Method for detecting equipment abnormity through model in digital twinning

Abstract

The invention provides a method for detecting equipment abnormity by a model in digital twinning. The method comprises the following steps of: modeling based on a simulation environment of mechanism and data coupling; defining an action space and a state space of an equipment simulation environment; preprocessing and visualizing the data; after analyzing and disassembling the data, carrying out univariate anomaly detection; evaluating a detection result of the model; and calling the model on the digital twin platform, accessing real-time data to train the model, and analyzing abnormal data to process problems in time. According to the method, the KNN and Prophet models are applied to the equipment digital twin system for calling, abnormal points can be effectively recognized, and the problem of single-variable time series data anomaly detection is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of equipment anomaly detection, and in particular to a method for detecting equipment anomaly using a model in a digital twin. Background Art

[0002] The current common background technologies for device anomaly detection involve multiple aspects, including anomaly detection algorithms, digital twin technology, machine learning and deep learning, data-driven methods, etc. These technologies are often organically combined to form an advanced technical support system for device anomaly detection.

[0003] Since digital twin is a technology that achieves accurate simulation of real-world objects or systems by creating virtual copies of physical systems, it has shown significant advantages in anomaly detection. Digital twin technology enables the model to reflect the working status and environmental changes of actual equipment by receiving data from equipment sensors in real time, such as temperature and vibration. This allows predictions to be made before equipment failure occurs, so that maintenance can be carried out in advance to avoid unexpected downtime and production losses. Digital twin models can also be used to simulate and analyze the operating status of equipment to identify and diagnose potential faults. In addition, digital twin models have the characteristics of self-correction, and the model can be corrected by verifying the diagnosis results. This self-correction mechanism ensures the synchronization of physical equipment and virtual models, and improves the accuracy and scope of fault diagnosis. Taking a continuous casting machine as an example, the present invention introduces a method for detecting equipment anomalies in a model in a digital twin.

[0004] The production process of continuous casting machine is to continuously pour high-temperature molten steel into one or a group of water-cooled copper crystallizers. The molten steel gradually solidifies into a billet shell along the periphery of the crystallizer. When the molten steel level rises to a certain height and the billet shell solidifies to a certain thickness, the straightening machine pulls the billet out and sprays water in the secondary cooling area to completely solidify the billet. The cutting device cuts it into a fixed length according to the requirements of steel rolling. This process of directly pouring high-temperature molten steel into billets is called continuous casting. Nowadays, the equipment of continuous casting machines has almost become a necessity for steelmaking enterprises, whether they are long-process steelmaking or short-process steelmaking.

[0005] In the equipment production process, there are a series of business pain points and challenges, such as frequent failures, scattered data, and difficult maintenance. The current traditional equipment anomaly detection has the following shortcomings:

[0006] 1. The system performance is not intelligent enough and relies heavily on manual observation and empirical judgment, and the level of automation and intelligence is not high.

[0007] 2. Lack of real-time performance. Abnormal signals cannot be captured and processed quickly enough, which may result in the problem being discovered only after it has worsened.

[0008] 3. The accuracy needs to be improved. The detection algorithm may produce false positives or missed positives due to environmental interference or equipment aging.

[0009] 4. Poor adaptability. The detection system may need to be adjusted frequently for ingots of different specifications or changing process parameters.

[0010] 5. Difficult maintenance. Complex hardware configuration and maintenance process increase the operating cost of the system.

[0011] 6. In terms of data processing, the phenomenon of information islands is serious, and there is a lack of effective data sharing and integration between various sensors and control systems.

[0012] 7. Limited data analysis capabilities, lack of deep learning and big data analysis technology to explore potential failure modes.

[0013] Moreover, historical data is underutilized, and the value of historical fault data is often ignored and cannot be used to optimize detection models.

[0014] 8. Lack of predictive maintenance, mainly focusing on post-processing rather than predicting and preventing failures in advance. This will not only affect economic benefits, but also pose certain safety hazards. For example, some old equipment does not meet the electrical safety standards and there is a risk of electric shock.

[0015] 9. In terms of human-computer interaction, the operation interface is not friendly and the user interface is complex, which is not conducive to rapid mastery and operation.

[0016] 10. High training cost: New employees need long-term training to become familiar with the system operation and exception handling process.

[0017] Patent document CN115289837A discloses a rotary kiln state anomaly detection system based on digital twin, including a digital twin module and an anomaly detection module; the digital twin module includes a physical layer, a data layer and a model layer; the anomaly detection module includes an algorithm library and an online prediction layer.

[0018] However, the digital twin modeling method of patent document CN115289837A is to integrate the digital twin module and the anomaly detection module, and then realize the browsing and display of the rotary kiln based on the three-dimensional visualization technology of Unity3D, which focuses more on intuitive display and real-time operation, and is suitable for rapid application and scenes with high requirements for visual effects. This method is based on simulation environment modeling coupled with mechanism and data, which requires in-depth understanding of the working principle, key components and process parameters of the equipment. The model is more accurate and has better scalability. It emphasizes the accuracy of the model and the depth of scientific research, and is suitable for fields that require in-depth analysis and long-term research. Summary of the invention

[0019] In view of the defects in the prior art, the purpose of the present invention is to provide a method for detecting device anomalies by a model in a digital twin.

[0020] A method for detecting device anomalies using a model in a digital twin provided by the present invention includes:

[0021] Step S1: Modeling the simulation environment based on the coupling of mechanism and data to obtain a device simulation model;

[0022] Step S2: defining the action space and state space of the device simulation model;

[0023] Step S3: obtaining raw data and performing raw data preprocessing and visualization;

[0024] Step S4: after analyzing and decomposing the data, a univariate anomaly detection model is obtained, and univariate anomaly detection is performed using the univariate anomaly detection model;

[0025] Step S5: Evaluate the detection result;

[0026] Step S6: Call the model on the digital twin platform, access real-time data to train the single variable anomaly detection model, analyze the abnormal data, and then deal with the problem in time through the equipment simulation model.

[0027] Preferably, the step S1 comprises:

[0028] Step S1.1: Establish a mechanism model based on the working principle of the equipment and the interactions and influences between key components;

[0029] Step S1.2: collecting data generated during the actual operation of the device and preprocessing the data;

[0030] Step S1.3: Establish a data-driven model based on the preprocessed data;

[0031] Step S1.4: Combine the mechanism model and the data-driven model to obtain a coupling model.

[0032] Preferably, the raw data preprocessing includes denoising and interpolation;

[0033] The raw data includes environmental parameters, planning information and process information.

[0034] Preferably, the state space refers to a state space that can describe the current operation of the equipment, including the state and process parameters of each key component of the equipment;

[0035] The action space refers to the set of all actions that the model can take.

[0036] Preferably, the step S3 comprises:

[0037] Step S3.1: describe the original data;

[0038] Step S3.2: preprocessing the described data;

[0039] Step S3.3: extract features related to the target variable from the original data and remove redundant features;

[0040] Step S3.4: Visualize the processed data.

[0041] Preferably, the original data includes point data of each device;

[0042] Key point data include operating parameters, equipment status and process quality;

[0043] The method of preprocessing the described data includes removing duplicate values, processing missing values, merging and regularizing data, and marking fault types;

[0044] The methods of marking fault types include decision trees, support vector machines, and neural networks.

[0045] Preferably, step S4 comprises:

[0046] Step S4.1: Use KNN supervised learning algorithm to classify outliers;

[0047] Step S4.2: Use the Prophet algorithm to process the outliers.

[0048] Preferably, step S5 includes integrating all models into the algorithm library of the digital twin platform, and providing the functions of loading and calling the models through an interface. Subsequent iterations of the models will be based on the algorithm library, and only require updating and replacing the model weight files and modifying the loading and calling methods to complete the iterative deployment of the model.

[0049] Preferably, it also includes model optimization, verifying the effect of the model by comparison with the actual equipment, including optimizing process parameters, improving product quality, and increasing production efficiency;

[0050] Optimize and adjust the model based on actual verification results, including adjusting the neural network structure and training parameters.

[0051] A system for detecting device anomalies in a model in a digital twin provided by the present invention includes:

[0052] Module M1: Modeling the simulation environment based on the coupling of mechanism and data to obtain the equipment simulation model;

[0053] Module M2: defining the action space and state space of the obtained device simulation model;

[0054] Module M3: Obtaining raw data and performing raw data preprocessing and visualization;

[0055] Module M4: after analyzing and decomposing the data, a univariate anomaly detection model is obtained, and univariate anomaly detection is performed using the univariate anomaly detection model;

[0056] Module M5: evaluating the detection results;

[0057] Module M6: Call the model on the digital twin platform, access real-time data to train the single variable anomaly detection model, analyze the abnormal data, and then deal with the problem in time through the equipment simulation model.

[0058] Compared with the prior art, the present invention has the following beneficial effects:

[0059] 1. The present invention uses digital twin technology to provide a virtual "mirror" for the continuous casting machine. Through real-time data acquisition, modeling and simulation, the physical state and process of the continuous casting machine can be accurately reproduced. Then, by using machine learning and other algorithms, the digital twin model can learn the normal operating mode and automatically identify abnormal signals that deviate from normal behavior, thereby improving the accuracy and adaptability of abnormality detection.

[0060] 2. The present invention applies KNN and Prophet models to the digital twin system of the continuous casting machine for calling, which can effectively identify abnormal points and solve the problem of abnormal detection of single variable time series data. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0062] Figure 1 It is a schematic diagram of the working method of the present invention. DETAILED DESCRIPTION

[0063] The present invention is described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those of ordinary skill in the art, several changes and improvements can also be made without departing from the concept of the present invention. These all belong to the protection scope of the present invention.

[0064] Embodiment 1

[0065] According to a method for detecting device anomalies in a model in a digital twin provided by the present invention, Figure 1 As shown, including:

[0066] Step S1: Model the simulation environment based on the coupling of mechanism and data to obtain the equipment simulation model. Taking the continuous casting machine equipment as an example, we first have a deep understanding of the working principle and key components of the continuous casting machine, including the crystallizer, the tundish, the fire cutter, etc., as well as the interactions and influences between them. Based on the understanding of the continuous casting machine process, a mathematical model is established to describe the physical process and interaction of each component. Collect the data generated during the actual operation of the continuous casting machine, including parameters such as temperature, pressure, and speed. Preprocess the data, including denoising and interpolation, for subsequent modeling. Use machine learning techniques, such as deep learning and regression analysis, to model the data and obtain a data-driven model. These models can be used to describe the behavior of the continuous casting machine, such as temperature distribution, fluid dynamics, etc. Combine the established mechanism model with the data-driven model to form a coupled model. In this way, the physical process and actual data can be comprehensively considered to improve the accuracy and reliability of the model. Use historical data or experimental data to verify the established simulation model, and adjust and optimize the model according to the verification results to improve the accuracy and reliability of the simulation.

[0067] Step S2: Define the action space and state space of the equipment simulation model. The state space of the continuous casting machine simulation environment includes the states and process parameters of each key component of the continuous casting machine, and the key components include the ladle turret, crystallizer, straightening machine, tundish, secondary cooling device, etc. The states and process parameters of key components such as the temperature distribution of the crystallizer, the speed of the straightening machine, the liquid level change of the tundish, etc. It should be noted that the state space should be able to fully describe the current situation of the continuous casting process so that the reinforcement learning model can make decisions. The action space of the continuous casting machine includes adjustable operating parameters, such as the height position of the stopper rod, the temperature of the molten steel in the crystallizer, etc. The action space should contain all possible operations that the model can take so that the model can select the optimal operating strategy.

[0068] Step S3: raw data preprocessing and visualization. Step S3 includes:

[0069] Step S3.1: Describe the original data. The original data includes the point data of each device on the continuous casting machine. The key point data includes operating parameters, equipment status and process quality. The specific description is as follows:

[0070] 1. Operation parameters: including billet temperature, molten steel temperature, billet size, billet pulling speed, billet curvature, etc. The changes in these parameters can reflect the operating status of the continuous casting machine, timely discover and warn potential faults, thereby avoiding or reducing the occurrence of accidents.

[0071] 2. Equipment status: Involving the continuous casting machine motor, continuous casting machine reducer, continuous casting machine hydraulic system, continuous casting machine cooling system, etc. The status of these equipment directly affects the production efficiency and product quality of the continuous casting machine. Therefore, by monitoring the status of these equipment, abnormal conditions of the equipment can be discovered in time, and measures can be taken in time for maintenance or replacement.

[0072] 3. Process quality: including billet surface quality, billet internal quality, billet mechanical properties, etc. These quality indicators directly reflect the product quality of the continuous casting machine. Therefore, by monitoring these quality indicators, product quality problems of the continuous casting machine can be discovered in time, and timely measures can be taken to make adjustments.

[0073] There are a total of 96 complete data points. After verification, 14 of them are type 01 signals. Subsequent processing and training processes will no longer focus on such data points.

[0074] Step S3.2: Preprocess the raw data after description. Taking the mold liquid level of the continuous casting machine as an example, the data in the working state fluctuates around 800, and the data in the non-working state is 720. For the anomaly detection algorithm, the data should be selected for training under normal working conditions to learn the normal mode of the working state. For the prediction algorithm, the working state or the working state combined with the non-working state is selected for training according to the time frequency, and the change trend of the point data is learned according to the time node. The data processing methods of different models are different. The following is the standard practice:

[0075] The procedure is divided into four steps. The first step is to remove duplicate values: the original data contains a large number of duplicate values ​​at the second level. First, the data is deduplicated and only the first data in the same second is retained. The second step is to handle missing values: for missing values ​​in the original data, interpolation is used to fill in the missing values. The third step is to merge and regularize data: after data deduplication and missing value processing, the data of all points are regularized according to TimeStampStr (timestamp string) into a table containing TimeStampStr and data columns of each point. The fourth step is to use decision trees, support vector machines, neural networks and other methods to mark fault types and improve data quality and accuracy.

[0076] Step S3.3: Extract useful features from the original data, select features related to the target variable, remove redundant features, and improve the efficiency and interpretability of the model. This is used to establish a fault prediction model. At the same time, remove irrelevant and redundant features to improve the accuracy and interpretability of the model. First, based on domain knowledge and data analysis, select features related to the fault. For example, based on the understanding and analysis of equipment and business, the stopper rod, crystallizer liquid level, crystallizer width adjustment, and crystallizer vibration are the key points. According to the different point characteristics, the points are divided into the following groups: ① stopper rod point group; ② crystallizer liquid level, outlet flow point group; ③ upper and lower roller current point group; ④ driving torque group. Then it is necessary to use feature engineering technology (1, principal component analysis (PCA): transform the data to a new coordinate system through linear transformation, and the new coordinates (principal components) are sorted in descending order of variance; 2, linear discriminant analysis (LDA): not only dimensionality reduction is performed, but also category information is considered to maximize the separation between categories) to extract meaningful features from the original data.

[0077] Step S3.4: Visualize the processed data, such as forming a displayable and describable curve graph and histogram.

[0078] After preprocessing and visualizing the raw data, we analyzed which point data had obvious periodic trends while other points showed randomness. The histogram also helped us understand the distribution and anomalies of the data. These works will provide valuable guidance and foundation for subsequent data analysis, model building, and the development of anomaly detection systems.

[0079] Step S4: Based on the analysis of data and actual business needs, we decomposed complex problems and found that many of them are single variable anomaly detection problems, so we first consider the algorithm design of this part.

[0080] Step S4.1: KNN (K-Nearest Neighbors) is a supervised learning algorithm commonly used in regression and classification tasks. Its algorithm is used to classify outliers in anomaly detection. The core idea of ​​KNN is to use a given K value (i.e., how many groups of samples are calculated) to calculate the average distance between each sample point and its nearest K samples in turn, and finally compare the calculated distance with the set threshold to determine whether it is an outlier. KNN can predict the value of a new data point and assign a value to the new point based on the similarity between the new point and the points in the training set.

[0081] The core steps of KNN anomaly detection are as follows:

[0082] 1) Data standardization: Eliminate the dimensional influence between different features and standardize the scale of data.

[0083] 2) Calculate sample distance: Calculate the distance between each sample point and other sample points (generally using Euclidean distance, Manhattan distance, Chebyshev distance and angle cosine distance).

[0084] 3) Find neighboring samples: Sort in ascending order according to the calculated distances, and then find the K samples closest to each sample point.

[0085] 4) Calculate the average distance: Calculate the average distance between each sample point and its nearest neighbor sample point. Calculate the distance between the current point and its K neighbors, and take the mean, median or maximum value as the outlier. If the calculated outlier value is greater than the preset threshold, the point is marked as an outlier.

[0086] Step S4.2: Because the Prophet algorithm is very sensitive to missing data and trend changes and can usually handle outliers well, Prophet is used for anomaly detection. Prophet is a time series data prediction algorithm based on an additive model, in which nonlinear trends are matched with annual, seasonal and holiday effects, so it is very suitable for time series with seasonal effects. The Prophet model consists of trend terms, periodicity, holidays, regression terms and noise terms, and can comprehensively consider various factors. The main steps of Prophet for anomaly detection include:

[0087] 1) Use the Prophet model to fit historical time series data.

[0088] 2) Generate model predictions, including point predictions and uncertainty intervals around the predictions.

[0089] 3) For a new time point, calculate the difference between the observed value and the point forecast. If this difference exceeds a certain threshold, it can be marked as an anomaly.

[0090] 4) The uncertainty interval of the model can be used to further refine the anomaly detection. If the observation falls outside the uncertainty interval, it can also be marked as an anomaly.

[0091] Step S5: Model evaluation. The accuracy of anomaly prediction is mainly evaluated through evaluation indicators such as Accuracy, Precision, Recall, F1-Score, and the relationship between PR curve, ROC curve and AUC. The iterative deployment of the model is specifically described as follows: The disadvantage of KNN is that it needs to store all training data, the amount of calculation is large and it is sensitive to outliers, while Prophet has high requirements for data and requires proper processing for non-stationary sequences, which increases the complexity of practical applications. According to different fault types, different data characteristics, and corresponding business scenarios, the two models can be combined for anomaly judgment with higher accuracy. Therefore, the models are integrated into the digital twin platform algorithm library, and the functions of loading and calling models are provided in the form of interfaces. Subsequent iterations of the model will be based on the algorithm library. It only needs to update and replace the model weight file and modify the loading and calling methods to complete the iterative deployment of the model.

[0092] Step S6: Call the model on the digital twin platform, access real-time data to train the single variable anomaly detection model, analyze the abnormal data, and deal with the problem in a timely manner. You can also perform historical backtracking, simulate historical data, summarize experience, and save key information into the knowledge base for management and utilization.

[0093] Step S7: Model optimization. The effect of the model is verified by comparison with the actual continuous casting machine, including the effects of optimizing process parameters, improving product quality, and improving production efficiency. The model is optimized and adjusted according to the actual verification results, including adjusting the neural network structure and training parameters, so as to further improve the performance and adaptability of the model.

[0094] The present invention establishes a real-time monitoring system through digital twin technology, which can integrate and display the data flows of various systems and points of complex equipment. The KNN and Prophet models are applied to the digital twin system of the continuous casting machine for calling, which can effectively identify abnormal points and solve the problem of anomaly detection of single variable time series data.

[0095] Embodiment 2

[0096] The present invention also provides a system for detecting device abnormalities of a model in a digital twin. The system for detecting device abnormalities of a model in a digital twin can be implemented by executing the process steps of the method for detecting device abnormalities of a model in the digital twin, that is, those skilled in the art can understand the method for detecting device abnormalities of a model in the digital twin as a preferred implementation of the system for detecting device abnormalities of a model in the digital twin.

[0097] A system for detecting device anomalies in a model in a digital twin provided by the present invention includes:

[0098] Module M1: Modeling the simulation environment based on the coupling of mechanism and data to obtain a device simulation model. The module M1 includes: Module M1.1: Establishing a mechanism model based on the working principle of the equipment and the interaction and influence between key components. Module M1.2: Collecting the data generated during the actual operation of the equipment and preprocessing the data. The data preprocessing includes denoising and interpolation. The data includes environmental parameters, planning information and process information. Module M1.3: Establishing a data-driven model based on the preprocessed data. Module M1.4: Combining the mechanism model with the data-driven model to obtain a coupling model.

[0099] Module M2: Define the action space and state space of the equipment simulation model. The state space refers to the current situation of the equipment operation, including the status and process parameters of each key component of the equipment. The action space refers to the set of all operations that the model can take.

[0100] Module M3: Acquire raw data and perform preprocessing and visualization of raw data. The module M3 includes: Module M3.1: Describe the raw data. The raw data includes the point data of each device. The key point data includes operating parameters, equipment status and process quality. Module M3.2: Preprocess the described data. The method of preprocessing the described data includes removing duplicate values, processing missing values, merging regular data, and marking fault types. The method of marking fault types includes decision trees, support vector machines, and neural networks. Module M3.3: Extract features related to the target variable from the raw data and remove redundant features. Module M3.4: Visualize the processed data.

[0101] Module M4: after analyzing and decomposing the data, single variable anomaly detection is performed. The module M4 includes: Module M4.1: using KNN supervised learning algorithm to classify anomalies. Module M4.2: using Prophet algorithm to process the anomalies.

[0102] Module M5: Evaluate the detection results of the model. The module M5 includes integrating the models into the algorithm library of the digital twin platform, providing the functions of loading and calling the models through the interface. Subsequent iterations of the models will be based on the algorithm library. It only needs to update and replace the model weight file and modify the loading and calling methods to complete the iterative deployment of the model.

[0103] Module M6: Call the model on the digital twin platform, access real-time data to train the model, analyze abnormal data, and deal with problems in a timely manner.

[0104] The present invention also includes model optimization, which verifies the effect of the model by comparing it with the actual equipment, including optimizing process parameters, improving product quality, and increasing production efficiency. The model is optimized and adjusted according to the actual verification results, including adjusting the neural network structure and adjusting the training parameters.

[0105] Those skilled in the art know that, in addition to realizing the system and its various devices, modules, and units provided by the present invention in a purely computer-readable program code, the system and its various devices, modules, and units provided by the present invention can be completely realized in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by performing logic programming on the method step S. Therefore, the system and its various devices, modules, and units provided by the present invention can be considered as a hardware component, and the devices, modules, and units included therein for realizing various functions can also be regarded as structures within the hardware component; the devices, modules, and units for realizing various functions can also be regarded as both software modules for realizing the method and structures within the hardware component.

[0106] The above describes the specific embodiments of the present invention. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essence of the present invention. In the absence of conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.

Claims

1. A method for detecting device anomalies using a model in a digital twin, characterized in that: include: Step S1: Modeling the simulation environment based on the coupling of mechanism and data to obtain a device simulation model; Step S2: defining the action space and state space of the device simulation model; Step S3: obtaining raw data and performing raw data preprocessing and visualization; Step S4: after analyzing and decomposing the data, a univariate anomaly detection model is obtained, and univariate anomaly detection is performed using the univariate anomaly detection model; Step S5: Evaluate the detection result; Step S6: Call the model on the digital twin platform, access real-time data to train the univariate anomaly detection model, analyze the abnormal data, and then use the equipment simulation model to handle the problem in a timely manner.

2. The method for detecting device anomalies using a model in a digital twin according to claim 1, characterized in that: The step S1 comprises: Step S1.1: Establish a mechanism model based on the working principle of the equipment and the interactions and influences between key components; Step S1.2: collecting data generated during the actual operation of the device and preprocessing the data; Step S1.3: Establish a data-driven model based on the preprocessed data; Step S1.4: Combine the mechanism model and the data-driven model to obtain a coupling model.

3. The method for detecting device anomalies using a model in a digital twin according to claim 2, characterized in that: The raw data preprocessing includes denoising and interpolation; The raw data includes environmental parameters, planning information and process information.

4. The method for detecting device anomalies using a model in a digital twin according to claim 1, characterized in that: The state space refers to the current situation of the equipment operation, including the state and process parameters of each key component of the equipment; The action space refers to the set of all actions that the model can take.

5. The method for detecting device anomalies using a model in a digital twin according to claim 1, characterized in that: The step S3 comprises: Step S3.1: describe the original data; Step S3.2: preprocessing the described data; Step S3.3: extract features related to the target variable from the original data and remove redundant features; Step S3.4: Visualize the processed data.

6. The method for detecting device anomalies using a model in a digital twin according to claim 5, characterized in that: The original data includes point data of each device; Key point data include operating parameters, equipment status and process quality; The method of preprocessing the described data includes removing duplicate values, processing missing values, merging and regularizing data, and marking fault types; The methods of marking fault types include decision trees, support vector machines, and neural networks.

7. The method for detecting device anomalies using a model in a digital twin according to claim 1, characterized in that: The step S4 comprises: Step S4.1: Use KNN supervised learning algorithm to classify outliers; Step S4.2: Use the Prophet algorithm to process the outliers.

8. The method for detecting device anomalies using a model in a digital twin according to claim 1, characterized in that: The step S5 includes integrating all models into the algorithm library of the digital twin platform, and providing the functions of loading and calling models through an interface. Subsequent iterations of the models will be based on the algorithm library. It only requires updating and replacing the model weight file and modifying the loading and calling methods to complete the iterative deployment of the model.

9. The method for detecting device anomalies using a model in a digital twin according to claim 1, characterized in that: It also includes model optimization, which verifies the effect of the model by comparing it with the actual equipment, including optimizing process parameters, improving product quality, and increasing production efficiency; Optimize and adjust the model based on actual verification results, including adjusting the neural network structure and training parameters.

10. A system for detecting device anomalies in a model in a digital twin, characterized in that: include: Module M1: Modeling the simulation environment based on the coupling of mechanism and data to obtain the equipment simulation model; Module M2: defining the action space and state space of the obtained device simulation model; Module M3: Obtaining raw data and performing raw data preprocessing and visualization; Module M4: after analyzing and decomposing the data, a univariate anomaly detection model is obtained, and univariate anomaly detection is performed using the univariate anomaly detection model; Module M5: evaluating the detection results; Module M6: Call the model on the digital twin platform, access real-time data to train the single variable anomaly detection model, analyze the abnormal data, and then deal with the problem in time through the equipment simulation model.

Images