and the values ​​of the input variables are converted into fuzzy sets, the output variables are fuzzified and the output variables are converted into fuzzy sets, fuzzy rules are formulated to describe the optimization requirements under different data type combinations, the fuzzified input variables are inferred through fuzzy rules, and the type of optimization that needs to be performed on the target inefficient running component is determined.

6. The method for managing a smart factory based on digital twins according to claim 5 is characterized in that: For efficient operation components, the actual sampling frequency is intelligently regulated based on its initial sampling frequency. The intelligent regulation formula is: ; represents the actual sampling frequency after regulation, represents the initial sampling frequency before regulation, k is the preset proportional coefficient, and F is the ratio of the preset operation reference threshold to the comprehensive operation quality coefficient.

7. A smart factory management platform based on digital twins, based on a smart factory management method based on digital twins according to any one of claims 1 to 6, characterized in that: It includes information collection module, digital twin module, preliminary analysis module, evaluation module and optimization module; The digital twin module is used to build a smart factory that is a digital twin of the actual factory; The information collection module is used to divide the digital twin smart factory into multiple digital twin model components, and then collect the twin operation information of multiple digital twin model components respectively; The preliminary analysis module is used to conduct preliminary analysis on each digital twin model component to determine whether there is a hidden danger of incompatibility between the digital twin model component and the current actual factory component; The evaluation module is used to extract features and perform anomaly analysis on digital twin model components that have hidden dangers of being incompatible with current actual factory components, and obtain the data source quality index and the prediction performance quality index respectively. The data source quality index and the prediction performance quality index are then substituted into the pre-trained machine learning model to evaluate the comprehensive operation quality of the digital twin model components. The optimization module is used to divide the digital twin model components into inefficient operating components and efficient operating components based on the output data of the machine learning model. Then, for the inefficient operating components, the optimization method is judged according to the data source quality index, the prediction performance quality index, and the output data of the machine learning model. For the efficient operating components, the actual sampling frequency is intelligently regulated based on its initial sampling frequency.

Images