Fault self-diagnosis system for industrial production line based on digital twinning

Abstract

This invention discloses a digital twin-based industrial production line fault self-diagnosis system, belonging to the field of fault diagnosis technology. The system includes a status monitoring module for continuously collecting equipment operation data; a data analysis module for converting historical equipment operation data under different historical characteristics of raw materials to obtain historical equipment operation data under the same raw material characteristics, and training a fault self-diagnosis model based on this data; a data transmission module for connecting the status monitoring module and the data analysis module; and a data storage module for storing equipment operation data and raw material characteristic data. By analyzing the impact of different raw material characteristics on equipment operating parameters, the system corrects the collected equipment operation data, establishes a conversion model, and reduces the impact of raw material characteristic differences on the diagnostic model and diagnostic results. Based on the matching degree, the system accelerates the diagnosis of fault types and reduces the harm caused by faults to the production line.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis technology, specifically to a self-diagnosis system for industrial production line faults based on digital twins. Background Technology

[0002] Industrial production line fault self-diagnosis refers to the use of various technical means and methods to monitor and analyze the equipment, process parameters, and production process of industrial production lines in real time, to promptly detect faults and determine their causes and locations, so as to achieve the purpose of rapid repair and reduce downtime.

[0003] However, different batches of raw materials may have different physical and chemical properties, such as hardness, viscosity, and purity. For example, different batches of plastic granules have different melt flow indices, which will affect parameters such as melting pressure and temperature during processing. Diagnostic models based on fixed raw material properties may not be able to adapt to such changes. Training a fault diagnosis model by putting processing parameter data of raw materials with different properties together will affect the judgment of processing status. Therefore, how to improve the effectiveness of fault diagnosis models has become an urgent problem to be solved. Summary of the Invention

[0004] The purpose of this invention is to provide a self-diagnostic system for industrial production line faults based on digital twins, in order to solve the problems raised in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a digital twin-based industrial production line fault self-diagnosis system, comprising a status monitoring module, a data analysis module, a data transmission module, and a data storage module; the status monitoring module is used to continuously collect equipment operation data; the data analysis module is used to convert the historical operation data of the equipment under different historical characteristic data of raw materials to obtain the historical operation data of the equipment under the characteristic data of the same raw material, and to train a fault self-diagnosis model based on the historical operation data of the equipment under the characteristic data of the same raw material; the data transmission module is used to connect the status monitoring module and the data analysis module; the data storage module is used to store the equipment operation data and the raw material characteristic data.

[0006] The status monitoring module further includes a sensor unit, a detection unit, and a data acquisition unit; the sensor unit is used to acquire the operating data of the equipment; the detection unit is used to acquire the characteristic data of the raw materials; and the data acquisition unit is used to acquire the physical characteristics of the equipment.

[0007] The data analysis module further includes a raw material characteristic conversion unit, a raw material characteristic classification unit, and a fault diagnosis unit. The raw material characteristic conversion unit is used to convert the historical operating data of the equipment under different historical characteristic data of the raw material to obtain the historical operating data of the equipment under the characteristic data of the same raw material. The raw material characteristic classification unit is used to establish a classification model of the raw material characteristic data. The fault diagnosis unit trains a fault self-diagnosis model based on the historical operating data of the equipment under the characteristic data of the same raw material, and uses the fault self-diagnosis model to diagnose faults in the industrial production line.

[0008] The raw material characteristic conversion unit acquires the characteristic data of the raw materials and the operating data of the equipment under the characteristic data of the raw materials to form a dataset. The acquired dataset is then divided into a training set and a test set, let a... ij Let y represent the j-th feature under the i-th raw material characteristic data in the training set. i This represents the equipment's operating data under the i-th raw material characteristic data in the training set; it also represents the equipment's operating data y under the k-th raw material characteristic data in the test set. k and the j-th feature a under the k-th raw material characteristic data in the test set kj We obtain the j-th feature a under the k-th raw material characteristic data in the test set. kj and the j-th feature a under the i-th raw material characteristic data in the training set ij The difference between them is concatenated with the characteristic data of the raw materials and the operating data of the equipment to obtain the input vector (..., a). kj -a ij , ..., y i ), will use the equipment's operating data y under the k-th raw material characteristic data in the test set. k As output, a neural network model for raw material characteristic conversion is trained; the trained neural network model for raw material characteristic conversion is validated using test set data, and the difference and the equipment operation data under the characteristic data of the k-th raw material in the test set are concatenated to obtain the input vector (…, a ij -a kj , ..., y k The input vector is fed into the raw material property conversion neural network model to obtain y. i Fitted values, based on y i Fitted values ​​and y i The neural network model for raw material property conversion was validated.

[0009] The raw material property conversion unit trains a class-centered raw material property conversion neural network model through the following steps:

[0010] S11, obtain the conversion error vector [e] for converting equipment operation data under the i-th raw material characteristic data to equipment operation data under other raw material characteristic data. i1 e i2 , ..., e in In the formula, n represents the number of raw material characteristic data, and e i1 e i2 , ..., e in Indicates conversion error, e i1 =|y i -ny i The formula is |+|y1-ny1|, where y1 represents the equipment's operating data under the first raw material characteristic data, and ny1 represents the fitted value of y1 obtained by inputting the input vector obtained based on the i-th raw material characteristic data into the raw material characteristic transformation neural network model. i This means that the input vector obtained based on the first raw material characteristic data is input into the raw material characteristic transformation neural network model, and the resulting y... i Fitted value; e i2 , ..., e in Determination method and e i1 Same; the sum of errors e obtained by using the i-th raw material characteristic data as the class center. i1 +e i2 +…+e in Set the initial value for the number of class centers, num.

[0011] S12, based on the total error of raw material characteristic data as class centers, num raw material characteristic data are randomly selected as initial class centers. The probability of raw material characteristic data being selected is inversely proportional to the total error of raw material characteristic data as class centers.

[0012] S13, obtain the conversion error between other raw material characteristic data and all class centers, and assign the other raw material characteristic data to the class center with the smallest conversion error;

[0013] S14, for the raw material characteristic data of each class center, update the raw material characteristic data of the class center to minimize the total conversion error between all raw material characteristic data under the same class center;

[0014] S15, return to step S12 until the class center no longer changes or the number of iterations reaches the set value; determine whether the number of samples of all raw material characteristic data under the same class center meets the requirements of model training. If it meets the requirements, increment the value of num by 1 and return to step S12; if it does not meet the requirements, obtain the class center when the number of class centers is num-1.

[0015] S16, Obtain raw material characteristic data and equipment operation data under the same type of center, and retrain the raw material characteristic conversion neural network model under the same type of center.

[0016] The raw material characteristic classification unit obtains the class centers of raw material characteristic data and assigns labels based on the class centers to which the raw material characteristic data belongs; it uses the raw material characteristic data as input and the labels as output to train the raw material characteristic data classification model; it obtains the raw material characteristic data of the current industrial production line and inputs it into the raw material characteristic data classification model to obtain the class centers to which the raw material characteristic data of the current industrial production line belongs; it inputs the equipment operation data under the raw material characteristic data of the current industrial production line into the raw material characteristic transformation neural network model of the class center to obtain the equipment operation data of the current industrial production line under the raw material characteristics of the class center.

[0017] The fault diagnosis unit acquires the physical attributes and performance parameters of historical equipment and assigns them to a virtual model; it acquires historical operating data and raw material characteristic data of the equipment, and converts the equipment operating data under different raw material characteristic data into equipment operating data under the class center; it uses machine learning and deep learning algorithms to learn the patterns and trends in the historical data under the class center and trains the time series prediction model for equipment operating data; it acquires the physical attributes and performance parameters of the equipment in the current industrial production line, and inputs the equipment operating data of the current industrial production line under the raw material characteristics of the class center into the time series prediction model to obtain the equipment operating data prediction results.

[0018] The fault diagnosis unit retrieves equipment operating data for each fault type from the data storage module and constructs operating data curves for each fault type; it also analyzes the root mean square error R between the predicted values ​​of the operating parameters and operating data of the current industrial production line equipment and the equipment operating data for each fault type. uj In the formula, u represents the fault type label. The matching degree F between the current industrial production line equipment operating parameters and the u-th fault type is obtained based on the root mean square error analysis. u F=ΣW j ×(1 / R uj ), where W j The weight of the j-th feature in the raw material characteristic data; the matching degree F between the operating parameters of the current industrial production line equipment and the u-th fault type. u Supervised classification is performed to identify the fault type.

[0019] Compared with the prior art, the beneficial effects of the present invention are: by analyzing the influence of different raw material characteristics on equipment operating parameters, the collected equipment operating data is corrected, a conversion model is established, and the influence of raw material characteristic differences on the diagnostic model and diagnostic results is reduced; the diagnosis of fault types is accelerated based on the matching degree, and the harm caused by faults to the production line is reduced. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the industrial production line fault self-diagnosis system based on digital twins according to the present invention;

[0021] Figure 2 This is a flowchart of the industrial production line fault self-diagnosis method based on digital twins according to the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Example: Figure 1 As shown, this invention provides a digital twin-based industrial production line fault self-diagnosis system, including a condition monitoring module, a data analysis module, a data transmission module, and a data storage module. The condition monitoring module is used to continuously collect equipment operation data. The data analysis module is used to convert the historical operation data of the equipment under different historical characteristic data of raw materials to obtain the historical operation data of the equipment under the characteristic data of the same raw material, and to train a fault self-diagnosis model based on the historical operation data of the equipment under the characteristic data of the same raw material. The data transmission module is used to connect the condition monitoring module and the data analysis module. The data storage module is used to store the equipment operation data and the raw material characteristic data.

[0024] The status monitoring module further includes a sensor unit, a detection unit, and a data acquisition unit; the sensor unit is used to acquire the operating data of the equipment; the detection unit is used to acquire the characteristic data of the raw materials; and the data acquisition unit is used to acquire the physical characteristics of the equipment.

[0025] The data analysis module further includes a raw material characteristic conversion unit, a raw material characteristic classification unit, and a fault diagnosis unit. The raw material characteristic conversion unit is used to convert the historical operating data of the equipment under different historical characteristic data of the raw material to obtain the historical operating data of the equipment under the characteristic data of the same raw material. The raw material characteristic classification unit is used to establish a classification model of the raw material characteristic data. The fault diagnosis unit trains a fault self-diagnosis model based on the historical operating data of the equipment under the characteristic data of the same raw material, and uses the fault self-diagnosis model to diagnose faults in the industrial production line.

[0026] In another embodiment of the present invention, the present invention provides a method for self-diagnosis of faults in industrial production lines based on digital twins, comprising the following steps:

[0027] S1, deploy sensors on physical equipment to acquire equipment operation data and raw material characteristic data;

[0028] S2, acquire historical characteristic data of raw materials, and convert historical operating data of equipment based on historical characteristic data of raw materials;

[0029] S3, convert the historical operating data of the equipment under different historical characteristic data of raw materials to obtain the historical operating data of the equipment under the characteristic data of the same raw material, and train the fault self-diagnosis model based on the historical operating data of the equipment under the characteristic data of the same raw material.

[0030] S4: Obtain the current equipment's operating data, convert it to the characteristic data of the raw materials, and then use the fault self-diagnosis model to diagnose faults in the industrial production line.

[0031] The conversion of historical operating data of equipment based on historical characteristic data of raw materials also includes the following steps:

[0032] A dataset is formed by acquiring characteristic data of raw materials and operational data of equipment under the condition of raw material characteristic data. The acquired dataset is then divided into a training set and a test set, let a... ij Let y represent the j-th feature under the i-th raw material characteristic data in the training set. i This represents the equipment's operating data under the i-th raw material characteristic data in the training set; it also represents the equipment's operating data y under the k-th raw material characteristic data in the test set. k and the j-th feature a under the k-th raw material characteristic data in the test set kj We obtain the j-th feature a under the k-th raw material characteristic data in the test set. kj and the j-th feature a under the i-th raw material characteristic data in the training set ij The difference between them is concatenated with the characteristic data of the raw materials and the operating data of the equipment to obtain the input vector (..., a). kj-a ij , ..., y i ), will use the equipment's operating data y under the k-th raw material characteristic data in the test set. k As output, a neural network model for raw material characteristic conversion is trained; the trained neural network model for raw material characteristic conversion is validated using test set data, and the difference and the equipment operation data under the characteristic data of the k-th raw material in the test set are concatenated to obtain the input vector (…, a ij -a kj , ..., y k The input vector is fed into the raw material property conversion neural network model to obtain y. i Fitted values, based on y i Fitted values ​​and y i The neural network model for raw material property conversion was validated.

[0033] Different batches of raw materials may differ in physical and chemical properties, such as hardness, viscosity, and purity. For example, different batches of plastic granules may have different melt flow indices, affecting parameters like melt pressure and temperature during processing. Diagnostic models based on fixed raw material properties may not adapt to these variations. Training a fault diagnosis model by combining processing parameter data from raw materials with different properties can negatively impact the model's assessment of processing status. For instance, as the hardness of raw materials increases, the temperature rises during processing. Training a fault diagnosis model by combining temperatures from two different raw material hardnesses can negatively affect the model's training effectiveness. Therefore, a raw material property conversion model is used to convert temperatures from different raw material hardnesses to temperatures from the same raw material hardness, improving the model's performance and accuracy. The raw material property conversion neural network model is a reversible model. For the five properties of the first raw material property [a...] 11 a 12 , ..., a 15 ] and the corresponding equipment operating parameters y1, the five characteristics under the second raw material characteristics [a 21 a 22 , ..., a 25 ] and the corresponding equipment operating parameter y2, according to [a 11 a 12 , ..., a 15 ] and [a 21 a 22 , ..., a 25 The error vector is obtained, and the error vector and y1 are used as inputs to train the neural network model; when validating the model, the error vector and y2 are used as inputs to the model for validation.

[0034] Analyzing the impact of raw material characteristic data on equipment operating data also includes the following steps:

[0035] S11, obtain the conversion error vector [e] for converting equipment operation data under the i-th raw material characteristic data to equipment operation data under other raw material characteristic data. i1 e i2 , ..., e in In the formula, n represents the number of raw material characteristic data, and e i1 e i2 , ..., e in Indicates conversion error, e i1 =|y i -ny i The formula is |+|y1-ny1|, where y1 represents the equipment's operating data under the first raw material characteristic data, and ny1 represents the fitted value of y1 obtained by inputting the input vector obtained based on the i-th raw material characteristic data into the raw material characteristic transformation neural network model. i This means that the input vector obtained based on the first raw material characteristic data is input into the raw material characteristic transformation neural network model, and the resulting y... i Fitted value; e i2 , ..., e in Determination method and e i1 Same; the sum of errors e obtained by using the i-th raw material characteristic data as the class center. i1 +e i2 +…+e in Set the initial value for the number of class centers, num.

[0036] S12, based on the total error of raw material characteristic data as class centers, num raw material characteristic data are randomly selected as initial class centers. The probability of raw material characteristic data being selected is inversely proportional to the total error of raw material characteristic data as class centers.

[0037] S13, obtain the conversion error between other raw material characteristic data and all class centers, and assign the other raw material characteristic data to the class center with the smallest conversion error;

[0038] S14, for the raw material characteristic data of each class center, update the raw material characteristic data of the class center to minimize the total conversion error between all raw material characteristic data under the same class center;

[0039] S15, return to step S12 until the class center no longer changes or the number of iterations reaches the set value; determine whether the number of samples of all raw material characteristic data under the same class center meets the requirements of model training. If it meets the requirements, increment the value of num by 1 and return to step S12; if it does not meet the requirements, obtain the class center when the number of class centers is num-1.

[0040] S16, Obtain raw material characteristic data and equipment operation data under the same type of center, and retrain the raw material characteristic conversion neural network model under the same type of center.

[0041] The initial value of num can be set to a small value. For five raw material characteristic data points, the conversion error vector in decibels of the obtained equipment operation data is: [e 11 e 12 , ..., e 15 ]、…、[e 51 e 52 , ..., e 55 ], where e 11 ..., e 55 The value is 0, so no calculation is needed. Using the equipment operation data under the first and second raw material characteristic data, a raw material characteristic conversion neural network model is used to convert the equipment operation data between the two sets of data, yielding the conversion error e. 12 and e 21 e 12 and e 21 Equal; when the initial number of class centers is 2, e 11 e 12 , ..., e 15 The sum of the transformation errors obtained from the first raw material characteristic data is used as the class center. Two class centers are then selected using random numbers. For example, if the first and second raw material characteristic data are selected as class centers, the transformation errors e between the third, fourth, and fifth raw material characteristic data and the first raw material characteristic data can be used as the basis for selection. 13 e 14 e 15 The conversion error e between the 3rd, 4th, and 5th raw material characteristic data and the 2nd raw material characteristic data. 23 e 24 e 25 Based on the conversion error, the 3rd, 4th, and 5th raw material characteristic data are assigned to the two existing class centers. If e 13 Less than e 23 If the third raw material characteristic data is assigned to the class center of the first raw material characteristic data, then the assignment continues in this manner. After the assignment is complete, the third, fourth, and fifth raw material characteristic data are all assigned to the class center of the first raw material characteristic data. Therefore, the first class center needs to be updated, and the total conversion error between the first, third, fourth, and fifth raw material characteristic data under the first class center is calculated. For the first raw material characteristic data, the total error is e. 11 +e 13 +e 14 +e 15 For the third raw material characteristic data, the total error is e.31 +e 33 +e 34 +e 35 The raw material characteristic data with the smallest total error is selected as the new class center. Since the second class center only contains one raw material characteristic data, no update is needed. The raw material characteristic data outside the class center are redistributed according to the updated class center. When the class center no longer changes or the number of iterations reaches the set value, it is determined whether the number of raw material characteristic data samples under the same class center meets the requirements of model training. If the requirements are met, it means that the number of class centers can be increased. If not, the result when the number of class centers is num-1 is taken.

[0042] The sample data for different raw material characteristics are not the same. For raw material characteristics with a small number of samples, the fault diagnosis learning model is prone to overfitting due to the lack of sufficient sample training data, and misclassifies normal states as faults. Therefore, it is necessary to ensure that the number of raw material characteristic data samples under each class center is sufficient for model training to improve the model training effect.

[0043] Establishing a classification model for raw material characteristic data also includes the following steps:

[0044] Obtain the class centers of raw material characteristic data and assign labels based on the class centers to which the raw material characteristic data belongs; use the raw material characteristic data as input and the labels as output to train the raw material characteristic data classification model; obtain the raw material characteristic data of the current industrial production line and input the raw material characteristic data of the current industrial production line into the raw material characteristic data classification model to obtain the class centers to which the raw material characteristic data of the current industrial production line belongs; input the equipment operation data under the raw material characteristic data of the current industrial production line into the raw material characteristic transformation neural network model of the class centers to obtain the equipment operation data of the current industrial production line under the raw material characteristics of the class centers.

[0045] To diagnose faults in the current industrial production line, it is necessary to convert the equipment data under the current raw material characteristics to a class center. Then, the fault diagnosis model under the class center is used to diagnose faults in the current industrial production line. This achieves the training of the fault model on the same scale and the use of the trained model to diagnose faults on the operating data on the same scale, thus eliminating the interference of raw material characteristics on the operating data.

[0046] Using a fault self-diagnosis model for fault diagnosis in industrial production lines also includes the following steps:

[0047] Acquire the physical attributes and performance parameters of historical equipment and assign them to a virtual model; acquire historical operating data and raw material characteristic data of the equipment, and convert the equipment operating data under different raw material characteristic data into equipment operating data under the class center;

[0048] Using machine learning and deep learning algorithms, patterns and trends in historical data under the category center are learned, and a time series prediction model for equipment operation data is trained.

[0049] The physical attributes and performance parameters of the equipment in the current industrial production line will be obtained. The equipment operation data of the current industrial production line under the raw material characteristics of the class center will be input into the time series prediction model to obtain the equipment operation data prediction results.

[0050] For equipment operation data under the same class center, they are all converted to the raw material characteristics of the class center. Then, the converted equipment operation data is used to train a time series prediction model for equipment operation data. The input is the virtual model of the equipment and the equipment operation parameter y(t). The output is the equipment operation parameter y(t+step), where step is the prediction time step and t represents time. The predicted value of the operation parameter is obtained based on the current operation parameter of the industrial production line equipment. The fault diagnosis of the industrial production line is performed based on the current operation parameter of the industrial production line equipment and the predicted value of the operation data.

[0051] Using a fault self-diagnosis model for fault diagnosis in industrial production lines also includes the following steps:

[0052] Equipment operating data for each fault type is retrieved from the data storage module, and operating data curves for each fault type are constructed. The root mean square error R between the predicted values ​​of the operating parameters and operating data of the current industrial production line equipment and the equipment operating data for each fault type is analyzed. uj In the formula, u represents the fault type label. The matching degree F between the current industrial production line equipment operating parameters and the u-th fault type is obtained based on the root mean square error analysis. u F=ΣW j ×(1 / R uj ), where W j The weight of the j-th feature in the raw material characteristic data; the matching degree F between the operating parameters of the current industrial production line equipment and the u-th fault type. u Supervised classification is performed to identify the fault type.

[0053] Based on the class center to which the current raw material characteristic data belongs, operational data curves are obtained for each fault type within that class center. These curves are formed by transforming the operational parameters within the class center to the raw material characteristic data, as the same fault can exhibit differences depending on the raw material characteristics. Based on the equipment operational data prediction results, a matching degree sequence can be obtained. For example, given the current equipment operational parameter y(t), the matching degree is obtained by combining y(t) with previous operational parameters, leading to a further prediction of y(t+1). Similarly, the matching degree is obtained by combining y(t+1), y(t), and previous operational parameters, leading to another prediction of y(t+2), generating a matching degree sequence. This matching degree sequence is then input into a supervised classification model to identify the fault type, enabling early fault type determination. The supervised classification model uses historical operational data sequences prior to fault occurrences as output and fault labels as output for training. Similarly, this supervised classification model is trained using the transformed operational parameters.

[0054] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A self-diagnostic system for industrial production line faults based on digital twins, characterized in that: The system includes a status monitoring module, a data analysis module, a data transmission module, and a data storage module. The status monitoring module continuously collects equipment operation data. The data transmission module connects the status monitoring module and the data analysis module. The data storage module stores equipment operation data and raw material characteristic data. The data analysis module converts historical equipment operation data under different historical characteristics of raw materials to obtain historical equipment operation data under the same raw material characteristic data, and trains a fault self-diagnosis model based on this historical equipment operation data. The data analysis module includes a raw material characteristic conversion unit, which converts historical equipment operation data under different historical characteristics of raw materials to obtain historical equipment operation data under the same raw material characteristic data. The raw material characteristic conversion unit acquires the characteristic data of the raw materials and the operating data of the equipment under the characteristic data of the raw materials to form a dataset. The acquired dataset is divided into a training set and a test set, let a... ij Let y represent the j-th feature under the i-th raw material characteristic data in the training set. i This represents the equipment operation data under the i-th raw material characteristic data; it also retrieves the j-th feature a under the k-th raw material characteristic data. kj With a ij The difference between them, and the difference with y i Concatenating the input vectors (..., a) yields the input vector. kj -a ij , ..., y i ), and the equipment operation data y under the kth raw material characteristic data. k As output, a neural network model for raw material characteristic conversion is trained; the neural network model is a reversible model to convert historical equipment operation data under different batches of raw material characteristic data to equipment operation data under the same raw material characteristic data; the trained neural network model for raw material characteristic conversion is validated using test set data, and the difference and the equipment operation data under the characteristic data of the kth raw material in the test set are concatenated to obtain the input vector (…, a ij -a kj , ..., y k The input vector is fed into the raw material property conversion neural network model to obtain y. i Fitted values, based on y i Fitted values ​​and y i Validate the neural network model for raw material property conversion; The raw material characteristic conversion unit also clusters the raw material characteristic data through the following steps: S11, obtain the conversion error vector [e] for converting equipment operation data under the i-th raw material characteristic data to equipment operation data under other raw material characteristic data. i1 e i2 , ..., e in In the formula, n represents the number of raw material characteristic data, and e i1 e i2 , ..., e in Indicates conversion error, e i1 =|y i -ny i The formula is |+|y1-ny1|, where y1 represents the equipment's operating data under the first raw material characteristic data, and ny1 represents the fitted value of y1 obtained by inputting the input vector obtained based on the i-th raw material characteristic data into the raw material characteristic transformation neural network model. i This means that the input vector obtained based on the first raw material characteristic data is input into the raw material characteristic transformation neural network model, and the resulting y... i Fitted value; e i2 , ..., e in Determination method and e i1 Same; the sum of errors e obtained by using the i-th raw material characteristic data as the class center. i1 +e i2 +…+e in Set the initial value for the number of class centers, num. S12, based on the total error of raw material characteristic data as class centers, num raw material characteristic data are randomly selected as initial class centers. The probability of raw material characteristic data being selected is inversely proportional to the total error of raw material characteristic data as class centers. S13, obtain the conversion error between other raw material characteristic data and all class centers, and assign the other raw material characteristic data to the class center with the smallest conversion error; S14, for the raw material characteristic data of each class center, update the raw material characteristic data of the class center to minimize the total conversion error between all raw material characteristic data under the same class center; S15, return to step S12 until the class center no longer changes or the number of iterations reaches the set value; determine whether the number of samples of all raw material characteristic data under the same class center meets the requirements of model training. If the requirements are met, increment the value of num by 1 and return to step S12. If the requirement is not met, then obtain the class centers when the number of class centers is num-1; S16, Obtain raw material characteristic data and equipment operation data under the same type of center, and retrain the raw material characteristic conversion neural network model under the same type of center.

2. The industrial production line fault self-diagnosis system based on digital twins according to claim 1, characterized in that, The status monitoring module further includes a sensor unit, a detection unit, and a data acquisition unit; the sensor unit is used to acquire the operating data of the equipment; the detection unit is used to acquire the characteristic data of the raw materials; and the data acquisition unit is used to acquire the physical characteristics of the equipment.

3. The industrial production line fault self-diagnosis system based on digital twins according to claim 1, characterized in that, The data analysis module also includes a raw material characteristic classification unit and a fault diagnosis unit; the raw material characteristic classification unit is used to establish a classification model for raw material characteristic data; the fault diagnosis unit trains a fault self-diagnosis model based on the historical operating data of the equipment under the characteristic data of the same raw material, and uses the fault self-diagnosis model to diagnose faults in the industrial production line. The raw material characteristic classification unit assigns labels to each raw material characteristic data based on the cluster centers obtained from clustering, trains a raw material characteristic data classification model, obtains the cluster centers to which the raw material characteristic data of the current production line belongs, inputs the current equipment operation data into the transformation neural network model corresponding to the cluster center, and obtains the equipment operation data under the raw material characteristics of the cluster center. The fault diagnosis unit trains a time series prediction model based on historical operating data of equipment under the class center, constructs operating data curves for each fault type, and calculates the root mean square error R between the predicted values ​​of the current equipment operating parameters and the operating data for each fault type. uj The matching degree F is obtained. u =ΣW j ×(1 / R uj In the formula, u represents the fault type label, and W j The weight of the j-th feature in the raw material characteristic data is used to perform supervised classification on the matching degree sequence to identify the fault type.

4. The industrial production line fault self-diagnosis system based on digital twins according to claim 3, characterized in that, The raw material characteristic classification unit obtains the class centers of raw material characteristic data and assigns labels based on the class centers to which the raw material characteristic data belongs; it uses the raw material characteristic data as input and the labels as output to train the raw material characteristic data classification model; it obtains the raw material characteristic data of the current industrial production line and inputs it into the raw material characteristic data classification model to obtain the class centers to which the raw material characteristic data of the current industrial production line belongs; it inputs the equipment operation data under the raw material characteristic data of the current industrial production line into the raw material characteristic transformation neural network model of the class center to obtain the equipment operation data of the current industrial production line under the raw material characteristics of the class center.

5. The industrial production line fault self-diagnosis system based on digital twins according to claim 3, characterized in that, The fault diagnosis unit acquires the physical attributes and performance parameters of historical equipment and assigns them to a virtual model; it acquires historical operating data and raw material characteristic data of the equipment, and converts the equipment operating data under different raw material characteristic data into equipment operating data under the class center; it uses machine learning and deep learning algorithms to learn the patterns and trends in the historical data under the class center and trains the time series prediction model for equipment operating data; it acquires the physical attributes and performance parameters of the equipment in the current industrial production line, and inputs the equipment operating data of the current industrial production line under the raw material characteristics of the class center into the time series prediction model to obtain the equipment operating data prediction results.

Images