Automatic equipment predictive maintenance system based on digital twinning

Abstract

The invention is suitable for the technical field of equipment maintenance, and provides an automatic equipment predictive maintenance system based on digital twinning, which comprises a reference curve library module used for constructing a reference curve library; the reference curve determination module is used for determining a reference curve of each parameter item in the equipment twinborn model based on the scheduling information; the reference curve correction module is used for collecting environment data and correcting the reference curve of each parameter item based on the environment data; the actual curve determination module is used for obtaining an actual curve of each parameter item; the deviation information determination module is used for determining deviation information of each parameter item according to the reference curve and the actual curve; the prediction curve combination module is used for predicting three deviation curves in the deviation information to obtain a prediction curve combination; and the risk prediction maintenance module is used for analyzing the prediction curve combination and carrying out risk prediction. The prediction result can more comprehensively capture the change characteristics of the equipment state, and the effect is better.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of equipment maintenance, in particular to an automatic equipment predictive maintenance system based on digital twinning. BACKGROUND

[0002] With the continuous improvement of industrial automation, automation equipment plays a more and more key role in the production process. The traditional equipment maintenance method is mainly divided into post-maintenance and regular maintenance. Post-maintenance is to repair the equipment after it fails, which will cause production interruption and cause great economic loss; regular maintenance is to overhaul the equipment according to the predetermined time interval, which cannot timely find potential faults and may still cause equipment sudden failure, and cannot accurately maintain each equipment according to its specific situation, which may cause resource waste or equipment failure hidden danger. Therefore, an automatic equipment predictive maintenance system based on digital twinning is needed to solve the above problems. SUMMARY

[0003] In view of the defects in the prior art, the purpose of the present application is to provide an automatic equipment predictive maintenance system based on digital twinning to solve the problems in the background art.

[0004] The present application is implemented as follows: an automatic equipment predictive maintenance system based on digital twinning, the system comprises: a reference curve library module for constructing a reference curve library based on historical operation data, the reference curve library comprising a plurality of parameter items, each parameter item corresponding to a product model and a reference curve; a twin model calling module for calling a device twin model, the device twin model displaying a plurality of parameter items; a reference curve determination module for determining the reference curve of each parameter item in the device twin model based on production scheduling information, and displaying the reference curve on the device twin model; a reference curve correction module for collecting environmental data and correcting the reference curve of each parameter item based on the environmental data; an actual curve determination module for collecting parameter time sequence information of the device body in real time to obtain the actual curve of each parameter item; a deviation information determination module for determining the deviation information of each parameter item according to the reference curve and the actual curve, the deviation information comprising an absolute deviation curve, a relative deviation curve and a trend deviation curve; a prediction curve combination module for predicting the three deviation curves in the deviation information to obtain a prediction curve combination; A risk prediction maintenance module is configured to analyze the prediction curve combination and determine whether there is a risk anomaly, and generate maintenance information when there is a risk anomaly.

[0005] As a further scheme of the present application, the reference curve determination module comprises: A production scheduling information extraction unit is configured to determine a production scheduling model and a production scheduling time period according to production scheduling information; A reference curve output unit is configured to input the production scheduling model into a reference curve library for matching, and output a corresponding reference curve, each of which is marked with a parameter item; A curve length cutting unit is configured to cut the length of the reference curve according to the production scheduling time period, and modify the abscissa of the reference curve.

[0006] As a further scheme of the present application, the reference curve correction module comprises: An environmental correction formula unit is configured to input the parameter item into an environmental correction model to determine an environmental correction formula corresponding to the parameter item; An influence factor value unit is configured to determine an environmental influence factor based on the environmental correction formula, and obtain a specific value of each environmental influence factor according to environmental data; A reference curve correction unit is configured to correct the reference curve of the parameter item according to the environmental correction formula with the specific value.

[0007] As a further scheme of the present application, the deviation information determination module comprises: An absolute deviation curve unit is configured to determine a real-time value in the actual curve and a reference value in the reference curve, and obtain an absolute deviation curve according to the difference between the real-time value and the reference value; A relative deviation curve unit is configured to obtain a relative deviation curve according to the difference ratio between the real-time value and the reference value; A trend deviation curve unit is configured to determine an actual change rate of the real-time value and a reference change rate of the reference value, and obtain a trend deviation curve according to the actual change rate and the reference change rate.

[0008] As a further scheme of the present application, the prediction curve combination module comprises: A model architecture determination unit is configured to determine that the prediction model architecture is a hybrid architecture of LSTM and Transformer, wherein the LSTM layer is used to capture the time dynamic relationship of the three deviation curves, and the Transformer encoder can capture the correlation across curves through the self-attention mechanism; A prediction model design unit is configured to design three parallel fully connected layers, which respectively output three prediction curves; and a weighted multi-task loss function is used; The prediction model training unit is configured to divide the historical deviation information into a training set, a verification set and a test set, and determine a prediction model; The prediction curve combination unit is configured to intercept three deviation curves at the latest N time points, construct an input sequence, input the input sequence into the prediction model, and output a prediction curve combination.

[0009] As a further scheme of the present application, the risk prediction maintenance module comprises: The abnormal curve calling unit is configured to call all abnormal curve images in the abnormal curve library, wherein the abnormal curve images comprise absolute deviation curves, relative deviation curves and trend deviation curves, and each abnormal curve image is labeled with an abnormal type; The abnormal curve slicing unit is configured to slice each abnormal curve image to obtain a plurality of curve sub-images according to the time length of the prediction curve combination; The risk abnormality determination unit is configured to perform graphic similarity calculation on the prediction curve combination and all the curve sub-images, determine a highest similarity value, and when the highest similarity value is greater than a similarity threshold value, determine that there is a risk abnormality, and output the abnormal type corresponding to the highest similarity value.

[0010] As a further scheme of the present application, the risk prediction maintenance module further comprises: The abnormal type input unit is configured to input the output abnormal type into the equipment maintenance library; The maintenance information output unit is configured to output maintenance information.

[0011] As a further scheme of the present application, the prediction curve combination comprises an absolute deviation prediction curve, a relative deviation prediction curve and a trend deviation prediction curve, and the three prediction curves are expressed in the same curve image, wherein the horizontal axis of the curve image is time and the vertical axis is a deviation value.

[0012] Compared with the prior art, the present application has the following advantages: The present application constructs an equipment twin model based on digital twin technology, and constructs a benchmark curve library combined with historical operation data, can accurately determine the benchmark curve of each parameter item according to the actual situation and production scheduling information of each equipment. At the same time, the benchmark curve is corrected by collecting environmental data, so that the benchmark curve is more in line with the current actual operation environment of the equipment. In addition, the deviation information comprises three kinds of deviation curves, which reflect the deviation of the equipment state from different angles. Predicting them is equivalent to comprehensively considering the multiple aspects of the equipment state information. The prediction result can more comprehensively capture the change characteristics of the equipment state, reduce the prediction error and improve the prediction accuracy. When the risk is predicted, maintenance information can be sent before the fault occurs or has an impact on production, and the effect is better. BRIEF DESCRIPTION OF DRAWINGS

[0013] Figure 1 Structure diagram of a benchmark curve determination module in an automated equipment predictive maintenance system based on digital twinning.

[0014] Figure 2 Structure diagram of a benchmark curve determination module in an automated equipment predictive maintenance system based on digital twinning.

[0015] Figure 3 Structure diagram of a benchmark curve correction module in an automated equipment predictive maintenance system based on digital twinning.

[0016] Figure 4 Structure diagram of a deviation information determination module in an automated equipment predictive maintenance system based on digital twinning.

[0017] Figure 5 Structure diagram of a prediction curve combination module in an automated equipment predictive maintenance system based on digital twinning.

[0018] Figure 6 Structure diagram of a risk prediction maintenance module in an automated equipment predictive maintenance system based on digital twinning.

[0019] Figure 7 Flowchart of an automated equipment predictive maintenance method based on digital twinning. DETAILED DESCRIPTION

[0020] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings and specific 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.

[0021] The specific implementation of the present application will be described in detail below with reference to specific embodiments.

[0022] As shown in Figure 1 The present application provides an automated equipment predictive maintenance system based on digital twinning, which comprises: A benchmark curve library module 100 is configured to construct a benchmark curve library based on historical operation data, wherein the benchmark curve library comprises a plurality of parameter items, and each parameter item corresponds to a product model and a benchmark curve. A twin model calling module 200 is configured to call a device twin model, wherein the device twin model displays a plurality of parameter items. A benchmark curve determination module 300 is configured to determine the benchmark curve of each parameter item in the device twin model based on production scheduling information, and display the benchmark curve on the device twin model. A benchmark curve correction module 400 is configured to collect environmental data and correct the benchmark curve of each parameter item based on the environmental data. An actual curve determination module 500 is configured to acquire parameter time sequence information of the equipment body in real time, and obtain an actual curve of each parameter item; A deviation information determination module 600 is configured to determine deviation information of each parameter item according to the reference curve and the actual curve, wherein the deviation information is composed of an absolute deviation curve, a relative deviation curve and a trend deviation curve; A prediction curve combination module 700 is configured to predict the three deviation curves in the deviation information to obtain a prediction curve combination; A risk prediction maintenance module 800 is configured to analyze the prediction curve combination, and determine whether there is a risk exception. When there is a risk exception, maintenance information is generated.

[0023] It should be noted that the traditional maintenance method mainly relies on manual experience and periodic inspection to find equipment failure, and it is difficult to detect potential and gradually developing failure in advance. Often, the failure is not discovered until it has a significant impact on production, at which point maintenance will not only increase the difficulty and cost of maintenance, but also cause long-term interruption of production. Digital twin technology is gradually emerging and being applied in the industrial field. By constructing a digital twin model of the equipment, the running state of the equipment can be monitored in real time, providing a more accurate basis for decision-making for equipment maintenance. At the same time, with the development of Internet of Things technology, various parameter data and environmental data during the operation of the equipment can be collected in real time, which provides a rich data basis for predictive maintenance of equipment based on data analysis. In this context, an automated equipment predictive maintenance system based on digital twin emerges as the times require, aiming to solve many problems existing in the traditional equipment maintenance method and improve the reliability and maintenance efficiency of equipment operation.

[0024] In the embodiment of the present application, first, a benchmark curve library is constructed according to historical operation data of the equipment. The historical operation data is normal operation data, and the benchmark curve library contains a plurality of parameter items, each of which corresponds to a product model and a benchmark curve. It is easy to understand that for the same parameter item, the corresponding benchmark curve is different when producing different models of products. The horizontal axis of the benchmark curve represents the time when production starts, and the vertical axis represents the numerical value of the parameter item. When performing predictive maintenance of a certain automated equipment, the twin model of the equipment needs to be called. Then the benchmark curve of each parameter item in the twin model of the equipment is automatically determined according to the production scheduling information, and the benchmark curve is displayed on the twin model of the equipment to realize visual expression. At the same time, environmental data is collected through various sensors, and the benchmark curve of each parameter item is corrected according to the environmental data, so that the benchmark curve is more in line with the current actual operating environment of the equipment, thereby providing more accurate reference for equipment maintenance. The embodiment of the present application fully considers the influence of environmental factors on equipment operation, so that the evaluation of the equipment operating state is more scientific and reasonable, and the real operating condition of the equipment in the actual environment can be more accurately reflected, thereby improving the accuracy and reliability of predictive maintenance. In addition, the time sequence information of each parameter of the equipment body is also collected in real time to obtain the actual curve of each parameter item. Then the deviation information of each parameter item is determined according to the benchmark curve and the actual curve. The deviation information includes an absolute deviation curve, a relative deviation curve and a trend deviation curve. The three types of deviation curves reflect the deviation of the equipment state from different angles, and the prediction of them is equivalent to considering the information of the equipment state from multiple aspects. Compared with relying only on a single type of deviation curve for prediction, this comprehensive prediction method can more comprehensively capture the characteristics of the change of the equipment state, reduce the prediction error and improve the accuracy of the prediction. Moreover, there is a certain correlation and complementarity between different types of deviation curves. For example, the absolute deviation curve may show that the equipment parameter deviates greatly at a certain time, but the relative deviation curve and the trend deviation curve may indicate that this deviation is temporary or within the normal fluctuation range. Through comprehensive analysis of the three types of curves, they can be verified and supplemented with each other to avoid prediction errors due to the limitations of a single curve. Then the prediction curve combination is obtained by predicting the three types of deviation curves in the deviation information, and the prediction curve combination is analyzed to determine whether there is a risk anomaly. When there is a risk anomaly, corresponding maintenance information is generated. In this way, maintenance information can be issued before a fault occurs or has an impact on production, so that maintenance personnel can take timely measures to avoid the expansion of the fault and reduce production losses.

[0025] As shown in Figure 2 As a preferred embodiment of the present application, the benchmark curve determination module 300 includes: A production scheduling information extraction unit 301 is configured to determine a production scheduling model and a production scheduling time period according to production scheduling information. The reference curve output unit 302 is configured to input the production model into the reference curve library for matching, and output a corresponding reference curve, wherein each reference curve is marked with a parameter item; The curve length cutting unit 303 is configured to cut the length of the reference curve according to the production time period, and modify the abscissa of the reference curve.

[0026] In the embodiment of the present application, the production information contains the production time period corresponding to the production model, and the production operation needs to be strictly performed according to the production information. The production model is input into the reference curve library for matching, and a reference curve matched with the production model is obtained, wherein each reference curve is marked with a corresponding parameter item. In addition, the length of the reference curve is cut according to the production time period, so that the length value of the reference curve is consistent with the production time period, and the abscissa of the reference curve is adaptively modified. For example, the abscissa of the original reference curve is from 0 to 4 hours, and the production time period is 8:00-12:00, so the abscissa of the reference curve is modified to be from 8:00 to 12:00.

[0027] As shown in Figure 3 As a preferred embodiment of the present application, the reference curve correction module 400 includes: The environment correction formula unit 401 is configured to input the parameter item into the environment correction model to determine the environment correction formula corresponding to the parameter item; The influence factor value unit 402 is configured to determine the environmental influence factor based on the environment correction formula, and obtain the specific value of each environmental influence factor according to the environmental data; The reference curve correction unit 403 is configured to correct the reference curve of the parameter item according to the environment correction formula with the specific value.

[0028] In the embodiment of the present application, the corresponding relationship between the historical environmental data and the equipment parameter curve needs to be extracted, the correlation analysis (such as Pearson coefficient) or causal inference (such as Granger test) is used to screen the key environmental factors, the regression model (such as multiple linear regression, support vector regression) is selected for training to obtain the environment correction model, so that each parameter item corresponds to a regression correction formula (environment correction formula). Then, the environmental influence factor is determined according to the environment correction formula, and the specific value of each environmental influence factor is obtained combined with the collected environmental data. The environment correction formula is quantitatively expressed, so that the reference curve can be corrected.

[0029] As shown in Figure 4 As a preferred embodiment of the present application, the deviation information determination module 600 includes: The absolute deviation curve unit 601 is used to determine the real-time value in the actual curve and the reference value in the reference curve, and to obtain the absolute deviation curve based on the difference between the real-time value and the reference value. The relative deviation curve unit 602 is used to obtain a relative deviation curve based on the ratio of the difference between the real-time value and the reference value; The trend deviation curve unit 603 is used to determine the actual rate of change of the real-time value and the baseline rate of change of the baseline value, and to obtain the trend deviation curve based on the actual rate of change and the baseline rate of change.

[0030] In this embodiment of the invention, before determining the deviation information, it is necessary to determine the real-time value at each time point in the actual curve and the baseline value at each time point in the reference curve. The absolute deviation curve can be obtained based on the difference between the real-time value and the baseline value. Simultaneously, the relative deviation curve can be obtained based on the ratio of the difference between the real-time value and the baseline value, where the difference ratio = (real-time value - baseline value) / baseline value. Finally, it is also necessary to determine the actual rate of change of the real-time value and the baseline rate of change of the baseline value. The trend deviation curve is obtained based on the difference between the actual rate of change and the baseline rate of change, where the rate of change = (value at this time point - value at the previous time point) / value at the previous time point.

[0031] like Figure 5 As shown, in a preferred embodiment of the present invention, the prediction curve combination module 700 includes: The model architecture determination unit 701 is used to determine that the prediction model architecture is a hybrid LSTM-Transformer architecture. The prediction model design unit 702 is used to design three parallel fully connected layers, which output three prediction curves respectively; a weighted multi-task loss function is adopted. The prediction model training unit 703 is used to divide historical deviation information into training set, validation set and test set to determine the prediction model; The prediction curve combination unit 704 is used to extract three deviation curves from the latest N time points, construct an input sequence, input the input sequence into the prediction model, and output the prediction curve combination.

[0032] In the embodiment of the present application, before constructing the prediction model, it is necessary to first independently perform Z-Score standardization on each deviation curve to eliminate dimensional differences. Then model selection is needed, and a hybrid architecture of LSTM-Transformer is adopted, wherein the LSTM layer is used to capture the time dynamic relationship of the three deviation curves, and the Transformer encoder can capture the correlation across curves through the self-attention mechanism, for example, the influence of the upward trend deviation on the absolute deviation. In addition, three parallel fully connected layers need to be designed to output three predicted curves respectively, output layer 1: absolute deviation prediction value; output layer 2: relative deviation prediction value; output layer 3: trend deviation prediction value. The embodiment of the present application adopts a weighted multi-task loss function to balance the prediction errors of the three curves. The historical deviation information is divided into a training set, a validation set and a test set, and formal training is started. In the early stage of training, the real historical deviation is used as the input of the LSTM to avoid error accumulation. As the number of training rounds increases, the model's own prediction value is gradually used to replace the real historical data to improve the robustness of the model, and finally the prediction model is obtained. Finally, the three deviation curves of the latest N time points are intercepted, N is a predetermined value, an input sequence is constructed, and the input sequence is input into the prediction model to automatically output the predicted curve combination. The predicted curve combination includes an absolute deviation prediction curve, a relative deviation prediction curve and a trend deviation prediction curve, and the three predicted curves are expressed in the same curve image. The horizontal axis of the curve image is time, and the vertical axis is deviation value. Preferably, the three curves are expressed using different colors.

[0033] As shown in Figure 6 As a preferred embodiment of the present application, the risk prediction maintenance module 800 includes: An abnormal curve calling unit 801 is configured to call all abnormal curve images in an abnormal curve library, wherein the abnormal curve images include absolute deviation curves, relative deviation curves and trend deviation curves, and each abnormal curve image is labeled with an abnormal type; An abnormal curve slicing unit 802 is configured to slice each abnormal curve image to obtain a plurality of curve sub-images according to the time length of the predicted curve combination; A risk abnormality determination unit 803 is configured to perform graphic similarity calculation on the predicted curve combination and all curve sub-images to determine a highest similarity value, and when the highest similarity value is greater than a similarity threshold value, it is determined that there is a risk abnormality, and the abnormal type corresponding to the highest similarity value is output.

[0034] In the embodiment of the present application, an abnormal curve library is established in advance, the abnormal curve library is obtained according to historical data, the abnormal curve library contains abnormal curve images of all cases, and the abnormal curve images contain absolute deviation curves, relative deviation curves and trend deviation curves. When performing risk prediction, a plurality of curve sub-images are obtained by slicing each abnormal curve image according to the time length of the prediction curve combination. When slicing, a slicing interval is set, for example, the slicing interval is 20 seconds, the time length of the prediction curve combination is 20 minutes, and the time length of the abnormal curve image is 30 minutes. The time lengths of the plurality of curve sub-images are 0-20 minutes, 0 minutes 20 seconds-20 minutes 20 seconds, 0 minutes 40 seconds-20 minutes 40 seconds, 1 minute-21 minutes, 1 minute 20 seconds-21 minutes 20 seconds, 1 minute 40 seconds-21 minutes 40 seconds, and the like. Then, the prediction curve combination and all the curve sub-images are sequentially subjected to graphic similarity calculation, the similarity calculation between the curve images can use the Frechet distance method, and then the highest similarity value is determined. When the highest similarity value is greater than a similarity threshold value, it is determined that there is a risk anomaly, and the curve sub-image corresponding to the highest similarity value is output, and the abnormal type of the abnormal curve image in which the curve sub-image is located is determined.

[0035] As shown in Figure 6 , as a preferred embodiment of the present application, the risk prediction maintenance module 800 further comprises: an abnormal type input unit 804, configured to input the output abnormal type into a device maintenance library; a maintenance information output unit 805, configured to output maintenance information.

[0036] In the embodiment of the present application, a device maintenance library is established in advance, and the device maintenance library contains maintenance information corresponding to all abnormal types.

[0037] As shown in Figure 7 , the embodiment of the present application further provides an automatic device predictive maintenance method based on digital twinning, and the method comprises the following steps: S100, constructing a benchmark curve library based on historical operation data, the benchmark curve library containing a plurality of parameter items, each parameter item corresponding to a product model and a benchmark curve; S200, calling a device twin model, the device twin model displaying a plurality of parameter items; S300, determining the benchmark curve of each parameter item in the device twin model based on production scheduling information, and displaying the benchmark curve on the device twin model; S400, collecting environmental data, and correcting the benchmark curve of each parameter item based on the environmental data; S500, collecting parameter time sequence information of a device body in real time to obtain an actual curve of each parameter item; S600, determining deviation information of each parameter item according to the reference curve and the actual curve, the deviation information consisting of an absolute deviation curve, a relative deviation curve and a trend deviation curve; S700, predicting the three kinds of deviation curves in the deviation information to obtain a predicted curve combination; S800, analyzing the predicted curve combination to determine whether there is a risk exception, and generating maintenance information when there is.

[0038] The above only describes the preferred embodiments of the present application in detail, and does not limit the present application. Any modification, equivalent replacement and improvement within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0039] It should be understood that, although each step in the flowchart of each embodiment of the present application is displayed in sequence according to the arrow, these steps are not necessarily executed in sequence according to the arrow. Unless explicitly stated herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other sequences. Moreover, at least part of the steps in each embodiment can include multiple sub-steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution sequence of these sub-steps or stages is not necessarily sequential, but can be executed in rotation or alternation with at least part of other steps or sub-steps or stages of other steps.

[0040] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The program can be stored in a non-volatile computer readable storage medium, and when the program is executed, the processes of the above-mentioned embodiments of the methods can be included. Any reference to memory, storage, databases, or other media in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in many forms such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), Synchlink DRAM (SLDRAM), Rambus DRAM (RDRAM), direct Rambus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.

[0041] Other embodiments of the present disclosure will be apparent to those skilled in the art with the disclosure herein. The present application is intended to cover any variations, uses, or adaptations of the present disclosure, including those that are deemed to be equivalents of the generic principles of the present disclosure and including those that are within the known and customary practice of the art. The specification and examples are to be considered exemplary only, with the true scope and spirit of the present disclosure being indicated by the claims.

Claims

1. A digital-twin-based automated-equipment predictive-maintenance system, characterized in that, The system comprises: A benchmark curve library module for constructing a benchmark curve library based on historical operation data, the benchmark curve library containing a plurality of parameter items, each parameter item corresponding to a product model and a benchmark curve; A twin model calling module for calling a device twin model, the device twin model displaying a plurality of parameter items; A benchmark curve determination module for determining the benchmark curve of each parameter item in the device twin model based on scheduling information, and displaying the benchmark curve on the device twin model; A benchmark curve correction module for collecting environmental data and correcting the benchmark curve of each parameter item based on the environmental data; An actual curve determination module for collecting parameter time sequence information of the device in real time to obtain an actual curve of each parameter item; A deviation information determination module for determining deviation information of each parameter item according to the benchmark curve and the actual curve, the deviation information consisting of an absolute deviation curve, a relative deviation curve and a trend deviation curve; A predicted curve combination module for predicting the three deviation curves in the deviation information to obtain a predicted curve combination; A risk prediction and maintenance module for analyzing the predicted curve combination to determine whether there is a risk anomaly, and generating maintenance information when there is a risk anomaly.

2. The digital-twin-based automated-equipment predictive-maintenance system of claim 1, wherein, The benchmark curve determination module comprises: A scheduling information extraction unit for determining a scheduling model and a scheduling time period according to scheduling information; A benchmark curve output unit for inputting the scheduling model into the benchmark curve library for matching, and outputting the corresponding benchmark curve, each benchmark curve being marked with a parameter item; A curve length cutting unit for cutting the length of the benchmark curve according to the scheduling time period, and modifying the abscissa of the benchmark curve.

3. The digital-twin-based automated-equipment predictive-maintenance system of claim 1, wherein, The benchmark curve correction module comprises: An environmental correction formula unit for inputting the parameter item into an environmental correction model to determine an environmental correction formula corresponding to the parameter item; An influence factor value unit for determining an environmental influence factor based on the environmental correction formula, and obtaining the specific value of each environmental influence factor according to the environmental data; A benchmark curve correction unit for correcting the benchmark curve of the parameter item according to the environmental correction formula with the specific value.

4. The digital-twin-based automated-equipment predictive-maintenance system of claim 1, wherein, The deviation information determination module comprises: An absolute deviation curve unit for determining the real-time value in the actual curve and the benchmark value in the benchmark curve, and obtaining the absolute deviation curve according to the difference between the real-time value and the benchmark value; A relative deviation curve unit for obtaining the relative deviation curve according to the difference ratio between the real-time value and the benchmark value; A trend deviation curve unit for determining the actual change rate of the real-time value and the benchmark change rate of the benchmark value, and obtaining the trend deviation curve according to the actual change rate and the benchmark change rate.

5. The digital-twin-based automated-equipment predictive-maintenance system of claim 1, wherein, The predicted curve combination module comprises: A model architecture determination unit for determining that the prediction model architecture is an LSTM-Transformer hybrid architecture, wherein the LSTM layer is used to capture the time dynamic relationship of the three deviation curves, and the Transformer encoder can capture the correlation across curves through the self-attention mechanism. The prediction model design unit is configured to design three parallel full connection layers, and output three prediction curves respectively; and a weighted multi-task loss function is used; The prediction model training unit is configured to divide the historical deviation information into a training set, a verification set and a test set, and determine the prediction model; The prediction curve combination unit is configured to intercept three deviation curves at the latest N time points, construct an input sequence, input the input sequence into the prediction model, and output a prediction curve combination.

6. The digital-twin-based automated-equipment predictive-maintenance system of claim 1, wherein, The risk prediction maintenance module comprises: The abnormal curve calling unit is configured to call all abnormal curve images in an abnormal curve library, wherein the abnormal curve images comprise absolute deviation curves, relative deviation curves and trend deviation curves, and each abnormal curve image is labeled with an abnormal type; The abnormal curve slicing unit is configured to slice each abnormal curve image to obtain a plurality of curve sub-images according to a time length of the prediction curve combination; The risk abnormality determination unit is configured to perform graphic similarity calculation on the prediction curve combination and all the curve sub-images, determine a highest similarity value, and determine that there is a risk abnormality when the highest similarity value is greater than a similarity threshold value, and output an abnormal type corresponding to the highest similarity value.

7. The digital-twin-based automated-equipment predictive-maintenance system of claim 6, wherein, The risk prediction maintenance module further comprises: The abnormal type input unit is configured to input the output abnormal type into a device maintenance library; The maintenance information output unit is configured to output maintenance information.

8. The digital-twin-based automated-equipment predictive-maintenance system of claim 1, wherein, The prediction curve combination comprises an absolute deviation prediction curve, a relative deviation prediction curve and a trend deviation prediction curve, and the three prediction curves are expressed in the same curve image, wherein the horizontal axis of the curve image is time, and the vertical axis is a deviation value.

Images