A neural network-based self-learning detection system for industrial control data anomalies

Abstract

The application relates to the field of self-learning detection, in particular to a self-learning detection system for abnormal industrial control data based on a neural network, which comprises a continuous command module used for sending operation commands to an industrial control system at intervals of a preset time, a data collection module used for collecting real-time industrial control data of the industrial control system, an abnormality prediction module used for selecting a plurality of learning features, pre-processing the real-time industrial control data, learning the industrial control pre-processed data by using a neural network model, and generating a corresponding abnormality possibility prediction graph, and a fault adjustment module used for comparing a possibility value in the abnormality possibility prediction graph with an abnormality possibility threshold value, positioning an abnormal axis of a machine tool when the fault adjustment module determines that the industrial control data is abnormal, and realizing full-process automation from data collection to abnormality detection and fault positioning, so that the precision and efficiency of abnormality detection are improved, and the detection model is continuously optimized through a self-learning function to adapt to dynamic changes of the industrial control system.

Description

TECHNICAL FIELD

[0001] The application relates to the field of self-learning detection, in particular to a self-learning detection system for industrial control data anomalies based on a neural network. BACKGROUND

[0002] Industrial control systems are systems used for monitoring and managing physical devices and processes in industrial environments, widely used in manufacturing, energy, chemical industry, transportation and other fields. These systems usually include programmable logic controllers (PLCs), distributed control systems (DCSs), supervisory control and data acquisition systems (SCADAs) and the like. With the rapid development of industrial automation and informatization, the complexity and scale of industrial control systems are increasing, and higher requirements are put forward for the reliability and security of the systems. With the rapid development of industrial automation and informatization, the complexity and scale of industrial control systems are increasing, and traditional anomaly detection methods have been difficult to meet the needs of modern industrial production. The self-learning detection system for industrial control data anomalies based on a neural network emerges as the times require, which can monitor and automatically detect anomalies in industrial control systems in real time through self-learning ability and nonlinear modeling advantages, and improve the reliability, security and production efficiency of the system.

[0003] Chinese Patent No. CN107370732B discloses an industrial control system abnormal behavior discovery system based on neural network and optimal recommendation, which analyzes industrial control system flow data, reduces high-dimensional data in the industrial control system, realizes irrelevant attribute deletion and redundant attribute redundancy reduction, and realizes dynamic selection of intrusion features while following the inherent relationship between real input and output samples. The system models different scene detected flow data, compares sample flow and modeled flow to detect the difference between the characteristics of the detected flow data and the characteristics of normal flow, and discovers abnormal flow. The invention is based on the optimal recommendation theory to compare and analyze the flow, establishes a malicious flow (scene) optimal recommendation model, uses cloud computing technology, and designs and realizes a cloud platform-based on-demand customized industrial control device and system malicious attack behavior discovery and detection function.

[0004] Chinese Patent Authorization Announcement No. CN115081585B discloses a method for detecting abnormal states of human-machine-object collaboration using a reinforced heterogeneous graph neural network. The method includes collecting data from multiple sensor measurement and control systems on intelligent production lines in industrial production to form raw data for state detection; determining the associations between multi-source heterogeneous raw data, establishing a heterogeneous information network graph, and using a reinforced graph neural network for representation learning and classification to obtain an abnormal state detection discriminator; and automatically exploring meta-structures, aggregating information based on the meta-structures, to achieve abnormal state detection. This invention aims to ensure the production safety of workshop employees and achieve full-cycle monitoring and timely maintenance of production line equipment and assembly products. It can simultaneously detect abnormal states of human-machine-object collaboration using multi-source heterogeneous data, thereby meeting the monitoring needs of the production line data lifecycle in intelligent manufacturing.

[0005] However, the above method has the following problems: it is impossible to trace the fault location in real time, thereby failing to improve the efficiency of data anomaly detection. Summary of the Invention

[0006] To this end, the present invention provides a self-learning detection system for industrial control data anomalies based on a neural network, which is used to overcome the problem in the prior art that the fault location cannot be traced in real time, thereby failing to improve the efficiency of data anomaly detection.

[0007] To achieve the above objectives, the present invention provides a self-learning detection system for industrial control data anomalies based on a neural network, comprising:

[0008] Continuous command module, used to send operation commands to the industrial control system at preset intervals;

[0009] a data collection module connected to the continuous command module and configured to collect real-time industrial control data during the process of the industrial control system executing the operation command;

[0010] an abnormality prediction module, connected to the data collection module, configured to select a number of learning features based on the real-time industrial control data, pre-process the real-time industrial control data to form corresponding industrial control pre-processed data, and learn the industrial control pre-processed data using a neural network model to generate a corresponding abnormality possibility prediction map;

[0011] a fault adjustment module connected to the abnormality prediction module, configured to compare the probability value in the abnormality probability prediction map with an abnormality probability threshold value; when the probability value is greater than the abnormality probability threshold value, the fault adjustment module determines that the industrial control data is abnormal and locates the abnormal axis of the machine tool;

[0012] Wherein, the preset time is related to the operating performance of the industrial control system;

[0013] The real-time industrial control data is an industrial control parameter generated by each axis of the machine tool in the process of executing the operation command by the industrial control system.

[0014] The industrial control parameter includes a speed, an acceleration, and a torque of each axis of the machine tool.

[0015] The learning feature includes a speed change rate, an acceleration change rate, and a torque change rate.

[0016] The neural network model is generated by training an industrial control training set formed by the industrial control pre-processing data.

[0017] The abnormality possibility threshold is a critical value of the possibility of the abnormality of each axis of the machine tool, and is related to the synchronism of each axis of the machine tool during operation.

[0018] Further, the continuous command module comprises:

[0019] a storage device configured to store the operation command;

[0020] a command transmitting device connected to the storage device and configured to send the corresponding operation command to the industrial control system;

[0021] a timing device connected to the command transmitting device and configured to trigger the command transmitting device at intervals of the preset time and to time mark the operation command in the order of transmission time.

[0022] Further, the data collection module comprises:

[0023] a data sensor configured to monitor and collect the real-time industrial control data corresponding to each axis of the machine tool;

[0024] a data adjuster connected to the data sensor and configured to correct data deviation of the real-time industrial control data;

[0025] a data converter connected to the data adjuster and configured to perform digital signal conversion processing on the real-time industrial control data and to transmit the real-time industrial control data to the abnormality prediction module.

[0026] Further, the abnormality prediction module comprises:

[0027] a feature selector configured to select a plurality of learning features corresponding to the real-time industrial control data;

[0028] a pre-processor connected to the feature selector and configured to pre-process the real-time industrial control data and to generate corresponding industrial control pre-processing data;

[0029] a learner connected with the preprocessor, configured to learn the industrial control preprocessed data according to the learning features and generate a corresponding abnormal possibility prediction graph.

[0030] Further, the fault adjustment module comprises:

[0031] a comparator configured to extract a possibility value in the abnormal possibility prediction graph and compare the possibility value with the abnormal possibility threshold to form a corresponding comparison result;

[0032] a determiner connected with the comparator, configured to determine the running state of the industrial control data according to the comparison result;

[0033] a tracer connected with the determiner, configured to locate the abnormal axis of the machine tool according to the emission time sequence when the industrial control data is abnormal.

[0034] Further, the timing device generates a trigger signal according to a corresponding preset time, and the command emission device receives the trigger signal from the timing device, reads the running command from the storage device according to the order of the trigger signal and sends it to the industrial control system.

[0035] Further, the data sensor is placed on each axis of the machine tool and connected with each axis of the machine tool through a network, monitors and collects corresponding real-time industrial control data of each axis of the machine tool, the data adjuster reads the collected real-time industrial control data and performs data deviation correction, the data converter converts the analog signal in the real-time industrial control data into a digital signal, and transmits the converted real-time industrial control data to the abnormal prediction module.

[0036] Wherein, the data deviation correction is to read the missing value in the real-time industrial control data and perform linear interpolation filling on the missing value.

[0037] Further, the feature selector selects a plurality of learning features according to the real-time industrial control data, the preprocessor cuts the real-time industrial control data according to a sampling rate to form a plurality of industrial control preprocessed data with a standard sampling rate, the industrial control preprocessed data enters the neural network model for learning and generates a corresponding abnormal possibility prediction graph.

[0038] Wherein, the standard sampling rate is a sampling rate that can be recognized by the neural network model, and the standard sampling rate remains unchanged in the learning process of the neural network model.

[0039] Further, when the possibility value is less than the abnormal possibility threshold, the determiner determines that the industrial control data is normal, and the fault adjustment module does not adjust the industrial control system.

[0040] Further, when the possibility value is greater than the abnormal possibility threshold, the determinator determines that the industrial control data is abnormal, and the tracer traces the current possibility value according to the emission time sequence and locates the abnormal axis of the machine tool.

[0041] Compared with the prior art, the present application uses a continuous command module to send operation commands to the industrial control system at intervals of a preset time, a data collection module to collect real-time industrial control data of the industrial control system, an abnormality prediction module to select a plurality of learning features, pre-process the real-time industrial control data, and use a neural network model to learn the industrial pre-processed data to generate a corresponding abnormality possibility prediction graph, and a fault adjustment module to compare the possibility value in the abnormality possibility prediction graph with the abnormality possibility threshold, locate the abnormal axis of the machine tool when the fault adjustment module determines that the industrial control data is abnormal, and realize the full-process automation from data collection to abnormality detection and fault location, thereby improving the accuracy and efficiency of abnormality detection, and continuously optimizing the detection model through self-learning function to adapt to the dynamic changes of the industrial control system.

[0042] Further, the timing device is used to trigger the command emission device at intervals of a preset time, and the operation commands are time-labeled in emission time sequence, and the storage device, command emission device and timing device are used to work together to realize accurate control and sending of the operation commands of the industrial control system, thereby improving the automation level of the industrial control system and providing basic support for subsequent abnormality detection and fault prediction.

[0043] Further, the real-time industrial control data is subjected to data deviation correction processing, and the data converter is used to perform digital signal conversion processing on the real-time industrial control data, thereby realizing real-time monitoring, data processing and abnormality prediction of the machine tool operation state, and improving the reliability and operation efficiency of the equipment.

[0044] Further, the feature selector is used to filter out key features from the collected real-time industrial control data, and the pre-processed data is input into the neural network model to predict the possibility of abnormality of the industrial control system, thereby reducing the data dimension, reducing the complexity of calculation, removing irrelevant features, improving the identification ability of the neural network model for abnormal data, and converting the original data into a format suitable for subsequent learning and analysis, thereby improving the data quality and consistency.

[0045] Further, the comparator is used to extract the possibility value from the abnormality possibility prediction graph, compare it with the preset abnormality possibility threshold, generate a comparison result, and determine the operation state of the industrial control data according to the comparison result, and locate the abnormal axis of the machine tool according to the emission time sequence, thereby efficiently handling abnormal situations, quickly locating the fault source, and improving the overall performance and reliability of the system. BRIEF DESCRIPTION OF DRAWINGS

[0046] Figure 1 FIG. 1 is a structural schematic diagram of a self-learning detection system for abnormal industrial control data based on a neural network according to an embodiment of the present application;

[0047] Figure 2 FIG. 2 is a structural schematic diagram of a continuous command module according to an embodiment of the present application;

[0048] Figure 3 FIG. 3 is a structural schematic diagram of a data collection module according to an embodiment of the present application;

[0049] Figure 4 FIG. 4 is a structural schematic diagram of a fault adjustment module according to an embodiment of the present application. DETAILED DESCRIPTION

[0050] In order to make the objects and advantages of the present application clearer, the present application will be further described in conjunction with embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and should not be used to limit the present application.

[0051] The preferred embodiments of the present application will be described below with reference to the accompanying drawings. It should be understood by those skilled in the art that the embodiments are only used to explain the technical principles of the present application and are not used to limit the protection scope of the present application.

[0052] It should be noted that, in the description of the present application, the terms of direction or position relationship such as "upper", "lower", "left", "right", "inner", "outer" and the like are based on the direction or position relationship shown in the drawings, which is only for the convenience of description and does not indicate or imply that the device or element must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present application.

[0053] In addition, it should also be noted that, in the description of the present application, unless otherwise explicitly specified and limited, the terms "mounting", "connecting", "connection" should be understood in a broad sense, for example, it can be fixed connection, or detachable connection, or integrally connected; it can be mechanical connection, or electrical connection; it can be directly connected, or indirectly connected through an intermediate medium, or the communication inside two elements. Those skilled in the art can understand the specific meaning of the above terms in the present application according to the specific circumstances.

[0054] Please refer to Figure 1 FIG. 1 is a structural schematic diagram of a self-learning detection system for abnormal industrial control data based on a neural network according to an embodiment of the present application, which comprises:

[0055] The continuous command module is used to send a running command to the industrial control system at a preset interval;

[0056] a data collection module connected with the continuous command module, configured to collect real-time industrial control data in the process of the industrial control system executing the running command;

[0057] an anomaly prediction module connected with the data collection module, configured to select a plurality of learning features according to the real-time industrial control data, pre-process the real-time industrial control data to form corresponding industrial control pre-processed data, and learn the industrial control pre-processed data by using a neural network model to generate a corresponding anomaly likelihood prediction graph;

[0058] a fault adjustment module connected with the anomaly prediction module, configured to compare the likelihood value in the anomaly likelihood prediction graph with an anomaly likelihood threshold value, and when the likelihood value is greater than the anomaly likelihood threshold value, the fault adjustment module determines that the industrial control data is abnormal and locates the abnormal axis of the machine tool;

[0059] wherein the preset time is related to the running performance of the industrial control system;

[0060] The real-time industrial control data is an industrial control parameter generated by each axis of the machine tool in the process of the industrial control system executing the running command;

[0061] The industrial control parameter includes the speed, acceleration and torque of each axis of the machine tool;

[0062] The learning features include the speed change rate, acceleration change rate and torque change rate;

[0063] The neural network model is generated by training an industrial control training set formed by the industrial control pre-processed data;

[0064] The anomaly likelihood threshold value is a critical value of the possibility of the anomaly of each axis of the machine tool, and is related to the synchronism of each axis of the machine tool when working.

[0065] By setting the continuous command module, the industrial control system is sent a running command at intervals of a preset time, the data collection module is configured to collect real-time industrial control data of the industrial control system, the anomaly prediction module is configured to select a plurality of learning features, pre-process the real-time industrial control data, and learn the industrial control pre-processed data by using a neural network model to generate a corresponding anomaly likelihood prediction graph, and the fault adjustment module is configured to compare the likelihood value in the anomaly likelihood prediction graph with an anomaly likelihood threshold value, and when the fault adjustment module determines that the industrial control data is abnormal, the abnormal axis of the machine tool is located, realizing the full-process automation from data collection to anomaly detection and fault positioning, not only improving the accuracy and efficiency of anomaly detection, but also continuously optimizing the detection model through the self-learning function to adapt to the dynamic changes of the industrial control system.

[0066] Please refer to Figure 2 shown, which is a continuous command module structure schematic diagram of the embodiment of the present application, comprising:

[0067] A storage device is used to store operation commands;

[0068] A command transmitting device is connected with the storage device, and is used to send corresponding operation commands to the industrial control system;

[0069] A timing device is connected with the command transmitting device, and is used to trigger the command transmitting device at a preset time interval, and time mark the operation commands in the order of transmission time.

[0070] In the specific implementation, the operation commands are basic instructions for the operation of the industrial control system, and the storage device is used to save the instructions, so as to ensure that the instructions can be accurately called when needed, and preferably, a memory or a storage chip is usually used to realize the storage process, so as to store operation commands in various formats, such as binary codes or ASCII codes.

[0071] The command transmitting device sends the commands to the target equipment through the communication interface (such as RS232, RS485, Ethernet, etc.) of the industrial computer. For example, the operation frequency command of the Fuji frequency converter can be sent in the ASCII code format.

[0072] The timing device can be realized by a hardware clock or a software timer, and can accurately control the sending frequency of the commands.

[0073] In the industrial control system, the sending of the operation commands is a key step for realizing the automatic control of the equipment. Through the continuous command module, the equipment can be ensured to perform tasks in the preset frequency and order.

[0074] Through the setting of the timing device to trigger the command transmitting device at a preset time interval, and the time marking of the operation commands in the order of transmission time, the accurate control and sending of the operation commands of the industrial control system are realized by the cooperative work of the storage device, the command transmitting device and the timing device, which not only improves the automation level of the industrial control system, but also provides a basis support for subsequent abnormal detection and fault prediction.

[0075] Please refer to Figure 3 The data collection module structure schematic diagram of the embodiment of the present application is shown in the figure, which comprises:

[0076] A data sensor is used to monitor and collect real-time industrial control data corresponding to each axis of the machine tool;

[0077] A data adjuster is connected with the data sensor, and is used to correct the data deviation of the real-time industrial control data;

[0078] A data converter is connected with the data adjuster, and is used to perform digital signal conversion processing on the real-time industrial control data, and transmit the real-time industrial control data to the abnormal prediction module.

[0079] In specific implementation, the data sensor is the front-end device of the entire data collection module, and its core role is to obtain real-time data generated by each axis of the machine tool during operation. These data include position information, speed information, acceleration information, load information, etc., and are the basis for subsequent data processing and analysis.

[0080] The role of the data adjuster is to correct the deviations in the data collected by the data sensor. Such deviations are caused by the accuracy of the sensor itself, environmental interference, noise in the signal transmission process, etc. Through the data adjuster, the accuracy and reliability of the data can be improved, providing higher quality input for subsequent data processing.

[0081] The main task of the data converter is to convert the data after deviation correction into a digital signal format suitable for subsequent module processing. This involves operations such as analog-to-digital conversion (A / D conversion), ensuring that the data can be correctly identified and processed by the anomaly prediction module.

[0082] Through data deviation correction processing of real-time industrial control data and digital signal conversion processing of real-time industrial control data using the data converter, real-time monitoring, data processing and anomaly prediction of the machine tool running state are achieved, thereby improving the reliability and operating efficiency of the equipment.

[0083] Specifically, the anomaly prediction module includes:

[0084] The feature selector is used to select a plurality of learning features corresponding to the real-time industrial control data;

[0085] The preprocessor is connected to the feature selector and is used to preprocess the real-time industrial control data and generate corresponding industrial control preprocessed data;

[0086] The learner is connected to the preprocessor and is used to learn the industrial control preprocessed data according to the learning features and generate a corresponding anomaly likelihood prediction map.

[0087] In specific implementation, the feature selector selects the most valuable features for anomaly detection from a large amount of raw data by calculating the correlation or importance between the features and the target variables, and dynamically adjusts the feature selection strategy according to the running state and historical data of the industrial control system to adapt to different working conditions.

[0088] Through the feature selector, key features are selected from the collected real-time industrial control data, and the preprocessed data is input into the neural network model to predict the likelihood of the industrial control system occurring anomalies, reducing the data dimension, reducing the complexity of calculation, removing irrelevant features, improving the recognition ability of the neural network model for abnormal data, and converting the original data into a format suitable for subsequent learning and analysis, improving the data quality and consistency.

[0089] Please refer to Figure 4 As shown in FIG. 1, which is a schematic diagram of a fault adjustment module structure according to an embodiment of the present application, comprising:

[0090] a comparator configured to extract a possibility value from the anomaly possibility prediction graph and compare the possibility value with an anomaly possibility threshold to form a corresponding comparison result;

[0091] a determiner connected to the comparator and configured to determine the running state of the industrial control data according to the comparison result;

[0092] a tracer connected to the determiner and configured to locate the abnormal axis of the machine tool according to the emission time sequence when the industrial control data is abnormal.

[0093] In a specific implementation, if the industrial control data is determined to be abnormal, the tracer locates the abnormal axis of the machine tool according to the emission time sequence and outputs the locating result to the control system or the operator interface so that corresponding measures can be taken.

[0094] By using the comparator to extract the possibility value from the anomaly possibility prediction graph, compare the possibility value with the preset anomaly possibility threshold to generate the comparison result, determine the running state of the industrial control data according to the comparison result, and locate the abnormal axis of the machine tool according to the emission time sequence, not only can the abnormal situation be efficiently handled, but also the fault source can be quickly located, thereby improving the overall performance and reliability of the system.

[0095] Specifically, the timing device generates a trigger signal according to the corresponding preset time, the emission device receives the trigger signal from the timing device, reads the running command from the storage device according to the order of the trigger signal, and sends the running command to the industrial control system.

[0096] In a specific implementation, the timing device sets a preset time interval according to the running performance and requirements of the industrial control system. This time interval can be adjusted according to the actual working conditions to ensure the real-time performance and stability of the system.

[0097] The timing device generates a trigger signal according to the preset time interval. This trigger signal is an accurate time marker used to inform the command emission device to send the running command.

[0098] Specifically, the data sensor is placed on each axis of the machine tool and connected to each axis of the machine tool through a network to monitor and collect the corresponding real-time industrial control data of each axis of the machine tool, the data adjuster reads the collected real-time industrial control data and performs data deviation correction, the data converter converts the analog signal in the real-time industrial control data into a digital signal, and the converted real-time industrial control data is transmitted to the anomaly prediction module;

[0099] In this embodiment, the data deviation correction is to read the missing values in the real-time industrial control data and perform linear interpolation filling on the missing values.

[0100] In specific implementation, data sensors are installed on each axis of the machine tool and connected to the data collection system through a network. These sensors include position sensors, speed sensors, acceleration sensors, temperature sensors, current sensors, etc.

[0101] The data sensors monitor the operating status of each axis of the machine tool in real time, convert physical quantities such as position, speed, temperature, etc. into electrical signals, and sample at a certain frequency to generate real-time data streams. The collected real-time data streams are transmitted to the data adjuster through a network such as Ethernet, industrial bus, etc.

[0102] Data bias correction includes reading missing values in real-time industrial control data and filling missing values with linear interpolation. Among them, linear interpolation filling includes: detecting missing values in real-time industrial control data, which may be caused by sensor failure, data transmission error or inconsistent sampling interval, etc. For the detected missing values, linear interpolation method is used for filling. Linear interpolation estimates the missing values by calculating the linear relationship between adjacent data points. By correcting data bias, false positives caused by data quality problems are reduced.

[0103] The data converter converts the collected analog signals (such as voltage, current, etc.) into digital signals. This process is usually achieved through an analog-to-digital converter (ADC).

[0104] Specifically, the feature selector selects a number of learning features from the real-time industrial control data, the preprocessor cuts the real-time industrial control data according to the sampling rate to form a number of industrial control preprocessing data with a standard sampling rate, and the industrial control preprocessing data enters the neural network model for learning and generates the corresponding anomaly likelihood prediction map.

[0105] Among them, the standard sampling rate is the sampling rate that the neural network model can identify, and the standard sampling rate remains unchanged in the learning process of the neural network model.

[0106] In specific implementation, when the standard sampling rate is 1700 / s, the learning effect of the neural network model on the industrial control pre-data is best. According to the standard sampling rate that the neural network model can identify, the real-time industrial control data is cut according to the sampling rate. The standard sampling rate is the sampling rate that the neural network model can identify in the training process, and remains unchanged in the learning process. The cut industrial control preprocessing data is formatted into the input format required by the neural network model, such as converting the industrial control preprocessing data into a fixed-length sequence or matrix.

[0107] The feature selector selects the most useful features by calculating the correlation or importance between the features and the target variable. For example, statistical analysis, correlation analysis, or model-based feature selection methods such as Lasso regression, random forest, etc. can be used.

[0108] Specifically, when the possibility value is less than the abnormal possibility threshold, the determiner determines that the industrial control data is normal, and the fault adjustment module does not adjust the industrial control system.

[0109] Specifically, when the possibility value is greater than the abnormal possibility threshold, the determiner determines that the industrial control data is abnormal, the tracer traces the current possibility value according to the emission time sequence, and locates the abnormal axis of the machine tool.

[0110] So far, the technical solutions of the present application have been described in combination with the preferred embodiments shown in the drawings, but it is easy for those skilled in the art to understand that the protection scope of the present application is obviously not limited to these specific embodiments. Those skilled in the art can make equivalent changes or replacements to the related technical features without departing from the principles of the present application, and the technical solutions after the changes or replacements will fall within the protection scope of the present application.

[0111] The above description is only the preferred embodiments of the present application and is not intended to limit the present application; for those skilled in the art, the present application can have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A self-learning detection system for industrial control data anomalies based on neural networks, characterized in that: include: Continuous command module, used to send operation commands to the industrial control system at preset intervals; A data collection module, which is connected to the continuous command module and is used to collect real-time industrial control data during the process of the industrial control system executing the operation command; The abnormality prediction module is connected to the data collection module and is used to select a number of learning features based on the real-time industrial control data, pre-process the real-time industrial control data to form corresponding industrial control pre-processed data, and use the neural network model to learn the industrial control pre-processed data to generate the corresponding abnormality possibility prediction map; The fault adjustment module is connected to the anomaly prediction module and includes: A comparator, which is used to extract a possibility value from the abnormal possibility prediction map and compare the possibility value with the abnormal possibility threshold to form a corresponding comparison result; A determiner, connected to the comparator, for determining the operating status of the industrial control data based on the comparison result; The tracer is connected to the determiner and is used to locate the abnormal axis of the machine tool according to the transmission time sequence when the industrial control data is abnormal; The timing device generates a trigger signal according to the corresponding preset time, and the command transmitting device receives the trigger signal from the timing device, reads the operation command from the storage device according to the order of the trigger signal, and sends it to the industrial control system; Data sensors are placed on each axis of the machine tool and connected to each axis through a network to monitor and collect real-time industrial control data corresponding to each axis of the machine tool. The data adjuster reads the collected real-time industrial control data and performs data deviation correction. The data converter converts the analog signals in the real-time industrial control data into digital signals and transmits the converted real-time industrial control data to the abnormality prediction module. Among them, data deviation correction is to read the missing values ​​in the real-time industrial control data and fill the missing values ​​with linear interpolation; Among them, the preset time is related to the operating performance of the industrial control system; Real-time industrial control data refers to the industrial control parameters generated by each axis of the machine tool during the execution of the operation command by the industrial control system; Industrial control parameters include the speed, acceleration and torque of each axis of the machine tool; The learning features include the rate of change of velocity, rate of change of acceleration, and rate of change of torque; The neural network model is trained and generated using the industrial control training set formed by industrial control preprocessing data; The abnormality possibility threshold is the critical value of the possibility of abnormality occurring in each axis of the machine tool, and is related to the synchronization of each axis of the machine tool when working.

2. The neural network-based self-learning detection system for industrial control data anomalies according to claim 1 is characterized in that: Continuous command module, including: A storage device for storing an operation command; A command transmitting device, which is connected to the storage device and is used to send corresponding operation commands to the industrial control system; The timing device is connected to the command transmitting device and is used to trigger the command transmitting device at preset time intervals and mark the running commands in the order of the transmitting time.

3. The neural network-based self-learning detection system for industrial control data anomalies according to claim 2 is characterized in that: Data collection module, including: Data sensors, which are used to monitor and collect real-time industrial control data corresponding to each axis of the machine tool; A data adjuster, which is connected to the data sensor and is used to correct data deviations of real-time industrial control data; The data converter is connected to the data adjuster and is used to perform digital signal conversion processing on the real-time industrial control data and transmit the real-time industrial control data to the abnormality prediction module.

4. The neural network-based self-learning detection system for industrial control data anomalies according to claim 3 is characterized in that: Anomaly prediction module, including: A feature selector, which is used to select corresponding learning features based on real-time industrial control data; A preprocessor, which is connected to the feature selector, is used to preprocess the real-time industrial control data and generate corresponding industrial control preprocessed data; The learner is connected to the preprocessor and is used to learn the industrial control preprocessing data according to the learning features and generate a corresponding abnormality possibility prediction map.

5. The neural network-based self-learning detection system for industrial control data anomalies according to claim 4 is characterized in that: The feature selector selects several learning features based on real-time industrial control data. The preprocessor cuts the real-time industrial control data according to the sampling rate to form several industrial control preprocessed data with a standard sampling rate. The industrial control preprocessed data enters the neural network model for learning and generates the corresponding abnormality possibility prediction map; The standard sampling rate is a sampling rate that can be recognized by the neural network model, and the standard sampling rate remains unchanged during the learning process of the neural network model.

6. The neural network-based self-learning detection system for industrial control data anomalies according to claim 5 is characterized in that: When the probability value is less than the abnormal probability threshold, the determiner determines that the industrial control data is normal, and the fault adjustment module does not adjust the industrial control system.

7. The neural network-based self-learning detection system for industrial control data anomalies according to claim 6 is characterized in that: When the probability value is greater than the abnormal probability threshold, the determiner determines that the industrial control data is abnormal, and the tracer traces the current probability value according to the emission time sequence and locates the abnormal axis of the machine tool.

Images