Intelligent production line digital twin modeling method based on Internet of Things modeling

Abstract

The invention discloses an intelligent production line digital twin modeling method based on Internet of Things modeling, and belongs to the field of industrial production, and the modeling method comprises the following specific steps: I, collecting and preprocessing multi-dimensional data in an intelligent production line in real time; iI, performing multi-dimensional feature extraction on the processed intelligent production line data; iII, constructing a virtual model based on the extracted feature data to simulate the production process; the method can greatly improve the production efficiency, can rapidly make an optimal response when the production demand changes, effectively guarantees the real-time performance and adaptability of an algorithm, guarantees the stability and efficiency of a scheduling system, facilitates the discovery of hidden association, carries out the effective fault prevention, can achieve the continuous improvement in the operation process of the system, and improves the operation efficiency of the system. A more accurate maintenance plan is provided, accidental shutdown of the production line is reduced, the maintenance cost can be effectively reduced, and the overall resource utilization efficiency of the production line is improved.

Description

Technical Field

[0001] The present invention relates to the field of industrial production, and in particular to a digital twin modeling method for an intelligent production line based on Internet of Things modeling. Background Art

[0002] In the modern intelligent manufacturing system, the digitization and intelligence of production lines have become key means to improve manufacturing efficiency and cope with complex production needs. With the development of Internet of Things technology, the data interaction between production equipment and systems is becoming increasingly close, and real-time monitoring and intelligent decision-making of production lines have become particularly important. As a bridge connecting the physical world and the virtual world, digital twin technology can reflect the status of physical equipment in real time by creating a virtual model of the production line, and perform dynamic analysis and optimization. Traditional production line monitoring and optimization solutions often rely on the analysis of historical data, lack the ability to respond to real-time data, and cannot accurately predict potential problems in the production line; therefore, it is particularly important to invent a digital twin modeling method for intelligent production lines based on Internet of Things modeling.

[0003] The existing digital twin modeling method for intelligent production lines has low production efficiency, cannot make the best response quickly when production demand changes, cannot guarantee the real-time and adaptability of the algorithm, and reduces the stability and efficiency of the scheduling system. In addition, the existing digital twin modeling method for intelligent production lines is not conducive to discovering hidden associations and preventing faults. There are many cases of unexpected shutdowns of production lines, which increases maintenance costs and reduces the overall resource utilization efficiency of the production line. To this end, we propose a digital twin modeling method for intelligent production lines based on Internet of Things modeling. Summary of the invention

[0004] The purpose of the present invention is to solve the defects in the prior art and propose a digital twin modeling method for an intelligent production line based on Internet of Things modeling.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] The digital twin modeling method of the intelligent production line based on the Internet of Things modeling has the following specific steps:

[0007] Ⅰ. Real-time collection and pre-processing of multi-dimensional data in intelligent production lines;

[0008] Ⅱ. Extract multi-dimensional features from the processed intelligent production line data;

[0009] III. Construct a virtual model based on the extracted feature data to simulate the production process;

[0010] IV. Optimize production line logistics and equipment scheduling and predict future production trends in real time;

[0011] V. Infer the potential failure risk of equipment and verify and optimize the virtual model.

[0012] As a further solution of the present invention, the specific steps of preprocessing the multi-dimensional data in the intelligent production line in step I are as follows:

[0013] S1.1: The deployed sensor network collects data from each group of intelligent production lines in real time, detects missing values ​​in the collected production line data, and then fills the detected missing values ​​by inference through mean filling, and then sorts the data groups after the missing values ​​are processed from large to small;

[0014] S1.2: Calculate the first quartile, second quartile, and third quartile of each group of sorted data, and calculate the interquartile range between the first quartile and the third quartile. Then, define the upper and lower boundaries of the outliers based on the calculated interquartile range, mark the data that is less than the lower boundary or greater than the upper boundary as outliers, and replace the outliers by interpolation;

[0015] S1.3: The noise in each group of production line data is removed by the sliding average method, and then the data of different dimensions are unified into the range of [0,1] by the Z-score normalization method, and the data units of each group of processed data are unified.

[0016] As a further solution of the present invention, the specific steps of performing multi-dimensional feature extraction on the processed smart production line data in step II are as follows:

[0017] S2.1: Based on the preprocessed production line data of each group, a corresponding data matrix is ​​constructed, where each row is a sample and each column is a feature. The mean vector of each column feature is calculated, and then the corresponding mean vector is subtracted from each group of features. The corresponding covariance matrix is ​​calculated to obtain the correlation between each group of features.

[0018] S2.2: Perform eigenvalue decomposition on the covariance matrix to obtain the corresponding eigenvectors and eigenvalues, select multiple groups of eigenvectors with the largest eigenvalues ​​as the principal component directions, and use the selected eigenvectors to construct low-dimensional data. The linear correlation between the two groups of features is measured by the Pearson correlation coefficient. If the correlation coefficient of the two groups of features exceeds the preset threshold, one of the groups of features is removed;

[0019] S2.3: Randomly select multiple groups of feature data points as the initial centers of the clusters, assign each group of feature data points to the cluster center with the closest Euclidean distance to it, recalculate the mean of all feature data points in each cluster, and use the mean as the new cluster center. Repeat the steps of assigning data points and updating cluster centers until the preset number of iterations is reached.

[0020] As a further solution of the present invention, the specific steps of optimizing production line logistics and equipment scheduling in step IV are as follows:

[0021] S3.1: Collect information about logistics sorting points, equipment sites, and work centers on the intelligent production line, and mark the corresponding nodes on the intelligent production line based on the collected information, initialize the path pheromone concentration and heuristic factor between the nodes, and randomly generate multiple groups of exploration individuals participating in the scheduling problem;

[0022] S3.2: Randomly distribute the exploration individuals to different nodes in the production line, calculate the selection probability of each path according to the pheromone concentration and heuristic factor on each group of paths connecting the node, and select the next moving path by random selection based on the selection probability of each group of paths;

[0023] S3.3: Each group of exploration individuals repeatedly selects a path and moves until a logistics route or a complete equipment scheduling plan is completed, and the logistics cost or scheduling time of the plan is calculated. After all exploration individuals complete the path selection, the pheromones on the path are updated;

[0024] S3.4: Repeat the path construction and pheromone update until the preset maximum number of iterations is reached, and output the optimal production line logistics path and equipment scheduling plan with a smaller logistics cost or the best scheduling efficiency, and feed it back to the staff for review.

[0025] As a further solution of the present invention, the specific steps of real-time prediction of future production trends in step IV are as follows:

[0026] S4.1: Build a trend prediction model using the TensorFlow deep learning framework, initialize the generator network and discriminator network in the trend prediction model, and then randomly collect multiple groups of noise vectors from the standard normal distribution;

[0027] S4.2: Input the collected multiple groups of noise vectors into the trend prediction model. The generator network in the model receives each group of noise vectors and maps the noise into realistic virtual data through a multi-layer neural network. Then, the generated virtual data and the pre-processed production line data are input into the discriminator network in the model.

[0028] S4.3: The discriminator network discriminates virtual data and real production line data through a multi-layer neural network, and outputs the probability that the virtual data and the real production line data are real data respectively. Based on the output result of the discriminator network, the judgment error of the discriminator network is calculated through the cross entropy loss function. Based on the loss function value, the parameters of the discriminator network are updated through the gradient descent method;

[0029] S4.4: The probability that the generator network receives virtual data is judged as real data, and the generation error of the generator network is calculated through the cross entropy loss function. Then, the parameters of the generator network are updated through Adam optimization, and the alternating iterations of the discriminator and the generator are repeated until the errors of the discriminator network and the generator network converge to the preset range;

[0030] S4.5: After training is completed, the virtual data generated by the generator network is used to complete the missing values ​​in the original production line data. After the data is completed, based on the distribution law of historical data, random noise input is used to generate multiple groups of production scenarios through the trained generator to predict production trends.

[0031] As a further solution of the present invention, the specific steps of inferring the potential failure risk of the equipment in step V are as follows:

[0032] S5.1: Collect data on device status, sensor data, fault categories, and external environment, and analyze the causal relationship between the collected data sets based on the knowledge of domain experts or historical data, and connect them with directed edges to construct a Bayesian network in the form of a directed acyclic graph, where each node in the directed acyclic graph represents a set of data;

[0033] S5.2: Optimize the Bayesian network structure through data-driven algorithms, adjust the dependencies between variables, and then determine the parent node corresponding to each node in the network structure based on the causal relationship between each group of data. Then, through statistical analysis of a large amount of historical data, obtain the probability distribution of the node under different parent node conditions;

[0034] S5.3: Collect each group of probability distribution values ​​and generate a conditional probability table corresponding to each node. At the same time, the Bayesian network receives the latest collected production line data in real time, and updates each node data and the conditional probability table based on the latest collected production line data. When the equipment status is updated, the future equipment failure is predicted based on the current equipment status and the corresponding conditional probability table, and the probability of a specific type of equipment failure is calculated.

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] 1. The digital twin modeling method of the intelligent production line based on the Internet of Things modeling collects the information of logistics sorting points, equipment sites and work centers on the intelligent production line, and marks the corresponding nodes on the intelligent production line according to the collected information, initializes the path pheromone concentration and heuristic factor between each node, and randomly generates multiple groups of exploration individuals participating in the scheduling problem. The exploration individuals are randomly distributed to different nodes in the production line, and the selection probability of each path is calculated according to the pheromone concentration and heuristic factor on each group of paths connecting the node. Based on the selection probability of each group of paths, the next moving path is selected by random selection, and each group of exploration individuals repeats the selection. The path is constructed and moved until a logistics route or a complete equipment scheduling plan is completed, and the logistics cost or scheduling time of the plan is calculated. After all exploration individuals have completed the path selection, the pheromones on the path are updated, and the path construction and pheromone update are repeated until the preset maximum number of iterations is reached. The optimal production line logistics path and equipment scheduling plan with a smaller logistics cost or the best scheduling efficiency is output, and the plan is fed back to the staff for review, which can greatly improve production efficiency. At the same time, when production demand changes, the optimal response can be made quickly, effectively ensuring the real-time and adaptability of the algorithm, and ensuring the stability and efficiency of the scheduling system.

[0037] 2. The digital twin modeling method of the intelligent production line based on the Internet of Things modeling collects various groups of data on equipment status, sensor data, fault categories, and external environment, and analyzes the causal relationship between the collected groups of data based on the knowledge of domain experts or historical data. By connecting them with directed edges, a Bayesian network in the form of a directed acyclic graph is constructed, in which each node in the directed acyclic graph represents a group of data. The Bayesian network structure is optimized through a data-driven algorithm, and the dependency relationship between variables is adjusted. Then, according to the causal relationship between each group of data, the parent node corresponding to each node in the network structure is determined, and by statistically analyzing a large amount of historical data, the node in different parent nodes is obtained. The probability distribution under node conditions collects each group of probability distribution values ​​and generates a conditional probability table corresponding to each node. At the same time, the Bayesian network receives the latest collected production line data in real time, and updates each node data and the conditional probability table based on the latest collected production line data. When the equipment status is updated, the future equipment failure is predicted based on the current equipment status and the corresponding conditional probability table, and the probability of a specific type of equipment failure is calculated, which helps to discover hidden associations and effectively prevent faults. It can continuously improve during system operation, provide more accurate maintenance plans, reduce unexpected downtime of production lines, effectively reduce maintenance costs and improve the overall resource utilization efficiency of production lines. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0039] Figure 1 This is a flowchart of the digital twin modeling method for intelligent production lines based on Internet of Things modeling proposed in the present invention. DETAILED DESCRIPTION

[0040] Example 1

[0041] Reference Figure 1 , a digital twin modeling method for intelligent production lines based on IoT modeling, the specific steps of this modeling method are as follows:

[0042] Real-time collection and preprocessing of multi-dimensional data in intelligent production lines.

[0043] Specifically, each group of data of the intelligent production line is collected in real time through the deployed sensor network, and the missing values ​​in the collected production line data are detected. Then, the detected missing values ​​are inferred and filled through mean filling, and then the groups of data after the missing values ​​are processed are sorted from large to small, and the first quartile, second quartile and third quartile of each group of sorted data are calculated, and the interquartile range between the first quartile and the third quartile is calculated. Then, the upper and lower boundaries of the outliers are defined based on the calculated interquartile range, and the data less than the lower boundary or greater than the upper boundary are marked as outliers, and the outliers are replaced by interpolation. The noise in each group of production line data is removed by the sliding average method, and then the data of different dimensions are unified into the range of [0,1] through the Z-score standardization method, and the data units of each group of processed data are unified.

[0044] Perform multi-dimensional feature extraction on the processed intelligent production line data.

[0045] Specifically, according to each group of production line data after preprocessing, a corresponding data matrix is ​​constructed, in which each row is a sample and each column is a feature, and the mean vector of each column feature is calculated, and then the corresponding mean vector is subtracted from each group of features, and the corresponding covariance matrix is ​​calculated to obtain the correlation between each group of features, and the covariance matrix is ​​eigenvalue decomposition is performed to obtain the corresponding eigenvectors and eigenvalues, and multiple groups of eigenvectors with the largest eigenvalues ​​are selected as the principal component directions, and low-dimensional data is constructed using the selected eigenvectors. The linear correlation between two groups of features is measured by the Pearson correlation coefficient. If the correlation coefficient of the two groups of features exceeds a preset threshold, one of the features is removed, and multiple groups of feature data points are randomly selected as the initial center of the cluster, and each group of feature data points is assigned to the cluster center with the closest Euclidean distance to it, and the mean of all feature data points in each cluster is recalculated, and the mean is used as the new cluster center, and the steps of allocating data points and updating cluster centers are repeated until the preset number of iterations is reached.

[0046] A virtual model is constructed based on the extracted feature data to simulate the production process.

[0047] Example 2

[0048] Reference Figure 1 , a digital twin modeling method for intelligent production lines based on IoT modeling, the specific steps of this modeling method are as follows:

[0049] Optimize production line logistics and equipment scheduling and predict future production trends in real time.

[0050] Specifically, the information of logistics sorting points, equipment sites and work centers on the intelligent production line is collected, and the corresponding nodes on the intelligent production line are marked according to the collected groups of information, the path pheromone concentration and heuristic factor between each node are initialized, and multiple groups of exploration individuals participating in the scheduling problem are randomly generated, and the exploration individuals are randomly distributed to different nodes in the production line. According to the pheromone concentration and heuristic factor on each group of paths connecting the node, the selection probability of each path is calculated, and the next moving path is selected by random selection based on the selection probability of each group of paths. Each group of exploration individuals repeatedly selects the path and moves until a logistics route or a complete equipment scheduling plan is completed, and the logistics cost or scheduling time of the plan is calculated. After all exploration individuals complete the path selection, the pheromones on the path are updated, and the path construction and pheromone update are repeated until the preset maximum number of iterations is reached, and the optimal production line logistics path and equipment scheduling plan with a smaller logistics cost or the best scheduling efficiency is output, and it is fed back to the staff for review.

[0051] Specifically, a trend prediction model is constructed through the TensorFlow deep learning framework, and the generator network and the discriminator network in the trend prediction model are initialized. Then, multiple groups of noise vectors are randomly collected from the standard normal distribution, and the collected multiple groups of noise vectors are input into the trend prediction model. The generator network in the model receives each group of noise vectors and maps the noise into realistic virtual data through a multi-layer neural network. Then, the generated virtual data and the pre-processed production line data are input into the discriminator network in the model. The discriminator network discriminates the virtual data and the real production line data through a multi-layer neural network, and outputs the probability that the virtual data and the real production line data are real data respectively. According to the output result of the discriminator network, The cross-entropy loss function calculates the judgment error of the discriminator network. Based on the loss function value, the parameters of the discriminator network are updated through the gradient descent method. The probability that the generator network receives virtual data is judged as real data, and the generation error of the producer network is calculated through the cross-entropy loss function. The parameters of the generator network are updated through Adam optimization, and the alternating iterations of the discriminator and the generator are repeated until the errors of the discriminator network and the generator network converge to the preset range. After the training is completed, the virtual data generated by the generator network is used to complete the missing values ​​in the original production line data. After the data is completed, based on the distribution law of historical data, random noise input is used to generate multiple groups of production scenarios through the trained generator to predict production trends.

[0052] Infer potential equipment failure risks and verify and optimize virtual models.

[0053] Specifically, the device status, sensor data, fault category and external environment data are collected, and based on the knowledge of domain experts or historical data, the causal relationship between the collected data groups is analyzed, and the Bayesian network in the form of a directed acyclic graph is constructed by connecting them with directed edges, where each node in the directed acyclic graph represents a group of data. The Bayesian network structure is optimized by a data-driven algorithm, and the dependencies between the variables are adjusted. Then, based on the causal relationship between the data groups, the parent node corresponding to each node in the network structure is determined, and the probability distribution of the node under different parent node conditions is obtained by statistically analyzing a large amount of historical data, collecting the probability distribution values ​​of each group, and generating the conditional probability table corresponding to each node. At the same time, the Bayesian network receives the latest collected production line data in real time, and updates the data of each node and the conditional probability table based on the latest collected production line data. When the equipment status is updated, future equipment failures are predicted based on the current equipment status and the corresponding conditional probability table, and the probability of a specific type of equipment failure is calculated.

Claims

1. A digital twin modeling method for intelligent production lines based on Internet of Things modeling, characterized in that: The specific steps of the modeling method are as follows: Ⅰ. Real-time collection and pre-processing of multi-dimensional data in intelligent production lines; Ⅱ. Extract multi-dimensional features from the processed intelligent production line data; III. Construct a virtual model based on the extracted feature data to simulate the production process; IV. Optimize production line logistics and equipment scheduling and predict future production trends in real time; V. Infer the potential failure risk of equipment and verify and optimize the virtual model.

2. The method for modeling a digital twin of an intelligent production line based on Internet of Things modeling according to claim 1 is characterized in that: The specific steps for preprocessing the multi-dimensional data in the intelligent production line in step I are as follows: S1.1: The deployed sensor network collects data from each group of intelligent production lines in real time, detects missing values ​​in the collected production line data, and then fills the detected missing values ​​by inference through mean filling, and then sorts the data groups after the missing values ​​are processed from large to small; S1.2: Calculate the first quartile, second quartile, and third quartile of each group of sorted data, and calculate the interquartile range between the first quartile and the third quartile. Then, define the upper and lower boundaries of the outliers based on the calculated interquartile range, mark the data that is less than the lower boundary or greater than the upper boundary as outliers, and replace the outliers by interpolation; S1.3: The noise in each group of production line data is removed by the sliding average method, and then the data of different dimensions are unified into the range of [0,1] by the Z-score normalization method, and the data units of each group of processed data are unified.

3. The method for modeling a digital twin of an intelligent production line based on Internet of Things modeling according to claim 2 is characterized in that: The specific steps for multi-dimensional feature extraction of the processed intelligent production line data in step II are as follows: S2.1: Based on the preprocessed production line data of each group, a corresponding data matrix is ​​constructed, where each row is a sample and each column is a feature. The mean vector of each column feature is calculated, and then the corresponding mean vector is subtracted from each group of features. The corresponding covariance matrix is ​​calculated to obtain the correlation between each group of features. S2.2: Perform eigenvalue decomposition on the covariance matrix to obtain the corresponding eigenvectors and eigenvalues, select multiple groups of eigenvectors with the largest eigenvalues ​​as the principal component directions, and use the selected eigenvectors to construct low-dimensional data. The linear correlation between the two groups of features is measured by the Pearson correlation coefficient. If the correlation coefficient of the two groups of features exceeds the preset threshold, one of the groups of features is removed; S2.3: Randomly select multiple groups of feature data points as the initial centers of the clusters, assign each group of feature data points to the cluster center with the closest Euclidean distance to it, recalculate the mean of all feature data points in each cluster, and use the mean as the new cluster center. Repeat the steps of assigning data points and updating cluster centers until the preset number of iterations is reached.

4. The method for modeling a digital twin of an intelligent production line based on Internet of Things modeling according to claim 1 is characterized in that: The specific steps for optimizing production line logistics and equipment scheduling described in step IV are as follows: S3.1: Collect information about logistics sorting points, equipment sites, and work centers on the intelligent production line, and mark the corresponding nodes on the intelligent production line based on the collected information, initialize the path pheromone concentration and heuristic factor between the nodes, and randomly generate multiple groups of exploration individuals participating in the scheduling problem; S3.2: Randomly distribute the exploration individuals to different nodes in the production line, calculate the selection probability of each path according to the pheromone concentration and heuristic factor on each group of paths connecting the node, and select the next moving path by random selection based on the selection probability of each group of paths; S3.3: Each group of exploration individuals repeatedly selects a path and moves until a logistics route or a complete equipment scheduling plan is completed, and the logistics cost or scheduling time of the plan is calculated. After all exploration individuals complete the path selection, the pheromones on the path are updated; S3.4: Repeat the path construction and pheromone update until the preset maximum number of iterations is reached, and output the optimal production line logistics path and equipment scheduling plan with a smaller logistics cost or the best scheduling efficiency, and feed it back to the staff for review.

5. The method for modeling a digital twin of an intelligent production line based on Internet of Things modeling according to claim 3 is characterized in that: The specific steps of real-time prediction of future production trends described in step IV are as follows: S4.1: Build a trend prediction model using the TensorFlow deep learning framework, initialize the generator network and discriminator network in the trend prediction model, and then randomly collect multiple groups of noise vectors from the standard normal distribution; S4.2: Input the collected multiple groups of noise vectors into the trend prediction model. The generator network in the model receives each group of noise vectors and maps the noise into realistic virtual data through a multi-layer neural network. Then, the generated virtual data and the pre-processed production line data are input into the discriminator network in the model. S4.3: The discriminator network discriminates virtual data and real production line data through a multi-layer neural network, and outputs the probability that the virtual data and the real production line data are real data respectively. Based on the output result of the discriminator network, the judgment error of the discriminator network is calculated through the cross entropy loss function. Based on the loss function value, the parameters of the discriminator network are updated through the gradient descent method; S4.4: The probability that the generator network receives virtual data is judged as real data, and the generation error of the generator network is calculated through the cross entropy loss function. Then, the parameters of the generator network are updated through Adam optimization, and the alternating iterations of the discriminator and the generator are repeated until the errors of the discriminator network and the generator network converge to the preset range; S4.5: After training is completed, the virtual data generated by the generator network is used to complete the missing values ​​in the original production line data. After the data is completed, based on the distribution law of historical data, random noise input is used to generate multiple groups of production scenarios through the trained generator to predict production trends.

6. The method for modeling a digital twin of an intelligent production line based on Internet of Things modeling according to claim 1, characterized in that: The specific steps of inferring the potential failure risk of the equipment in step V are as follows: S5.1: Collect data on device status, sensor data, fault categories, and external environment, and analyze the causal relationship between the collected data sets based on the knowledge of domain experts or historical data, and connect them with directed edges to construct a Bayesian network in the form of a directed acyclic graph, where each node in the directed acyclic graph represents a set of data; S5.2: Optimize the Bayesian network structure through data-driven algorithms, adjust the dependencies between variables, and then determine the parent node corresponding to each node in the network structure based on the causal relationship between each group of data. Then, through statistical analysis of a large amount of historical data, obtain the probability distribution of the node under different parent node conditions; S5.3: Collect each group of probability distribution values ​​and generate a conditional probability table corresponding to each node. At the same time, the Bayesian network receives the latest collected production line data in real time, and updates each node data and the conditional probability table based on the latest collected production line data. When the equipment status is updated, the future equipment failure is predicted based on the current equipment status and the corresponding conditional probability table, and the probability of a specific type of equipment failure is calculated.

Images