Equipment performance prediction method based on digital twinning

Abstract

The invention discloses an equipment performance prediction method based on digital twinning, particularly relates to the field of digital twinning, is used for solving the problems of hidden fluctuation and burst interference in equipment operation data, and comprises the following steps: dividing the equipment operation data into different clusters through clustering analysis; irrelevant fluctuation and interference signals are accurately identified and removed by combining a fluctuation trajectory stability index and a sudden-change derivative acceleration index, so that the data input into the model can reflect the real state of the equipment more accurately; hidden fluctuation caused by equipment microscopic change, external interference or environmental fluctuation is stripped, so that the interference of the change which is not easy to perceive on a prediction result is avoided; according to the process, the data quality is remarkably improved, the influence of noise which is difficult to perceive in a short time on the stability of the model is prevented, fluctuation and jumping fluctuation in the prediction process are effectively reduced, and it is ensured that the model keeps high prediction reliability and consistency in a complex equipment operation environment and long-term data.

Description

Technical Field

[0001] The present invention relates to the field of digital twins, and more specifically, to a device performance prediction method based on digital twins. Background Art

[0002] In modern industry, digital twin technology has been widely used in equipment performance prediction and health monitoring. However, the operation of equipment is often accompanied by hidden fluctuations and sudden interference. These changes are often not easy to detect in the short term, but can have a serious impact on the prediction results of equipment status. Especially when the equipment is under extreme working conditions or frequently switches operating modes, small fluctuations or environmental interferences are often amplified by the digital twin model, resulting in instability of the prediction, and even causing jump fluctuations, which greatly reduces the accuracy and reliability of the prediction. If these hidden fluctuations, external environmental changes and occasional errors hidden in the equipment operation data are not identified and removed in time, they will directly lead to model failure and fail to truly reflect the actual status of the equipment, thereby affecting equipment maintenance decisions, increasing the risk of failures, and even causing equipment downtime and production losses in severe cases. Therefore, how to effectively identify and remove these hidden fluctuations and interference signals and improve the prediction accuracy and stability of digital twin models has become a key issue that needs to be solved urgently. Summary of the invention

[0003] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a device performance prediction method based on digital twins. Through cluster analysis, the device operation data is divided into different clusters, and the fluctuation trajectory stability index and the mutation-derived acceleration index are combined to accurately identify and remove irrelevant fluctuations and interference signals, so that the data input to the model can more accurately reflect the true state of the device. By stripping off the implicit fluctuations caused by microscopic changes in the equipment, external interference or environmental fluctuations, the interference of these imperceptible changes on the prediction results is avoided. This process not only significantly improves the data quality and prevents the impact of imperceptible noise on the stability of the model in the short term, but also effectively reduces the volatility and jump fluctuations in the prediction process, ensuring that the model maintains a high prediction reliability and consistency in complex equipment operating environments and long-term data, so as to solve the problems raised in the above-mentioned background technology.

[0004] To achieve the above object, the present invention provides the following technical solutions:

[0005] The device performance prediction method based on digital twins is characterized by comprising the steps of: S1, collects the operation data of the equipment in real time through sensors in the physical workshop, and transmits the collected equipment operation data to the digital workshop.

[0006] S2, detects hidden dangers in the data transmitted to the digital workshop, identifies potential hidden fluctuations and sudden interference signals, and divides the data into different clusters through cluster analysis.

[0007] S3, for each cluster, by evaluating the stability of fluctuations in the data and the deviation from periodic changes, as well as the differences in acceleration and amplitude of sudden changes, calculates the feature gap and determines whether to switch to the micro-fluctuation stripping mode or the sudden interference stripping mode.

[0008] S4, retrain the digital twin model using the stripped data.

[0009] S5, equipment performance prediction based on the optimized digital twin model.

[0010] In a preferred embodiment, step S2 includes the following contents: S2.1, perform preliminary cleaning and preprocessing on the raw data transmitted from the physical workshop to the digital workshop.

[0011] S2.2, use fast Fourier transform to convert time domain data to frequency domain and analyze the frequency components therein; perform periodic analysis on the time series data of the equipment, use the autoregressive model to predict the future equipment operation status, and compare it with the actual data to detect whether there is a deviation in periodic fluctuations; if the standard deviation of the residual exceeds the set threshold, it means that there are large fluctuations in the equipment data, and an implicit fluctuation prompt is issued.

[0012] S2.3, use the autoregressive integrated moving average model to model the time series data and calculate the prediction error; if the prediction error exceeds the set threshold, it is considered that a sudden interference has occurred.

[0013] S2.4, classify the detected implicit fluctuations and sudden interference signals to determine their impact on the equipment performance prediction; cluster the fluctuation characteristics through clustering algorithms to evaluate the impact of different categories.

[0014] In a preferred embodiment, step S3 further includes the following contents:

[0015] S3.1, through cluster analysis, the data within the cluster has been divided into multiple groups, each cluster represents a specific signal pattern; the data within each cluster shows a certain degree of similarity; next, the signal of each cluster is feature extracted, and the fluctuation trajectory stability index and mutation-derived acceleration index are used to measure the fluctuation stability and sudden change characteristics of the cluster.

[0016] S3.2, Fluctuation trajectory stability index The stability of the clustered data points is evaluated by calculating the curve fitting error, aiming to reveal the stability or irregularity of the periodic fluctuations; the regularity of the fluctuations is detected by fitting the trajectory of the curve to reflect whether the periodic changes are disturbed by external factors; for each data point in the cluster, Perform polynomial regression fitting to obtain the fitting curve ; Use curve regression error To measure the difference between each data point and the fitted curve; calculate the fluctuation trajectory stability index value within the cluster: ;in, is the normalized amplitude of the data within the cluster, is the amplitude feature within the cluster, is the period of periodic fluctuation, The first The regression coefficient for each data point is is the number of data points in the cluster, It expresses the absolute value of the ratio of amplitude to period.

[0017] S3.3, Mutation-derived acceleration index The second-order difference is used to capture sudden changes in acceleration and detect the violent fluctuations of the equipment at a certain moment. Sudden interference is accompanied by a violent change in acceleration, and the instantaneous impact of the data is identified by the mutation-derived acceleration index. The second-order difference of the data points within the cluster is used to calculate the acceleration change: ;in, is the time interval of the data points; the sudden fluctuations are weighted to obtain the mutation-derived acceleration index value: ;in, is the mean amplitude of the data within the cluster, The function is used to smooth the effects of sudden fluctuations.

[0018] S3.4, after obtaining the fluctuation trajectory stability index and the mutation-derived acceleration index, the gap between the two indicators is calculated to determine whether there is hidden fluctuation or sudden interference; gap assessment The calculation formula is: ;in, and are the standardized parameters of the fluctuation trajectory stability index and the mutation-derived acceleration index respectively; value, set the threshold , determine whether to switch to peeling mode: if , indicating that there is a large fluctuation difference within the cluster, and it is necessary to switch to the sudden interference stripping mode; if , indicating that the intra-cluster fluctuations are stable and continue to maintain the micro-fluctuation peeling mode without special treatment.

[0019] In a preferred embodiment, step S4 includes the following contents: The digital twin model is retrained using the stripped data: first, relevant features are selected from the stripped device data and the data is standardized; then, the corresponding training model is selected for training based on the data characteristics; next, the data is divided into training set, validation set and test set using the cross-validation method, the generalization ability of the model during the training process is evaluated, and the loss function is minimized through the optimization algorithm to improve the accuracy of the model prediction; after the training is completed, the model is evaluated and the prediction effect of the model is quantified using evaluation indicators; finally, the optimized digital twin model will be deployed in the actual system to predict and monitor the equipment status through real-time data.

[0020] In a preferred embodiment, step S5 includes the following contents: The prediction process of the digital twin model first collects real-time operating data from the equipment; from the preprocessed data, key features that have a sufficient impact on the prediction results are extracted; then the digital twin model receives these feature data, performs calculations, and obtains the prediction results of the future state of the equipment.

[0021] Technical effects and advantages of the device performance prediction method based on digital twins of the present invention: The present invention significantly improves the accuracy and stability of equipment performance prediction based on digital twins by stripping away the implicit fluctuations and sudden interference signals in the equipment operation data. The microscopic changes of the equipment under extreme working conditions, the fluctuations of the external environment, and the occasional errors in the acquisition process usually have a significant impact on the prediction of equipment performance. Although these fluctuations may not be easy to detect in the short term, they can lead to inaccurate prediction results and even cause jump fluctuations, reducing the credibility of the prediction model. Through the technical method of the present invention, these implicit fluctuations can be accurately identified and stripped, thereby eliminating their interference with model prediction. Especially when the equipment frequently switches operating modes or is in a complex working state, the stripped data can better reflect the true state of the equipment, avoid the amplification effect of small fluctuations, and improve the model's adaptability to long-term operating data. This method ensures that the digital twin model maintains a high degree of stability during the operation of the equipment, and provides a more accurate prediction of equipment health, providing reliable decision support for equipment maintenance, fault warning and performance optimization, and significantly improving the intelligence level and accuracy of equipment management. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a schematic diagram of the process of the device performance prediction method based on digital twins of the present invention; Figure 2 This is a schematic diagram of step S2 of the device performance prediction method based on digital twins of the present invention. DETAILED DESCRIPTION

[0023] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0024] Embodiment 1: Figure 1 The present invention provides a device performance prediction method based on digital twins, including: S1, collects the operation data of the equipment in real time through sensors in the physical workshop, and transmits the collected equipment operation data to the digital workshop.

[0025] S2, detects hidden dangers in the data transmitted to the digital workshop, identifies potential hidden fluctuations and sudden interference signals, and divides the data into different clusters through cluster analysis.

[0026] S3, for each cluster, by evaluating the stability of fluctuations in the data and the deviation from periodic changes, as well as the differences in acceleration and amplitude of sudden changes, calculates the feature gap and determines whether to switch to the micro-fluctuation stripping mode or the sudden interference stripping mode.

[0027] S4, retrain the digital twin model using the stripped data.

[0028] S5, equipment performance prediction based on the optimized digital twin model.

[0029] Step S1 includes the following contents: S1.1, in the physical workshop, first install a variety of sensors at key parts of the equipment, including but not limited to temperature sensors, pressure sensors, vibration sensors, current sensors, etc. Each sensor sets the sampling frequency and accuracy according to different equipment operating parameters and monitoring requirements to ensure that the real-time changes of the equipment can be captured. The data collection frequency should be customized according to the working nature and forecasting needs of the equipment. It is usually required that the data collection frequency is at least 10 times the frequency of changes in the equipment's operating status to ensure sufficient real-time performance.

[0030] S1.2, the collected raw data will be transmitted in real time through edge computing devices or local data transmission systems. Data transmission uses a low-latency, high-bandwidth communication protocol to ensure that equipment operation data can be quickly and accurately transmitted to the digital workshop. During data transmission, data encryption and integrity verification need to be implemented to prevent data loss or tampering during transmission. At the same time, the transmission process must be highly reliable to avoid data interruption or loss due to network fluctuations or failures.

[0031] S1.3, after the data is received in the digital workshop, a series of preprocessing operations are first performed to ensure the validity and consistency of the data in subsequent analysis: Data cleaning: Clean the received raw data to remove invalid, duplicate and abnormal data. The identification of abnormal data is usually based on set thresholds, such as temperature exceeding the maximum tolerance range of the equipment, vibration exceeding the normal vibration range of the equipment, etc. For occasional sensor failures, data loss, etc., supplementation or repair processing is required, and interpolation or prediction methods based on historical data are used to fill in missing values.

[0032] Time synchronization and alignment: When different types of sensors collect data, inconsistent timestamps may occur. In this case, the sensor data needs to be time synchronized to ensure that all data points are aligned with the same timestamp to avoid errors in subsequent analysis due to data time deviation.

[0033] Noise filtering: Filter the noise signal during the transmission process, especially when the working environment of the equipment is relatively complex, external environmental factors (such as electromagnetic interference, vibration, etc.) may affect the collected signals. Filtering methods may include low-pass filtering, Kalman filtering, etc. According to the working characteristics of the equipment, select the appropriate filtering algorithm to remove irrelevant interference signals.

[0034] S1.4, after completing data cleaning and preprocessing, perform a data integrity check to ensure that all equipment operation data is intact. Use algorithms to verify that the sensor output data matches the preset standards and confirm whether the data meets the basic requirements for further analysis. For data that does not meet the integrity requirements, it needs to be repaired through re-collection or compensation mechanisms.

[0035] S1.5, standardize or normalize the processed data to unify the data ranges of different sensors and measurement dimensions. For example, map data of different dimensions such as temperature, pressure, and vibration to a unified dimension or standard range, so that subsequent multivariate analysis and model training can effectively process different types of data.

[0036] S1.6, real-time monitoring and early warning mechanisms should also be embedded in the data preprocessing process to promptly detect abnormal situations during data collection and transmission, such as sensor failures, data transmission delays, etc. Once a problem is found, the system can immediately issue an alarm to remind the operation and maintenance personnel to check and deal with it, preventing erroneous data from entering the digital workshop.

[0037] In digital twin technology, the accuracy of equipment performance prediction is directly affected by the quality of input data. The real-time data collected by the equipment through sensors must undergo strict hidden danger detection and processing, especially during the long-term operation of the equipment, where slight changes in the equipment's operating status or external interference signals may affect the prediction of equipment performance. If these hidden fluctuations and sudden interference signals are not identified and removed in time, they may lead to inaccurate subsequent analysis and prediction results, thereby affecting the stability and prediction accuracy of the digital twin model. Therefore, it is necessary to conduct comprehensive hidden danger detection on the data transmitted from the physical workshop to the digital workshop, accurately identify and process potential influencing factors, and ensure that the data quality is suitable for subsequent analysis and modeling.

[0038] like Figure 2 As shown, step S2 includes the following contents: S2.1, before hidden danger detection, the original data transmitted from the physical workshop to the digital workshop needs to be initially cleaned and preprocessed to ensure the validity and availability of the data.

[0039] S2.2, hidden fluctuations usually manifest as small, periodic changes in equipment under normal operating conditions. These fluctuations are difficult to detect directly through the equipment monitoring system, but may affect the long-term equipment performance prediction. Therefore, the detection of hidden fluctuations is an important part of hidden danger detection. The processing process is as follows: Use fast Fourier transform to convert time domain data to frequency domain and analyze the frequency components. Frequency domain analysis can reveal small fluctuations in equipment operation, such as the frequency of periodic changes in vibration or temperature. By performing frequency domain decomposition on the multi-dimensional data of the equipment, the frequency components related to equipment performance are identified, and the amplitude, period and other characteristics of the fluctuations are analyzed. The specific processing steps are as follows: Compute the spectral density of a signal: ;in, is the original device data, is the Fourier transform operation, is the spectral density.

[0040] Identify the main frequency component in the signal and extract its amplitude and frequency : ; The amplitude and frequency reflect the amplitude and frequency of periodic fluctuations during the operation of the equipment.

[0041] Perform periodic analysis on the time series data of the equipment, use the autoregressive model to predict the future equipment operation status, and compare it with the actual data to detect whether there is a deviation in periodic fluctuations. The specific processing steps are as follows: Fitting device data based on an autoregressive model: ;in, For the moment Device data, are the parameters of the autoregressive model, is the model order, is a white noise term. Calculate the residual , and determine whether there is a significant fluctuation deviation: ; If the standard deviation of the residual exceeds the set threshold, it means that there is a large fluctuation in the device data, and an implicit fluctuation prompt is issued, which may be caused by implicit fluctuations.

[0042] S2.3, sudden interference signals usually appear as sudden abnormal fluctuations in equipment operation, which may be caused by changes in the external environment, equipment operation errors or sudden failures. The detection of sudden interference signals includes the following steps: Use statistical methods (such as IQR test based on box plots) to identify sudden fluctuations in the data. These fluctuations often appear as transient outliers, so they can be detected by the following formula: ;in, and are the first and third quartiles of the data, respectively. If a data point exceeds the upper or lower bound, it is marked as a sudden disturbance.

[0043] Using autoregressive integrated moving average The model models the time series data and calculates the prediction error. If the prediction error exceeds the set threshold, it is considered that a sudden interference has occurred. The specific calculation process is as follows: ;if (Set threshold), it is judged as a sudden interference signal.

[0044] S2.4, classify the detected hidden fluctuations and sudden interference signals to determine their impact on the equipment performance prediction. Cluster the fluctuation characteristics through clustering algorithms (such as K-means) to evaluate the impact of different categories. Prioritize the signals with greater impact. The specific steps are as follows: Cluster analysis of fluctuation characteristics (such as amplitude, frequency, residual, etc.) is performed using the Euclidean distance metric: ;in, and are two sample points of the fluctuation characteristics, is the feature dimension. By clustering features according to distance, data with similar features can be clustered together, making it easier to use the same model for batch processing.

[0045] Based on step S2, the cluster analysis of the data is completed to obtain multiple clusters, and the data in each cluster shows certain similarities. Each cluster represents a specific device operation mode, which may contain periodic fluctuation signals or sudden interference signals. The task of this stage is to further analyze each cluster and extract the periodic fluctuation characteristics and sudden change characteristics within the cluster in order to prepare for the subsequent mode switching process.

[0046] Step S3 includes the following contents: After the equipment data is transferred to the digital workshop, the data has been divided into multiple clusters through cluster analysis, and the signals in each cluster show similar fluctuation characteristics. In order to further process these clusters, the data characteristics of each cluster must be analyzed in detail to determine whether there are hidden fluctuations or sudden interference signals. The traditional processing method determines the abnormality of the data through simple standardization and weighted summation, but this method often ignores the inherent complex changes in the data and the dynamic characteristics of sudden fluctuations. Therefore, this step calculates the innovative fluctuation trajectory stability index and mutation-derived acceleration index, and combines the results of these two complex indexes to determine whether it is necessary to switch to the micro-fluctuation stripping mode or the sudden interference stripping mode. This process can more accurately distinguish between periodic fluctuations and sudden interferences, and provide an accurate basis for mode switching for subsequent processing.

[0047] S3.1, through the cluster analysis in step S2, the data in the cluster has been divided into multiple groups, each cluster represents a specific signal pattern. The data in each cluster shows a certain degree of similarity, which may contain periodic fluctuations, sudden changes or other potential interference signals. Next, the signal of each cluster is feature extracted, and the fluctuation trajectory stability index and mutation-derived acceleration index are used to measure the fluctuation stability and sudden change characteristics of the cluster.

[0048] S3.2, Fluctuation trajectory stability index The stability of the cluster is evaluated by calculating the curve fitting error of the data points within the cluster, aiming to reveal the stability or irregularity of the periodic fluctuations. The regularity of the fluctuations is detected by fitting the trajectory of the curve to reflect whether the periodic changes are disturbed by external factors.

[0049] For each data point in the cluster Perform polynomial regression fitting to obtain the fitting curve .

[0050] Using the curve regression error to measure the difference between each data point and the fitted curve.

[0051] Calculate the volatility trajectory stability index value within the cluster: ;in, is the normalized amplitude of the data within the cluster, is the amplitude feature within the cluster, is the period of periodic fluctuation, The first The regression coefficient for each data point is is the number of data points in the cluster, Represents the absolute value of the ratio of amplitude to period. This calculation quantifies the periodic stability of the data within the cluster. A smaller value of the fluctuation trajectory stability index indicates that the periodicity of the data is relatively stable, while a larger value implies that the data fluctuation is irregular, which may be caused by sudden interference or periodic instability.

[0052] S3.3, Mutation-derived acceleration index It uses the second-order difference to capture sudden changes in acceleration and detect the dramatic fluctuations of the device at a certain moment. Sudden interference is usually accompanied by a dramatic change in acceleration, and the sudden acceleration index can be used to identify the instantaneous impact of the data.

[0053] Use second-order differences of the data points within the cluster to calculate the change in acceleration: ;in, is the time interval between data points. The second-order difference reflects the acceleration of the data, and sudden fluctuations usually lead to rapid changes in acceleration.

[0054] The sudden fluctuations are weighted to obtain the mutation-derived acceleration index value: ;in, is the mean amplitude of the data within the cluster, The function is used to smooth the impact of sudden fluctuations so that larger changes in acceleration can more significantly affect the value of the mutation-derived acceleration index. A larger mutation-derived acceleration index value indicates that the sudden fluctuation is more severe and may have a significant impact on the performance of the device.

[0055] S3.4, after obtaining the fluctuation trajectory stability index and the mutation-derived acceleration index, the gap between the two indicators is calculated to determine whether there is hidden fluctuation or sudden interference. Gap Assessment The calculation formula is: ;in, and are the normalized parameters of the fluctuation trajectory stability index and the mutation-derived acceleration index, respectively, and are used to balance the impact of these two indices at different scales. The value indicates that the difference between periodic fluctuations and sudden changes is significant, which may indicate the presence of sudden interference signals.

[0056] Obtained by calculation value, set the threshold , determine whether to switch to peeling mode: if , indicating that there is a large fluctuation difference within the cluster, and it is necessary to switch to the sudden interference stripping mode. , indicating that the intra-cluster fluctuations are stable and continue to maintain the micro-fluctuation peeling mode without special treatment.

[0057] The sudden interference stripping mode is used to process sudden, instantaneous and severe interference signals that occur during equipment operation. This type of interference is usually caused by changes in the external environment, operating errors, short-term equipment overload or other sudden events, and manifests as obvious short-term abnormal fluctuations in equipment data. The characteristics of sudden interference are large amplitude, short duration, and may have a great impact on equipment status prediction. Therefore, in this mode, the processing logic first identifies and strips these sudden interference signals. The specific processing process is to detect the instantaneous change of fluctuations by calculating the acceleration change in the data (such as the second-order difference or CUSUM algorithm). When the speed of change of the data exceeds the set threshold, it is considered that there is a sudden interference. Subsequently, these sudden fluctuations are removed from the data or smoothed by filtering methods to remove the impact of interference signals on subsequent prediction models, thereby ensuring that the model can still maintain accuracy and stability when facing long-term operation data.

[0058] The micro-fluctuation stripping mode is mainly used to process subtle fluctuations in equipment operation, which are usually caused by normal changes in the equipment's internal operation or slight changes in the external environment. Such fluctuations usually do not affect the long-term performance of the equipment, but may be misjudged as abnormal signals in data analysis, affecting the prediction results of the digital twin model. Micro-fluctuations are characterized by small fluctuation amplitudes, long periods, and slow changes. The processing logic of this mode first detects periodic fluctuations in the data through frequency domain analysis and time domain analysis, and calculates the frequency components and fluctuation amplitudes of the data. If the fluctuation amplitude is small and the frequency is within the normal operating range of the equipment, it is considered to be a micro-fluctuation, and then low-pass filters, smoothing filters and other methods are used to remove these small fluctuations. In this way, the stripped data is closer to the long-term stable operation state of the equipment, which helps to improve the accuracy and stability of the prediction model in predicting the health status of the equipment.

[0059] This step effectively evaluates the gap between periodic fluctuations and sudden disturbances in the data by extracting the fluctuation trajectory stability index and mutation-derived acceleration index of each cluster. The combination of these two highly creative complex indexes can not only accurately identify micro-fluctuations and sudden changes in the equipment, but also provide a strong basis for mode switching. This processing logic avoids the limitations of traditional simple weighting methods and ensures that in a dynamic and complex equipment operating environment, it can accurately judge and switch to the most suitable processing mode, thereby improving the accuracy and stability of the prediction.

[0060] After feature evaluation and mode switching, the implicit fluctuations and sudden interference signals in the data have been identified and stripped. The stripped data is more consistent with the actual operating status of the equipment. Therefore, the next step is to use these "clean" data to retrain the digital twin model. The retraining of the digital twin model is not only to improve the prediction accuracy of the model, but also to enhance the stability of the model, ensuring that it can maintain a high prediction accuracy when facing complex factors such as different equipment conditions and environmental changes.

[0061] Step S4 includes the following contents: Retrain the digital twin model using the stripped data to improve the model’s predictive stability and accuracy.

[0062] In this process, relevant features are first selected from the stripped device data, and the data is standardized or normalized to ensure that different features are trained at the same scale to eliminate the impact of dimensional differences on model training.

[0063] Next, select the appropriate training model based on the data characteristics, including regression models, classification models or deep learning models. When selecting a regression model, linear regression or neural network regression is usually used to predict the continuous output of the device; when selecting a classification model, support vector machines, random forests and other algorithms are used to predict the state category of the device; for time series data, deep learning algorithms such as LSTM or GRU are used to capture the long-term dependencies of the device's operating status.

[0064] Next, the cross-validation method is used to divide the data into training sets, validation sets, and test sets, evaluate the generalization ability of the model during the training process, and minimize the loss function through optimization algorithms (such as gradient descent) to improve the accuracy of model prediction. To avoid overfitting, regularization techniques (such as L2 regularization and dropout) are applied to enhance the generalization ability of the model. If multiple models are used for prediction, the results of different models can be fused through ensemble learning methods (such as weighted average and voting method) to further improve the prediction performance.

[0065] After training is completed, the model is evaluated and the prediction effect of the model is quantified using evaluation indicators (such as accuracy, mean square error, etc.). If necessary, the model is optimized by adjusting hyperparameters, adding training data, etc. As the equipment continues to run, new data will continue to be generated. Using online learning or incremental learning methods, the model can be dynamically updated according to real-time data to maintain high accuracy and stability.

[0066] Finally, the optimized digital twin model will be deployed in the actual system to predict and monitor equipment status through real-time data, provide real-time decision support for operation and maintenance personnel, detect potential failures or performance degradation risks in advance, and avoid equipment failures and production stagnation.

[0067] Step S5 includes the following contents: The prediction process of the digital twin model is a dynamic calculation process based on real-time data and historical data, which aims to provide accurate predictions of the health status and future performance of the equipment. The process includes data input, feature extraction, model calculation, prediction output, decision support, and feedback mechanism, and each link is crucial to the prediction accuracy and stability. The following is the detailed logic and steps of the process: S5.1, the prediction process of the digital twin model first collects real-time operation data (such as temperature, pressure, vibration, power consumption, etc.) from the equipment. These data are collected in real time by sensors and transmitted to the digital workshop for processing. For historical data, equipment operation logs and historical status data are usually used.

[0068] The data are preprocessed as necessary, including standardization, normalization, missing value filling, and outlier removal, to ensure that the data is analyzed on a consistent scale. The purpose of data preprocessing is to eliminate dimensional differences and noise interference, so that the data is more in line with the requirements of subsequent model training.

[0069] S5.2, extract key features that have a significant impact on the prediction results from the preprocessed data. Usually, these features include equipment operating status indicators (such as temperature, pressure, speed, etc.), and equipment environmental factors (such as external temperature, humidity, etc.). In addition, trend features and volatility features in historical data can also be used as input features.

[0070] The sliding window method is used to divide the data into multiple time periods (for example, every hour or every minute), and the data in each time period is extracted as a set of features for input into the model for analysis.

[0071] S5.3, the digital twin model (usually a deep learning model, such as LSTM or GRU, or a traditional regression / classification model) receives these feature data, performs calculations, and obtains the prediction results of the future state of the equipment. For regression models, the predicted value is usually a continuous variable, such as equipment temperature, pressure, remaining life, etc.; for classification models, it is a classification prediction of the equipment health status (such as normal, warning, fault).

[0072] For example, if a LSTM-based deep learning model is used to predict the future health status of a device, the model processes the input time series data through a long short-term memory unit (LSTM), learns the temporal dependencies, and thus predicts the device status at future moments. The calculation process of the LSTM model is based on the following formula: ;in, Indicates at time The hidden state of It's time The input features of and They are the hidden state and memory unit of the previous moment respectively. LSTM captures the long-term dependencies in the time series by updating these states.

[0073] S5.4, the predicted values ​​output by the model are used for equipment health assessment and status judgment. For example, the predicted temperature and pressure values ​​may exceed the set threshold range, or the remaining life of the equipment may be lower than the preset standard. At this time, the system will give a classification of the current status of the equipment (such as normal, warning, fault) according to the set health status judgment rules.

[0074] Assuming the predicted temperature Exceeds the predetermined safety temperature threshold , or the predicted pressure Exceeding safety pressure , the system determines that the equipment is in a warning or fault state and triggers an early warning to remind the operation and maintenance personnel to take necessary measures.

[0075] S5.5, based on real-time input data and prediction results, the system will regularly or in real time update the health assessment of the equipment to provide decision support. If the health status of the equipment is abnormal, the system will issue a warning to the operation and maintenance personnel through the dashboard, alarm notification or operation and maintenance management system, reminding the potential failure risk of the equipment and giving corresponding maintenance suggestions.

[0076] During the long-term operation of the equipment, the model is not only used to monitor the health of the equipment in real time, but also to predict the status of the equipment in the next few days or weeks based on the equipment's operating trends, providing early warnings for operation and maintenance personnel to avoid equipment failures at critical moments.

[0077] The prediction results of the digital twin model are usually presented in a visual form. Operation and maintenance personnel can view the current status of the equipment, prediction results, and potential risks through dashboards or reports. The report will list the prediction results, equipment operation trends, and health assessments in detail to help decision makers predict faults, plan maintenance, and schedule resources.

[0078] The equipment performance prediction of the digital twin model can accurately combine the real-time data and historical data of the equipment to predict the future health status, potential failures or performance degradation of the equipment. The entire process achieves comprehensive monitoring and prediction of the equipment health status through data input, feature extraction, model calculation, output prediction, real-time feedback and online learning. The model continuously learns and adjusts to ensure that it can adapt to the dynamic changes of the equipment, provide timely and reliable decision-making support for operation and maintenance personnel, reduce equipment failures and downtime, and improve production efficiency and the accuracy of equipment management.

[0079] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0080] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

[0081] It should be noted that, in this article, if there are relational terms such as first and second, etc., they are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "including a..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0082] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A device performance prediction method based on digital twins, characterized in that: Includes steps: S1, collects equipment operation data in real time through sensors in the physical workshop and transmits the collected equipment operation data to the digital workshop; S2, detects hidden dangers in the data transmitted to the digital workshop, identifies potential hidden fluctuations and sudden interference signals, and divides the data into different clusters through cluster analysis; S3, for each cluster, by evaluating the stability of fluctuations in the data and the deviation from periodic changes, as well as the differences in acceleration and amplitude of sudden changes, calculate the characteristic gap and determine whether to switch to the micro-fluctuation stripping mode or the sudden interference stripping mode; S4, retraining the digital twin model using the stripped data; S5, equipment performance prediction based on the optimized digital twin model.

2. The device performance prediction method based on digital twin according to claim 1 is characterized in that: Step S2 includes the following contents: S2.1, preliminary cleaning and preprocessing of raw data transferred from the physical workshop to the digital workshop; S2.2, use fast Fourier transform to convert time domain data to frequency domain and analyze the frequency components; perform periodic analysis on the time series data of the equipment, use the autoregressive model to predict the future equipment operation status, and compare it with the actual data to detect whether there is a deviation in periodic fluctuations; if the standard deviation of the residual exceeds the set threshold, it means that there is a large fluctuation in the equipment data, and an implicit fluctuation prompt is issued; S2.3, use the autoregressive integrated moving average model to model the time series data and calculate the prediction error; if the prediction error exceeds the set threshold, it is considered that a sudden interference has occurred; S2.4, classify the detected implicit fluctuations and sudden interference signals to determine their impact on the equipment performance prediction; cluster the fluctuation characteristics through clustering algorithms to evaluate the impact of different categories.

3. The device performance prediction method based on digital twin according to claim 2 is characterized in that: Step S3 also Includes the following: S3.1, through cluster analysis, the data within the cluster has been divided into multiple groups, each cluster represents a specific signal pattern; the data within each cluster shows a certain degree of similarity; next, the signal of each cluster is feature extracted, and the fluctuation trajectory stability index and mutation-derived acceleration index are used to measure the fluctuation stability and sudden change characteristics of the cluster.

4. The device performance prediction method based on digital twin according to claim 3 is characterized in that: Step S3 also Includes the following: S3.2, Fluctuation trajectory stability index The stability of the cluster data points is evaluated by calculating the curve fitting error, aiming to reveal the stability or irregularity of periodic fluctuations; the regularity of fluctuations is detected by fitting the trajectory of the curve, reflecting whether the periodic changes are disturbed by external factors; For each data point in the cluster Perform polynomial regression fitting to obtain the fitting curve ; Use curve regression error to measure the difference between each data point and the fitted curve; Calculate the volatility trajectory stability index value within the cluster: ;in, is the normalized amplitude of the data within the cluster, is the amplitude feature within the cluster, is the period of periodic fluctuation, The first The regression coefficient for each data point is is the number of data points in the cluster, It expresses the absolute value of the ratio of amplitude to period.

5. The device performance prediction method based on digital twin according to claim 4 is characterized in that: Step S3 also Includes the following: S3.3, Mutation-derived acceleration index The second-order difference is used to capture sudden changes in acceleration and detect the violent fluctuations of the equipment at a certain moment. Sudden interference is accompanied by a violent change in acceleration, and the instantaneous impact of the data is identified by the mutation-derived acceleration index. The second-order difference of the data points within the cluster is used to calculate the acceleration change: ;in, is the time interval of the data points; the sudden fluctuations are weighted to obtain the mutation-derived acceleration index value: ;in, is the mean amplitude of the data within the cluster, The function is used to smooth the effects of sudden fluctuations.

6. The device performance prediction method based on digital twin according to claim 5 is characterized in that: Step S3 also Includes the following: S3.4, after obtaining the fluctuation trajectory stability index and the mutation-derived acceleration index, the gap between the two indicators is calculated to determine whether there is hidden fluctuation or sudden interference; gap assessment The calculation formula is: ;in, and are the standardized parameters of the fluctuation trajectory stability index and the mutation-derived acceleration index respectively; value, set the threshold , determine whether to switch to peeling mode: if , indicating that there is a large fluctuation difference within the cluster, and it is necessary to switch to the sudden interference stripping mode; if , indicating that the intra-cluster fluctuations are stable and continue to maintain the micro-fluctuation peeling mode without special treatment.

7. The device performance prediction method based on digital twin according to claim 6 is characterized in that: Step S4 includes the following contents: The digital twin model is retrained using the stripped data: first, relevant features are selected from the stripped device data and the data is standardized; then, the corresponding training model is selected for training based on the data characteristics; next, the data is divided into training set, validation set and test set using the cross-validation method, the generalization ability of the model during the training process is evaluated, and the loss function is minimized through the optimization algorithm to improve the accuracy of the model prediction; after the training is completed, the model is evaluated and the prediction effect of the model is quantified using evaluation indicators; finally, the optimized digital twin model will be deployed in the actual system to predict and monitor the equipment status through real-time data.

8. The device performance prediction method based on digital twin according to claim 7 is characterized in that: Step S5 includes the following contents: The prediction process of the digital twin model first collects real-time operating data from the equipment; from the preprocessed data, key features that have a sufficient impact on the prediction results are extracted; then the digital twin model receives these feature data, performs calculations, and obtains the prediction results of the future state of the equipment.

Images