Production process monitoring system based on industrial internet

By adopting cloud-edge collaborative architecture and ellipsoid description model in the production process monitoring system, the problem of insufficient acquisition of abnormal situation data sets in the prior art is solved, and anomaly detection with high accuracy and efficiency is achieved, reducing production costs.

CN118915649AInactive Publication Date: 2024-11-08WUXI UNIV

Patent Information

Application Number
CN202410949243.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-15
Publication Date
2024-11-08
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

Due to the inability of complete and effective acquisition of abnormal situation data sets, machine learning methods frequently misjudgment in abnormal situation detection, delaying production lines, increasing production costs, and insufficient practicality.

Method used

Adopting a cloud-edge collaborative architecture based on industrial Internet, the adversarial network model and support vector data description model are generated through conditional generation, equipment operation data that meets qualified production indicators, expand the positive sample set, and use the ellipsoid description model to improve the accuracy of abnormal detection. , reduce the misjudgment rate.

Benefits of technology

It improves the accuracy and real-time nature of abnormal situation identification, reduces the frequency of production line shutdown, reduces production costs, and enhances the competitiveness of production enterprises.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118915649A_ABST
    Figure CN118915649A_ABST
Patent Text Reader

Abstract

The invention discloses a production process monitoring system based on the industrial internet, and belongs to the technical field of data processing systems, and the system comprises a collection module which is used for collecting equipment operation data; the uploading module is used for uploading the collected equipment operation data to the cloud; the first construction module is used for constructing a conditional generative adversarial network model; the first training module is used for training the conditional generative adversarial grid model; the first generation module is used for generating equipment operation data meeting qualified production indexes; the second construction module is used for constructing a support vector data description model; the second generation module is used for training the support vector data description model to generate a minimum ellipsoid; the issuing module is used for issuing the trained support vector data description model to an edge end; and the output module is used for collecting real-time equipment operation data, inputting the real-time equipment operation data into the trained support vector data description model and outputting abnormal data. And the accuracy and the real-time performance of production process monitoring are improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of data processing systems, and specifically relates to a production process monitoring system based on industrial Internet. Background Art

[0002] The Industrial Internet is a system that combines advanced computing, analysis and network technologies with traditional industrial systems. It aims to improve industrial production efficiency, optimize resource allocation, reduce production costs, and improve product quality and safety through data collection, transmission, processing and analysis. The Industrial Internet includes the application of technologies such as the Internet of Things (IoT), big data analysis, cloud computing, and artificial intelligence, which can achieve the interconnection, real-time monitoring and intelligent control of industrial equipment.

[0003] Production process monitoring refers to ensuring the stability, efficiency and quality of the production process through real-time monitoring and data analysis of production equipment and production processes in industrial production. The production process monitoring system collects and processes various data in the production process (such as temperature, pressure, speed, quality, etc.) in real time through sensors, control systems, data acquisition and analysis tools, and performs real-time control and adjustment based on the data analysis results. Production process monitoring ensures production efficiency, product quality and safety through real-time data collection and analysis, improves the automation and optimization level of the production process, reduces costs, and ensures the normal operation of equipment.

[0004] However, due to the existence of a large amount of high-dimensional data in the production process, manual monitoring is extremely difficult to achieve and inefficient, and it is impossible to quickly identify abnormal situations in the production process, which leads to the inability of the production process to proceed normally. Existing production process monitoring methods also use machine learning methods to monitor the production process. However, since the abnormal situation data set in the production process cannot be completely and effectively obtained, the machine learning method cannot learn the complete production abnormalities. In this case, although the monitoring efficiency is improved, it is very easy to make misjudgments, delay production of the production line, increase production costs of manufacturers, lack practicality, and have low accuracy in abnormal situation detection. Summary of the invention

[0005] In order to solve the technical problems in the prior art that the abnormal situation data set in the production process cannot be completely and effectively obtained, which leads to the inability of the machine learning method to learn the complete production abnormal situation, in this case, although the monitoring efficiency is improved, it is very easy to make misjudgments, delay production line production, increase production costs of manufacturers, lack of practicality, and low accuracy of abnormal situation detection, the present invention provides a production process monitoring system based on Industrial Internet.

[0006] The present invention provides a production process monitoring system based on the industrial Internet, which is applied to a cloud-edge collaborative architecture, wherein the cloud-edge collaborative architecture comprises a cloud end, an edge end, and a production terminal connected in sequence, wherein the cloud end is connected to the production terminal;

[0007] The production process monitoring system based on the Industrial Internet includes:

[0008] The collection module is used to collect equipment operation data under qualified production indicators;

[0009] An uploading module, used to upload the collected equipment operation data to the cloud;

[0010] A first construction module is used to construct a conditional generative adversarial network model with the qualified production index as a condition;

[0011] A first training module is used to train the conditional generative adversarial grid model using the equipment operation data and the corresponding qualified production indicators as a training set;

[0012] A first generation module is used to generate equipment operation data that meets the qualified production index through a trained generative adversarial network model, wherein the collected equipment operation data and the generated equipment operation data constitute a positive sample set of the production process;

[0013] A second construction module is used to construct a support vector data description model that uses an ellipsoid to describe the positive sample set of the production process;

[0014] A second generating module is used to train the support vector data description model using the production process positive sample set to generate a minimum ellipsoid corresponding to the production process positive sample set;

[0015] A sending module, used for sending the trained support vector data description model to the edge end;

[0016] The output module is used to collect the real-time equipment operation data of the production terminal, input the real-time equipment operation data into the trained support vector data description model, and output abnormal data, wherein the abnormal data is the real-time equipment operation data that exceeds the minimum ellipsoid.

[0017] Compared with the prior art, the present invention has at least the following beneficial technical effects:

[0018] In the present invention, a multi-class sample training process is converted into a single-class sample training process. In this process, a conditional generative adversarial network model is constructed with the qualified production index as a condition. By learning the correlation and change law between the collected real equipment operation data, normal equipment operation data that is difficult to collect completely is generated to expand the positive sample set. Then, based on the positive sample, a support vector data description model that uses an ellipsoid to describe the positive sample set of the production process is constructed. The distribution shape of the normal production process data can be accurately described, thereby improving the accuracy of anomaly detection. The ellipsoid has higher flexibility and can better adapt to complex and asymmetric data distributions. The boundary of the ellipsoid fits the actual distribution of the data more closely, reducing the situation of surrounding too much space without data points, so as to more effectively eliminate abnormal data points in reverse, thereby ensuring the efficiency of production process monitoring while increasing the accuracy of abnormal situation recognition, and can quickly and accurately locate abnormal points, reduce the frequency of production line shutdowns, reduce production costs, and has high practicality, thereby enhancing the competitiveness of manufacturing enterprises. In addition, the use of cloud-edge collaborative architecture makes full use of the high computing performance of the cloud to ensure that the hidden correlations of the positive sample set are fully captured, further ensuring the accuracy of abnormal situation identification. Moreover, the edge layout can effectively reduce the abnormal identification process from abnormal identification delays caused by long-distance transmission. The abnormal identification process is highly real-time, ensuring that abnormal data can be accurately and quickly identified, avoiding damage to production equipment and further reducing production costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The preferred implementation modes will be described below in a clear and understandable manner with reference to the accompanying drawings to further illustrate the above-mentioned characteristics, technical features, advantages and implementation methods of the present invention.

[0020] Figure 1 It is a structural schematic diagram of a production process monitoring system based on industrial Internet provided by the present invention;

[0021] Figure 2 It is a structural diagram of a cloud-edge collaborative architecture provided by the present invention. DETAILED DESCRIPTION

[0022] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the specific implementation methods of the present invention will be described below with reference to the accompanying drawings. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings and other implementation methods can be obtained based on these drawings without creative work.

[0023] In order to simplify the drawings, only the parts related to the invention are schematically shown in each figure, and they do not represent the actual structure of the product. In addition, in order to simplify the drawings and facilitate understanding, in some figures, only one of the parts with the same structure or function is schematically drawn or marked. In this article, "one" not only means "only one", but also means "more than one".

[0024] It should be further understood that the term "and / or" used in the present description and the appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0025] In this document, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection. It can be a mechanical connection or an electrical connection. It can be directly connected or indirectly connected through an intermediate medium, and it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0026] In addition, in the description of the present invention, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0027] In one embodiment, referring to Figure 1 , showing a schematic diagram of the structure of a production process monitoring system based on the industrial Internet provided by the present invention. Figure 2 , showing a structural diagram of a cloud-edge collaborative architecture provided by the present invention.

[0028] A production process monitoring system based on industrial Internet provided by the present invention is applied to a cloud-edge collaborative architecture, wherein the cloud-edge collaborative architecture includes a cloud end, an edge end and a production terminal connected in sequence, and the cloud end is connected to the production terminal.

[0029] The cloud refers to large data centers or cloud computing platforms, which have powerful computing and storage capabilities and perform complex computing and big data analysis, such as machine learning model training, data mining, and predictive analysis; the edge is a computing node close to the production terminal, which has strong real-time processing capabilities, monitors and processes the data of the production terminal in real time, and provides fast response and low-latency services; the production terminal is the production line, including many categories of production equipment, which collects various production parameters and status data such as temperature, pressure, speed, etc. through sensors. The cloud-edge collaborative architecture combines the powerful computing and storage capabilities of the cloud with the real-time response and preprocessing functions of the edge, significantly improving the data processing efficiency and system robustness, reducing data transmission delays and bandwidth pressure, and ensuring the continuity and safety of the production process.

[0030] In a possible implementation, the cloud-edge collaborative architecture further includes: a DTS module;

[0031] The edge end is connected to the cloud via the DTS module.

[0032] Among them, the DTS (Data Transfer Service) module is a service module used to efficiently, securely and reliably transmit large amounts of data. It is usually used to migrate and synchronize data between different computing environments (such as the cloud and the edge) to ensure that the data remains consistent and complete during transmission. The DTS module can support multiple transmission protocols and provide functions such as data encryption, compression, and transmission optimization. In the cloud-edge collaborative architecture, connecting the edge and the cloud through the DTS module can achieve efficient and secure data transmission, reduce data transmission delays, and ensure data consistency and integrity, thereby improving the reliability and real-time performance of the overall system.

[0033] The production process monitoring system based on the Industrial Internet includes:

[0034] Collection module 1, used to collect equipment operation data under qualified production indicators;

[0035] Among them, the equipment operation data refers to the basic operating parameters of the equipment involved in the production process. The equipment operation data under the qualified production index is the equipment operation data that ensures that the production quality and production speed meet the requirements of the enterprise, that is, the equipment operation data under the normal operation of the production line, including electrical parameters such as voltage, current or power consumption, mechanical performance parameters such as vibration, temperature or speed, operating parameters such as pressure or flow, and state indication data such as open state or closed state. The production data includes production quality, that is, the quality of the products produced under different equipment operation data, and production speed, that is, the output speed of the products under different equipment operation data. In the actual application process, according to the different production equipment, other dimensions of equipment operation data are also included, not just limited to the listed data. Specifically, the collected equipment operation data includes equipment operation data under two or more different production qualities and two or more different production speeds, so as to capture the qualified production index, that is, the changing rules and correlations of each equipment operation parameter under changing conditions.

[0036] In a possible implementation manner, the qualified production index includes a qualified production quality index and a qualified production speed index.

[0037] Among them, the production qualification index is the index quantified by the enterprise itself, that is, the production quality qualification index is the first-level product, the second-level product and the third-level product. Products that do not meet these indicators are unqualified products. The same is true for the production speed qualification index, which is the production speed qualification range specified by the enterprise itself, such as one piece per minute, two pieces per minute, and three pieces per minute. Within this production speed range, it is qualified and meets the requirements. If it is too fast or too slow, the equipment may have malfunction problems. In the actual process, the production quality qualification index may be subdivided into dozens of levels according to different enterprises, that is, these levels of production are all qualified, and the same is true for production speed.

[0038] Uploading module 2, used to upload the collected equipment operation data to the cloud;

[0039] It is understandable that for data processing that requires a lot of computing resources, uploading to the cloud for fast and efficient analysis.

[0040] In a possible implementation, it further includes:

[0041] The equipment operation data is preprocessed to remove abnormal values, wherein the abnormal values ​​are data exceeding the upper and lower limits of the speed equipment operation data.

[0042] Among them, the data beyond the upper and lower limits of equipment operation data is the range of abnormal equipment operation. During the production process, equipment operation data may be affected by various factors, resulting in abnormal values. Abnormal values ​​are usually values ​​that exceed the upper and lower limits of normal equipment operation data, indicating that the equipment may be in an abnormal operating state. Preprocessing the equipment operation data to remove abnormal values ​​can ensure the accuracy and reliability of subsequent data analysis and model training.

[0043] The first construction module 3 is used to construct a conditional generative adversarial network model with the qualified production index as a condition;

[0044] Among them, the conditional generative adversarial network model is a variant based on the generative adversarial network. By introducing conditional variables in the generator and the discriminator, the generated samples are not only authentic, but also meet specific conditions, namely, qualified production index requirements. On the basis of the original GAN, additional information (such as labels or conditions) is provided to the generator and the discriminator to guide the data generation and discrimination process.

[0045] In a possible implementation, the conditional generative adversarial network model includes a generator and a discriminator.

[0046] Among them, the generator is used to accept random noise and conditional variables (production quality and production speed) to generate new equipment operation data. The goal of the generator is to generate samples that can deceive the discriminator so that the discriminator cannot distinguish between generated data and real data. The discriminator is used to determine whether the equipment operation data generated under given conditional variables is real data, and to distinguish whether the input data is real data or generated data. The goal of the discriminator is to maximize the judgment results of real data while minimizing the wrong judgment of generated data. In the production process monitoring system based on the industrial Internet, the conditional generative adversarial network model realizes the generation and accurate judgment of high-quality data by introducing two components, the generator and the discriminator. By using CGAN, data samples that meet specific production conditions can be generated, the training data set can be expanded, the accuracy and reliability of the monitoring system can be improved, resource utilization can be optimized, and the performance of the overall system can be improved.

[0047] A first training module 4 is used to train the conditional generative adversarial grid model using the equipment operation data and the corresponding qualified production indicators as a training set;

[0048] It should be noted that during the training process, the generative adversarial network model captures the changes in the qualified production indicators, captures how the equipment operation data of each dimension changes, and takes the equipment operation data of each dimension as a whole, capturing the hidden correlation between the production indicators of each dimension when the conditions change. Through the training process, the generative adversarial grid model learns this change law, and then can obtain complete equipment operation data under different qualified production indicators, making up for the difficulty in collecting complete equipment operation data of the entire normal production process in reality. Through the above process, the conditional generative adversarial grid model can generate equipment operation data that meets the requirements under different conditions, thereby capturing and simulating the dynamic changes and correlations of equipment operation data in the production process.

[0049] In a possible implementation manner, the first training module 4 is specifically used for:

[0050] S401: using the generator to generate equipment operation data under given qualified production indicators;

[0051] S402: Using the generated equipment operation data as generated data, and using the collected equipment operation data corresponding to the same qualified production index as the generated process as real data, training the discriminator in the conditional generative adversarial network model with the goal of maximizing the loss function of the discriminator:

[0052]

[0053] Among them, max means taking the maximum value, L D represents the loss function of the discriminator D, and x represents the distribution of real data p data (x), c represents the data sampled from the conditional distribution p c The conditional variables sampled in (c) are represents the expected value of the real data, represents the expected value of the generated data, G(z,c) represents the equipment operation data generated under the random noise vector z and the conditional variable c representing the qualified production index, and D(G(z,c) represents the discriminant result of the discriminator on the generated equipment operation data;

[0054] S403: The generator is trained with the goal of minimizing the discrimination error of the trained discriminator on the generated data output by the generator:

[0055]

[0056] Among them, min means taking the minimum value, L G represents the loss function of the generator G,

[0057] represents the expected value of the data generated under the random noise vector z and the conditional variable c, z:p z (z) and c:p c (c) The distribution of random noise vector and conditional variable, respectively, D represents the discriminator;

[0058] The distribution of the random noise vector is usually a standard normal distribution (i.e. a normal distribution with a mean of 0 and a variance of 1). The generator receives this random noise as input to generate new data samples. The distribution of the conditional variable. The conditional variable can be any additional information, i.e., production quality and production speed. Both the generator and the discriminator receive this conditional variable as input.

[0059] The hyperparameters of the discriminator and generator are fixed during repeated training.

[0060] S404: The number of iterations is increased by 1, and the process returns to step S401 until all the collected equipment operation data under the qualified production indicators are consumed or the number of iterations is greater than the preset number of iterations.

[0061] Specifically, first, in step S401, the generator uses random noise and a given qualified production index to generate equipment operation data. Next, in step S402, the generated data is used as the generated data, and the real collected equipment operation data corresponding to the same qualified production index as the generation process is used as the real data, and the discriminator is trained with the goal of maximizing the loss function of the discriminator, that is, the goal of the discriminator is to distinguish between real data and generated data. Then, in step S403, the generator is trained with the goal of minimizing the discrimination error of the trained discriminator on the generated data output by the generator, that is, the goal of the generator is to generate data that can deceive the discriminator. During the training process, the distribution of the random noise vector is usually a standard normal distribution, and both the generator and the discriminator receive conditional variables as input. The above process is repeated with fixed discriminator and generator hyperparameters until the collected equipment operation data under the qualified production index is consumed or the number of iterations exceeds the preset value. In this way, the conditional generative adversarial network model can generate high-quality equipment operation data that meets specific production indicators, improving the diversity of the data set and the generalization ability of the model.

[0062] The first generating module 5 is used to generate equipment operation data that meets the qualified production indicators through the trained generative adversarial network model.

[0063] Among them, the collected equipment operation data and the generated equipment operation data constitute the positive sample set of the production process;

[0064] It should be noted that in the production process, the identification of abnormal data is related to whether the production process can operate normally. There is a large amount of high-dimensional production data in the production process, and the efficiency of manual monitoring is extremely low. In the existing technology, abnormal data identification through machine learning has been widely popularized, but the machine model is often trained through a complete training set. The complete abnormal data set in the production process is extremely difficult to obtain. This directly leads to the current model training based on the complete data set. The model has limited knowledge of abnormal data learned, insufficient generalization ability, and cannot effectively identify abnormal production data. The positive sample set of the production process generated by the trained generative adversarial network model covers the normal production status in most cases, greatly expanding the positive sample set. Compared with the traditional method of directly collecting positive sample sets, this method is faster and more efficient, and can cover a wider range of normal operation, thereby effectively improving the ability to reversely identify abnormal conditions.

[0065] A second construction module 6 is used to construct a support vector data description model that uses an ellipsoid to describe the positive sample set of the production process;

[0066] It should be noted that in high-dimensional space, data often has sparse and complex distribution characteristics. Sparsity means that in high-dimensional space, the distance between data points is usually large and the data points are more dispersed. A hypersphere may not be able to effectively surround all normal data points in high-dimensional space while excluding abnormal points. As the dimension increases, the volume of the hypersphere grows too fast, resulting in a decrease in the ability to surround normal data. The hypersphere constructed by the conventional support vector data description model tends to surround too much space, that is, in order to contain all normal data points, the volume of the hypersphere will be too large, surrounding a large amount of space without data points. This results in the abnormal points being contained in the hypersphere due to the overly simple boundary of the hypersphere, affecting the subsequent misjudgment of abnormal data points. This scheme uses an ellipsoid to describe the geometric shape of the distribution of positive sample data. By calculating the center and covariance matrix of the ellipsoid, the distribution shape of the data can be accurately described, thereby improving the accuracy of anomaly detection. Compared with the hypersphere, the ellipsoid has higher flexibility and can better adapt to complex and asymmetric data distribution. The boundary of the ellipsoid fits the actual distribution of the data more closely, reducing the situation of surrounding too much space without data points, so as to more effectively reversely exclude abnormal data points.

[0067] A second generating module 7, configured to train the support vector data description model using the production process positive sample set, and generate a minimum ellipsoid corresponding to the production process positive sample set;

[0068] It should be noted that the support vector data description model is trained using the positive sample set of the production process to generate the minimum ellipsoid corresponding to the positive sample set of the production process. By training the model, a minimum ellipsoid is determined that can enclose all the equipment operation data under normal operating conditions, thereby accurately describing the distribution shape of normal data and improving the accuracy and reliability of subsequent anomaly detection.

[0069] In a possible implementation manner, the second generating module 7 is specifically configured to:

[0070] S701: Standardizing the positive sample set of the production process;

[0071] S702: Determine the minimum ellipsoid through the standardized production process positive sample set, and the objective function and constraint conditions of the minimum ellipsoid are:

[0072] The objective function:

[0073] Among them, min represents the ellipsoid center a and the ellipsoid covariance matrix A under the minimum value of the function, ξ i represents the slack variable that allows the i-th data point in the standardized production process positive sample set to deviate from the ellipsoid, det(A) represents the determinant of the covariance matrix A, C represents the penalty coefficient, and n represents the total number of data points in the standardized production process positive sample set;

[0074] The logarithm of the determinant is used to minimize the volume of the ellipsoid.

[0075] The constraints: (x i -a) T A(x i -a)≤1+ξ i ,ξ i ≥0

[0076] Among them, x i Represents the i-th data point in the normalized production process positive sample set;

[0077] S703: Solve the objective function to obtain the minimum ellipsoid.

[0078] Specifically, first, the positive sample set of the production process is standardized to eliminate the dimensional differences between different data dimensions. Next, the minimum ellipsoid is determined by the standardized positive sample set, where the objective function is to minimize the logarithm of the determinant of the ellipsoid center and the covariance matrix, and includes slack variables and penalty coefficients to control the tolerance to outliers. The constraints ensure that each data point after standardization is within the ellipsoid or within the allowable deviation range. The objective function is then solved to obtain the minimum ellipsoid. This process accurately describes the distribution of positive sample data by minimizing the volume of the ellipsoid, thereby improving the accuracy of anomaly detection. The advantage of using the minimum ellipsoid to describe the positive sample set is that it can more accurately adapt to the true distribution shape of the data, especially in high-dimensional space. The ellipsoid has high flexibility and can better surround normal data, reduce the possibility of mis-encapsulating anomalies, and improve the accuracy and robustness of anomaly detection.

[0079] In a possible implementation manner, the S703 is specifically used to:

[0080] S703A: Introduce Lagrangian multipliers, and construct a Lagrangian function based on the objective function and the constraints of the objective function:

[0081]

[0082] Among them, L(a, A, ξ, α, η) represents the ellipsoid center a, the ellipsoid covariance matrix A, the slack variable ξ, and the constraint (x i -a) T A(x i -a)≤1+ξ i The first Lagrange multiplier α is used to constrain ξ i The Lagrangian function of the second Lagrangian multiplier η ≥ 0, the subscript i represents the i-th data point x in the normalized production process positive sample set i association;

[0083] S703B: Establish the optimization target of the Lagrangian function:

[0084]

[0085] Among them, max α represents the first Lagrange multiplier when the function is maximum, <x i ,x i > represents the i-th data point x i The inner product of represents the double summation, that is, the i-th data point x i With the jth data point x j The inner product between is the inner product between all data points;

[0086] S703C: jointly solving the optimization objective and the Lagrangian function to obtain the ellipsoid center and the ellipsoid covariance matrix, that is, to obtain the minimum ellipsoid.

[0087] Among them, the center a of the ellipsoid and the covariance matrix A can accurately describe the distribution of the data. The center a represents the geometric center of the data, while the covariance matrix A describes the distribution of the data in all directions. By determining the optimal ellipsoid center and covariance matrix, we can build an ellipsoid model for detecting abnormal data points, and any data point outside the ellipsoid can be considered an anomaly. By optimizing the center and covariance matrix, we can ensure that the ellipsoid can tightly surround normal data points while having a certain tolerance for abnormal points. This makes the model more robust and practical. The center and covariance matrix are because these two parameters can most accurately describe the distribution of the data and build an effective anomaly detection model. By optimizing the center and covariance matrix, it can be ensured that the model has good recognition capabilities for both normal and abnormal data points.

[0088] A sending module 8, used for sending the trained support vector data description model to the edge end;

[0089] It should be noted that the edge is a computing terminal that is very close to the production terminal and has a certain computing capability. Sending the support vector data description model to the edge for abnormal data identification can improve the speed and real-time performance of identifying abnormal conditions in equipment operation data, avoid abnormal identification delays caused by long-distance data transmission, and subsequently cause damage to production equipment and corporate losses.

[0090] The output module 9 is used to collect the real-time equipment operation data of the production terminal, input the real-time equipment operation data into the trained support vector data description model, and output abnormal data.

[0091] The abnormal data is real-time equipment operation data that exceeds the minimum ellipsoid.

[0092] It should be noted that this method can quickly locate abnormal data and abnormal equipment from massive equipment operation data, improve equipment maintenance efficiency, and collect real-time equipment operation data from production terminals and input it into the trained support vector data description model for analysis. Abnormal data that exceeds the minimum ellipsoid range can be identified and output in real time. This method can quickly locate abnormal situations and abnormal equipment from a large amount of equipment operation data, thereby reducing the maintenance time of equipment failures and improving production efficiency and benefits.

[0093] In a possible implementation manner, the output module 9 is specifically used for:

[0094] S901: Calculate the Mahalanobis distance between the data point in the real-time device operation data and the ellipsoid center of the minimum ellipsoid:

[0095] d 2 (x new )=(x new -a) T A -1 (x new -a)

[0096] Among them, d 2 (x new ) represents the data point x in the real-time device operation data new The Mahalanobis distance from the ellipsoid center a of the minimum ellipsoid, A -1 represents the inverse matrix of the covariance matrix of the minimum ellipsoid, and the subscript T represents the transpose;

[0097] S902: When the Mahalanobis distance is less than or greater than 1, the data point x in the real-time device operation data is new As abnormal data output, otherwise, the data point x in the real-time equipment operation data new It is normal data.

[0098] Among them, the Mahalanobis distance is an effective method to measure the distance between a data point and the center of distribution, and is particularly suitable for high-dimensional data. Unlike the Euclidean distance, the Mahalanobis distance takes into account the covariance structure of the data, so it can measure the correlation of data points in multidimensional space. Specifically, the Mahalanobis distance can standardize data of different scales, consider the correlation between variables, and provide a scale-independent distance measure. By calculating the Mahalanobis distance between the real-time equipment operation data and the center of the minimum ellipsoid, it is determined whether the data point is abnormal. The Mahalanobis distance takes into account the correlation and distribution shape of the data, and can accurately identify abnormal data beyond the normal range, thereby improving the accuracy and real-time performance of anomaly detection, quickly locating problems that may occur in the production process, and ensuring the stability and efficiency of the production process.

[0099] In actual application, a single-class sample training set is used, that is, only normal samples with normal production process and qualified production quality are used to train the model. In this process, a normal data ellipsoid is constructed that can accommodate conventional normal data and individual normal outlier data, so that the production process data is converted into a simple two-classification task, that is, normal data or abnormal data. When the boundary range of normal data is determined, it is also possible to judge whether the production data is abnormal data. That is, if it does not belong to the normal data range, it is abnormal data, thereby completing the monitoring of the production process.

[0100] In a possible implementation, it further includes:

[0101] The conditional generative adversarial network model and the support vector data description model are retrained at preset time intervals.

[0102] It is understandable that the system also includes retraining the conditional generative adversarial network model and the support vector data description model at preset intervals to adapt to new production data and environmental changes, thereby ensuring the continued accuracy and effectiveness of the monitoring system.

[0103] It should be noted that those skilled in the art can set the preset duration according to actual needs, and the present invention is not limited thereto.

[0104] Compared with the prior art, the present invention has at least the following beneficial technical effects:

[0105] In the present invention, a multi-class sample training process is converted into a single-class sample training process. In this process, a conditional generative adversarial network model is constructed with the qualified production index as a condition. By learning the correlation and change law between the collected real equipment operation data, normal equipment operation data that is difficult to collect completely is generated to expand the positive sample set. Then, based on the positive sample, a support vector data description model that uses an ellipsoid to describe the positive sample set of the production process is constructed. The distribution shape of the normal production process data can be accurately described, thereby improving the accuracy of anomaly detection. The ellipsoid has higher flexibility and can better adapt to complex and asymmetric data distributions. The boundary of the ellipsoid fits the actual distribution of the data more closely, reducing the situation of surrounding too much space without data points, so as to more effectively eliminate abnormal data points in reverse, thereby ensuring the efficiency of production process monitoring while increasing the accuracy of abnormal situation recognition, and can quickly and accurately locate abnormal points, reduce the frequency of production line shutdowns, reduce production costs, and has high practicality, thereby enhancing the competitiveness of manufacturing enterprises. In addition, the use of cloud-edge collaborative architecture makes full use of the high computing performance of the cloud to ensure that the hidden correlations of the positive sample set are fully captured, further ensuring the accuracy of abnormal situation identification. Moreover, the edge layout can effectively reduce the abnormal identification process from abnormal identification delays caused by long-distance transmission. The abnormal identification process is highly real-time, ensuring that abnormal data can be accurately and quickly identified, avoiding damage to production equipment and further reducing production costs.

[0106] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0107] The above embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for those of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A production process monitoring system based on industrial Internet, characterized in that: Applied to a cloud-edge collaborative architecture, the cloud-edge collaborative architecture includes a cloud, an edge, and a production terminal connected in sequence, the cloud is connected to the production terminal, and the production process monitoring system based on the industrial Internet includes: The collection module is used to collect equipment operation data under qualified production indicators; An uploading module, used to upload the collected equipment operation data to the cloud; A first construction module is used to construct a conditional generative adversarial network model with the qualified production index as a condition; A first training module is used to train the conditional generative adversarial grid model using the equipment operation data and the corresponding qualified production indicators as a training set; A first generation module is used to generate equipment operation data that meets the qualified production index through a trained generative adversarial network model, wherein the collected equipment operation data and the generated equipment operation data constitute a positive sample set of the production process; A second construction module is used to construct a support vector data description model that uses an ellipsoid to describe the positive sample set of the production process; A second generating module is used to train the support vector data description model using the production process positive sample set to generate a minimum ellipsoid corresponding to the production process positive sample set; A sending module, used for sending the trained support vector data description model to the edge end; The output module is used to collect the real-time equipment operation data of the production terminal, input the real-time equipment operation data into the trained support vector data description model, and output abnormal data, wherein the abnormal data is the real-time equipment operation data that exceeds the minimum ellipsoid.

2. The production process monitoring system based on the industrial Internet according to claim 1 is characterized in that: The cloud-edge collaborative architecture also includes: a DTS module; The edge end is connected to the cloud via the DTS module.

3. The production process monitoring system based on industrial Internet according to claim 1 is characterized in that: The qualified production indicators include qualified production quality indicators and qualified production speed indicators.

4. The production process monitoring system based on industrial Internet according to claim 1 is characterized in that: Also includes: The equipment operation data is preprocessed to remove abnormal values, wherein the abnormal values ​​are data exceeding the upper and lower limits of the speed equipment operation data.

5. The production process monitoring system based on industrial Internet according to claim 1 is characterized in that: The conditional generative adversarial network model includes a generator and a discriminator.

6. The production process monitoring system based on the industrial Internet according to claim 5 is characterized in that: The first training module is specifically used for: S401: using the generator to generate equipment operation data under given qualified production indicators; S402: Using the generated equipment operation data as generated data, and using the collected equipment operation data corresponding to the same qualified production index as the generated process as real data, training the discriminator in the conditional generative adversarial network model with the goal of maximizing the loss function of the discriminator: Among them, max means taking the maximum value, L D represents the loss function of the discriminator D, and x represents the distribution of real data p data (x), c represents the data sampled from the conditional distribution p c The conditional variables sampled in (c) are represents the expected value of the real data, represents the expected value of the generated data, G(z,c) represents the equipment operation data generated under the random noise vector z and the conditional variable c representing the qualified production index, and D(G(z,c) represents the discriminant result of the discriminator on the generated equipment operation data; S403: The generator is trained with the goal of minimizing the discrimination error of the trained discriminator on the generated data output by the generator: Among them, min means taking the minimum value, L G represents the loss function of the generator G, represents the expected value of the data generated under the random noise vector z and the conditional variable c, z:p z (z) and c:p c (c) The distribution of random noise vector and conditional variable, respectively, D represents the discriminator; S404: The number of iterations is increased by 1, and the process returns to step S401 until all the collected equipment operation data under the qualified production indicators are consumed or the number of iterations is greater than the preset number of iterations.

7. The production process monitoring system based on industrial Internet according to claim 1 is characterized in that: The second generating module is specifically used for: S701: Standardizing the positive sample set of the production process; S702: Determine the minimum ellipsoid through the standardized production process positive sample set, and the objective function and constraint conditions of the minimum ellipsoid are: The objective function: Among them, min represents the ellipsoid center a and the ellipsoid covariance matrix A under the minimum value of the function, ξ i represents the slack variable that allows the i-th data point in the standardized production process positive sample set to deviate from the ellipsoid, det(A) represents the determinant of the covariance matrix A, C represents the penalty coefficient, and n represents the total number of data points in the standardized production process positive sample set; The constraints: (x i -a) T A(x i -a)≤1+ξ i ,ξ i ≥0 Among them, x i Represents the i-th data point in the normalized production process positive sample set; S703: Solve the objective function to obtain the minimum ellipsoid.

8. The production process monitoring system based on industrial Internet according to claim 7 is characterized in that: The S703 is specifically used for: S703A: Introduce Lagrangian multipliers, and construct a Lagrangian function based on the objective function and the constraints of the objective function: Among them, L(a, A, ξ, α, η) represents the ellipsoid center a, the ellipsoid covariance matrix A, the slack variable ξ, and the constraint (x i -a) T A(x i -a)≤1+ξ i The first Lagrange multiplier α is used to constrain ξ i The Lagrangian function of the second Lagrangian multiplier η ≥ 0, the subscript i represents the i-th data point x in the normalized production process positive sample set i association; S703B: Establish the optimization target of the Lagrangian function: Among them, max α represents the first Lagrange multiplier when the function is maximum, <x i ,x i > represents the i-th data point x i The inner product of represents the double summation, that is, the i-th data point x i With the jth data point x j The inner product between is the inner product between all data points; S703C: jointly solving the optimization objective and the Lagrangian function to obtain the ellipsoid center and the ellipsoid covariance matrix, that is, to obtain the minimum ellipsoid.

9. The production process monitoring system based on industrial Internet according to claim 1 is characterized in that: The output module is specifically used for: S901: Calculate the Mahalanobis distance between the data point in the real-time device operation data and the ellipsoid center of the minimum ellipsoid: d 2 (x new )=(x new -a) T A -1 (x new -a) Among them, d 2 (x new ) represents the data point x in the real-time device operation data new The Mahalanobis distance from the ellipsoid center a of the minimum ellipsoid, A -1 represents the inverse matrix of the covariance matrix of the minimum ellipsoid, and the subscript T represents the transpose; S902: When the Mahalanobis distance is less than or greater than 1, the data point x in the real-time device operation data is new As abnormal data output, otherwise, the data point x in the real-time equipment operation data new It is normal data.

10. The production process monitoring system based on industrial Internet according to claim 1 is characterized in that: Also includes: The conditional generative adversarial network model and the support vector data description model are retrained at preset time intervals.

Citation Information

Patent Citations

  • Distribution network operation state identification method and system based on data enhancement technology

    CN112016774A

  • Complex industrial process monitoring method and system based on cloud edge collaborative dictionary learning

    CN115840419A

  • Abnormality detection method and application thereof in aero-engine component fault early warning

    CN116611310A

  • Rotary machine few-sample fault diagnosis system based on meta learning under cloud-edge cooperation

    CN117093910A

Cited By

  • Comprehensive pipe gallery dynamic risk prediction method and system based on Internet of Things

    CN119557569A

  • An Internet of Things-based dynamic risk prediction method and system for integrated pipe corridors

    CN119557569B

  • Network data asset security discovery system based on distributed machine learning

    CN119728210A

  • A network data asset security discovery system based on distributed machine learning

    CN119728210B